vdev.c revision 249195
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) 2013 by Delphix. All rights reserved. 26 * Copyright 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved. 27 */ 28 29#include <sys/zfs_context.h> 30#include <sys/fm/fs/zfs.h> 31#include <sys/spa.h> 32#include <sys/spa_impl.h> 33#include <sys/dmu.h> 34#include <sys/dmu_tx.h> 35#include <sys/vdev_impl.h> 36#include <sys/uberblock_impl.h> 37#include <sys/metaslab.h> 38#include <sys/metaslab_impl.h> 39#include <sys/space_map.h> 40#include <sys/zio.h> 41#include <sys/zap.h> 42#include <sys/fs/zfs.h> 43#include <sys/arc.h> 44#include <sys/zil.h> 45#include <sys/dsl_scan.h> 46#include <sys/trim_map.h> 47 48SYSCTL_DECL(_vfs_zfs); 49SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV"); 50 51/* 52 * Virtual device management. 53 */ 54 55static vdev_ops_t *vdev_ops_table[] = { 56 &vdev_root_ops, 57 &vdev_raidz_ops, 58 &vdev_mirror_ops, 59 &vdev_replacing_ops, 60 &vdev_spare_ops, 61#ifdef _KERNEL 62 &vdev_geom_ops, 63#else 64 &vdev_disk_ops, 65#endif 66 &vdev_file_ops, 67 &vdev_missing_ops, 68 &vdev_hole_ops, 69 NULL 70}; 71 72 73/* 74 * Given a vdev type, return the appropriate ops vector. 75 */ 76static vdev_ops_t * 77vdev_getops(const char *type) 78{ 79 vdev_ops_t *ops, **opspp; 80 81 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 82 if (strcmp(ops->vdev_op_type, type) == 0) 83 break; 84 85 return (ops); 86} 87 88/* 89 * Default asize function: return the MAX of psize with the asize of 90 * all children. This is what's used by anything other than RAID-Z. 91 */ 92uint64_t 93vdev_default_asize(vdev_t *vd, uint64_t psize) 94{ 95 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 96 uint64_t csize; 97 98 for (int c = 0; c < vd->vdev_children; c++) { 99 csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 100 asize = MAX(asize, csize); 101 } 102 103 return (asize); 104} 105 106/* 107 * Get the minimum allocatable size. We define the allocatable size as 108 * the vdev's asize rounded to the nearest metaslab. This allows us to 109 * replace or attach devices which don't have the same physical size but 110 * can still satisfy the same number of allocations. 111 */ 112uint64_t 113vdev_get_min_asize(vdev_t *vd) 114{ 115 vdev_t *pvd = vd->vdev_parent; 116 117 /* 118 * If our parent is NULL (inactive spare or cache) or is the root, 119 * just return our own asize. 120 */ 121 if (pvd == NULL) 122 return (vd->vdev_asize); 123 124 /* 125 * The top-level vdev just returns the allocatable size rounded 126 * to the nearest metaslab. 127 */ 128 if (vd == vd->vdev_top) 129 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 130 131 /* 132 * The allocatable space for a raidz vdev is N * sizeof(smallest child), 133 * so each child must provide at least 1/Nth of its asize. 134 */ 135 if (pvd->vdev_ops == &vdev_raidz_ops) 136 return (pvd->vdev_min_asize / pvd->vdev_children); 137 138 return (pvd->vdev_min_asize); 139} 140 141void 142vdev_set_min_asize(vdev_t *vd) 143{ 144 vd->vdev_min_asize = vdev_get_min_asize(vd); 145 146 for (int c = 0; c < vd->vdev_children; c++) 147 vdev_set_min_asize(vd->vdev_child[c]); 148} 149 150vdev_t * 151vdev_lookup_top(spa_t *spa, uint64_t vdev) 152{ 153 vdev_t *rvd = spa->spa_root_vdev; 154 155 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 156 157 if (vdev < rvd->vdev_children) { 158 ASSERT(rvd->vdev_child[vdev] != NULL); 159 return (rvd->vdev_child[vdev]); 160 } 161 162 return (NULL); 163} 164 165vdev_t * 166vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 167{ 168 vdev_t *mvd; 169 170 if (vd->vdev_guid == guid) 171 return (vd); 172 173 for (int c = 0; c < vd->vdev_children; c++) 174 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 175 NULL) 176 return (mvd); 177 178 return (NULL); 179} 180 181void 182vdev_add_child(vdev_t *pvd, vdev_t *cvd) 183{ 184 size_t oldsize, newsize; 185 uint64_t id = cvd->vdev_id; 186 vdev_t **newchild; 187 188 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 189 ASSERT(cvd->vdev_parent == NULL); 190 191 cvd->vdev_parent = pvd; 192 193 if (pvd == NULL) 194 return; 195 196 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 197 198 oldsize = pvd->vdev_children * sizeof (vdev_t *); 199 pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 200 newsize = pvd->vdev_children * sizeof (vdev_t *); 201 202 newchild = kmem_zalloc(newsize, KM_SLEEP); 203 if (pvd->vdev_child != NULL) { 204 bcopy(pvd->vdev_child, newchild, oldsize); 205 kmem_free(pvd->vdev_child, oldsize); 206 } 207 208 pvd->vdev_child = newchild; 209 pvd->vdev_child[id] = cvd; 210 211 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 212 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 213 214 /* 215 * Walk up all ancestors to update guid sum. 216 */ 217 for (; pvd != NULL; pvd = pvd->vdev_parent) 218 pvd->vdev_guid_sum += cvd->vdev_guid_sum; 219} 220 221void 222vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 223{ 224 int c; 225 uint_t id = cvd->vdev_id; 226 227 ASSERT(cvd->vdev_parent == pvd); 228 229 if (pvd == NULL) 230 return; 231 232 ASSERT(id < pvd->vdev_children); 233 ASSERT(pvd->vdev_child[id] == cvd); 234 235 pvd->vdev_child[id] = NULL; 236 cvd->vdev_parent = NULL; 237 238 for (c = 0; c < pvd->vdev_children; c++) 239 if (pvd->vdev_child[c]) 240 break; 241 242 if (c == pvd->vdev_children) { 243 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 244 pvd->vdev_child = NULL; 245 pvd->vdev_children = 0; 246 } 247 248 /* 249 * Walk up all ancestors to update guid sum. 250 */ 251 for (; pvd != NULL; pvd = pvd->vdev_parent) 252 pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 253} 254 255/* 256 * Remove any holes in the child array. 257 */ 258void 259vdev_compact_children(vdev_t *pvd) 260{ 261 vdev_t **newchild, *cvd; 262 int oldc = pvd->vdev_children; 263 int newc; 264 265 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 266 267 for (int c = newc = 0; c < oldc; c++) 268 if (pvd->vdev_child[c]) 269 newc++; 270 271 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 272 273 for (int c = newc = 0; c < oldc; c++) { 274 if ((cvd = pvd->vdev_child[c]) != NULL) { 275 newchild[newc] = cvd; 276 cvd->vdev_id = newc++; 277 } 278 } 279 280 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 281 pvd->vdev_child = newchild; 282 pvd->vdev_children = newc; 283} 284 285/* 286 * Allocate and minimally initialize a vdev_t. 287 */ 288vdev_t * 289vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 290{ 291 vdev_t *vd; 292 293 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 294 295 if (spa->spa_root_vdev == NULL) { 296 ASSERT(ops == &vdev_root_ops); 297 spa->spa_root_vdev = vd; 298 spa->spa_load_guid = spa_generate_guid(NULL); 299 } 300 301 if (guid == 0 && ops != &vdev_hole_ops) { 302 if (spa->spa_root_vdev == vd) { 303 /* 304 * The root vdev's guid will also be the pool guid, 305 * which must be unique among all pools. 306 */ 307 guid = spa_generate_guid(NULL); 308 } else { 309 /* 310 * Any other vdev's guid must be unique within the pool. 311 */ 312 guid = spa_generate_guid(spa); 313 } 314 ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 315 } 316 317 vd->vdev_spa = spa; 318 vd->vdev_id = id; 319 vd->vdev_guid = guid; 320 vd->vdev_guid_sum = guid; 321 vd->vdev_ops = ops; 322 vd->vdev_state = VDEV_STATE_CLOSED; 323 vd->vdev_ishole = (ops == &vdev_hole_ops); 324 325 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 326 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 327 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 328 for (int t = 0; t < DTL_TYPES; t++) { 329 space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0, 330 &vd->vdev_dtl_lock); 331 } 332 txg_list_create(&vd->vdev_ms_list, 333 offsetof(struct metaslab, ms_txg_node)); 334 txg_list_create(&vd->vdev_dtl_list, 335 offsetof(struct vdev, vdev_dtl_node)); 336 vd->vdev_stat.vs_timestamp = gethrtime(); 337 vdev_queue_init(vd); 338 vdev_cache_init(vd); 339 340 return (vd); 341} 342 343/* 344 * Allocate a new vdev. The 'alloctype' is used to control whether we are 345 * creating a new vdev or loading an existing one - the behavior is slightly 346 * different for each case. 347 */ 348int 349vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 350 int alloctype) 351{ 352 vdev_ops_t *ops; 353 char *type; 354 uint64_t guid = 0, islog, nparity; 355 vdev_t *vd; 356 357 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 358 359 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 360 return (SET_ERROR(EINVAL)); 361 362 if ((ops = vdev_getops(type)) == NULL) 363 return (SET_ERROR(EINVAL)); 364 365 /* 366 * If this is a load, get the vdev guid from the nvlist. 367 * Otherwise, vdev_alloc_common() will generate one for us. 368 */ 369 if (alloctype == VDEV_ALLOC_LOAD) { 370 uint64_t label_id; 371 372 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 373 label_id != id) 374 return (SET_ERROR(EINVAL)); 375 376 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 377 return (SET_ERROR(EINVAL)); 378 } else if (alloctype == VDEV_ALLOC_SPARE) { 379 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 380 return (SET_ERROR(EINVAL)); 381 } else if (alloctype == VDEV_ALLOC_L2CACHE) { 382 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 383 return (SET_ERROR(EINVAL)); 384 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 385 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 386 return (SET_ERROR(EINVAL)); 387 } 388 389 /* 390 * The first allocated vdev must be of type 'root'. 391 */ 392 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 393 return (SET_ERROR(EINVAL)); 394 395 /* 396 * Determine whether we're a log vdev. 397 */ 398 islog = 0; 399 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 400 if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 401 return (SET_ERROR(ENOTSUP)); 402 403 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 404 return (SET_ERROR(ENOTSUP)); 405 406 /* 407 * Set the nparity property for RAID-Z vdevs. 408 */ 409 nparity = -1ULL; 410 if (ops == &vdev_raidz_ops) { 411 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 412 &nparity) == 0) { 413 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 414 return (SET_ERROR(EINVAL)); 415 /* 416 * Previous versions could only support 1 or 2 parity 417 * device. 418 */ 419 if (nparity > 1 && 420 spa_version(spa) < SPA_VERSION_RAIDZ2) 421 return (SET_ERROR(ENOTSUP)); 422 if (nparity > 2 && 423 spa_version(spa) < SPA_VERSION_RAIDZ3) 424 return (SET_ERROR(ENOTSUP)); 425 } else { 426 /* 427 * We require the parity to be specified for SPAs that 428 * support multiple parity levels. 429 */ 430 if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 431 return (SET_ERROR(EINVAL)); 432 /* 433 * Otherwise, we default to 1 parity device for RAID-Z. 434 */ 435 nparity = 1; 436 } 437 } else { 438 nparity = 0; 439 } 440 ASSERT(nparity != -1ULL); 441 442 vd = vdev_alloc_common(spa, id, guid, ops); 443 444 vd->vdev_islog = islog; 445 vd->vdev_nparity = nparity; 446 447 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 448 vd->vdev_path = spa_strdup(vd->vdev_path); 449 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 450 vd->vdev_devid = spa_strdup(vd->vdev_devid); 451 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 452 &vd->vdev_physpath) == 0) 453 vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 454 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 455 vd->vdev_fru = spa_strdup(vd->vdev_fru); 456 457 /* 458 * Set the whole_disk property. If it's not specified, leave the value 459 * as -1. 460 */ 461 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 462 &vd->vdev_wholedisk) != 0) 463 vd->vdev_wholedisk = -1ULL; 464 465 /* 466 * Look for the 'not present' flag. This will only be set if the device 467 * was not present at the time of import. 468 */ 469 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 470 &vd->vdev_not_present); 471 472 /* 473 * Get the alignment requirement. 