vdev.c revision 271776
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) 2011, 2014 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/space_reftree.h> 41#include <sys/zio.h> 42#include <sys/zap.h> 43#include <sys/fs/zfs.h> 44#include <sys/arc.h> 45#include <sys/zil.h> 46#include <sys/dsl_scan.h> 47#include <sys/trim_map.h> 48 49SYSCTL_DECL(_vfs_zfs); 50SYSCTL_NODE(_vfs_zfs, OID_AUTO, vdev, CTLFLAG_RW, 0, "ZFS VDEV"); 51 52/* 53 * Virtual device management. 54 */ 55 56/* 57 * The limit for ZFS to automatically increase a top-level vdev's ashift 58 * from logical ashift to physical ashift. 59 * 60 * Example: one or more 512B emulation child vdevs 61 * child->vdev_ashift = 9 (512 bytes) 62 * child->vdev_physical_ashift = 12 (4096 bytes) 63 * zfs_max_auto_ashift = 11 (2048 bytes) 64 * zfs_min_auto_ashift = 9 (512 bytes) 65 * 66 * On pool creation or the addition of a new top-level vdev, ZFS will 67 * increase the ashift of the top-level vdev to 2048 as limited by 68 * zfs_max_auto_ashift. 69 * 70 * Example: one or more 512B emulation child vdevs 71 * child->vdev_ashift = 9 (512 bytes) 72 * child->vdev_physical_ashift = 12 (4096 bytes) 73 * zfs_max_auto_ashift = 13 (8192 bytes) 74 * zfs_min_auto_ashift = 9 (512 bytes) 75 * 76 * On pool creation or the addition of a new top-level vdev, ZFS will 77 * increase the ashift of the top-level vdev to 4096 to match the 78 * max vdev_physical_ashift. 79 * 80 * Example: one or more 512B emulation child vdevs 81 * child->vdev_ashift = 9 (512 bytes) 82 * child->vdev_physical_ashift = 9 (512 bytes) 83 * zfs_max_auto_ashift = 13 (8192 bytes) 84 * zfs_min_auto_ashift = 12 (4096 bytes) 85 * 86 * On pool creation or the addition of a new top-level vdev, ZFS will 87 * increase the ashift of the top-level vdev to 4096 to match the 88 * zfs_min_auto_ashift. 89 */ 90static uint64_t zfs_max_auto_ashift = SPA_MAXASHIFT; 91static uint64_t zfs_min_auto_ashift = SPA_MINASHIFT; 92 93static int 94sysctl_vfs_zfs_max_auto_ashift(SYSCTL_HANDLER_ARGS) 95{ 96 uint64_t val; 97 int err; 98 99 val = zfs_max_auto_ashift; 100 err = sysctl_handle_64(oidp, &val, 0, req); 101 if (err != 0 || req->newptr == NULL) 102 return (err); 103 104 if (val > SPA_MAXASHIFT || val < zfs_min_auto_ashift) 105 return (EINVAL); 106 107 zfs_max_auto_ashift = val; 108 109 return (0); 110} 111SYSCTL_PROC(_vfs_zfs, OID_AUTO, max_auto_ashift, 112 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t), 113 sysctl_vfs_zfs_max_auto_ashift, "QU", 114 "Max ashift used when optimising for logical -> physical sectors size on " 115 "new top-level vdevs."); 116 117static int 118sysctl_vfs_zfs_min_auto_ashift(SYSCTL_HANDLER_ARGS) 119{ 120 uint64_t val; 121 int err; 122 123 val = zfs_min_auto_ashift; 124 err = sysctl_handle_64(oidp, &val, 0, req); 125 if (err != 0 || req->newptr == NULL) 126 return (err); 127 128 if (val < SPA_MINASHIFT || val > zfs_max_auto_ashift) 129 return (EINVAL); 130 131 zfs_min_auto_ashift = val; 132 133 return (0); 134} 135SYSCTL_PROC(_vfs_zfs, OID_AUTO, min_auto_ashift, 136 CTLTYPE_U64 | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(uint64_t), 137 sysctl_vfs_zfs_min_auto_ashift, "QU", 138 "Min ashift used when creating new top-level vdevs."); 139 140static vdev_ops_t *vdev_ops_table[] = { 141 &vdev_root_ops, 142 &vdev_raidz_ops, 143 &vdev_mirror_ops, 144 &vdev_replacing_ops, 145 &vdev_spare_ops, 146#ifdef _KERNEL 147 &vdev_geom_ops, 148#else 149 &vdev_disk_ops, 150#endif 151 &vdev_file_ops, 152 &vdev_missing_ops, 153 &vdev_hole_ops, 154 NULL 155}; 156 157 158/* 159 * Given a vdev type, return the appropriate ops vector. 160 */ 161static vdev_ops_t * 162vdev_getops(const char *type) 163{ 164 vdev_ops_t *ops, **opspp; 165 166 for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 167 if (strcmp(ops->vdev_op_type, type) == 0) 168 break; 169 170 return (ops); 171} 172 173/* 174 * Default asize function: return the MAX of psize with the asize of 175 * all children. This is what's used by anything other than RAID-Z. 176 */ 177uint64_t 178vdev_default_asize(vdev_t *vd, uint64_t psize) 179{ 180 uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 181 uint64_t csize; 182 183 for (int c = 0; c < vd->vdev_children; c++) { 184 csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 185 asize = MAX(asize, csize); 186 } 187 188 return (asize); 189} 190 191/* 192 * Get the minimum allocatable size. We define the allocatable size as 193 * the vdev's asize rounded to the nearest metaslab. This allows us to 194 * replace or attach devices which don't have the same physical size but 195 * can still satisfy the same number of allocations. 196 */ 197uint64_t 198vdev_get_min_asize(vdev_t *vd) 199{ 200 vdev_t *pvd = vd->vdev_parent; 201 202 /* 203 * If our parent is NULL (inactive spare or cache) or is the root, 204 * just return our own asize. 205 */ 206 if (pvd == NULL) 207 return (vd->vdev_asize); 208 209 /* 210 * The top-level vdev just returns the allocatable size rounded 211 * to the nearest metaslab. 212 */ 213 if (vd == vd->vdev_top) 214 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 215 216 /* 217 * The allocatable space for a raidz vdev is N * sizeof(smallest child), 218 * so each child must provide at least 1/Nth of its asize. 219 */ 220 if (pvd->vdev_ops == &vdev_raidz_ops) 221 return (pvd->vdev_min_asize / pvd->vdev_children); 222 223 return (pvd->vdev_min_asize); 224} 225 226void 227vdev_set_min_asize(vdev_t *vd) 228{ 229 vd->vdev_min_asize = vdev_get_min_asize(vd); 230 231 for (int c = 0; c < vd->vdev_children; c++) 232 vdev_set_min_asize(vd->vdev_child[c]); 233} 234 235vdev_t * 236vdev_lookup_top(spa_t *spa, uint64_t vdev) 237{ 238 vdev_t *rvd = spa->spa_root_vdev; 239 240 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 241 242 if (vdev < rvd->vdev_children) { 243 ASSERT(rvd->vdev_child[vdev] != NULL); 244 return (rvd->vdev_child[vdev]); 245 } 246 247 return (NULL); 248} 249 250vdev_t * 251vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 252{ 253 vdev_t *mvd; 254 255 if (vd->vdev_guid == guid) 256 return (vd); 257 258 for (int c = 0; c < vd->vdev_children; c++) 259 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 260 NULL) 261 return (mvd); 262 263 return (NULL); 264} 265 266void 267vdev_add_child(vdev_t *pvd, vdev_t *cvd) 268{ 269 size_t oldsize, newsize; 270 uint64_t id = cvd->vdev_id; 271 vdev_t **newchild; 272 273 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 274 ASSERT(cvd->vdev_parent == NULL); 275 276 cvd->vdev_parent = pvd; 277 278 if (pvd == NULL) 279 return; 280 281 ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 282 283 oldsize = pvd->vdev_children * sizeof (vdev_t *); 284 pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 285 newsize = pvd->vdev_children * sizeof (vdev_t *); 286 287 newchild = kmem_zalloc(newsize, KM_SLEEP); 288 if (pvd->vdev_child != NULL) { 289 bcopy(pvd->vdev_child, newchild, oldsize); 290 kmem_free(pvd->vdev_child, oldsize); 291 } 292 293 pvd->vdev_child = newchild; 294 pvd->vdev_child[id] = cvd; 295 296 cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 297 ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 298 299 /* 300 * Walk up all ancestors to update guid sum. 301 */ 302 for (; pvd != NULL; pvd = pvd->vdev_parent) 303 pvd->vdev_guid_sum += cvd->vdev_guid_sum; 304} 305 306void 307vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 308{ 309 int c; 310 uint_t id = cvd->vdev_id; 311 312 ASSERT(cvd->vdev_parent == pvd); 313 314 if (pvd == NULL) 315 return; 316 317 ASSERT(id < pvd->vdev_children); 318 ASSERT(pvd->vdev_child[id] == cvd); 319 320 pvd->vdev_child[id] = NULL; 321 cvd->vdev_parent = NULL; 322 323 for (c = 0; c < pvd->vdev_children; c++) 324 if (pvd->vdev_child[c]) 325 break; 326 327 if (c == pvd->vdev_children) { 328 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 329 pvd->vdev_child = NULL; 330 pvd->vdev_children = 0; 331 } 332 333 /* 334 * Walk up all ancestors to update guid sum. 335 */ 336 for (; pvd != NULL; pvd = pvd->vdev_parent) 337 pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 338} 339 340/* 341 * Remove any holes in the child array. 342 */ 343void 344vdev_compact_children(vdev_t *pvd) 345{ 346 vdev_t **newchild, *cvd; 347 int oldc = pvd->vdev_children; 348 int newc; 349 350 ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 351 352 for (int c = newc = 0; c < oldc; c++) 353 if (pvd->vdev_child[c]) 354 newc++; 355 356 newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 357 358 for (int c = newc = 0; c < oldc; c++) { 359 if ((cvd = pvd->vdev_child[c]) != NULL) { 360 newchild[newc] = cvd; 361 cvd->vdev_id = newc++; 362 } 363 } 364 365 kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 366 pvd->vdev_child = newchild; 367 pvd->vdev_children = newc; 368} 369 370/* 371 * Allocate and minimally initialize a vdev_t. 372 */ 373vdev_t * 374vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 375{ 376 vdev_t *vd; 377 378 vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 379 380 if (spa->spa_root_vdev == NULL) { 381 ASSERT(ops == &vdev_root_ops); 382 spa->spa_root_vdev = vd; 383 spa->spa_load_guid = spa_generate_guid(NULL); 384 } 385 386 if (guid == 0 && ops != &vdev_hole_ops) { 387 if (spa->spa_root_vdev == vd) { 388 /* 389 * The root vdev's guid will also be the pool guid, 390 * which must be unique among all pools. 