474 */ 475 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 476 477 /* 478 * Retrieve the vdev creation time. 479 */ 480 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 481 &vd->vdev_crtxg); 482 483 /* 484 * If we're a top-level vdev, try to load the allocation parameters. 485 */ 486 if (parent && !parent->vdev_parent && 487 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 488 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 489 &vd->vdev_ms_array); 490 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 491 &vd->vdev_ms_shift); 492 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 493 &vd->vdev_asize); 494 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 495 &vd->vdev_removing); 496 } 497 498 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { 499 ASSERT(alloctype == VDEV_ALLOC_LOAD || 500 alloctype == VDEV_ALLOC_ADD || 501 alloctype == VDEV_ALLOC_SPLIT || 502 alloctype == VDEV_ALLOC_ROOTPOOL); 503 vd->vdev_mg = metaslab_group_create(islog ? 504 spa_log_class(spa) : spa_normal_class(spa), vd); 505 } 506 507 /* 508 * If we're a leaf vdev, try to load the DTL object and other state. 509 */ 510 if (vd->vdev_ops->vdev_op_leaf && 511 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 512 alloctype == VDEV_ALLOC_ROOTPOOL)) { 513 if (alloctype == VDEV_ALLOC_LOAD) { 514 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 515 &vd->vdev_dtl_smo.smo_object); 516 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 517 &vd->vdev_unspare); 518 } 519 520 if (alloctype == VDEV_ALLOC_ROOTPOOL) { 521 uint64_t spare = 0; 522 523 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 524 &spare) == 0 && spare) 525 spa_spare_add(vd); 526 } 527 528 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 529 &vd->vdev_offline); 530 531 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING, 532 &vd->vdev_resilvering); 533 534 /* 535 * When importing a pool, we want to ignore the persistent fault 536 * state, as the diagnosis made on another system may not be 537 * valid in the current context. Local vdevs will 538 * remain in the faulted state. 539 */ 540 if (spa_load_state(spa) == SPA_LOAD_OPEN) { 541 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 542 &vd->vdev_faulted); 543 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 544 &vd->vdev_degraded); 545 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 546 &vd->vdev_removed); 547 548 if (vd->vdev_faulted || vd->vdev_degraded) { 549 char *aux; 550 551 vd->vdev_label_aux = 552 VDEV_AUX_ERR_EXCEEDED; 553 if (nvlist_lookup_string(nv, 554 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 555 strcmp(aux, "external") == 0) 556 vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 557 } 558 } 559 } 560 561 /* 562 * Add ourselves to the parent's list of children. 563 */ 564 vdev_add_child(parent, vd); 565 566 *vdp = vd; 567 568 return (0); 569} 570 571void 572vdev_free(vdev_t *vd) 573{ 574 spa_t *spa = vd->vdev_spa; 575 576 /* 577 * vdev_free() implies closing the vdev first. This is simpler than 578 * trying to ensure complicated semantics for all callers. 579 */ 580 vdev_close(vd); 581 582 ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 583 ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 584 585 /* 586 * Free all children. 587 */ 588 for (int c = 0; c < vd->vdev_children; c++) 589 vdev_free(vd->vdev_child[c]); 590 591 ASSERT(vd->vdev_child == NULL); 592 ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 593 594 /* 595 * Discard allocation state. 596 */ 597 if (vd->vdev_mg != NULL) { 598 vdev_metaslab_fini(vd); 599 metaslab_group_destroy(vd->vdev_mg); 600 } 601 602 ASSERT0(vd->vdev_stat.vs_space); 603 ASSERT0(vd->vdev_stat.vs_dspace); 604 ASSERT0(vd->vdev_stat.vs_alloc); 605 606 /* 607 * Remove this vdev from its parent's child list. 608 */ 609 vdev_remove_child(vd->vdev_parent, vd); 610 611 ASSERT(vd->vdev_parent == NULL); 612 613 /* 614 * Clean up vdev structure. 615 */ 616 vdev_queue_fini(vd); 617 vdev_cache_fini(vd); 618 619 if (vd->vdev_path) 620 spa_strfree(vd->vdev_path); 621 if (vd->vdev_devid) 622 spa_strfree(vd->vdev_devid); 623 if (vd->vdev_physpath) 624 spa_strfree(vd->vdev_physpath); 625 if (vd->vdev_fru) 626 spa_strfree(vd->vdev_fru); 627 628 if (vd->vdev_isspare) 629 spa_spare_remove(vd); 630 if (vd->vdev_isl2cache) 631 spa_l2cache_remove(vd); 632 633 txg_list_destroy(&vd->vdev_ms_list); 634 txg_list_destroy(&vd->vdev_dtl_list); 635 636 mutex_enter(&vd->vdev_dtl_lock); 637 for (int t = 0; t < DTL_TYPES; t++) { 638 space_map_unload(&vd->vdev_dtl[t]); 639 space_map_destroy(&vd->vdev_dtl[t]); 640 } 641 mutex_exit(&vd->vdev_dtl_lock); 642 643 mutex_destroy(&vd->vdev_dtl_lock); 644 mutex_destroy(&vd->vdev_stat_lock); 645 mutex_destroy(&vd->vdev_probe_lock); 646 647 if (vd == spa->spa_root_vdev) 648 spa->spa_root_vdev = NULL; 649 650 kmem_free(vd, sizeof (vdev_t)); 651} 652 653/* 654 * Transfer top-level vdev state from svd to tvd. 655 */ 656static void 657vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 658{ 659 spa_t *spa = svd->vdev_spa; 660 metaslab_t *msp; 661 vdev_t *vd; 662 int t; 663 664 ASSERT(tvd == tvd->vdev_top); 665 666 tvd->vdev_ms_array = svd->vdev_ms_array; 667 tvd->vdev_ms_shift = svd->vdev_ms_shift; 668 tvd->vdev_ms_count = svd->vdev_ms_count; 669 670 svd->vdev_ms_array = 0; 671 svd->vdev_ms_shift = 0; 672 svd->vdev_ms_count = 0; 673 674 if (tvd->vdev_mg) 675 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 676 tvd->vdev_mg = svd->vdev_mg; 677 tvd->vdev_ms = svd->vdev_ms; 678 679 svd->vdev_mg = NULL; 680 svd->vdev_ms = NULL; 681 682 if (tvd->vdev_mg != NULL) 683 tvd->vdev_mg->mg_vd = tvd; 684 685 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 686 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 687 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 688 689 svd->vdev_stat.vs_alloc = 0; 690 svd->vdev_stat.vs_space = 0; 691 svd->vdev_stat.vs_dspace = 0; 692 693 for (t = 0; t < TXG_SIZE; t++) { 694 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 695 (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 696 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 697 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 698 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 699 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 700 } 701 702 if (list_link_active(&svd->vdev_config_dirty_node)) { 703 vdev_config_clean(svd); 704 vdev_config_dirty(tvd); 705 } 706 707 if (list_link_active(&svd->vdev_state_dirty_node)) { 708 vdev_state_clean(svd); 709 vdev_state_dirty(tvd); 710 } 711 712 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 713 svd->vdev_deflate_ratio = 0; 714 715 tvd->vdev_islog = svd->vdev_islog; 716 svd->vdev_islog = 0; 717} 718 719static void 720vdev_top_update(vdev_t *tvd, vdev_t *vd) 721{ 722 if (vd == NULL) 723 return; 724 725 vd->vdev_top = tvd; 726 727 for (int c = 0; c < vd->vdev_children; c++) 728 vdev_top_update(tvd, vd->vdev_child[c]); 729} 730 731/* 732 * Add a mirror/replacing vdev above an existing vdev. 733 */ 734vdev_t * 735vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 736{ 737 spa_t *spa = cvd->vdev_spa; 738 vdev_t *pvd = cvd->vdev_parent; 739 vdev_t *mvd; 740 741 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 742 743 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 744 745 mvd->vdev_asize = cvd->vdev_asize; 746 mvd->vdev_min_asize = cvd->vdev_min_asize; 747 mvd->vdev_max_asize = cvd->vdev_max_asize; 748 mvd->vdev_ashift = cvd->vdev_ashift; 749 mvd->vdev_state = cvd->vdev_state; 750 mvd->vdev_crtxg = cvd->vdev_crtxg; 751 752 vdev_remove_child(pvd, cvd); 753 vdev_add_child(pvd, mvd); 754 cvd->vdev_id = mvd->vdev_children; 755 vdev_add_child(mvd, cvd); 756 vdev_top_update(cvd->vdev_top, cvd->vdev_top); 757 758 if (mvd == mvd->vdev_top) 759 vdev_top_transfer(cvd, mvd); 760 761 return (mvd); 762} 763 764/* 765 * Remove a 1-way mirror/replacing vdev from the tree. 766 */ 767void 768vdev_remove_parent(vdev_t *cvd) 769{ 770 vdev_t *mvd = cvd->vdev_parent; 771 vdev_t *pvd = mvd->vdev_parent; 772 773 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 774 775 ASSERT(mvd->vdev_children == 1); 776 ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 777 mvd->vdev_ops == &vdev_replacing_ops || 778 mvd->vdev_ops == &vdev_spare_ops); 779 cvd->vdev_ashift = mvd->vdev_ashift; 780 781 vdev_remove_child(mvd, cvd); 782 vdev_remove_child(pvd, mvd); 783 784 /* 785 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 786 * Otherwise, we could have detached an offline device, and when we 787 * go to import the pool we'll think we have two top-level vdevs, 788 * instead of a different version of the same top-level vdev. 789 */ 790 if (mvd->vdev_top == mvd) { 791 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 792 cvd->vdev_orig_guid = cvd->vdev_guid; 793 cvd->vdev_guid += guid_delta; 794 cvd->vdev_guid_sum += guid_delta; 795 } 796 cvd->vdev_id = mvd->vdev_id; 797 vdev_add_child(pvd, cvd); 798 vdev_top_update(cvd->vdev_top, cvd->vdev_top); 799 800 if (cvd == cvd->vdev_top) 801 vdev_top_transfer(mvd, cvd); 802 803 ASSERT(mvd->vdev_children == 0); 804 vdev_free(mvd); 805} 806 807int 808vdev_metaslab_init(vdev_t *vd, uint64_t txg) 809{ 810 spa_t *spa = vd->vdev_spa; 811 objset_t *mos = spa->spa_meta_objset; 812 uint64_t m; 813 uint64_t oldc = vd->vdev_ms_count; 814 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 815 metaslab_t **mspp; 816 int error; 817 818 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 819 820 /* 821 * This vdev is not being allocated from yet or is a hole. 822 */ 823 if (vd->vdev_ms_shift == 0) 824 return (0); 825 826 ASSERT(!vd->vdev_ishole); 827 828 /* 829 * Compute the raidz-deflation ratio. Note, we hard-code 830 * in 128k (1 << 17) because it is the current "typical" blocksize. 831 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change, 832 * or we will inconsistently account for existing bp's. 833 */ 834 vd->vdev_deflate_ratio = (1 << 17) / 835 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 836 837 ASSERT(oldc <= newc); 838 839 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 840 841 if (oldc != 0) { 842 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 843 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 844 } 845 846 vd->vdev_ms = mspp; 847 vd->vdev_ms_count = newc; 848 849 for (m = oldc; m < newc; m++) { 850 space_map_obj_t smo = { 0, 0, 0 }; 851 if (txg == 0) { 852 uint64_t object = 0; 853 error = dmu_read(mos, vd->vdev_ms_array, 854 m * sizeof (uint64_t), sizeof (uint64_t), &object, 855 DMU_READ_PREFETCH); 856 if (error) 857 return (error); 858 if (object != 0) { 859 dmu_buf_t *db; 860 error = dmu_bonus_hold(mos, object, FTAG, &db); 861 if (error) 862 return (error); 863 ASSERT3U(db->db_size, >=, sizeof (smo)); 864 bcopy(db->db_data, &smo, sizeof (smo)); 865 ASSERT3U(smo.smo_object, ==, object); 866 dmu_buf_rele(db, FTAG); 867 } 868 } 869 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo, 870 m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg); 871 } 872 873 if (txg == 0) 874 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 875 876 /* 877 * If the vdev is being removed we don't activate 878 * the metaslabs since we want to ensure that no new 879 * allocations are performed on this device. 880 */ 881 if (oldc == 0 && !vd->vdev_removing) 882 metaslab_group_activate(vd->vdev_mg); 883 884 if (txg == 0) 885 spa_config_exit(spa, SCL_ALLOC, FTAG); 886 887 return (0); 888} 889 890void 891vdev_metaslab_fini(vdev_t *vd) 892{ 893 uint64_t m; 894 uint64_t count = vd->vdev_ms_count; 895 896 if (vd->vdev_ms != NULL) { 897 metaslab_group_passivate(vd->vdev_mg); 898 for (m = 0; m < count; m++) 899 if (vd->vdev_ms[m] != NULL) 900 metaslab_fini(vd->vdev_ms[m]); 901 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 902 vd->vdev_ms = NULL; 903 } 904} 905 906typedef struct vdev_probe_stats { 907 boolean_t vps_readable; 908 boolean_t vps_writeable; 909 int vps_flags; 910} vdev_probe_stats_t; 911 912static void 913vdev_probe_done(zio_t *zio) 914{ 915 spa_t *spa = zio->io_spa; 916 vdev_t *vd = zio->io_vd; 917 vdev_probe_stats_t *vps = zio->io_private; 918 919 ASSERT(vd->vdev_probe_zio != NULL); 920 921 if (zio->io_type == ZIO_TYPE_READ) { 922 if (zio->io_error == 0) 923 vps->vps_readable = 1; 924 if (zio->io_error == 0 && spa_writeable(spa)) { 925 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 926 zio->io_offset, zio->io_size, zio->io_data, 927 ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 928 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 929 } else { 930 zio_buf_free(zio->io_data, zio->io_size); 931 } 932 } else if (zio->io_type == ZIO_TYPE_WRITE) { 933 if (zio->io_error == 0) 934 vps->vps_writeable = 1; 935 zio_buf_free(zio->io_data, zio->io_size); 936 } else if (zio->io_type == ZIO_TYPE_NULL) { 937 zio_t *pio; 938 939 vd->vdev_cant_read |= !vps->vps_readable; 940 vd->vdev_cant_write |= !vps->vps_writeable; 941 942 if (vdev_readable(vd) && 943 (vdev_writeable(vd) || !spa_writeable(spa))) { 944 zio->io_error = 0; 945 } else { 946 ASSERT(zio->io_error != 0); 947 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 948 spa, vd, NULL, 0, 0); 949 zio->io_error = SET_ERROR(ENXIO); 950 } 951 952 mutex_enter(&vd->vdev_probe_lock); 953 ASSERT(vd->vdev_probe_zio == zio); 954 vd->vdev_probe_zio = NULL; 955 mutex_exit(&vd->vdev_probe_lock); 956 957 while ((pio = zio_walk_parents(zio)) != NULL) 958 if (!vdev_accessible(vd, pio)) 959 pio->io_error = SET_ERROR(ENXIO); 960 961 kmem_free(vps, sizeof (*vps)); 962 } 963} 964 965/* 966 * Determine whether this device is accessible by reading and writing 967 * to several known locations: the pad regions of each vdev label 968 * but the first (which we leave alone in case it contains a VTOC). 969 */ 970zio_t * 971vdev_probe(vdev_t *vd, zio_t *zio) 972{ 973 spa_t *spa = vd->vdev_spa; 974 vdev_probe_stats_t *vps = NULL; 975 zio_t *pio; 976 977 ASSERT(vd->vdev_ops->vdev_op_leaf); 978 979 /* 980 * Don't probe the probe. 981 */ 982 if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 983 return (NULL); 984 985 /* 986 * To prevent 'probe storms' when a device fails, we create 987 * just one probe i/o at a time. All zios that want to probe 988 * this vdev will become parents of the probe io. 989 */ 990 mutex_enter(&vd->vdev_probe_lock); 991 992 if ((pio = vd->vdev_probe_zio) == NULL) { 993 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 994 995 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 996 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 997 ZIO_FLAG_TRYHARD; 998 999 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 1000 /* 1001 * vdev_cant_read and vdev_cant_write can only 1002 * transition from TRUE to FALSE when we have the 1003 * SCL_ZIO lock as writer; otherwise they can only 1004 * transition from FALSE to TRUE. This ensures that 1005 * any zio looking at these values can assume that 1006 * failures persist for the life of the I/O. That's 1007 * important because when a device has intermittent 1008 * connectivity problems, we want to ensure that 1009 * they're ascribed to the device (ENXIO) and not 1010 * the zio (EIO). 1011 * 1012 * Since we hold SCL_ZIO as writer here, clear both 1013 * values so the probe can reevaluate from first 1014 * principles. 1015 */ 1016 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1017 vd->vdev_cant_read = B_FALSE; 1018 vd->vdev_cant_write = B_FALSE; 1019 } 1020 1021 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1022 vdev_probe_done, vps, 1023 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1024 1025 /* 1026 * We can't change the vdev state in this context, so we 1027 * kick off an async task to do it on our behalf. 1028 */ 1029 if (zio != NULL) { 1030 vd->vdev_probe_wanted = B_TRUE; 1031 spa_async_request(spa, SPA_ASYNC_PROBE); 1032 } 1033 } 1034 1035 if (zio != NULL) 1036 zio_add_child(zio, pio); 1037 1038 mutex_exit(&vd->vdev_probe_lock); 1039 1040 if (vps == NULL) { 1041 ASSERT(zio != NULL); 1042 return (NULL); 1043 } 1044 1045 for (int l = 1; l < VDEV_LABELS; l++) { 1046 zio_nowait(zio_read_phys(pio, vd, 1047 vdev_label_offset(vd->vdev_psize, l, 1048 offsetof(vdev_label_t, vl_pad2)), 1049 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), 1050 ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1051 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1052 } 1053 1054 if (zio == NULL) 1055 return (pio); 1056 1057 zio_nowait(pio); 1058 return (NULL); 1059} 1060 1061static void 1062vdev_open_child(void *arg) 1063{ 1064 vdev_t *vd = arg; 1065 1066 vd->vdev_open_thread = curthread; 1067 vd->vdev_open_error = vdev_open(vd); 1068 vd->vdev_open_thread = NULL; 1069} 1070 1071boolean_t 1072vdev_uses_zvols(vdev_t *vd) 1073{ 1074 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1075 strlen(ZVOL_DIR)) == 0) 1076 return (B_TRUE); 1077 for (int c = 0; c < vd->vdev_children; c++) 1078 if (vdev_uses_zvols(vd->vdev_child[c])) 1079 return (B_TRUE); 1080 return (B_FALSE); 1081} 1082 1083void 1084vdev_open_children(vdev_t *vd) 1085{ 1086 taskq_t *tq; 1087 int children = vd->vdev_children; 1088 1089 /* 1090 * in order to handle pools on top of zvols, do the opens 1091 * in a single thread so that the same thread holds the 1092 * spa_namespace_lock 1093 */ 1094 if (B_TRUE || vdev_uses_zvols(vd)) { 1095 for (int c = 0; c < children; c++) 1096 vd->vdev_child[c]->vdev_open_error = 1097 vdev_open(vd->vdev_child[c]); 1098 return; 1099 } 1100 tq = taskq_create("vdev_open", children, minclsyspri, 1101 children, children, TASKQ_PREPOPULATE); 1102 1103 for (int c = 0; c < children; c++) 1104 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], 1105 TQ_SLEEP) != 0); 1106 1107 taskq_destroy(tq); 1108} 1109 1110/* 1111 * Prepare a virtual device for access. 1112 */ 1113int 1114vdev_open(vdev_t *vd) 1115{ 1116 spa_t *spa = vd->vdev_spa; 1117 int error; 1118 uint64_t osize = 0; 1119 uint64_t max_osize = 0; 1120 uint64_t asize, max_asize, psize; 1121 uint64_t ashift = 0; 1122 1123 ASSERT(vd->vdev_open_thread == curthread || 1124 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1125 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1126 vd->vdev_state == VDEV_STATE_CANT_OPEN || 1127 vd->vdev_state == VDEV_STATE_OFFLINE); 1128 1129 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1130 vd->vdev_cant_read = B_FALSE; 1131 vd->vdev_cant_write = B_FALSE; 1132 vd->vdev_min_asize = vdev_get_min_asize(vd); 1133 1134 /* 1135 * If this vdev is not removed, check its fault status. If it's 1136 * faulted, bail out of the open. 1137 */ 1138 if (!vd->vdev_removed && vd->vdev_faulted) { 1139 ASSERT(vd->vdev_children == 0); 1140 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1141 vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1142 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1143 vd->vdev_label_aux); 1144 return (SET_ERROR(ENXIO)); 1145 } else if (vd->vdev_offline) { 1146 ASSERT(vd->vdev_children == 0); 1147 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1148 return (SET_ERROR(ENXIO)); 1149 } 1150 1151 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); 1152 1153 /* 1154 * Reset the vdev_reopening flag so that we actually close 1155 * the vdev on error. 1156 */ 1157 vd->vdev_reopening = B_FALSE; 1158 if (zio_injection_enabled && error == 0) 1159 error = zio_handle_device_injection(vd, NULL, ENXIO); 1160 1161 if (error) { 1162 if (vd->vdev_removed && 1163 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 1164 vd->vdev_removed = B_FALSE; 1165 1166 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1167 vd->vdev_stat.vs_aux); 1168 return (error); 1169 } 1170 1171 vd->vdev_removed = B_FALSE; 1172 1173 /* 1174 * Recheck the faulted flag now that we have confirmed that 1175 * the vdev is accessible. If we're faulted, bail. 1176 */ 1177 if (vd->vdev_faulted) { 1178 ASSERT(vd->vdev_children == 0); 1179 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1180 vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1181 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1182 vd->vdev_label_aux); 1183 return (SET_ERROR(ENXIO)); 1184 } 1185 1186 if (vd->vdev_degraded) { 1187 ASSERT(vd->vdev_children == 0); 1188 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1189 VDEV_AUX_ERR_EXCEEDED); 1190 } else { 1191 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 1192 } 1193 1194 /* 1195 * For hole or missing vdevs we just return success. 1196 */ 1197 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 1198 return (0); 1199 1200 if (vd->vdev_ops->vdev_op_leaf) { 1201 vd->vdev_notrim = B_FALSE; 1202 trim_map_create(vd); 1203 } 1204 1205 for (int c = 0; c < vd->vdev_children; c++) { 1206 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1207 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1208 VDEV_AUX_NONE); 1209 break; 1210 } 1211 } 1212 1213 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 1214 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1215 1216 if (vd->vdev_children == 0) { 1217 if (osize < SPA_MINDEVSIZE) { 1218 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1219 VDEV_AUX_TOO_SMALL); 1220 return (SET_ERROR(EOVERFLOW)); 1221 } 1222 psize = osize; 1223 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 1224 max_asize = max_osize - (VDEV_LABEL_START_SIZE + 1225 VDEV_LABEL_END_SIZE); 1226 } else { 1227 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1228 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1229 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1230 VDEV_AUX_TOO_SMALL); 1231 return (SET_ERROR(EOVERFLOW)); 1232 } 1233 psize = 0; 1234 asize = osize; 1235 max_asize = max_osize; 1236 } 1237 1238 vd->vdev_psize = psize; 1239 1240 /* 1241 * Make sure the allocatable size hasn't shrunk. 1242 */ 1243 if (asize < vd->vdev_min_asize) { 1244 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1245 VDEV_AUX_BAD_LABEL); 1246 return (SET_ERROR(EINVAL)); 1247 } 1248 1249 if (vd->vdev_asize == 0) { 1250 /* 1251 * This is the first-ever open, so use the computed values. 1252 * For testing purposes, a higher ashift can be requested. 1253 */ 1254 vd->vdev_asize = asize; 1255 vd->vdev_max_asize = max_asize; 1256 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 1257 } else { 1258 /* 1259 * Make sure the alignment requirement hasn't increased. 1260 */ 1261 if (ashift > vd->vdev_top->vdev_ashift) { 1262 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1263 VDEV_AUX_BAD_LABEL); 1264 return (EINVAL); 1265 } 1266 vd->vdev_max_asize = max_asize; 1267 } 1268 1269 /* 1270 * If all children are healthy and the asize has increased, 1271 * then we've experienced dynamic LUN growth. If automatic 1272 * expansion is enabled then use the additional space. 1273 */ 1274 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && 1275 (vd->vdev_expanding || spa->spa_autoexpand)) 1276 vd->vdev_asize = asize; 1277 1278 vdev_set_min_asize(vd); 1279 1280 /* 1281 * Ensure we can issue some IO before declaring the 1282 * vdev open for business. 1283 */ 1284 if (vd->vdev_ops->vdev_op_leaf && 1285 (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 1286 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1287 VDEV_AUX_ERR_EXCEEDED); 1288 return (error); 1289 } 1290 1291 /* 1292 * If a leaf vdev has a DTL, and seems healthy, then kick off a 1293 * resilver. But don't do this if we are doing a reopen for a scrub, 1294 * since this would just restart the scrub we are already doing. 