391 */ 392 guid = spa_generate_guid(NULL); 393 } else { 394 /* 395 * Any other vdev's guid must be unique within the pool. 396 */ 397 guid = spa_generate_guid(spa); 398 } 399 ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 400 } 401 402 vd->vdev_spa = spa; 403 vd->vdev_id = id; 404 vd->vdev_guid = guid; 405 vd->vdev_guid_sum = guid; 406 vd->vdev_ops = ops; 407 vd->vdev_state = VDEV_STATE_CLOSED; 408 vd->vdev_ishole = (ops == &vdev_hole_ops); 409 410 mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 411 mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 412 mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 413 for (int t = 0; t < DTL_TYPES; t++) { 414 vd->vdev_dtl[t] = range_tree_create(NULL, NULL, 415 &vd->vdev_dtl_lock); 416 } 417 txg_list_create(&vd->vdev_ms_list, 418 offsetof(struct metaslab, ms_txg_node)); 419 txg_list_create(&vd->vdev_dtl_list, 420 offsetof(struct vdev, vdev_dtl_node)); 421 vd->vdev_stat.vs_timestamp = gethrtime(); 422 vdev_queue_init(vd); 423 vdev_cache_init(vd); 424 425 return (vd); 426} 427 428/* 429 * Allocate a new vdev. The 'alloctype' is used to control whether we are 430 * creating a new vdev or loading an existing one - the behavior is slightly 431 * different for each case. 432 */ 433int 434vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 435 int alloctype) 436{ 437 vdev_ops_t *ops; 438 char *type; 439 uint64_t guid = 0, islog, nparity; 440 vdev_t *vd; 441 442 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 443 444 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 445 return (SET_ERROR(EINVAL)); 446 447 if ((ops = vdev_getops(type)) == NULL) 448 return (SET_ERROR(EINVAL)); 449 450 /* 451 * If this is a load, get the vdev guid from the nvlist. 452 * Otherwise, vdev_alloc_common() will generate one for us. 453 */ 454 if (alloctype == VDEV_ALLOC_LOAD) { 455 uint64_t label_id; 456 457 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 458 label_id != id) 459 return (SET_ERROR(EINVAL)); 460 461 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 462 return (SET_ERROR(EINVAL)); 463 } else if (alloctype == VDEV_ALLOC_SPARE) { 464 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 465 return (SET_ERROR(EINVAL)); 466 } else if (alloctype == VDEV_ALLOC_L2CACHE) { 467 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 468 return (SET_ERROR(EINVAL)); 469 } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 470 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 471 return (SET_ERROR(EINVAL)); 472 } 473 474 /* 475 * The first allocated vdev must be of type 'root'. 476 */ 477 if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 478 return (SET_ERROR(EINVAL)); 479 480 /* 481 * Determine whether we're a log vdev. 482 */ 483 islog = 0; 484 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 485 if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 486 return (SET_ERROR(ENOTSUP)); 487 488 if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 489 return (SET_ERROR(ENOTSUP)); 490 491 /* 492 * Set the nparity property for RAID-Z vdevs. 493 */ 494 nparity = -1ULL; 495 if (ops == &vdev_raidz_ops) { 496 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 497 &nparity) == 0) { 498 if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 499 return (SET_ERROR(EINVAL)); 500 /* 501 * Previous versions could only support 1 or 2 parity 502 * device. 503 */ 504 if (nparity > 1 && 505 spa_version(spa) < SPA_VERSION_RAIDZ2) 506 return (SET_ERROR(ENOTSUP)); 507 if (nparity > 2 && 508 spa_version(spa) < SPA_VERSION_RAIDZ3) 509 return (SET_ERROR(ENOTSUP)); 510 } else { 511 /* 512 * We require the parity to be specified for SPAs that 513 * support multiple parity levels. 514 */ 515 if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 516 return (SET_ERROR(EINVAL)); 517 /* 518 * Otherwise, we default to 1 parity device for RAID-Z. 519 */ 520 nparity = 1; 521 } 522 } else { 523 nparity = 0; 524 } 525 ASSERT(nparity != -1ULL); 526 527 vd = vdev_alloc_common(spa, id, guid, ops); 528 529 vd->vdev_islog = islog; 530 vd->vdev_nparity = nparity; 531 532 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 533 vd->vdev_path = spa_strdup(vd->vdev_path); 534 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 535 vd->vdev_devid = spa_strdup(vd->vdev_devid); 536 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 537 &vd->vdev_physpath) == 0) 538 vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 539 if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 540 vd->vdev_fru = spa_strdup(vd->vdev_fru); 541 542 /* 543 * Set the whole_disk property. If it's not specified, leave the value 544 * as -1. 545 */ 546 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 547 &vd->vdev_wholedisk) != 0) 548 vd->vdev_wholedisk = -1ULL; 549 550 /* 551 * Look for the 'not present' flag. This will only be set if the device 552 * was not present at the time of import. 553 */ 554 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 555 &vd->vdev_not_present); 556 557 /* 558 * Get the alignment requirement. 559 */ 560 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 561 562 /* 563 * Retrieve the vdev creation time. 564 */ 565 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 566 &vd->vdev_crtxg); 567 568 /* 569 * If we're a top-level vdev, try to load the allocation parameters. 570 */ 571 if (parent && !parent->vdev_parent && 572 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 573 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 574 &vd->vdev_ms_array); 575 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 576 &vd->vdev_ms_shift); 577 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 578 &vd->vdev_asize); 579 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 580 &vd->vdev_removing); 581 } 582 583 if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { 584 ASSERT(alloctype == VDEV_ALLOC_LOAD || 585 alloctype == VDEV_ALLOC_ADD || 586 alloctype == VDEV_ALLOC_SPLIT || 587 alloctype == VDEV_ALLOC_ROOTPOOL); 588 vd->vdev_mg = metaslab_group_create(islog ? 589 spa_log_class(spa) : spa_normal_class(spa), vd); 590 } 591 592 /* 593 * If we're a leaf vdev, try to load the DTL object and other state. 594 */ 595 if (vd->vdev_ops->vdev_op_leaf && 596 (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 597 alloctype == VDEV_ALLOC_ROOTPOOL)) { 598 if (alloctype == VDEV_ALLOC_LOAD) { 599 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 600 &vd->vdev_dtl_object); 601 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 602 &vd->vdev_unspare); 603 } 604 605 if (alloctype == VDEV_ALLOC_ROOTPOOL) { 606 uint64_t spare = 0; 607 608 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 609 &spare) == 0 && spare) 610 spa_spare_add(vd); 611 } 612 613 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 614 &vd->vdev_offline); 615 616 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 617 &vd->vdev_resilver_txg); 618 619 /* 620 * When importing a pool, we want to ignore the persistent fault 621 * state, as the diagnosis made on another system may not be 622 * valid in the current context. Local vdevs will 623 * remain in the faulted state. 624 */ 625 if (spa_load_state(spa) == SPA_LOAD_OPEN) { 626 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 627 &vd->vdev_faulted); 628 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 629 &vd->vdev_degraded); 630 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 631 &vd->vdev_removed); 632 633 if (vd->vdev_faulted || vd->vdev_degraded) { 634 char *aux; 635 636 vd->vdev_label_aux = 637 VDEV_AUX_ERR_EXCEEDED; 638 if (nvlist_lookup_string(nv, 639 ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 640 strcmp(aux, "external") == 0) 641 vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 642 } 643 } 644 } 645 646 /* 647 * Add ourselves to the parent's list of children. 648 */ 649 vdev_add_child(parent, vd); 650 651 *vdp = vd; 652 653 return (0); 654} 655 656void 657vdev_free(vdev_t *vd) 658{ 659 spa_t *spa = vd->vdev_spa; 660 661 /* 662 * vdev_free() implies closing the vdev first. This is simpler than 663 * trying to ensure complicated semantics for all callers. 664 */ 665 vdev_close(vd); 666 667 ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 668 ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 669 670 /* 671 * Free all children. 672 */ 673 for (int c = 0; c < vd->vdev_children; c++) 674 vdev_free(vd->vdev_child[c]); 675 676 ASSERT(vd->vdev_child == NULL); 677 ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 678 679 /* 680 * Discard allocation state. 681 */ 682 if (vd->vdev_mg != NULL) { 683 vdev_metaslab_fini(vd); 684 metaslab_group_destroy(vd->vdev_mg); 685 } 686 687 ASSERT0(vd->vdev_stat.vs_space); 688 ASSERT0(vd->vdev_stat.vs_dspace); 689 ASSERT0(vd->vdev_stat.vs_alloc); 690 691 /* 692 * Remove this vdev from its parent's child list. 693 */ 694 vdev_remove_child(vd->vdev_parent, vd); 695 696 ASSERT(vd->vdev_parent == NULL); 697 698 /* 699 * Clean up vdev structure. 