1295 */ 1296 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 1297 vdev_resilver_needed(vd, NULL, NULL)) 1298 spa_async_request(spa, SPA_ASYNC_RESILVER); 1299 1300 return (0); 1301} 1302 1303/* 1304 * Called once the vdevs are all opened, this routine validates the label 1305 * contents. This needs to be done before vdev_load() so that we don't 1306 * inadvertently do repair I/Os to the wrong device. 1307 * 1308 * If 'strict' is false ignore the spa guid check. This is necessary because 1309 * if the machine crashed during a re-guid the new guid might have been written 1310 * to all of the vdev labels, but not the cached config. The strict check 1311 * will be performed when the pool is opened again using the mos config. 1312 * 1313 * This function will only return failure if one of the vdevs indicates that it 1314 * has since been destroyed or exported. This is only possible if 1315 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1316 * will be updated but the function will return 0. 1317 */ 1318int 1319vdev_validate(vdev_t *vd, boolean_t strict) 1320{ 1321 spa_t *spa = vd->vdev_spa; 1322 nvlist_t *label; 1323 uint64_t guid = 0, top_guid; 1324 uint64_t state; 1325 1326 for (int c = 0; c < vd->vdev_children; c++) 1327 if (vdev_validate(vd->vdev_child[c], strict) != 0) 1328 return (SET_ERROR(EBADF)); 1329 1330 /* 1331 * If the device has already failed, or was marked offline, don't do 1332 * any further validation. Otherwise, label I/O will fail and we will 1333 * overwrite the previous state. 1334 */ 1335 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 1336 uint64_t aux_guid = 0; 1337 nvlist_t *nvl; 1338 uint64_t txg = spa_last_synced_txg(spa) != 0 ? 1339 spa_last_synced_txg(spa) : -1ULL; 1340 1341 if ((label = vdev_label_read_config(vd, txg)) == NULL) { 1342 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1343 VDEV_AUX_BAD_LABEL); 1344 return (0); 1345 } 1346 1347 /* 1348 * Determine if this vdev has been split off into another 1349 * pool. If so, then refuse to open it. 1350 */ 1351 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 1352 &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 1353 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1354 VDEV_AUX_SPLIT_POOL); 1355 nvlist_free(label); 1356 return (0); 1357 } 1358 1359 if (strict && (nvlist_lookup_uint64(label, 1360 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 || 1361 guid != spa_guid(spa))) { 1362 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1363 VDEV_AUX_CORRUPT_DATA); 1364 nvlist_free(label); 1365 return (0); 1366 } 1367 1368 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 1369 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 1370 &aux_guid) != 0) 1371 aux_guid = 0; 1372 1373 /* 1374 * If this vdev just became a top-level vdev because its 1375 * sibling was detached, it will have adopted the parent's 1376 * vdev guid -- but the label may or may not be on disk yet. 1377 * Fortunately, either version of the label will have the 1378 * same top guid, so if we're a top-level vdev, we can 1379 * safely compare to that instead. 1380 * 1381 * If we split this vdev off instead, then we also check the 1382 * original pool's guid. We don't want to consider the vdev 1383 * corrupt if it is partway through a split operation. 1384 */ 1385 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 1386 &guid) != 0 || 1387 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, 1388 &top_guid) != 0 || 1389 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && 1390 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { 1391 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1392 VDEV_AUX_CORRUPT_DATA); 1393 nvlist_free(label); 1394 return (0); 1395 } 1396 1397 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1398 &state) != 0) { 1399 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1400 VDEV_AUX_CORRUPT_DATA); 1401 nvlist_free(label); 1402 return (0); 1403 } 1404 1405 nvlist_free(label); 1406 1407 /* 1408 * If this is a verbatim import, no need to check the 1409 * state of the pool. 1410 */ 1411 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 1412 spa_load_state(spa) == SPA_LOAD_OPEN && 1413 state != POOL_STATE_ACTIVE) 1414 return (SET_ERROR(EBADF)); 1415 1416 /* 1417 * If we were able to open and validate a vdev that was 1418 * previously marked permanently unavailable, clear that state 1419 * now. 1420 */ 1421 if (vd->vdev_not_present) 1422 vd->vdev_not_present = 0; 1423 } 1424 1425 return (0); 1426} 1427 1428/* 1429 * Close a virtual device. 1430 */ 1431void 1432vdev_close(vdev_t *vd) 1433{ 1434 spa_t *spa = vd->vdev_spa; 1435 vdev_t *pvd = vd->vdev_parent; 1436 1437 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1438 1439 /* 1440 * If our parent is reopening, then we are as well, unless we are 1441 * going offline. 1442 */ 1443 if (pvd != NULL && pvd->vdev_reopening) 1444 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 1445 1446 vd->vdev_ops->vdev_op_close(vd); 1447 1448 vdev_cache_purge(vd); 1449 1450 if (vd->vdev_ops->vdev_op_leaf) 1451 trim_map_destroy(vd); 1452 1453 /* 1454 * We record the previous state before we close it, so that if we are 1455 * doing a reopen(), we don't generate FMA ereports if we notice that 1456 * it's still faulted. 1457 */ 1458 vd->vdev_prevstate = vd->vdev_state; 1459 1460 if (vd->vdev_offline) 1461 vd->vdev_state = VDEV_STATE_OFFLINE; 1462 else 1463 vd->vdev_state = VDEV_STATE_CLOSED; 1464 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1465} 1466 1467void 1468vdev_hold(vdev_t *vd) 1469{ 1470 spa_t *spa = vd->vdev_spa; 1471 1472 ASSERT(spa_is_root(spa)); 1473 if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1474 return; 1475 1476 for (int c = 0; c < vd->vdev_children; c++) 1477 vdev_hold(vd->vdev_child[c]); 1478 1479 if (vd->vdev_ops->vdev_op_leaf) 1480 vd->vdev_ops->vdev_op_hold(vd); 1481} 1482 1483void 1484vdev_rele(vdev_t *vd) 1485{ 1486 spa_t *spa = vd->vdev_spa; 1487 1488 ASSERT(spa_is_root(spa)); 1489 for (int c = 0; c < vd->vdev_children; c++) 1490 vdev_rele(vd->vdev_child[c]); 1491 1492 if (vd->vdev_ops->vdev_op_leaf) 1493 vd->vdev_ops->vdev_op_rele(vd); 1494} 1495 1496/* 1497 * Reopen all interior vdevs and any unopened leaves. We don't actually 1498 * reopen leaf vdevs which had previously been opened as they might deadlock 1499 * on the spa_config_lock. Instead we only obtain the leaf's physical size. 1500 * If the leaf has never been opened then open it, as usual. 1501 */ 1502void 1503vdev_reopen(vdev_t *vd) 1504{ 1505 spa_t *spa = vd->vdev_spa; 1506 1507 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1508 1509 /* set the reopening flag unless we're taking the vdev offline */ 1510 vd->vdev_reopening = !vd->vdev_offline; 1511 vdev_close(vd); 1512 (void) vdev_open(vd); 1513 1514 /* 1515 * Call vdev_validate() here to make sure we have the same device. 1516 * Otherwise, a device with an invalid label could be successfully 1517 * opened in response to vdev_reopen(). 1518 */ 1519 if (vd->vdev_aux) { 1520 (void) vdev_validate_aux(vd); 1521 if (vdev_readable(vd) && vdev_writeable(vd) && 1522 vd->vdev_aux == &spa->spa_l2cache && 1523 !l2arc_vdev_present(vd)) 1524 l2arc_add_vdev(spa, vd); 1525 } else { 1526 (void) vdev_validate(vd, B_TRUE); 1527 } 1528 1529 /* 1530 * Reassess parent vdev's health. 1531 */ 1532 vdev_propagate_state(vd); 1533} 1534 1535int 1536vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 1537{ 1538 int error; 1539 1540 /* 1541 * Normally, partial opens (e.g. of a mirror) are allowed. 1542 * For a create, however, we want to fail the request if 1543 * there are any components we can't open. 1544 */ 1545 error = vdev_open(vd); 1546 1547 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 1548 vdev_close(vd); 1549 return (error ? error : ENXIO); 1550 } 1551 1552 /* 1553 * Recursively initialize all labels. 1554 */ 1555 if ((error = vdev_label_init(vd, txg, isreplacing ? 1556 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 1557 vdev_close(vd); 1558 return (error); 1559 } 1560 1561 return (0); 1562} 1563 1564void 1565vdev_metaslab_set_size(vdev_t *vd) 1566{ 1567 /* 1568 * Aim for roughly 200 metaslabs per vdev. 1569 */ 1570 vd->vdev_ms_shift = highbit(vd->vdev_asize / 200); 1571 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); 1572} 1573 1574void 1575vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 1576{ 1577 ASSERT(vd == vd->vdev_top); 1578 ASSERT(!vd->vdev_ishole); 1579 ASSERT(ISP2(flags)); 1580 ASSERT(spa_writeable(vd->vdev_spa)); 1581 1582 if (flags & VDD_METASLAB) 1583 (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 1584 1585 if (flags & VDD_DTL) 1586 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 1587 1588 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 1589} 1590 1591/* 1592 * DTLs. 1593 * 1594 * A vdev's DTL (dirty time log) is the set of transaction groups for which 1595 * the vdev has less than perfect replication. There are four kinds of DTL: 1596 * 1597 * DTL_MISSING: txgs for which the vdev has no valid copies of the data 1598 * 1599 * DTL_PARTIAL: txgs for which data is available, but not fully replicated 1600 * 1601 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 1602 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 1603 * txgs that was scrubbed. 1604 * 1605 * DTL_OUTAGE: txgs which cannot currently be read, whether due to 1606 * persistent errors or just some device being offline. 1607 * Unlike the other three, the DTL_OUTAGE map is not generally 1608 * maintained; it's only computed when needed, typically to 1609 * determine whether a device can be detached. 1610 * 1611 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 1612 * either has the data or it doesn't. 1613 * 1614 * For interior vdevs such as mirror and RAID-Z the picture is more complex. 1615 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 1616 * if any child is less than fully replicated, then so is its parent. 1617 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 1618 * comprising only those txgs which appear in 'maxfaults' or more children; 1619 * those are the txgs we don't have enough replication to read. For example, 1620 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 1621 * thus, its DTL_MISSING consists of the set of txgs that appear in more than 1622 * two child DTL_MISSING maps. 1623 * 1624 * It should be clear from the above that to compute the DTLs and outage maps 1625 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 1626 * Therefore, that is all we keep on disk. When loading the pool, or after 1627 * a configuration change, we generate all other DTLs from first principles. 1628 */ 1629void 1630vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1631{ 1632 space_map_t *sm = &vd->vdev_dtl[t]; 1633 1634 ASSERT(t < DTL_TYPES); 1635 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1636 ASSERT(spa_writeable(vd->vdev_spa)); 1637 1638 mutex_enter(sm->sm_lock); 1639 if (!space_map_contains(sm, txg, size)) 1640 space_map_add(sm, txg, size); 1641 mutex_exit(sm->sm_lock); 1642} 1643 1644boolean_t 1645vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1646{ 1647 space_map_t *sm = &vd->vdev_dtl[t]; 1648 boolean_t dirty = B_FALSE; 1649 1650 ASSERT(t < DTL_TYPES); 1651 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1652 1653 mutex_enter(sm->sm_lock); 1654 if (sm->sm_space != 0) 1655 dirty = space_map_contains(sm, txg, size); 1656 mutex_exit(sm->sm_lock); 1657 1658 return (dirty); 1659} 1660 1661boolean_t 1662vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 1663{ 1664 space_map_t *sm = &vd->vdev_dtl[t]; 1665 boolean_t empty; 1666 1667 mutex_enter(sm->sm_lock); 1668 empty = (sm->sm_space == 0); 1669 mutex_exit(sm->sm_lock); 1670 1671 return (empty); 1672} 1673 1674/* 1675 * Reassess DTLs after a config change or scrub completion. 