700 */ 701 vdev_queue_fini(vd); 702 vdev_cache_fini(vd); 703 704 if (vd->vdev_path) 705 spa_strfree(vd->vdev_path); 706 if (vd->vdev_devid) 707 spa_strfree(vd->vdev_devid); 708 if (vd->vdev_physpath) 709 spa_strfree(vd->vdev_physpath); 710 if (vd->vdev_fru) 711 spa_strfree(vd->vdev_fru); 712 713 if (vd->vdev_isspare) 714 spa_spare_remove(vd); 715 if (vd->vdev_isl2cache) 716 spa_l2cache_remove(vd); 717 718 txg_list_destroy(&vd->vdev_ms_list); 719 txg_list_destroy(&vd->vdev_dtl_list); 720 721 mutex_enter(&vd->vdev_dtl_lock); 722 space_map_close(vd->vdev_dtl_sm); 723 for (int t = 0; t < DTL_TYPES; t++) { 724 range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); 725 range_tree_destroy(vd->vdev_dtl[t]); 726 } 727 mutex_exit(&vd->vdev_dtl_lock); 728 729 mutex_destroy(&vd->vdev_dtl_lock); 730 mutex_destroy(&vd->vdev_stat_lock); 731 mutex_destroy(&vd->vdev_probe_lock); 732 733 if (vd == spa->spa_root_vdev) 734 spa->spa_root_vdev = NULL; 735 736 kmem_free(vd, sizeof (vdev_t)); 737} 738 739/* 740 * Transfer top-level vdev state from svd to tvd. 741 */ 742static void 743vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 744{ 745 spa_t *spa = svd->vdev_spa; 746 metaslab_t *msp; 747 vdev_t *vd; 748 int t; 749 750 ASSERT(tvd == tvd->vdev_top); 751 752 tvd->vdev_ms_array = svd->vdev_ms_array; 753 tvd->vdev_ms_shift = svd->vdev_ms_shift; 754 tvd->vdev_ms_count = svd->vdev_ms_count; 755 756 svd->vdev_ms_array = 0; 757 svd->vdev_ms_shift = 0; 758 svd->vdev_ms_count = 0; 759 760 if (tvd->vdev_mg) 761 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 762 tvd->vdev_mg = svd->vdev_mg; 763 tvd->vdev_ms = svd->vdev_ms; 764 765 svd->vdev_mg = NULL; 766 svd->vdev_ms = NULL; 767 768 if (tvd->vdev_mg != NULL) 769 tvd->vdev_mg->mg_vd = tvd; 770 771 tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 772 tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 773 tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 774 775 svd->vdev_stat.vs_alloc = 0; 776 svd->vdev_stat.vs_space = 0; 777 svd->vdev_stat.vs_dspace = 0; 778 779 for (t = 0; t < TXG_SIZE; t++) { 780 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 781 (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 782 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 783 (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 784 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 785 (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 786 } 787 788 if (list_link_active(&svd->vdev_config_dirty_node)) { 789 vdev_config_clean(svd); 790 vdev_config_dirty(tvd); 791 } 792 793 if (list_link_active(&svd->vdev_state_dirty_node)) { 794 vdev_state_clean(svd); 795 vdev_state_dirty(tvd); 796 } 797 798 tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 799 svd->vdev_deflate_ratio = 0; 800 801 tvd->vdev_islog = svd->vdev_islog; 802 svd->vdev_islog = 0; 803} 804 805static void 806vdev_top_update(vdev_t *tvd, vdev_t *vd) 807{ 808 if (vd == NULL) 809 return; 810 811 vd->vdev_top = tvd; 812 813 for (int c = 0; c < vd->vdev_children; c++) 814 vdev_top_update(tvd, vd->vdev_child[c]); 815} 816 817/* 818 * Add a mirror/replacing vdev above an existing vdev. 819 */ 820vdev_t * 821vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 822{ 823 spa_t *spa = cvd->vdev_spa; 824 vdev_t *pvd = cvd->vdev_parent; 825 vdev_t *mvd; 826 827 ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 828 829 mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 830 831 mvd->vdev_asize = cvd->vdev_asize; 832 mvd->vdev_min_asize = cvd->vdev_min_asize; 833 mvd->vdev_max_asize = cvd->vdev_max_asize; 834 mvd->vdev_ashift = cvd->vdev_ashift; 835 mvd->vdev_logical_ashift = cvd->vdev_logical_ashift; 836 mvd->vdev_physical_ashift = cvd->vdev_physical_ashift; 837 mvd->vdev_state = cvd->vdev_state; 838 mvd->vdev_crtxg = cvd->vdev_crtxg; 839 840 vdev_remove_child(pvd, cvd); 841 vdev_add_child(pvd, mvd); 842 cvd->vdev_id = mvd->vdev_children; 843 vdev_add_child(mvd, cvd); 844 vdev_top_update(cvd->vdev_top, cvd->vdev_top); 845 846 if (mvd == mvd->vdev_top) 847 vdev_top_transfer(cvd, mvd); 848 849 return (mvd); 850} 851 852/* 853 * Remove a 1-way mirror/replacing vdev from the tree. 854 */ 855void 856vdev_remove_parent(vdev_t *cvd) 857{ 858 vdev_t *mvd = cvd->vdev_parent; 859 vdev_t *pvd = mvd->vdev_parent; 860 861 ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 862 863 ASSERT(mvd->vdev_children == 1); 864 ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 865 mvd->vdev_ops == &vdev_replacing_ops || 866 mvd->vdev_ops == &vdev_spare_ops); 867 cvd->vdev_ashift = mvd->vdev_ashift; 868 cvd->vdev_logical_ashift = mvd->vdev_logical_ashift; 869 cvd->vdev_physical_ashift = mvd->vdev_physical_ashift; 870 871 vdev_remove_child(mvd, cvd); 872 vdev_remove_child(pvd, mvd); 873 874 /* 875 * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 876 * Otherwise, we could have detached an offline device, and when we 877 * go to import the pool we'll think we have two top-level vdevs, 878 * instead of a different version of the same top-level vdev. 879 */ 880 if (mvd->vdev_top == mvd) { 881 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 882 cvd->vdev_orig_guid = cvd->vdev_guid; 883 cvd->vdev_guid += guid_delta; 884 cvd->vdev_guid_sum += guid_delta; 885 } 886 cvd->vdev_id = mvd->vdev_id; 887 vdev_add_child(pvd, cvd); 888 vdev_top_update(cvd->vdev_top, cvd->vdev_top); 889 890 if (cvd == cvd->vdev_top) 891 vdev_top_transfer(mvd, cvd); 892 893 ASSERT(mvd->vdev_children == 0); 894 vdev_free(mvd); 895} 896 897int 898vdev_metaslab_init(vdev_t *vd, uint64_t txg) 899{ 900 spa_t *spa = vd->vdev_spa; 901 objset_t *mos = spa->spa_meta_objset; 902 uint64_t m; 903 uint64_t oldc = vd->vdev_ms_count; 904 uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 905 metaslab_t **mspp; 906 int error; 907 908 ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 909 910 /* 911 * This vdev is not being allocated from yet or is a hole. 912 */ 913 if (vd->vdev_ms_shift == 0) 914 return (0); 915 916 ASSERT(!vd->vdev_ishole); 917 918 /* 919 * Compute the raidz-deflation ratio. Note, we hard-code 920 * in 128k (1 << 17) because it is the current "typical" blocksize. 921 * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change, 922 * or we will inconsistently account for existing bp's. 923 */ 924 vd->vdev_deflate_ratio = (1 << 17) / 925 (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 926 927 ASSERT(oldc <= newc); 928 929 mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 930 931 if (oldc != 0) { 932 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 933 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 934 } 935 936 vd->vdev_ms = mspp; 937 vd->vdev_ms_count = newc; 938 939 for (m = oldc; m < newc; m++) { 940 uint64_t object = 0; 941 942 if (txg == 0) { 943 error = dmu_read(mos, vd->vdev_ms_array, 944 m * sizeof (uint64_t), sizeof (uint64_t), &object, 945 DMU_READ_PREFETCH); 946 if (error) 947 return (error); 948 } 949 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, m, object, txg); 950 } 951 952 if (txg == 0) 953 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 954 955 /* 956 * If the vdev is being removed we don't activate 957 * the metaslabs since we want to ensure that no new 958 * allocations are performed on this device. 959 */ 960 if (oldc == 0 && !vd->vdev_removing) 961 metaslab_group_activate(vd->vdev_mg); 962 963 if (txg == 0) 964 spa_config_exit(spa, SCL_ALLOC, FTAG); 965 966 return (0); 967} 968 969void 970vdev_metaslab_fini(vdev_t *vd) 971{ 972 uint64_t m; 973 uint64_t count = vd->vdev_ms_count; 974 975 if (vd->vdev_ms != NULL) { 976 metaslab_group_passivate(vd->vdev_mg); 977 for (m = 0; m < count; m++) { 978 metaslab_t *msp = vd->vdev_ms[m]; 979 980 if (msp != NULL) 981 metaslab_fini(msp); 982 } 983 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 984 vd->vdev_ms = NULL; 985 } 986} 987 988typedef struct vdev_probe_stats { 989 boolean_t vps_readable; 990 boolean_t vps_writeable; 991 int vps_flags; 992} vdev_probe_stats_t; 993 994static void 995vdev_probe_done(zio_t *zio) 996{ 997 spa_t *spa = zio->io_spa; 998 vdev_t *vd = zio->io_vd; 999 vdev_probe_stats_t *vps = zio->io_private; 1000 1001 ASSERT(vd->vdev_probe_zio != NULL); 1002 1003 if (zio->io_type == ZIO_TYPE_READ) { 1004 if (zio->io_error == 0) 1005 vps->vps_readable = 1; 1006 if (zio->io_error == 0 && spa_writeable(spa)) { 1007 zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 1008 zio->io_offset, zio->io_size, zio->io_data, 1009 ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1010 ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 1011 } else { 1012 zio_buf_free(zio->io_data, zio->io_size); 1013 } 1014 } else if (zio->io_type == ZIO_TYPE_WRITE) { 1015 if (zio->io_error == 0) 1016 vps->vps_writeable = 1; 1017 zio_buf_free(zio->io_data, zio->io_size); 1018 } else if (zio->io_type == ZIO_TYPE_NULL) { 1019 zio_t *pio; 1020 1021 vd->vdev_cant_read |= !vps->vps_readable; 1022 vd->vdev_cant_write |= !vps->vps_writeable; 1023 1024 if (vdev_readable(vd) && 1025 (vdev_writeable(vd) || !spa_writeable(spa))) { 1026 zio->io_error = 0; 1027 } else { 1028 ASSERT(zio->io_error != 0); 1029 zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 1030 spa, vd, NULL, 0, 0); 1031 zio->io_error = SET_ERROR(ENXIO); 1032 } 1033 1034 mutex_enter(&vd->vdev_probe_lock); 1035 ASSERT(vd->vdev_probe_zio == zio); 1036 vd->vdev_probe_zio = NULL; 1037 mutex_exit(&vd->vdev_probe_lock); 1038 1039 while ((pio = zio_walk_parents(zio)) != NULL) 1040 if (!vdev_accessible(vd, pio)) 1041 pio->io_error = SET_ERROR(ENXIO); 1042 1043 kmem_free(vps, sizeof (*vps)); 1044 } 1045} 1046 1047/* 1048 * Determine whether this device is accessible. 1049 * 1050 * Read and write to several known locations: the pad regions of each 1051 * vdev label but the first, which we leave alone in case it contains 1052 * a VTOC. 1053 */ 1054zio_t * 1055vdev_probe(vdev_t *vd, zio_t *zio) 1056{ 1057 spa_t *spa = vd->vdev_spa; 1058 vdev_probe_stats_t *vps = NULL; 1059 zio_t *pio; 1060 1061 ASSERT(vd->vdev_ops->vdev_op_leaf); 1062 1063 /* 1064 * Don't probe the probe. 1065 */ 1066 if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 1067 return (NULL); 1068 1069 /* 1070 * To prevent 'probe storms' when a device fails, we create 1071 * just one probe i/o at a time. All zios that want to probe 1072 * this vdev will become parents of the probe io. 1073 */ 1074 mutex_enter(&vd->vdev_probe_lock); 1075 1076 if ((pio = vd->vdev_probe_zio) == NULL) { 1077 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 1078 1079 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 1080 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 1081 ZIO_FLAG_TRYHARD; 1082 1083 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 1084 /* 1085 * vdev_cant_read and vdev_cant_write can only 1086 * transition from TRUE to FALSE when we have the 1087 * SCL_ZIO lock as writer; otherwise they can only 1088 * transition from FALSE to TRUE. This ensures that 1089 * any zio looking at these values can assume that 1090 * failures persist for the life of the I/O. That's 1091 * important because when a device has intermittent 1092 * connectivity problems, we want to ensure that 1093 * they're ascribed to the device (ENXIO) and not 1094 * the zio (EIO). 1095 * 1096 * Since we hold SCL_ZIO as writer here, clear both 1097 * values so the probe can reevaluate from first 1098 * principles. 1099 */ 1100 vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1101 vd->vdev_cant_read = B_FALSE; 1102 vd->vdev_cant_write = B_FALSE; 1103 } 1104 1105 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1106 vdev_probe_done, vps, 1107 vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1108 1109 /* 1110 * We can't change the vdev state in this context, so we 1111 * kick off an async task to do it on our behalf. 1112 */ 1113 if (zio != NULL) { 1114 vd->vdev_probe_wanted = B_TRUE; 1115 spa_async_request(spa, SPA_ASYNC_PROBE); 1116 } 1117 } 1118 1119 if (zio != NULL) 1120 zio_add_child(zio, pio); 1121 1122 mutex_exit(&vd->vdev_probe_lock); 1123 1124 if (vps == NULL) { 1125 ASSERT(zio != NULL); 1126 return (NULL); 1127 } 1128 1129 for (int l = 1; l < VDEV_LABELS; l++) { 1130 zio_nowait(zio_read_phys(pio, vd, 1131 vdev_label_offset(vd->vdev_psize, l, 1132 offsetof(vdev_label_t, vl_pad2)), 1133 VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE), 1134 ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1135 ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1136 } 1137 1138 if (zio == NULL) 1139 return (pio); 1140 1141 zio_nowait(pio); 1142 return (NULL); 1143} 1144 1145static void 1146vdev_open_child(void *arg) 1147{ 1148 vdev_t *vd = arg; 1149 1150 vd->vdev_open_thread = curthread; 1151 vd->vdev_open_error = vdev_open(vd); 1152 vd->vdev_open_thread = NULL; 1153} 1154 1155boolean_t 1156vdev_uses_zvols(vdev_t *vd) 1157{ 1158 if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1159 strlen(ZVOL_DIR)) == 0) 1160 return (B_TRUE); 1161 for (int c = 0; c < vd->vdev_children; c++) 1162 if (vdev_uses_zvols(vd->vdev_child[c])) 1163 return (B_TRUE); 1164 return (B_FALSE); 1165} 1166 1167void 1168vdev_open_children(vdev_t *vd) 1169{ 1170 taskq_t *tq; 1171 int children = vd->vdev_children; 1172 1173 /* 1174 * in order to handle pools on top of zvols, do the opens 1175 * in a single thread so that the same thread holds the 1176 * spa_namespace_lock 1177 */ 1178 if (B_TRUE || vdev_uses_zvols(vd)) { 1179 for (int c = 0; c < children; c++) 1180 vd->vdev_child[c]->vdev_open_error = 1181 vdev_open(vd->vdev_child[c]); 1182 return; 1183 } 1184 tq = taskq_create("vdev_open", children, minclsyspri, 1185 children, children, TASKQ_PREPOPULATE); 1186 1187 for (int c = 0; c < children; c++) 1188 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], 1189 TQ_SLEEP) != 0); 1190 1191 taskq_destroy(tq); 1192} 1193 1194/* 1195 * Prepare a virtual device for access. 1196 */ 1197int 1198vdev_open(vdev_t *vd) 1199{ 1200 spa_t *spa = vd->vdev_spa; 1201 int error; 1202 uint64_t osize = 0; 1203 uint64_t max_osize = 0; 1204 uint64_t asize, max_asize, psize; 1205 uint64_t logical_ashift = 0; 1206 uint64_t physical_ashift = 0; 1207 1208 ASSERT(vd->vdev_open_thread == curthread || 1209 spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1210 ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1211 vd->vdev_state == VDEV_STATE_CANT_OPEN || 1212 vd->vdev_state == VDEV_STATE_OFFLINE); 1213 1214 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1215 vd->vdev_cant_read = B_FALSE; 1216 vd->vdev_cant_write = B_FALSE; 1217 vd->vdev_min_asize = vdev_get_min_asize(vd); 1218 1219 /* 1220 * If this vdev is not removed, check its fault status. If it's 1221 * faulted, bail out of the open. 1222 */ 1223 if (!vd->vdev_removed && vd->vdev_faulted) { 1224 ASSERT(vd->vdev_children == 0); 1225 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1226 vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1227 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1228 vd->vdev_label_aux); 1229 return (SET_ERROR(ENXIO)); 1230 } else if (vd->vdev_offline) { 1231 ASSERT(vd->vdev_children == 0); 1232 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1233 return (SET_ERROR(ENXIO)); 1234 } 1235 1236 error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, 1237 &logical_ashift, &physical_ashift); 1238 1239 /* 1240 * Reset the vdev_reopening flag so that we actually close 1241 * the vdev on error. 1242 */ 1243 vd->vdev_reopening = B_FALSE; 1244 if (zio_injection_enabled && error == 0) 1245 error = zio_handle_device_injection(vd, NULL, ENXIO); 1246 1247 if (error) { 1248 if (vd->vdev_removed && 1249 vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 1250 vd->vdev_removed = B_FALSE; 1251 1252 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1253 vd->vdev_stat.vs_aux); 1254 return (error); 1255 } 1256 1257 vd->vdev_removed = B_FALSE; 1258 1259 /* 1260 * Recheck the faulted flag now that we have confirmed that 1261 * the vdev is accessible. If we're faulted, bail. 1262 */ 1263 if (vd->vdev_faulted) { 1264 ASSERT(vd->vdev_children == 0); 1265 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1266 vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1267 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1268 vd->vdev_label_aux); 1269 return (SET_ERROR(ENXIO)); 1270 } 1271 1272 if (vd->vdev_degraded) { 1273 ASSERT(vd->vdev_children == 0); 1274 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1275 VDEV_AUX_ERR_EXCEEDED); 1276 } else { 1277 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 1278 } 1279 1280 /* 1281 * For hole or missing vdevs we just return success. 1282 */ 1283 if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 1284 return (0); 1285 1286 if (vd->vdev_ops->vdev_op_leaf) { 1287 vd->vdev_notrim = B_FALSE; 1288 trim_map_create(vd); 1289 } 1290 1291 for (int c = 0; c < vd->vdev_children; c++) { 1292 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1293 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1294 VDEV_AUX_NONE); 1295 break; 1296 } 1297 } 1298 1299 osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 1300 max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1301 1302 if (vd->vdev_children == 0) { 1303 if (osize < SPA_MINDEVSIZE) { 1304 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1305 VDEV_AUX_TOO_SMALL); 1306 return (SET_ERROR(EOVERFLOW)); 1307 } 1308 psize = osize; 1309 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 1310 max_asize = max_osize - (VDEV_LABEL_START_SIZE + 1311 VDEV_LABEL_END_SIZE); 1312 } else { 1313 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1314 (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1315 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1316 VDEV_AUX_TOO_SMALL); 1317 return (SET_ERROR(EOVERFLOW)); 1318 } 1319 psize = 0; 1320 asize = osize; 1321 max_asize = max_osize; 1322 } 1323 1324 vd->vdev_psize = psize; 1325 1326 /* 1327 * Make sure the allocatable size hasn't shrunk. 1328 */ 1329 if (asize < vd->vdev_min_asize) { 1330 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1331 VDEV_AUX_BAD_LABEL); 1332 return (SET_ERROR(EINVAL)); 1333 } 1334 1335 vd->vdev_physical_ashift = 1336 MAX(physical_ashift, vd->vdev_physical_ashift); 1337 vd->vdev_logical_ashift = MAX(logical_ashift, vd->vdev_logical_ashift); 1338 vd->vdev_ashift = MAX(vd->vdev_logical_ashift, vd->vdev_ashift); 1339 1340 if (vd->vdev_logical_ashift > SPA_MAXASHIFT) { 1341 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1342 VDEV_AUX_ASHIFT_TOO_BIG); 1343 return (EINVAL); 1344 } 1345 1346 if (vd->vdev_asize == 0) { 1347 /* 1348 * This is the first-ever open, so use the computed values. 