1676 */ 1677void 1678vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 1679{ 1680 spa_t *spa = vd->vdev_spa; 1681 avl_tree_t reftree; 1682 int minref; 1683 1684 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1685 1686 for (int c = 0; c < vd->vdev_children; c++) 1687 vdev_dtl_reassess(vd->vdev_child[c], txg, 1688 scrub_txg, scrub_done); 1689 1690 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) 1691 return; 1692 1693 if (vd->vdev_ops->vdev_op_leaf) { 1694 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 1695 1696 mutex_enter(&vd->vdev_dtl_lock); 1697 if (scrub_txg != 0 && 1698 (spa->spa_scrub_started || 1699 (scn && scn->scn_phys.scn_errors == 0))) { 1700 /* 1701 * We completed a scrub up to scrub_txg. If we 1702 * did it without rebooting, then the scrub dtl 1703 * will be valid, so excise the old region and 1704 * fold in the scrub dtl. Otherwise, leave the 1705 * dtl as-is if there was an error. 1706 * 1707 * There's little trick here: to excise the beginning 1708 * of the DTL_MISSING map, we put it into a reference 1709 * tree and then add a segment with refcnt -1 that 1710 * covers the range [0, scrub_txg). This means 1711 * that each txg in that range has refcnt -1 or 0. 1712 * We then add DTL_SCRUB with a refcnt of 2, so that 1713 * entries in the range [0, scrub_txg) will have a 1714 * positive refcnt -- either 1 or 2. We then convert 1715 * the reference tree into the new DTL_MISSING map. 1716 */ 1717 space_map_ref_create(&reftree); 1718 space_map_ref_add_map(&reftree, 1719 &vd->vdev_dtl[DTL_MISSING], 1); 1720 space_map_ref_add_seg(&reftree, 0, scrub_txg, -1); 1721 space_map_ref_add_map(&reftree, 1722 &vd->vdev_dtl[DTL_SCRUB], 2); 1723 space_map_ref_generate_map(&reftree, 1724 &vd->vdev_dtl[DTL_MISSING], 1); 1725 space_map_ref_destroy(&reftree); 1726 } 1727 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 1728 space_map_walk(&vd->vdev_dtl[DTL_MISSING], 1729 space_map_add, &vd->vdev_dtl[DTL_PARTIAL]); 1730 if (scrub_done) 1731 space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 1732 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 1733 if (!vdev_readable(vd)) 1734 space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 1735 else 1736 space_map_walk(&vd->vdev_dtl[DTL_MISSING], 1737 space_map_add, &vd->vdev_dtl[DTL_OUTAGE]); 1738 mutex_exit(&vd->vdev_dtl_lock); 1739 1740 if (txg != 0) 1741 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 1742 return; 1743 } 1744 1745 mutex_enter(&vd->vdev_dtl_lock); 1746 for (int t = 0; t < DTL_TYPES; t++) { 1747 /* account for child's outage in parent's missing map */ 1748 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 1749 if (t == DTL_SCRUB) 1750 continue; /* leaf vdevs only */ 1751 if (t == DTL_PARTIAL) 1752 minref = 1; /* i.e. non-zero */ 1753 else if (vd->vdev_nparity != 0) 1754 minref = vd->vdev_nparity + 1; /* RAID-Z */ 1755 else 1756 minref = vd->vdev_children; /* any kind of mirror */ 1757 space_map_ref_create(&reftree); 1758 for (int c = 0; c < vd->vdev_children; c++) { 1759 vdev_t *cvd = vd->vdev_child[c]; 1760 mutex_enter(&cvd->vdev_dtl_lock); 1761 space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1); 1762 mutex_exit(&cvd->vdev_dtl_lock); 1763 } 1764 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref); 1765 space_map_ref_destroy(&reftree); 1766 } 1767 mutex_exit(&vd->vdev_dtl_lock); 1768} 1769 1770static int 1771vdev_dtl_load(vdev_t *vd) 1772{ 1773 spa_t *spa = vd->vdev_spa; 1774 space_map_obj_t *smo = &vd->vdev_dtl_smo; 1775 objset_t *mos = spa->spa_meta_objset; 1776 dmu_buf_t *db; 1777 int error; 1778 1779 ASSERT(vd->vdev_children == 0); 1780 1781 if (smo->smo_object == 0) 1782 return (0); 1783 1784 ASSERT(!vd->vdev_ishole); 1785 1786 if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0) 1787 return (error); 1788 1789 ASSERT3U(db->db_size, >=, sizeof (*smo)); 1790 bcopy(db->db_data, smo, sizeof (*smo)); 1791 dmu_buf_rele(db, FTAG); 1792 1793 mutex_enter(&vd->vdev_dtl_lock); 1794 error = space_map_load(&vd->vdev_dtl[DTL_MISSING], 1795 NULL, SM_ALLOC, smo, mos); 1796 mutex_exit(&vd->vdev_dtl_lock); 1797 1798 return (error); 1799} 1800 1801void 1802vdev_dtl_sync(vdev_t *vd, uint64_t txg) 1803{ 1804 spa_t *spa = vd->vdev_spa; 1805 space_map_obj_t *smo = &vd->vdev_dtl_smo; 1806 space_map_t *sm = &vd->vdev_dtl[DTL_MISSING]; 1807 objset_t *mos = spa->spa_meta_objset; 1808 space_map_t smsync; 1809 kmutex_t smlock; 1810 dmu_buf_t *db; 1811 dmu_tx_t *tx; 1812 1813 ASSERT(!vd->vdev_ishole); 1814 1815 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 1816 1817 if (vd->vdev_detached) { 1818 if (smo->smo_object != 0) { 1819 int err = dmu_object_free(mos, smo->smo_object, tx); 1820 ASSERT0(err); 1821 smo->smo_object = 0; 1822 } 1823 dmu_tx_commit(tx); 1824 return; 1825 } 1826 1827 if (smo->smo_object == 0) { 1828 ASSERT(smo->smo_objsize == 0); 1829 ASSERT(smo->smo_alloc == 0); 1830 smo->smo_object = dmu_object_alloc(mos, 1831 DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT, 1832 DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx); 1833 ASSERT(smo->smo_object != 0); 1834 vdev_config_dirty(vd->vdev_top); 1835 } 1836 1837 mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL); 1838 1839 space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift, 1840 &smlock); 1841 1842 mutex_enter(&smlock); 1843 1844 mutex_enter(&vd->vdev_dtl_lock); 1845 space_map_walk(sm, space_map_add, &smsync); 1846 mutex_exit(&vd->vdev_dtl_lock); 1847 1848 space_map_truncate(smo, mos, tx); 1849 space_map_sync(&smsync, SM_ALLOC, smo, mos, tx); 1850 space_map_vacate(&smsync, NULL, NULL); 1851 1852 space_map_destroy(&smsync); 1853 1854 mutex_exit(&smlock); 1855 mutex_destroy(&smlock); 1856 1857 VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)); 1858 dmu_buf_will_dirty(db, tx); 1859 ASSERT3U(db->db_size, >=, sizeof (*smo)); 1860 bcopy(smo, db->db_data, sizeof (*smo)); 1861 dmu_buf_rele(db, FTAG); 1862 1863 dmu_tx_commit(tx); 1864} 1865 1866/* 1867 * Determine whether the specified vdev can be offlined/detached/removed 1868 * without losing data. 1869 */ 1870boolean_t 1871vdev_dtl_required(vdev_t *vd) 1872{ 1873 spa_t *spa = vd->vdev_spa; 1874 vdev_t *tvd = vd->vdev_top; 1875 uint8_t cant_read = vd->vdev_cant_read; 1876 boolean_t required; 1877 1878 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1879 1880 if (vd == spa->spa_root_vdev || vd == tvd) 1881 return (B_TRUE); 1882 1883 /* 1884 * Temporarily mark the device as unreadable, and then determine 1885 * whether this results in any DTL outages in the top-level vdev. 1886 * If not, we can safely offline/detach/remove the device. 1887 */ 1888 vd->vdev_cant_read = B_TRUE; 1889 vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 1890 required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 1891 vd->vdev_cant_read = cant_read; 1892 vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 1893 1894 if (!required && zio_injection_enabled) 1895 required = !!zio_handle_device_injection(vd, NULL, ECHILD); 1896 1897 return (required); 1898} 1899 1900/* 1901 * Determine if resilver is needed, and if so the txg range. 1902 */ 1903boolean_t 1904vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 1905{ 1906 boolean_t needed = B_FALSE; 1907 uint64_t thismin = UINT64_MAX; 1908 uint64_t thismax = 0; 1909 1910 if (vd->vdev_children == 0) { 1911 mutex_enter(&vd->vdev_dtl_lock); 1912 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 && 1913 vdev_writeable(vd)) { 1914 space_seg_t *ss; 1915 1916 ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root); 1917 thismin = ss->ss_start - 1; 1918 ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root); 1919 thismax = ss->ss_end; 1920 needed = B_TRUE; 1921 } 1922 mutex_exit(&vd->vdev_dtl_lock); 1923 } else { 1924 for (int c = 0; c < vd->vdev_children; c++) { 1925 vdev_t *cvd = vd->vdev_child[c]; 1926 uint64_t cmin, cmax; 1927 1928 if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 1929 thismin = MIN(thismin, cmin); 1930 thismax = MAX(thismax, cmax); 1931 needed = B_TRUE; 1932 } 1933 } 1934 } 1935 1936 if (needed && minp) { 1937 *minp = thismin; 1938 *maxp = thismax; 1939 } 1940 return (needed); 1941} 1942 1943void 1944vdev_load(vdev_t *vd) 1945{ 1946 /* 1947 * Recursively load all children. 1948 */ 1949 for (int c = 0; c < vd->vdev_children; c++) 1950 vdev_load(vd->vdev_child[c]); 1951 1952 /* 1953 * If this is a top-level vdev, initialize its metaslabs. 1954 */ 1955 if (vd == vd->vdev_top && !vd->vdev_ishole && 1956 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || 1957 vdev_metaslab_init(vd, 0) != 0)) 1958 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1959 VDEV_AUX_CORRUPT_DATA); 1960 1961 /* 1962 * If this is a leaf vdev, load its DTL. 1963 */ 1964 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) 1965 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1966 VDEV_AUX_CORRUPT_DATA); 1967} 1968 1969/* 1970 * The special vdev case is used for hot spares and l2cache devices. Its 1971 * sole purpose it to set the vdev state for the associated vdev. To do this, 1972 * we make sure that we can open the underlying device, then try to read the 1973 * label, and make sure that the label is sane and that it hasn't been 1974 * repurposed to another pool. 1975 */ 1976int 1977vdev_validate_aux(vdev_t *vd) 1978{ 1979 nvlist_t *label; 1980 uint64_t guid, version; 1981 uint64_t state; 1982 1983 if (!vdev_readable(vd)) 1984 return (0); 1985 1986 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 1987 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1988 VDEV_AUX_CORRUPT_DATA); 1989 return (-1); 1990 } 1991 1992 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 1993 !SPA_VERSION_IS_SUPPORTED(version) || 1994 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 1995 guid != vd->vdev_guid || 1996 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 1997 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1998 VDEV_AUX_CORRUPT_DATA); 1999 nvlist_free(label); 2000 return (-1); 2001 } 2002 2003 /* 2004 * We don't actually check the pool state here. If it's in fact in 2005 * use by another pool, we update this fact on the fly when requested. 2006 */ 2007 nvlist_free(label); 2008 return (0); 2009} 2010 2011void 2012vdev_remove(vdev_t *vd, uint64_t txg) 2013{ 2014 spa_t *spa = vd->vdev_spa; 2015 objset_t *mos = spa->spa_meta_objset; 2016 dmu_tx_t *tx; 2017 2018 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 2019 2020 if (vd->vdev_dtl_smo.smo_object) { 2021 ASSERT0(vd->vdev_dtl_smo.smo_alloc); 2022 (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx); 2023 vd->vdev_dtl_smo.smo_object = 0; 2024 } 2025 2026 if (vd->vdev_ms != NULL) { 2027 for (int m = 0; m < vd->vdev_ms_count; m++) { 2028 metaslab_t *msp = vd->vdev_ms[m]; 2029 2030 if (msp == NULL || msp->ms_smo.smo_object == 0) 2031 continue; 2032 2033 ASSERT0(msp->ms_smo.smo_alloc); 2034 (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx); 2035 msp->ms_smo.smo_object = 0; 2036 } 2037 } 2038 2039 if (vd->vdev_ms_array) { 2040 (void) dmu_object_free(mos, vd->vdev_ms_array, tx); 2041 vd->vdev_ms_array = 0; 2042 vd->vdev_ms_shift = 0; 2043 } 2044 dmu_tx_commit(tx); 2045} 2046 2047void 2048vdev_sync_done(vdev_t *vd, uint64_t txg) 2049{ 2050 metaslab_t *msp; 2051 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 2052 2053 ASSERT(!vd->vdev_ishole); 2054 2055 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 2056 metaslab_sync_done(msp, txg); 2057 2058 if (reassess) 2059 metaslab_sync_reassess(vd->vdev_mg); 2060} 2061 2062void 2063vdev_sync(vdev_t *vd, uint64_t txg) 2064{ 2065 spa_t *spa = vd->vdev_spa; 2066 vdev_t *lvd; 2067 metaslab_t *msp; 2068 dmu_tx_t *tx; 2069 2070 ASSERT(!vd->vdev_ishole); 2071 2072 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { 2073 ASSERT(vd == vd->vdev_top); 2074 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2075 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 2076 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 2077 ASSERT(vd->vdev_ms_array != 0); 2078 vdev_config_dirty(vd); 2079 dmu_tx_commit(tx); 2080 } 2081 2082 /* 2083 * Remove the metadata associated with this vdev once it's empty. 