1349 * For testing purposes, a higher ashift can be requested. 1350 */ 1351 vd->vdev_asize = asize; 1352 vd->vdev_max_asize = max_asize; 1353 } else { 1354 /* 1355 * Make sure the alignment requirement hasn't increased. 1356 */ 1357 if (vd->vdev_ashift > vd->vdev_top->vdev_ashift && 1358 vd->vdev_ops->vdev_op_leaf) { 1359 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1360 VDEV_AUX_BAD_LABEL); 1361 return (EINVAL); 1362 } 1363 vd->vdev_max_asize = max_asize; 1364 } 1365 1366 /* 1367 * If all children are healthy and the asize has increased, 1368 * then we've experienced dynamic LUN growth. If automatic 1369 * expansion is enabled then use the additional space. 1370 */ 1371 if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize && 1372 (vd->vdev_expanding || spa->spa_autoexpand)) 1373 vd->vdev_asize = asize; 1374 1375 vdev_set_min_asize(vd); 1376 1377 /* 1378 * Ensure we can issue some IO before declaring the 1379 * vdev open for business. 1380 */ 1381 if (vd->vdev_ops->vdev_op_leaf && 1382 (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 1383 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1384 VDEV_AUX_ERR_EXCEEDED); 1385 return (error); 1386 } 1387 1388 /* 1389 * If a leaf vdev has a DTL, and seems healthy, then kick off a 1390 * resilver. But don't do this if we are doing a reopen for a scrub, 1391 * since this would just restart the scrub we are already doing. 1392 */ 1393 if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 1394 vdev_resilver_needed(vd, NULL, NULL)) 1395 spa_async_request(spa, SPA_ASYNC_RESILVER); 1396 1397 return (0); 1398} 1399 1400/* 1401 * Called once the vdevs are all opened, this routine validates the label 1402 * contents. This needs to be done before vdev_load() so that we don't 1403 * inadvertently do repair I/Os to the wrong device. 1404 * 1405 * If 'strict' is false ignore the spa guid check. This is necessary because 1406 * if the machine crashed during a re-guid the new guid might have been written 1407 * to all of the vdev labels, but not the cached config. The strict check 1408 * will be performed when the pool is opened again using the mos config. 1409 * 1410 * This function will only return failure if one of the vdevs indicates that it 1411 * has since been destroyed or exported. This is only possible if 1412 * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1413 * will be updated but the function will return 0. 1414 */ 1415int 1416vdev_validate(vdev_t *vd, boolean_t strict) 1417{ 1418 spa_t *spa = vd->vdev_spa; 1419 nvlist_t *label; 1420 uint64_t guid = 0, top_guid; 1421 uint64_t state; 1422 1423 for (int c = 0; c < vd->vdev_children; c++) 1424 if (vdev_validate(vd->vdev_child[c], strict) != 0) 1425 return (SET_ERROR(EBADF)); 1426 1427 /* 1428 * If the device has already failed, or was marked offline, don't do 1429 * any further validation. Otherwise, label I/O will fail and we will 1430 * overwrite the previous state. 1431 */ 1432 if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) { 1433 uint64_t aux_guid = 0; 1434 nvlist_t *nvl; 1435 uint64_t txg = spa_last_synced_txg(spa) != 0 ? 1436 spa_last_synced_txg(spa) : -1ULL; 1437 1438 if ((label = vdev_label_read_config(vd, txg)) == NULL) { 1439 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1440 VDEV_AUX_BAD_LABEL); 1441 return (0); 1442 } 1443 1444 /* 1445 * Determine if this vdev has been split off into another 1446 * pool. If so, then refuse to open it. 1447 */ 1448 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 1449 &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 1450 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1451 VDEV_AUX_SPLIT_POOL); 1452 nvlist_free(label); 1453 return (0); 1454 } 1455 1456 if (strict && (nvlist_lookup_uint64(label, 1457 ZPOOL_CONFIG_POOL_GUID, &guid) != 0 || 1458 guid != spa_guid(spa))) { 1459 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1460 VDEV_AUX_CORRUPT_DATA); 1461 nvlist_free(label); 1462 return (0); 1463 } 1464 1465 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 1466 != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 1467 &aux_guid) != 0) 1468 aux_guid = 0; 1469 1470 /* 1471 * If this vdev just became a top-level vdev because its 1472 * sibling was detached, it will have adopted the parent's 1473 * vdev guid -- but the label may or may not be on disk yet. 1474 * Fortunately, either version of the label will have the 1475 * same top guid, so if we're a top-level vdev, we can 1476 * safely compare to that instead. 1477 * 1478 * If we split this vdev off instead, then we also check the 1479 * original pool's guid. We don't want to consider the vdev 1480 * corrupt if it is partway through a split operation. 1481 */ 1482 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, 1483 &guid) != 0 || 1484 nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, 1485 &top_guid) != 0 || 1486 ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) && 1487 (vd->vdev_guid != top_guid || vd != vd->vdev_top))) { 1488 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1489 VDEV_AUX_CORRUPT_DATA); 1490 nvlist_free(label); 1491 return (0); 1492 } 1493 1494 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 1495 &state) != 0) { 1496 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1497 VDEV_AUX_CORRUPT_DATA); 1498 nvlist_free(label); 1499 return (0); 1500 } 1501 1502 nvlist_free(label); 1503 1504 /* 1505 * If this is a verbatim import, no need to check the 1506 * state of the pool. 1507 */ 1508 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 1509 spa_load_state(spa) == SPA_LOAD_OPEN && 1510 state != POOL_STATE_ACTIVE) 1511 return (SET_ERROR(EBADF)); 1512 1513 /* 1514 * If we were able to open and validate a vdev that was 1515 * previously marked permanently unavailable, clear that state 1516 * now. 1517 */ 1518 if (vd->vdev_not_present) 1519 vd->vdev_not_present = 0; 1520 } 1521 1522 return (0); 1523} 1524 1525/* 1526 * Close a virtual device. 1527 */ 1528void 1529vdev_close(vdev_t *vd) 1530{ 1531 spa_t *spa = vd->vdev_spa; 1532 vdev_t *pvd = vd->vdev_parent; 1533 1534 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1535 1536 /* 1537 * If our parent is reopening, then we are as well, unless we are 1538 * going offline. 1539 */ 1540 if (pvd != NULL && pvd->vdev_reopening) 1541 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 1542 1543 vd->vdev_ops->vdev_op_close(vd); 1544 1545 vdev_cache_purge(vd); 1546 1547 if (vd->vdev_ops->vdev_op_leaf) 1548 trim_map_destroy(vd); 1549 1550 /* 1551 * We record the previous state before we close it, so that if we are 1552 * doing a reopen(), we don't generate FMA ereports if we notice that 1553 * it's still faulted. 1554 */ 1555 vd->vdev_prevstate = vd->vdev_state; 1556 1557 if (vd->vdev_offline) 1558 vd->vdev_state = VDEV_STATE_OFFLINE; 1559 else 1560 vd->vdev_state = VDEV_STATE_CLOSED; 1561 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1562} 1563 1564void 1565vdev_hold(vdev_t *vd) 1566{ 1567 spa_t *spa = vd->vdev_spa; 1568 1569 ASSERT(spa_is_root(spa)); 1570 if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1571 return; 1572 1573 for (int c = 0; c < vd->vdev_children; c++) 1574 vdev_hold(vd->vdev_child[c]); 1575 1576 if (vd->vdev_ops->vdev_op_leaf) 1577 vd->vdev_ops->vdev_op_hold(vd); 1578} 1579 1580void 1581vdev_rele(vdev_t *vd) 1582{ 1583 spa_t *spa = vd->vdev_spa; 1584 1585 ASSERT(spa_is_root(spa)); 1586 for (int c = 0; c < vd->vdev_children; c++) 1587 vdev_rele(vd->vdev_child[c]); 1588 1589 if (vd->vdev_ops->vdev_op_leaf) 1590 vd->vdev_ops->vdev_op_rele(vd); 1591} 1592 1593/* 1594 * Reopen all interior vdevs and any unopened leaves. We don't actually 1595 * reopen leaf vdevs which had previously been opened as they might deadlock 1596 * on the spa_config_lock. Instead we only obtain the leaf's physical size. 1597 * If the leaf has never been opened then open it, as usual. 1598 */ 1599void 1600vdev_reopen(vdev_t *vd) 1601{ 1602 spa_t *spa = vd->vdev_spa; 1603 1604 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1605 1606 /* set the reopening flag unless we're taking the vdev offline */ 1607 vd->vdev_reopening = !vd->vdev_offline; 1608 vdev_close(vd); 1609 (void) vdev_open(vd); 1610 1611 /* 1612 * Call vdev_validate() here to make sure we have the same device. 1613 * Otherwise, a device with an invalid label could be successfully 1614 * opened in response to vdev_reopen(). 1615 */ 1616 if (vd->vdev_aux) { 1617 (void) vdev_validate_aux(vd); 1618 if (vdev_readable(vd) && vdev_writeable(vd) && 1619 vd->vdev_aux == &spa->spa_l2cache && 1620 !l2arc_vdev_present(vd)) 1621 l2arc_add_vdev(spa, vd); 1622 } else { 1623 (void) vdev_validate(vd, B_TRUE); 1624 } 1625 1626 /* 1627 * Reassess parent vdev's health. 1628 */ 1629 vdev_propagate_state(vd); 1630} 1631 1632int 1633vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 1634{ 1635 int error; 1636 1637 /* 1638 * Normally, partial opens (e.g. of a mirror) are allowed. 1639 * For a create, however, we want to fail the request if 1640 * there are any components we can't open. 1641 */ 1642 error = vdev_open(vd); 1643 1644 if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 1645 vdev_close(vd); 1646 return (error ? error : ENXIO); 1647 } 1648 1649 /* 1650 * Recursively load DTLs and initialize all labels. 