2084 */ 2085 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 2086 vdev_remove(vd, txg); 2087 2088 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 2089 metaslab_sync(msp, txg); 2090 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 2091 } 2092 2093 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 2094 vdev_dtl_sync(lvd, txg); 2095 2096 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 2097} 2098 2099uint64_t 2100vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 2101{ 2102 return (vd->vdev_ops->vdev_op_asize(vd, psize)); 2103} 2104 2105/* 2106 * Mark the given vdev faulted. A faulted vdev behaves as if the device could 2107 * not be opened, and no I/O is attempted. 2108 */ 2109int 2110vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2111{ 2112 vdev_t *vd, *tvd; 2113 2114 spa_vdev_state_enter(spa, SCL_NONE); 2115 2116 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2117 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2118 2119 if (!vd->vdev_ops->vdev_op_leaf) 2120 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2121 2122 tvd = vd->vdev_top; 2123 2124 /* 2125 * We don't directly use the aux state here, but if we do a 2126 * vdev_reopen(), we need this value to be present to remember why we 2127 * were faulted. 2128 */ 2129 vd->vdev_label_aux = aux; 2130 2131 /* 2132 * Faulted state takes precedence over degraded. 2133 */ 2134 vd->vdev_delayed_close = B_FALSE; 2135 vd->vdev_faulted = 1ULL; 2136 vd->vdev_degraded = 0ULL; 2137 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 2138 2139 /* 2140 * If this device has the only valid copy of the data, then 2141 * back off and simply mark the vdev as degraded instead. 2142 */ 2143 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 2144 vd->vdev_degraded = 1ULL; 2145 vd->vdev_faulted = 0ULL; 2146 2147 /* 2148 * If we reopen the device and it's not dead, only then do we 2149 * mark it degraded. 2150 */ 2151 vdev_reopen(tvd); 2152 2153 if (vdev_readable(vd)) 2154 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 2155 } 2156 2157 return (spa_vdev_state_exit(spa, vd, 0)); 2158} 2159 2160/* 2161 * Mark the given vdev degraded. A degraded vdev is purely an indication to the 2162 * user that something is wrong. The vdev continues to operate as normal as far 2163 * as I/O is concerned. 2164 */ 2165int 2166vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2167{ 2168 vdev_t *vd; 2169 2170 spa_vdev_state_enter(spa, SCL_NONE); 2171 2172 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2173 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2174 2175 if (!vd->vdev_ops->vdev_op_leaf) 2176 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2177 2178 /* 2179 * If the vdev is already faulted, then don't do anything. 2180 */ 2181 if (vd->vdev_faulted || vd->vdev_degraded) 2182 return (spa_vdev_state_exit(spa, NULL, 0)); 2183 2184 vd->vdev_degraded = 1ULL; 2185 if (!vdev_is_dead(vd)) 2186 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 2187 aux); 2188 2189 return (spa_vdev_state_exit(spa, vd, 0)); 2190} 2191 2192/* 2193 * Online the given vdev. If 'unspare' is set, it implies two things. First, 2194 * any attached spare device should be detached when the device finishes 2195 * resilvering. Second, the online should be treated like a 'test' online case, 2196 * so no FMA events are generated if the device fails to open. 2197 */ 2198int 2199vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 2200{ 2201 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 2202 2203 spa_vdev_state_enter(spa, SCL_NONE); 2204 2205 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2206 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2207 2208 if (!vd->vdev_ops->vdev_op_leaf) 2209 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2210 2211 tvd = vd->vdev_top; 2212 vd->vdev_offline = B_FALSE; 2213 vd->vdev_tmpoffline = B_FALSE; 2214 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 2215 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 2216 2217 /* XXX - L2ARC 1.0 does not support expansion */ 2218 if (!vd->vdev_aux) { 2219 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2220 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 2221 } 2222 2223 vdev_reopen(tvd); 2224 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 2225 2226 if (!vd->vdev_aux) { 2227 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2228 pvd->vdev_expanding = B_FALSE; 2229 } 2230 2231 if (newstate) 2232 *newstate = vd->vdev_state; 2233 if ((flags & ZFS_ONLINE_UNSPARE) && 2234 !vdev_is_dead(vd) && vd->vdev_parent && 2235 vd->vdev_parent->vdev_ops == &vdev_spare_ops && 2236 vd->vdev_parent->vdev_child[0] == vd) 2237 vd->vdev_unspare = B_TRUE; 2238 2239 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 2240 2241 /* XXX - L2ARC 1.0 does not support expansion */ 2242 if (vd->vdev_aux) 2243 return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 2244 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2245 } 2246 return (spa_vdev_state_exit(spa, vd, 0)); 2247} 2248 2249static int 2250vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 2251{ 2252 vdev_t *vd, *tvd; 2253 int error = 0; 2254 uint64_t generation; 2255 metaslab_group_t *mg; 2256 2257top: 2258 spa_vdev_state_enter(spa, SCL_ALLOC); 2259 2260 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2261 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2262 2263 if (!vd->vdev_ops->vdev_op_leaf) 2264 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2265 2266 tvd = vd->vdev_top; 2267 mg = tvd->vdev_mg; 2268 generation = spa->spa_config_generation + 1; 2269 2270 /* 2271 * If the device isn't already offline, try to offline it. 2272 */ 2273 if (!vd->vdev_offline) { 2274 /* 2275 * If this device has the only valid copy of some data, 2276 * don't allow it to be offlined. Log devices are always 2277 * expendable. 2278 */ 2279 if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2280 vdev_dtl_required(vd)) 2281 return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2282 2283 /* 2284 * If the top-level is a slog and it has had allocations 2285 * then proceed. We check that the vdev's metaslab group 2286 * is not NULL since it's possible that we may have just 2287 * added this vdev but not yet initialized its metaslabs. 2288 */ 2289 if (tvd->vdev_islog && mg != NULL) { 2290 /* 2291 * Prevent any future allocations. 2292 */ 2293 metaslab_group_passivate(mg); 2294 (void) spa_vdev_state_exit(spa, vd, 0); 2295 2296 error = spa_offline_log(spa); 2297 2298 spa_vdev_state_enter(spa, SCL_ALLOC); 2299 2300 /* 2301 * Check to see if the config has changed. 2302 */ 2303 if (error || generation != spa->spa_config_generation) { 2304 metaslab_group_activate(mg); 2305 if (error) 2306 return (spa_vdev_state_exit(spa, 2307 vd, error)); 2308 (void) spa_vdev_state_exit(spa, vd, 0); 2309 goto top; 2310 } 2311 ASSERT0(tvd->vdev_stat.vs_alloc); 2312 } 2313 2314 /* 2315 * Offline this device and reopen its top-level vdev. 2316 * If the top-level vdev is a log device then just offline 2317 * it. Otherwise, if this action results in the top-level 2318 * vdev becoming unusable, undo it and fail the request. 2319 */ 2320 vd->vdev_offline = B_TRUE; 2321 vdev_reopen(tvd); 2322 2323 if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2324 vdev_is_dead(tvd)) { 2325 vd->vdev_offline = B_FALSE; 2326 vdev_reopen(tvd); 2327 return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2328 } 2329 2330 /* 2331 * Add the device back into the metaslab rotor so that 2332 * once we online the device it's open for business. 2333 */ 2334 if (tvd->vdev_islog && mg != NULL) 2335 metaslab_group_activate(mg); 2336 } 2337 2338 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 2339 2340 return (spa_vdev_state_exit(spa, vd, 0)); 2341} 2342 2343int 2344vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 2345{ 2346 int error; 2347 2348 mutex_enter(&spa->spa_vdev_top_lock); 2349 error = vdev_offline_locked(spa, guid, flags); 2350 mutex_exit(&spa->spa_vdev_top_lock); 2351 2352 return (error); 2353} 2354 2355/* 2356 * Clear the error counts associated with this vdev. Unlike vdev_online() and 2357 * vdev_offline(), we assume the spa config is locked. We also clear all 2358 * children. If 'vd' is NULL, then the user wants to clear all vdevs. 2359 */ 2360void 2361vdev_clear(spa_t *spa, vdev_t *vd) 2362{ 2363 vdev_t *rvd = spa->spa_root_vdev; 2364 2365 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2366 2367 if (vd == NULL) 2368 vd = rvd; 2369 2370 vd->vdev_stat.vs_read_errors = 0; 2371 vd->vdev_stat.vs_write_errors = 0; 2372 vd->vdev_stat.vs_checksum_errors = 0; 2373 2374 for (int c = 0; c < vd->vdev_children; c++) 2375 vdev_clear(spa, vd->vdev_child[c]); 2376 2377 /* 2378 * If we're in the FAULTED state or have experienced failed I/O, then 2379 * clear the persistent state and attempt to reopen the device. We 2380 * also mark the vdev config dirty, so that the new faulted state is 2381 * written out to disk. 2382 */ 2383 if (vd->vdev_faulted || vd->vdev_degraded || 2384 !vdev_readable(vd) || !vdev_writeable(vd)) { 2385 2386 /* 2387 * When reopening in reponse to a clear event, it may be due to 2388 * a fmadm repair request. In this case, if the device is 2389 * still broken, we want to still post the ereport again. 2390 */ 2391 vd->vdev_forcefault = B_TRUE; 2392 2393 vd->vdev_faulted = vd->vdev_degraded = 0ULL; 2394 vd->vdev_cant_read = B_FALSE; 2395 vd->vdev_cant_write = B_FALSE; 2396 2397 vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 2398 2399 vd->vdev_forcefault = B_FALSE; 2400 2401 if (vd != rvd && vdev_writeable(vd->vdev_top)) 2402 vdev_state_dirty(vd->vdev_top); 2403 2404 if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 2405 spa_async_request(spa, SPA_ASYNC_RESILVER); 2406 2407 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); 2408 } 2409 2410 /* 2411 * When clearing a FMA-diagnosed fault, we always want to 2412 * unspare the device, as we assume that the original spare was 2413 * done in response to the FMA fault. 2414 */ 2415 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 2416 vd->vdev_parent->vdev_ops == &vdev_spare_ops && 2417 vd->vdev_parent->vdev_child[0] == vd) 2418 vd->vdev_unspare = B_TRUE; 2419} 2420 2421boolean_t 2422vdev_is_dead(vdev_t *vd) 2423{ 2424 /* 2425 * Holes and missing devices are always considered "dead". 2426 * This simplifies the code since we don't have to check for 2427 * these types of devices in the various code paths. 2428 * Instead we rely on the fact that we skip over dead devices 2429 * before issuing I/O to them. 2430 */ 2431 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || 2432 vd->vdev_ops == &vdev_missing_ops); 2433} 2434 2435boolean_t 2436vdev_readable(vdev_t *vd) 2437{ 2438 return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 2439} 2440 2441boolean_t 2442vdev_writeable(vdev_t *vd) 2443{ 2444 return (!vdev_is_dead(vd) && !vd->vdev_cant_write); 2445} 2446 2447boolean_t 2448vdev_allocatable(vdev_t *vd) 2449{ 2450 uint64_t state = vd->vdev_state; 2451 2452 /* 2453 * We currently allow allocations from vdevs which may be in the 2454 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 2455 * fails to reopen then we'll catch it later when we're holding 2456 * the proper locks. Note that we have to get the vdev state 2457 * in a local variable because although it changes atomically, 2458 * we're asking two separate questions about it. 2459 */ 2460 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 2461 !vd->vdev_cant_write && !vd->vdev_ishole); 2462} 2463 2464boolean_t 2465vdev_accessible(vdev_t *vd, zio_t *zio) 2466{ 2467 ASSERT(zio->io_vd == vd); 2468 2469 if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 2470 return (B_FALSE); 2471 2472 if (zio->io_type == ZIO_TYPE_READ) 2473 return (!vd->vdev_cant_read); 2474 2475 if (zio->io_type == ZIO_TYPE_WRITE) 2476 return (!vd->vdev_cant_write); 2477 2478 return (B_TRUE); 2479} 2480 2481/* 2482 * Get statistics for the given vdev. 2483 */ 2484void 2485vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 2486{ 2487 vdev_t *rvd = vd->vdev_spa->spa_root_vdev; 2488 2489 mutex_enter(&vd->vdev_stat_lock); 2490 bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 2491 vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 2492 vs->vs_state = vd->vdev_state; 2493 vs->vs_rsize = vdev_get_min_asize(vd); 2494 if (vd->vdev_ops->vdev_op_leaf) 2495 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 2496 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize; 2497 mutex_exit(&vd->vdev_stat_lock); 2498 2499 /* 2500 * If we're getting stats on the root vdev, aggregate the I/O counts 2501 * over all top-level vdevs (i.e. the direct children of the root). 