1651 */ 1652 if ((error = vdev_dtl_load(vd)) != 0 || 1653 (error = vdev_label_init(vd, txg, isreplacing ? 1654 VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 1655 vdev_close(vd); 1656 return (error); 1657 } 1658 1659 return (0); 1660} 1661 1662void 1663vdev_metaslab_set_size(vdev_t *vd) 1664{ 1665 /* 1666 * Aim for roughly 200 metaslabs per vdev. 1667 */ 1668 vd->vdev_ms_shift = highbit64(vd->vdev_asize / 200); 1669 vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT); 1670} 1671 1672/* 1673 * Maximize performance by inflating the configured ashift for top level 1674 * vdevs to be as close to the physical ashift as possible while maintaining 1675 * administrator defined limits and ensuring it doesn't go below the 1676 * logical ashift. 1677 */ 1678void 1679vdev_ashift_optimize(vdev_t *vd) 1680{ 1681 if (vd == vd->vdev_top) { 1682 if (vd->vdev_ashift < vd->vdev_physical_ashift) { 1683 vd->vdev_ashift = MIN( 1684 MAX(zfs_max_auto_ashift, vd->vdev_ashift), 1685 MAX(zfs_min_auto_ashift, vd->vdev_physical_ashift)); 1686 } else { 1687 /* 1688 * Unusual case where logical ashift > physical ashift 1689 * so we can't cap the calculated ashift based on max 1690 * ashift as that would cause failures. 1691 * We still check if we need to increase it to match 1692 * the min ashift. 1693 */ 1694 vd->vdev_ashift = MAX(zfs_min_auto_ashift, 1695 vd->vdev_ashift); 1696 } 1697 } 1698} 1699 1700void 1701vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 1702{ 1703 ASSERT(vd == vd->vdev_top); 1704 ASSERT(!vd->vdev_ishole); 1705 ASSERT(ISP2(flags)); 1706 ASSERT(spa_writeable(vd->vdev_spa)); 1707 1708 if (flags & VDD_METASLAB) 1709 (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 1710 1711 if (flags & VDD_DTL) 1712 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 1713 1714 (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 1715} 1716 1717void 1718vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) 1719{ 1720 for (int c = 0; c < vd->vdev_children; c++) 1721 vdev_dirty_leaves(vd->vdev_child[c], flags, txg); 1722 1723 if (vd->vdev_ops->vdev_op_leaf) 1724 vdev_dirty(vd->vdev_top, flags, vd, txg); 1725} 1726 1727/* 1728 * DTLs. 1729 * 1730 * A vdev's DTL (dirty time log) is the set of transaction groups for which 1731 * the vdev has less than perfect replication. There are four kinds of DTL: 1732 * 1733 * DTL_MISSING: txgs for which the vdev has no valid copies of the data 1734 * 1735 * DTL_PARTIAL: txgs for which data is available, but not fully replicated 1736 * 1737 * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 1738 * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 1739 * txgs that was scrubbed. 1740 * 1741 * DTL_OUTAGE: txgs which cannot currently be read, whether due to 1742 * persistent errors or just some device being offline. 1743 * Unlike the other three, the DTL_OUTAGE map is not generally 1744 * maintained; it's only computed when needed, typically to 1745 * determine whether a device can be detached. 1746 * 1747 * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 1748 * either has the data or it doesn't. 1749 * 1750 * For interior vdevs such as mirror and RAID-Z the picture is more complex. 1751 * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 1752 * if any child is less than fully replicated, then so is its parent. 1753 * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 1754 * comprising only those txgs which appear in 'maxfaults' or more children; 1755 * those are the txgs we don't have enough replication to read. For example, 1756 * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 1757 * thus, its DTL_MISSING consists of the set of txgs that appear in more than 1758 * two child DTL_MISSING maps. 1759 * 1760 * It should be clear from the above that to compute the DTLs and outage maps 1761 * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 1762 * Therefore, that is all we keep on disk. When loading the pool, or after 1763 * a configuration change, we generate all other DTLs from first principles. 1764 */ 1765void 1766vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1767{ 1768 range_tree_t *rt = vd->vdev_dtl[t]; 1769 1770 ASSERT(t < DTL_TYPES); 1771 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1772 ASSERT(spa_writeable(vd->vdev_spa)); 1773 1774 mutex_enter(rt->rt_lock); 1775 if (!range_tree_contains(rt, txg, size)) 1776 range_tree_add(rt, txg, size); 1777 mutex_exit(rt->rt_lock); 1778} 1779 1780boolean_t 1781vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 1782{ 1783 range_tree_t *rt = vd->vdev_dtl[t]; 1784 boolean_t dirty = B_FALSE; 1785 1786 ASSERT(t < DTL_TYPES); 1787 ASSERT(vd != vd->vdev_spa->spa_root_vdev); 1788 1789 mutex_enter(rt->rt_lock); 1790 if (range_tree_space(rt) != 0) 1791 dirty = range_tree_contains(rt, txg, size); 1792 mutex_exit(rt->rt_lock); 1793 1794 return (dirty); 1795} 1796 1797boolean_t 1798vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 1799{ 1800 range_tree_t *rt = vd->vdev_dtl[t]; 1801 boolean_t empty; 1802 1803 mutex_enter(rt->rt_lock); 1804 empty = (range_tree_space(rt) == 0); 1805 mutex_exit(rt->rt_lock); 1806 1807 return (empty); 1808} 1809 1810/* 1811 * Returns the lowest txg in the DTL range. 1812 */ 1813static uint64_t 1814vdev_dtl_min(vdev_t *vd) 1815{ 1816 range_seg_t *rs; 1817 1818 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 1819 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 1820 ASSERT0(vd->vdev_children); 1821 1822 rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); 1823 return (rs->rs_start - 1); 1824} 1825 1826/* 1827 * Returns the highest txg in the DTL. 1828 */ 1829static uint64_t 1830vdev_dtl_max(vdev_t *vd) 1831{ 1832 range_seg_t *rs; 1833 1834 ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 1835 ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 1836 ASSERT0(vd->vdev_children); 1837 1838 rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); 1839 return (rs->rs_end); 1840} 1841 1842/* 1843 * Determine if a resilvering vdev should remove any DTL entries from 1844 * its range. If the vdev was resilvering for the entire duration of the 1845 * scan then it should excise that range from its DTLs. Otherwise, this 1846 * vdev is considered partially resilvered and should leave its DTL 1847 * entries intact. The comment in vdev_dtl_reassess() describes how we 1848 * excise the DTLs. 1849 */ 1850static boolean_t 1851vdev_dtl_should_excise(vdev_t *vd) 1852{ 1853 spa_t *spa = vd->vdev_spa; 1854 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 1855 1856 ASSERT0(scn->scn_phys.scn_errors); 1857 ASSERT0(vd->vdev_children); 1858 1859 if (vd->vdev_resilver_txg == 0 || 1860 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0) 1861 return (B_TRUE); 1862 1863 /* 1864 * When a resilver is initiated the scan will assign the scn_max_txg 1865 * value to the highest txg value that exists in all DTLs. If this 1866 * device's max DTL is not part of this scan (i.e. it is not in 1867 * the range (scn_min_txg, scn_max_txg] then it is not eligible 1868 * for excision. 1869 */ 1870 if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { 1871 ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); 1872 ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); 1873 ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); 1874 return (B_TRUE); 1875 } 1876 return (B_FALSE); 1877} 1878 1879/* 1880 * Reassess DTLs after a config change or scrub completion. 1881 */ 1882void 1883vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 1884{ 1885 spa_t *spa = vd->vdev_spa; 1886 avl_tree_t reftree; 1887 int minref; 1888 1889 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 1890 1891 for (int c = 0; c < vd->vdev_children; c++) 1892 vdev_dtl_reassess(vd->vdev_child[c], txg, 1893 scrub_txg, scrub_done); 1894 1895 if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux) 1896 return; 1897 1898 if (vd->vdev_ops->vdev_op_leaf) { 1899 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 1900 1901 mutex_enter(&vd->vdev_dtl_lock); 1902 1903 /* 1904 * If we've completed a scan cleanly then determine 1905 * if this vdev should remove any DTLs. We only want to 1906 * excise regions on vdevs that were available during 1907 * the entire duration of this scan. 1908 */ 1909 if (scrub_txg != 0 && 1910 (spa->spa_scrub_started || 1911 (scn != NULL && scn->scn_phys.scn_errors == 0)) && 1912 vdev_dtl_should_excise(vd)) { 1913 /* 1914 * We completed a scrub up to scrub_txg. If we 1915 * did it without rebooting, then the scrub dtl 1916 * will be valid, so excise the old region and 1917 * fold in the scrub dtl. Otherwise, leave the 1918 * dtl as-is if there was an error. 1919 * 1920 * There's little trick here: to excise the beginning 1921 * of the DTL_MISSING map, we put it into a reference 1922 * tree and then add a segment with refcnt -1 that 1923 * covers the range [0, scrub_txg). This means 1924 * that each txg in that range has refcnt -1 or 0. 1925 * We then add DTL_SCRUB with a refcnt of 2, so that 1926 * entries in the range [0, scrub_txg) will have a 1927 * positive refcnt -- either 1 or 2. We then convert 1928 * the reference tree into the new DTL_MISSING map. 1929 */ 1930 space_reftree_create(&reftree); 1931 space_reftree_add_map(&reftree, 1932 vd->vdev_dtl[DTL_MISSING], 1); 1933 space_reftree_add_seg(&reftree, 0, scrub_txg, -1); 1934 space_reftree_add_map(&reftree, 1935 vd->vdev_dtl[DTL_SCRUB], 2); 1936 space_reftree_generate_map(&reftree, 1937 vd->vdev_dtl[DTL_MISSING], 1); 1938 space_reftree_destroy(&reftree); 1939 } 1940 range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 1941 range_tree_walk(vd->vdev_dtl[DTL_MISSING], 1942 range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); 1943 if (scrub_done) 1944 range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 1945 range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 1946 if (!vdev_readable(vd)) 1947 range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 1948 else 1949 range_tree_walk(vd->vdev_dtl[DTL_MISSING], 1950 range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); 1951 1952 /* 1953 * If the vdev was resilvering and no longer has any 1954 * DTLs then reset its resilvering flag and dirty 1955 * the top level so that we persist the change. 1956 */ 1957 if (vd->vdev_resilver_txg != 0 && 1958 range_tree_space(vd->vdev_dtl[DTL_MISSING]) == 0 && 1959 range_tree_space(vd->vdev_dtl[DTL_OUTAGE]) == 0) { 1960 vd->vdev_resilver_txg = 0; 1961 vdev_config_dirty(vd->vdev_top); 1962 } 1963 1964 mutex_exit(&vd->vdev_dtl_lock); 1965 1966 if (txg != 0) 1967 vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 1968 return; 1969 } 1970 1971 mutex_enter(&vd->vdev_dtl_lock); 1972 for (int t = 0; t < DTL_TYPES; t++) { 1973 /* account for child's outage in parent's missing map */ 1974 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 1975 if (t == DTL_SCRUB) 1976 continue; /* leaf vdevs only */ 1977 if (t == DTL_PARTIAL) 1978 minref = 1; /* i.e. non-zero */ 1979 else if (vd->vdev_nparity != 0) 1980 minref = vd->vdev_nparity + 1; /* RAID-Z */ 1981 else 1982 minref = vd->vdev_children; /* any kind of mirror */ 1983 space_reftree_create(&reftree); 1984 for (int c = 0; c < vd->vdev_children; c++) { 1985 vdev_t *cvd = vd->vdev_child[c]; 1986 mutex_enter(&cvd->vdev_dtl_lock); 1987 space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); 1988 mutex_exit(&cvd->vdev_dtl_lock); 1989 } 1990 space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); 1991 space_reftree_destroy(&reftree); 1992 } 1993 mutex_exit(&vd->vdev_dtl_lock); 1994} 1995 1996int 1997vdev_dtl_load(vdev_t *vd) 1998{ 1999 spa_t *spa = vd->vdev_spa; 2000 objset_t *mos = spa->spa_meta_objset; 2001 int error = 0; 2002 2003 if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { 2004 ASSERT(!vd->vdev_ishole); 2005 2006 error = space_map_open(&vd->vdev_dtl_sm, mos, 2007 vd->vdev_dtl_object, 0, -1ULL, 0, &vd->vdev_dtl_lock); 2008 if (error) 2009 return (error); 2010 ASSERT(vd->vdev_dtl_sm != NULL); 2011 2012 mutex_enter(&vd->vdev_dtl_lock); 2013 2014 /* 2015 * Now that we've opened the space_map we need to update 2016 * the in-core DTL. 2017 */ 2018 space_map_update(vd->vdev_dtl_sm); 2019 2020 error = space_map_load(vd->vdev_dtl_sm, 2021 vd->vdev_dtl[DTL_MISSING], SM_ALLOC); 2022 mutex_exit(&vd->vdev_dtl_lock); 2023 2024 return (error); 2025 } 2026 2027 for (int c = 0; c < vd->vdev_children; c++) { 2028 error = vdev_dtl_load(vd->vdev_child[c]); 2029 if (error != 0) 2030 break; 2031 } 2032 2033 return (error); 2034} 2035 2036void 2037vdev_dtl_sync(vdev_t *vd, uint64_t txg) 2038{ 2039 spa_t *spa = vd->vdev_spa; 2040 range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; 2041 objset_t *mos = spa->spa_meta_objset; 2042 range_tree_t *rtsync; 2043 kmutex_t rtlock; 2044 dmu_tx_t *tx; 2045 uint64_t object = space_map_object(vd->vdev_dtl_sm); 2046 2047 ASSERT(!vd->vdev_ishole); 2048 ASSERT(vd->vdev_ops->vdev_op_leaf); 2049 2050 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2051 2052 if (vd->vdev_detached || vd->vdev_top->vdev_removing) { 2053 mutex_enter(&vd->vdev_dtl_lock); 2054 space_map_free(vd->vdev_dtl_sm, tx); 2055 space_map_close(vd->vdev_dtl_sm); 2056 vd->vdev_dtl_sm = NULL; 2057 mutex_exit(&vd->vdev_dtl_lock); 2058 dmu_tx_commit(tx); 2059 return; 2060 } 2061 2062 if (vd->vdev_dtl_sm == NULL) { 2063 uint64_t new_object; 2064 2065 new_object = space_map_alloc(mos, tx); 2066 VERIFY3U(new_object, !=, 0); 2067 2068 VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 2069 0, -1ULL, 0, &vd->vdev_dtl_lock)); 2070 ASSERT(vd->vdev_dtl_sm != NULL); 2071 } 2072 2073 bzero(&rtlock, sizeof(rtlock)); 2074 mutex_init(&rtlock, NULL, MUTEX_DEFAULT, NULL); 2075 2076 rtsync = range_tree_create(NULL, NULL, &rtlock); 2077 2078 mutex_enter(&rtlock); 2079 2080 mutex_enter(&vd->vdev_dtl_lock); 2081 range_tree_walk(rt, range_tree_add, rtsync); 2082 mutex_exit(&vd->vdev_dtl_lock); 2083 2084 space_map_truncate(vd->vdev_dtl_sm, tx); 2085 space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx); 2086 range_tree_vacate(rtsync, NULL, NULL); 2087 2088 range_tree_destroy(rtsync); 2089 2090 mutex_exit(&rtlock); 2091 mutex_destroy(&rtlock); 2092 2093 /* 2094 * If the object for the space map has changed then dirty 2095 * the top level so that we update the config. 2096 */ 2097 if (object != space_map_object(vd->vdev_dtl_sm)) { 2098 zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, " 2099 "new object %llu", txg, spa_name(spa), object, 2100 space_map_object(vd->vdev_dtl_sm)); 2101 vdev_config_dirty(vd->vdev_top); 2102 } 2103 2104 dmu_tx_commit(tx); 2105 2106 mutex_enter(&vd->vdev_dtl_lock); 2107 space_map_update(vd->vdev_dtl_sm); 2108 mutex_exit(&vd->vdev_dtl_lock); 2109} 2110 2111/* 2112 * Determine whether the specified vdev can be offlined/detached/removed 2113 * without losing data. 2114 */ 2115boolean_t 2116vdev_dtl_required(vdev_t *vd) 2117{ 2118 spa_t *spa = vd->vdev_spa; 2119 vdev_t *tvd = vd->vdev_top; 2120 uint8_t cant_read = vd->vdev_cant_read; 2121 boolean_t required; 2122 2123 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2124 2125 if (vd == spa->spa_root_vdev || vd == tvd) 2126 return (B_TRUE); 2127 2128 /* 2129 * Temporarily mark the device as unreadable, and then determine 2130 * whether this results in any DTL outages in the top-level vdev. 2131 * If not, we can safely offline/detach/remove the device. 2132 */ 2133 vd->vdev_cant_read = B_TRUE; 2134 vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 2135 required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 2136 vd->vdev_cant_read = cant_read; 2137 vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 2138 2139 if (!required && zio_injection_enabled) 2140 required = !!zio_handle_device_injection(vd, NULL, ECHILD); 2141 2142 return (required); 2143} 2144 2145/* 2146 * Determine if resilver is needed, and if so the txg range. 2147 */ 2148boolean_t 2149vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 2150{ 2151 boolean_t needed = B_FALSE; 2152 uint64_t thismin = UINT64_MAX; 2153 uint64_t thismax = 0; 2154 2155 if (vd->vdev_children == 0) { 2156 mutex_enter(&vd->vdev_dtl_lock); 2157 if (range_tree_space(vd->vdev_dtl[DTL_MISSING]) != 0 && 2158 vdev_writeable(vd)) { 2159 2160 thismin = vdev_dtl_min(vd); 2161 thismax = vdev_dtl_max(vd); 2162 needed = B_TRUE; 2163 } 2164 mutex_exit(&vd->vdev_dtl_lock); 2165 } else { 2166 for (int c = 0; c < vd->vdev_children; c++) { 2167 vdev_t *cvd = vd->vdev_child[c]; 2168 uint64_t cmin, cmax; 2169 2170 if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 2171 thismin = MIN(thismin, cmin); 2172 thismax = MAX(thismax, cmax); 2173 needed = B_TRUE; 2174 } 2175 } 2176 } 2177 2178 if (needed && minp) { 2179 *minp = thismin; 2180 *maxp = thismax; 2181 } 2182 return (needed); 2183} 2184 2185void 2186vdev_load(vdev_t *vd) 2187{ 2188 /* 2189 * Recursively load all children. 2190 */ 2191 for (int c = 0; c < vd->vdev_children; c++) 2192 vdev_load(vd->vdev_child[c]); 2193 2194 /* 2195 * If this is a top-level vdev, initialize its metaslabs. 2196 */ 2197 if (vd == vd->vdev_top && !vd->vdev_ishole && 2198 (vd->vdev_ashift == 0 || vd->vdev_asize == 0 || 2199 vdev_metaslab_init(vd, 0) != 0)) 2200 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2201 VDEV_AUX_CORRUPT_DATA); 2202 2203 /* 2204 * If this is a leaf vdev, load its DTL. 2205 */ 2206 if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0) 2207 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2208 VDEV_AUX_CORRUPT_DATA); 2209} 2210 2211/* 2212 * The special vdev case is used for hot spares and l2cache devices. Its 2213 * sole purpose it to set the vdev state for the associated vdev. To do this, 2214 * we make sure that we can open the underlying device, then try to read the 2215 * label, and make sure that the label is sane and that it hasn't been 2216 * repurposed to another pool. 2217 */ 2218int 2219vdev_validate_aux(vdev_t *vd) 2220{ 2221 nvlist_t *label; 2222 uint64_t guid, version; 2223 uint64_t state; 2224 2225 if (!vdev_readable(vd)) 2226 return (0); 2227 2228 if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 2229 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2230 VDEV_AUX_CORRUPT_DATA); 2231 return (-1); 2232 } 2233 2234 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 2235 !SPA_VERSION_IS_SUPPORTED(version) || 2236 nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 2237 guid != vd->vdev_guid || 2238 nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 2239 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 2240 VDEV_AUX_CORRUPT_DATA); 2241 nvlist_free(label); 2242 return (-1); 2243 } 2244 2245 /* 2246 * We don't actually check the pool state here. If it's in fact in 2247 * use by another pool, we update this fact on the fly when requested. 