2502 */ 2503 if (vd == rvd) { 2504 for (int c = 0; c < rvd->vdev_children; c++) { 2505 vdev_t *cvd = rvd->vdev_child[c]; 2506 vdev_stat_t *cvs = &cvd->vdev_stat; 2507 2508 mutex_enter(&vd->vdev_stat_lock); 2509 for (int t = 0; t < ZIO_TYPES; t++) { 2510 vs->vs_ops[t] += cvs->vs_ops[t]; 2511 vs->vs_bytes[t] += cvs->vs_bytes[t]; 2512 } 2513 cvs->vs_scan_removing = cvd->vdev_removing; 2514 mutex_exit(&vd->vdev_stat_lock); 2515 } 2516 } 2517} 2518 2519void 2520vdev_clear_stats(vdev_t *vd) 2521{ 2522 mutex_enter(&vd->vdev_stat_lock); 2523 vd->vdev_stat.vs_space = 0; 2524 vd->vdev_stat.vs_dspace = 0; 2525 vd->vdev_stat.vs_alloc = 0; 2526 mutex_exit(&vd->vdev_stat_lock); 2527} 2528 2529void 2530vdev_scan_stat_init(vdev_t *vd) 2531{ 2532 vdev_stat_t *vs = &vd->vdev_stat; 2533 2534 for (int c = 0; c < vd->vdev_children; c++) 2535 vdev_scan_stat_init(vd->vdev_child[c]); 2536 2537 mutex_enter(&vd->vdev_stat_lock); 2538 vs->vs_scan_processed = 0; 2539 mutex_exit(&vd->vdev_stat_lock); 2540} 2541 2542void 2543vdev_stat_update(zio_t *zio, uint64_t psize) 2544{ 2545 spa_t *spa = zio->io_spa; 2546 vdev_t *rvd = spa->spa_root_vdev; 2547 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 2548 vdev_t *pvd; 2549 uint64_t txg = zio->io_txg; 2550 vdev_stat_t *vs = &vd->vdev_stat; 2551 zio_type_t type = zio->io_type; 2552 int flags = zio->io_flags; 2553 2554 /* 2555 * If this i/o is a gang leader, it didn't do any actual work. 2556 */ 2557 if (zio->io_gang_tree) 2558 return; 2559 2560 if (zio->io_error == 0) { 2561 /* 2562 * If this is a root i/o, don't count it -- we've already 2563 * counted the top-level vdevs, and vdev_get_stats() will 2564 * aggregate them when asked. This reduces contention on 2565 * the root vdev_stat_lock and implicitly handles blocks 2566 * that compress away to holes, for which there is no i/o. 2567 * (Holes never create vdev children, so all the counters 2568 * remain zero, which is what we want.) 2569 * 2570 * Note: this only applies to successful i/o (io_error == 0) 2571 * because unlike i/o counts, errors are not additive. 2572 * When reading a ditto block, for example, failure of 2573 * one top-level vdev does not imply a root-level error. 2574 */ 2575 if (vd == rvd) 2576 return; 2577 2578 ASSERT(vd == zio->io_vd); 2579 2580 if (flags & ZIO_FLAG_IO_BYPASS) 2581 return; 2582 2583 mutex_enter(&vd->vdev_stat_lock); 2584 2585 if (flags & ZIO_FLAG_IO_REPAIR) { 2586 if (flags & ZIO_FLAG_SCAN_THREAD) { 2587 dsl_scan_phys_t *scn_phys = 2588 &spa->spa_dsl_pool->dp_scan->scn_phys; 2589 uint64_t *processed = &scn_phys->scn_processed; 2590 2591 /* XXX cleanup? */ 2592 if (vd->vdev_ops->vdev_op_leaf) 2593 atomic_add_64(processed, psize); 2594 vs->vs_scan_processed += psize; 2595 } 2596 2597 if (flags & ZIO_FLAG_SELF_HEAL) 2598 vs->vs_self_healed += psize; 2599 } 2600 2601 vs->vs_ops[type]++; 2602 vs->vs_bytes[type] += psize; 2603 2604 mutex_exit(&vd->vdev_stat_lock); 2605 return; 2606 } 2607 2608 if (flags & ZIO_FLAG_SPECULATIVE) 2609 return; 2610 2611 /* 2612 * If this is an I/O error that is going to be retried, then ignore the 2613 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 2614 * hard errors, when in reality they can happen for any number of 2615 * innocuous reasons (bus resets, MPxIO link failure, etc). 2616 */ 2617 if (zio->io_error == EIO && 2618 !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 2619 return; 2620 2621 /* 2622 * Intent logs writes won't propagate their error to the root 2623 * I/O so don't mark these types of failures as pool-level 2624 * errors. 2625 */ 2626 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 2627 return; 2628 2629 mutex_enter(&vd->vdev_stat_lock); 2630 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 2631 if (zio->io_error == ECKSUM) 2632 vs->vs_checksum_errors++; 2633 else 2634 vs->vs_read_errors++; 2635 } 2636 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 2637 vs->vs_write_errors++; 2638 mutex_exit(&vd->vdev_stat_lock); 2639 2640 if (type == ZIO_TYPE_WRITE && txg != 0 && 2641 (!(flags & ZIO_FLAG_IO_REPAIR) || 2642 (flags & ZIO_FLAG_SCAN_THREAD) || 2643 spa->spa_claiming)) { 2644 /* 2645 * This is either a normal write (not a repair), or it's 2646 * a repair induced by the scrub thread, or it's a repair 2647 * made by zil_claim() during spa_load() in the first txg. 2648 * In the normal case, we commit the DTL change in the same 2649 * txg as the block was born. In the scrub-induced repair 2650 * case, we know that scrubs run in first-pass syncing context, 2651 * so we commit the DTL change in spa_syncing_txg(spa). 2652 * In the zil_claim() case, we commit in spa_first_txg(spa). 2653 * 2654 * We currently do not make DTL entries for failed spontaneous 2655 * self-healing writes triggered by normal (non-scrubbing) 2656 * reads, because we have no transactional context in which to 2657 * do so -- and it's not clear that it'd be desirable anyway. 2658 */ 2659 if (vd->vdev_ops->vdev_op_leaf) { 2660 uint64_t commit_txg = txg; 2661 if (flags & ZIO_FLAG_SCAN_THREAD) { 2662 ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2663 ASSERT(spa_sync_pass(spa) == 1); 2664 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 2665 commit_txg = spa_syncing_txg(spa); 2666 } else if (spa->spa_claiming) { 2667 ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2668 commit_txg = spa_first_txg(spa); 2669 } 2670 ASSERT(commit_txg >= spa_syncing_txg(spa)); 2671 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 2672 return; 2673 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2674 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 2675 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 2676 } 2677 if (vd != rvd) 2678 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 2679 } 2680} 2681 2682/* 2683 * Update the in-core space usage stats for this vdev, its metaslab class, 2684 * and the root vdev. 2685 */ 2686void 2687vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 2688 int64_t space_delta) 2689{ 2690 int64_t dspace_delta = space_delta; 2691 spa_t *spa = vd->vdev_spa; 2692 vdev_t *rvd = spa->spa_root_vdev; 2693 metaslab_group_t *mg = vd->vdev_mg; 2694 metaslab_class_t *mc = mg ? mg->mg_class : NULL; 2695 2696 ASSERT(vd == vd->vdev_top); 2697 2698 /* 2699 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 2700 * factor. We must calculate this here and not at the root vdev 2701 * because the root vdev's psize-to-asize is simply the max of its 2702 * childrens', thus not accurate enough for us. 2703 */ 2704 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 2705 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 2706 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 2707 vd->vdev_deflate_ratio; 2708 2709 mutex_enter(&vd->vdev_stat_lock); 2710 vd->vdev_stat.vs_alloc += alloc_delta; 2711 vd->vdev_stat.vs_space += space_delta; 2712 vd->vdev_stat.vs_dspace += dspace_delta; 2713 mutex_exit(&vd->vdev_stat_lock); 2714 2715 if (mc == spa_normal_class(spa)) { 2716 mutex_enter(&rvd->vdev_stat_lock); 2717 rvd->vdev_stat.vs_alloc += alloc_delta; 2718 rvd->vdev_stat.vs_space += space_delta; 2719 rvd->vdev_stat.vs_dspace += dspace_delta; 2720 mutex_exit(&rvd->vdev_stat_lock); 2721 } 2722 2723 if (mc != NULL) { 2724 ASSERT(rvd == vd->vdev_parent); 2725 ASSERT(vd->vdev_ms_count != 0); 2726 2727 metaslab_class_space_update(mc, 2728 alloc_delta, defer_delta, space_delta, dspace_delta); 2729 } 2730} 2731 2732/* 2733 * Mark a top-level vdev's config as dirty, placing it on the dirty list 2734 * so that it will be written out next time the vdev configuration is synced. 2735 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 2736 */ 2737void 2738vdev_config_dirty(vdev_t *vd) 2739{ 2740 spa_t *spa = vd->vdev_spa; 2741 vdev_t *rvd = spa->spa_root_vdev; 2742 int c; 2743 2744 ASSERT(spa_writeable(spa)); 2745 2746 /* 2747 * If this is an aux vdev (as with l2cache and spare devices), then we 2748 * update the vdev config manually and set the sync flag. 2749 */ 2750 if (vd->vdev_aux != NULL) { 2751 spa_aux_vdev_t *sav = vd->vdev_aux; 2752 nvlist_t **aux; 2753 uint_t naux; 2754 2755 for (c = 0; c < sav->sav_count; c++) { 2756 if (sav->sav_vdevs[c] == vd) 2757 break; 2758 } 2759 2760 if (c == sav->sav_count) { 2761 /* 2762 * We're being removed. There's nothing more to do. 2763 */ 2764 ASSERT(sav->sav_sync == B_TRUE); 2765 return; 2766 } 2767 2768 sav->sav_sync = B_TRUE; 2769 2770 if (nvlist_lookup_nvlist_array(sav->sav_config, 2771 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 2772 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 2773 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 2774 } 2775 2776 ASSERT(c < naux); 2777 2778 /* 2779 * Setting the nvlist in the middle if the array is a little 2780 * sketchy, but it will work. 2781 */ 2782 nvlist_free(aux[c]); 2783 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 2784 2785 return; 2786 } 2787 2788 /* 2789 * The dirty list is protected by the SCL_CONFIG lock. The caller 2790 * must either hold SCL_CONFIG as writer, or must be the sync thread 2791 * (which holds SCL_CONFIG as reader). There's only one sync thread, 2792 * so this is sufficient to ensure mutual exclusion. 2793 */ 2794 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2795 (dsl_pool_sync_context(spa_get_dsl(spa)) && 2796 spa_config_held(spa, SCL_CONFIG, RW_READER))); 2797 2798 if (vd == rvd) { 2799 for (c = 0; c < rvd->vdev_children; c++) 2800 vdev_config_dirty(rvd->vdev_child[c]); 2801 } else { 2802 ASSERT(vd == vd->vdev_top); 2803 2804 if (!list_link_active(&vd->vdev_config_dirty_node) && 2805 !vd->vdev_ishole) 2806 list_insert_head(&spa->spa_config_dirty_list, vd); 2807 } 2808} 2809 2810void 2811vdev_config_clean(vdev_t *vd) 2812{ 2813 spa_t *spa = vd->vdev_spa; 2814 2815 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 2816 (dsl_pool_sync_context(spa_get_dsl(spa)) && 2817 spa_config_held(spa, SCL_CONFIG, RW_READER))); 2818 2819 ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 2820 list_remove(&spa->spa_config_dirty_list, vd); 2821} 2822 2823/* 2824 * Mark a top-level vdev's state as dirty, so that the next pass of 2825 * spa_sync() can convert this into vdev_config_dirty(). We distinguish 2826 * the state changes from larger config changes because they require 2827 * much less locking, and are often needed for administrative actions. 2828 */ 2829void 2830vdev_state_dirty(vdev_t *vd) 2831{ 2832 spa_t *spa = vd->vdev_spa; 2833 2834 ASSERT(spa_writeable(spa)); 2835 ASSERT(vd == vd->vdev_top); 2836 2837 /* 2838 * The state list is protected by the SCL_STATE lock. The caller 2839 * must either hold SCL_STATE as writer, or must be the sync thread 2840 * (which holds SCL_STATE as reader). There's only one sync thread, 2841 * so this is sufficient to ensure mutual exclusion. 2842 */ 2843 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2844 (dsl_pool_sync_context(spa_get_dsl(spa)) && 2845 spa_config_held(spa, SCL_STATE, RW_READER))); 2846 2847 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) 2848 list_insert_head(&spa->spa_state_dirty_list, vd); 2849} 2850 2851void 2852vdev_state_clean(vdev_t *vd) 2853{ 2854 spa_t *spa = vd->vdev_spa; 2855 2856 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 2857 (dsl_pool_sync_context(spa_get_dsl(spa)) && 2858 spa_config_held(spa, SCL_STATE, RW_READER))); 2859 2860 ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 2861 list_remove(&spa->spa_state_dirty_list, vd); 2862} 2863 2864/* 2865 * Propagate vdev state up from children to parent. 2866 */ 2867void 2868vdev_propagate_state(vdev_t *vd) 2869{ 2870 spa_t *spa = vd->vdev_spa; 2871 vdev_t *rvd = spa->spa_root_vdev; 2872 int degraded = 0, faulted = 0; 2873 int corrupted = 0; 2874 vdev_t *child; 2875 2876 if (vd->vdev_children > 0) { 2877 for (int c = 0; c < vd->vdev_children; c++) { 2878 child = vd->vdev_child[c]; 2879 2880 /* 2881 * Don't factor holes into the decision. 2882 */ 2883 if (child->vdev_ishole) 2884 continue; 2885 2886 if (!vdev_readable(child) || 2887 (!vdev_writeable(child) && spa_writeable(spa))) { 2888 /* 2889 * Root special: if there is a top-level log 2890 * device, treat the root vdev as if it were 2891 * degraded. 2892 */ 2893 if (child->vdev_islog && vd == rvd) 2894 degraded++; 2895 else 2896 faulted++; 2897 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 2898 degraded++; 2899 } 2900 2901 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 2902 corrupted++; 2903 } 2904 2905 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 2906 2907 /* 2908 * Root special: if there is a top-level vdev that cannot be 2909 * opened due to corrupted metadata, then propagate the root 2910 * vdev's aux state as 'corrupt' rather than 'insufficient 2911 * replicas'. 2912 */ 2913 if (corrupted && vd == rvd && 2914 rvd->vdev_state == VDEV_STATE_CANT_OPEN) 2915 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 2916 VDEV_AUX_CORRUPT_DATA); 2917 } 2918 2919 if (vd->vdev_parent) 2920 vdev_propagate_state(vd->vdev_parent); 2921} 2922 2923/* 2924 * Set a vdev's state. If this is during an open, we don't update the parent 2925 * state, because we're in the process of opening children depth-first. 2926 * Otherwise, we propagate the change to the parent. 2927 * 2928 * If this routine places a device in a faulted state, an appropriate ereport is 2929 * generated. 2930 */ 2931void 2932vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 2933{ 2934 uint64_t save_state; 2935 spa_t *spa = vd->vdev_spa; 2936 2937 if (state == vd->vdev_state) { 2938 vd->vdev_stat.vs_aux = aux; 2939 return; 2940 } 2941 2942 save_state = vd->vdev_state; 2943 2944 vd->vdev_state = state; 2945 vd->vdev_stat.vs_aux = aux; 2946 2947 /* 2948 * If we are setting the vdev state to anything but an open state, then 2949 * always close the underlying device unless the device has requested 2950 * a delayed close (i.e. we're about to remove or fault the device). 2951 * Otherwise, we keep accessible but invalid devices open forever. 2952 * We don't call vdev_close() itself, because that implies some extra 2953 * checks (offline, etc) that we don't want here. This is limited to 2954 * leaf devices, because otherwise closing the device will affect other 2955 * children. 2956 */ 2957 if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 2958 vd->vdev_ops->vdev_op_leaf) 2959 vd->vdev_ops->vdev_op_close(vd); 2960 2961 /* 2962 * If we have brought this vdev back into service, we need 2963 * to notify fmd so that it can gracefully repair any outstanding 2964 * cases due to a missing device. We do this in all cases, even those 2965 * that probably don't correlate to a repaired fault. This is sure to 2966 * catch all cases, and we let the zfs-retire agent sort it out. If 2967 * this is a transient state it's OK, as the retire agent will 2968 * double-check the state of the vdev before repairing it. 2969 */ 2970 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 2971 vd->vdev_prevstate != state) 2972 zfs_post_state_change(spa, vd); 2973 2974 if (vd->vdev_removed && 2975 state == VDEV_STATE_CANT_OPEN && 2976 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 2977 /* 2978 * If the previous state is set to VDEV_STATE_REMOVED, then this 2979 * device was previously marked removed and someone attempted to 2980 * reopen it. If this failed due to a nonexistent device, then 2981 * keep the device in the REMOVED state. We also let this be if 2982 * it is one of our special test online cases, which is only 2983 * attempting to online the device and shouldn't generate an FMA 2984 * fault. 2985 */ 2986 vd->vdev_state = VDEV_STATE_REMOVED; 2987 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 2988 } else if (state == VDEV_STATE_REMOVED) { 2989 vd->vdev_removed = B_TRUE; 2990 } else if (state == VDEV_STATE_CANT_OPEN) { 2991 /* 2992 * If we fail to open a vdev during an import or recovery, we 2993 * mark it as "not available", which signifies that it was 2994 * never there to begin with. Failure to open such a device 2995 * is not considered an error. 2996 */ 2997 if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 2998 spa_load_state(spa) == SPA_LOAD_RECOVER) && 2999 vd->vdev_ops->vdev_op_leaf) 3000 vd->vdev_not_present = 1; 3001 3002 /* 3003 * Post the appropriate ereport. If the 'prevstate' field is 3004 * set to something other than VDEV_STATE_UNKNOWN, it indicates 3005 * that this is part of a vdev_reopen(). In this case, we don't 3006 * want to post the ereport if the device was already in the 3007 * CANT_OPEN state beforehand. 3008 * 3009 * If the 'checkremove' flag is set, then this is an attempt to 3010 * online the device in response to an insertion event. If we 3011 * hit this case, then we have detected an insertion event for a 3012 * faulted or offline device that wasn't in the removed state. 3013 * In this scenario, we don't post an ereport because we are 3014 * about to replace the device, or attempt an online with 3015 * vdev_forcefault, which will generate the fault for us. 3016 */ 3017 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 3018 !vd->vdev_not_present && !vd->vdev_checkremove && 3019 vd != spa->spa_root_vdev) { 3020 const char *class; 3021 3022 switch (aux) { 3023 case VDEV_AUX_OPEN_FAILED: 3024 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 3025 break; 3026 case VDEV_AUX_CORRUPT_DATA: 3027 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 3028 break; 3029 case VDEV_AUX_NO_REPLICAS: 3030 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 3031 break; 3032 case VDEV_AUX_BAD_GUID_SUM: 3033 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 3034 break; 3035 case VDEV_AUX_TOO_SMALL: 3036 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 3037 break; 3038 case VDEV_AUX_BAD_LABEL: 3039 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 3040 break; 3041 default: 3042 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 3043 } 3044 3045 zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 3046 } 3047 3048 /* Erase any notion of persistent removed state */ 3049 vd->vdev_removed = B_FALSE; 3050 } else { 3051 vd->vdev_removed = B_FALSE; 3052 } 3053 3054 if (!isopen && vd->vdev_parent) 3055 vdev_propagate_state(vd->vdev_parent); 3056} 3057 3058/* 3059 * Check the vdev configuration to ensure that it's capable of supporting 3060 * a root pool. 3061 * 3062 * On Solaris, we do not support RAID-Z or partial configuration. In 3063 * addition, only a single top-level vdev is allowed and none of the 3064 * leaves can be wholedisks. 3065 * 3066 * For FreeBSD, we can boot from any configuration. There is a 3067 * limitation that the boot filesystem must be either uncompressed or 3068 * compresses with lzjb compression but I'm not sure how to enforce 3069 * that here. 3070 */ 3071boolean_t 3072vdev_is_bootable(vdev_t *vd) 3073{ 3074#ifdef sun 3075 if (!vd->vdev_ops->vdev_op_leaf) { 3076 char *vdev_type = vd->vdev_ops->vdev_op_type; 3077 3078 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 3079 vd->vdev_children > 1) { 3080 return (B_FALSE); 3081 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || 3082 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 3083 return (B_FALSE); 3084 } 3085 } else if (vd->vdev_wholedisk == 1) { 3086 return (B_FALSE); 3087 } 3088 3089 for (int c = 0; c < vd->vdev_children; c++) { 3090 if (!vdev_is_bootable(vd->vdev_child[c])) 3091 return (B_FALSE); 3092 } 3093#endif /* sun */ 3094 return (B_TRUE); 3095} 3096 3097/* 3098 * Load the state from the original vdev tree (ovd) which 3099 * we've retrieved from the MOS config object. If the original 3100 * vdev was offline or faulted then we transfer that state to the 3101 * device in the current vdev tree (nvd). 3102 */ 3103void 3104vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) 3105{ 3106 spa_t *spa = nvd->vdev_spa; 3107 3108 ASSERT(nvd->vdev_top->vdev_islog); 3109 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 3110 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); 3111 3112 for (int c = 0; c < nvd->vdev_children; c++) 3113 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); 3114 3115 if (nvd->vdev_ops->vdev_op_leaf) { 3116 /* 3117 * Restore the persistent vdev state 3118 */ 3119 nvd->vdev_offline = ovd->vdev_offline; 3120 nvd->vdev_faulted = ovd->vdev_faulted; 3121 nvd->vdev_degraded = ovd->vdev_degraded; 3122 nvd->vdev_removed = ovd->vdev_removed; 3123 } 3124} 3125 3126/* 3127 * Determine if a log device has valid content. If the vdev was 3128 * removed or faulted in the MOS config then we know that 3129 * the content on the log device has already been written to the pool. 3130 */ 3131boolean_t 3132vdev_log_state_valid(vdev_t *vd) 3133{ 3134 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 3135 !vd->vdev_removed) 3136 return (B_TRUE); 3137 3138 for (int c = 0; c < vd->vdev_children; c++) 3139 if (vdev_log_state_valid(vd->vdev_child[c])) 3140 return (B_TRUE); 3141 3142 return (B_FALSE); 3143} 3144 3145/* 3146 * Expand a vdev if possible. 3147 */ 3148void 3149vdev_expand(vdev_t *vd, uint64_t txg) 3150{ 3151 ASSERT(vd->vdev_top == vd); 3152 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3153 3154 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { 3155 VERIFY(vdev_metaslab_init(vd, txg) == 0); 3156 vdev_config_dirty(vd); 3157 } 3158} 3159 3160/* 3161 * Split a vdev. 3162 */ 3163void 3164vdev_split(vdev_t *vd) 3165{ 3166 vdev_t *cvd, *pvd = vd->vdev_parent; 3167 3168 vdev_remove_child(pvd, vd); 3169 vdev_compact_children(pvd); 3170 3171 cvd = pvd->vdev_child[0]; 3172 if (pvd->vdev_children == 1) { 3173 vdev_remove_parent(cvd); 3174 cvd->vdev_splitting = B_TRUE; 3175 } 3176 vdev_propagate_state(cvd); 3177} 3178 3179void 3180vdev_deadman(vdev_t *vd) 3181{ 3182 for (int c = 0; c < vd->vdev_children; c++) { 3183 vdev_t *cvd = vd->vdev_child[c]; 3184 3185 vdev_deadman(cvd); 3186 } 3187 3188 if (vd->vdev_ops->vdev_op_leaf) { 3189 vdev_queue_t *vq = &vd->vdev_queue; 3190 3191 mutex_enter(&vq->vq_lock); 3192 if (avl_numnodes(&vq->vq_pending_tree) > 0) { 3193 spa_t *spa = vd->vdev_spa; 3194 zio_t *fio; 3195 uint64_t delta; 3196 3197 /* 3198 * Look at the head of all the pending queues, 3199 * if any I/O has been outstanding for longer than 3200 * the spa_deadman_synctime we panic the system. 3201 */ 3202 fio = avl_first(&vq->vq_pending_tree); 3203 delta = ddi_get_lbolt64() - fio->io_timestamp; 3204 if (delta > NSEC_TO_TICK(spa_deadman_synctime(spa))) { 3205 zfs_dbgmsg("SLOW IO: zio timestamp %llu, " 3206 "delta %llu, last io %llu", 3207 fio->io_timestamp, delta, 3208 vq->vq_io_complete_ts); 3209 fm_panic("I/O to pool '%s' appears to be " 3210 "hung on vdev guid %llu at '%s'.", 3211 spa_name(spa), 3212 (long long unsigned int) vd->vdev_guid, 3213 vd->vdev_path); 3214 } 3215 } 3216 mutex_exit(&vq->vq_lock); 3217 } 3218} 3219