2248 */ 2249 nvlist_free(label); 2250 return (0); 2251} 2252 2253void 2254vdev_remove(vdev_t *vd, uint64_t txg) 2255{ 2256 spa_t *spa = vd->vdev_spa; 2257 objset_t *mos = spa->spa_meta_objset; 2258 dmu_tx_t *tx; 2259 2260 tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 2261 2262 if (vd->vdev_ms != NULL) { 2263 metaslab_group_t *mg = vd->vdev_mg; 2264 2265 metaslab_group_histogram_verify(mg); 2266 metaslab_class_histogram_verify(mg->mg_class); 2267 2268 for (int m = 0; m < vd->vdev_ms_count; m++) { 2269 metaslab_t *msp = vd->vdev_ms[m]; 2270 2271 if (msp == NULL || msp->ms_sm == NULL) 2272 continue; 2273 2274 mutex_enter(&msp->ms_lock); 2275 /* 2276 * If the metaslab was not loaded when the vdev 2277 * was removed then the histogram accounting may 2278 * not be accurate. Update the histogram information 2279 * here so that we ensure that the metaslab group 2280 * and metaslab class are up-to-date. 2281 */ 2282 metaslab_group_histogram_remove(mg, msp); 2283 2284 VERIFY0(space_map_allocated(msp->ms_sm)); 2285 space_map_free(msp->ms_sm, tx); 2286 space_map_close(msp->ms_sm); 2287 msp->ms_sm = NULL; 2288 mutex_exit(&msp->ms_lock); 2289 } 2290 2291 metaslab_group_histogram_verify(mg); 2292 metaslab_class_histogram_verify(mg->mg_class); 2293 for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) 2294 ASSERT0(mg->mg_histogram[i]); 2295 2296 } 2297 2298 if (vd->vdev_ms_array) { 2299 (void) dmu_object_free(mos, vd->vdev_ms_array, tx); 2300 vd->vdev_ms_array = 0; 2301 } 2302 dmu_tx_commit(tx); 2303} 2304 2305void 2306vdev_sync_done(vdev_t *vd, uint64_t txg) 2307{ 2308 metaslab_t *msp; 2309 boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 2310 2311 ASSERT(!vd->vdev_ishole); 2312 2313 while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 2314 metaslab_sync_done(msp, txg); 2315 2316 if (reassess) 2317 metaslab_sync_reassess(vd->vdev_mg); 2318} 2319 2320void 2321vdev_sync(vdev_t *vd, uint64_t txg) 2322{ 2323 spa_t *spa = vd->vdev_spa; 2324 vdev_t *lvd; 2325 metaslab_t *msp; 2326 dmu_tx_t *tx; 2327 2328 ASSERT(!vd->vdev_ishole); 2329 2330 if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) { 2331 ASSERT(vd == vd->vdev_top); 2332 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2333 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 2334 DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 2335 ASSERT(vd->vdev_ms_array != 0); 2336 vdev_config_dirty(vd); 2337 dmu_tx_commit(tx); 2338 } 2339 2340 /* 2341 * Remove the metadata associated with this vdev once it's empty. 2342 */ 2343 if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) 2344 vdev_remove(vd, txg); 2345 2346 while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 2347 metaslab_sync(msp, txg); 2348 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 2349 } 2350 2351 while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 2352 vdev_dtl_sync(lvd, txg); 2353 2354 (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 2355} 2356 2357uint64_t 2358vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 2359{ 2360 return (vd->vdev_ops->vdev_op_asize(vd, psize)); 2361} 2362 2363/* 2364 * Mark the given vdev faulted. A faulted vdev behaves as if the device could 2365 * not be opened, and no I/O is attempted. 2366 */ 2367int 2368vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2369{ 2370 vdev_t *vd, *tvd; 2371 2372 spa_vdev_state_enter(spa, SCL_NONE); 2373 2374 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2375 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2376 2377 if (!vd->vdev_ops->vdev_op_leaf) 2378 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2379 2380 tvd = vd->vdev_top; 2381 2382 /* 2383 * We don't directly use the aux state here, but if we do a 2384 * vdev_reopen(), we need this value to be present to remember why we 2385 * were faulted. 2386 */ 2387 vd->vdev_label_aux = aux; 2388 2389 /* 2390 * Faulted state takes precedence over degraded. 2391 */ 2392 vd->vdev_delayed_close = B_FALSE; 2393 vd->vdev_faulted = 1ULL; 2394 vd->vdev_degraded = 0ULL; 2395 vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 2396 2397 /* 2398 * If this device has the only valid copy of the data, then 2399 * back off and simply mark the vdev as degraded instead. 2400 */ 2401 if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 2402 vd->vdev_degraded = 1ULL; 2403 vd->vdev_faulted = 0ULL; 2404 2405 /* 2406 * If we reopen the device and it's not dead, only then do we 2407 * mark it degraded. 2408 */ 2409 vdev_reopen(tvd); 2410 2411 if (vdev_readable(vd)) 2412 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 2413 } 2414 2415 return (spa_vdev_state_exit(spa, vd, 0)); 2416} 2417 2418/* 2419 * Mark the given vdev degraded. A degraded vdev is purely an indication to the 2420 * user that something is wrong. The vdev continues to operate as normal as far 2421 * as I/O is concerned. 2422 */ 2423int 2424vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2425{ 2426 vdev_t *vd; 2427 2428 spa_vdev_state_enter(spa, SCL_NONE); 2429 2430 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2431 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2432 2433 if (!vd->vdev_ops->vdev_op_leaf) 2434 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2435 2436 /* 2437 * If the vdev is already faulted, then don't do anything. 2438 */ 2439 if (vd->vdev_faulted || vd->vdev_degraded) 2440 return (spa_vdev_state_exit(spa, NULL, 0)); 2441 2442 vd->vdev_degraded = 1ULL; 2443 if (!vdev_is_dead(vd)) 2444 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 2445 aux); 2446 2447 return (spa_vdev_state_exit(spa, vd, 0)); 2448} 2449 2450/* 2451 * Online the given vdev. 2452 * 2453 * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached 2454 * spare device should be detached when the device finishes resilvering. 2455 * Second, the online should be treated like a 'test' online case, so no FMA 2456 * events are generated if the device fails to open. 2457 */ 2458int 2459vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 2460{ 2461 vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 2462 2463 spa_vdev_state_enter(spa, SCL_NONE); 2464 2465 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2466 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2467 2468 if (!vd->vdev_ops->vdev_op_leaf) 2469 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2470 2471 tvd = vd->vdev_top; 2472 vd->vdev_offline = B_FALSE; 2473 vd->vdev_tmpoffline = B_FALSE; 2474 vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 2475 vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 2476 2477 /* XXX - L2ARC 1.0 does not support expansion */ 2478 if (!vd->vdev_aux) { 2479 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2480 pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 2481 } 2482 2483 vdev_reopen(tvd); 2484 vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 2485 2486 if (!vd->vdev_aux) { 2487 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2488 pvd->vdev_expanding = B_FALSE; 2489 } 2490 2491 if (newstate) 2492 *newstate = vd->vdev_state; 2493 if ((flags & ZFS_ONLINE_UNSPARE) && 2494 !vdev_is_dead(vd) && vd->vdev_parent && 2495 vd->vdev_parent->vdev_ops == &vdev_spare_ops && 2496 vd->vdev_parent->vdev_child[0] == vd) 2497 vd->vdev_unspare = B_TRUE; 2498 2499 if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 2500 2501 /* XXX - L2ARC 1.0 does not support expansion */ 2502 if (vd->vdev_aux) 2503 return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 2504 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 2505 } 2506 return (spa_vdev_state_exit(spa, vd, 0)); 2507} 2508 2509static int 2510vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 2511{ 2512 vdev_t *vd, *tvd; 2513 int error = 0; 2514 uint64_t generation; 2515 metaslab_group_t *mg; 2516 2517top: 2518 spa_vdev_state_enter(spa, SCL_ALLOC); 2519 2520 if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2521 return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2522 2523 if (!vd->vdev_ops->vdev_op_leaf) 2524 return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2525 2526 tvd = vd->vdev_top; 2527 mg = tvd->vdev_mg; 2528 generation = spa->spa_config_generation + 1; 2529 2530 /* 2531 * If the device isn't already offline, try to offline it. 2532 */ 2533 if (!vd->vdev_offline) { 2534 /* 2535 * If this device has the only valid copy of some data, 2536 * don't allow it to be offlined. Log devices are always 2537 * expendable. 2538 */ 2539 if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2540 vdev_dtl_required(vd)) 2541 return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2542 2543 /* 2544 * If the top-level is a slog and it has had allocations 2545 * then proceed. We check that the vdev's metaslab group 2546 * is not NULL since it's possible that we may have just 2547 * added this vdev but not yet initialized its metaslabs. 2548 */ 2549 if (tvd->vdev_islog && mg != NULL) { 2550 /* 2551 * Prevent any future allocations. 2552 */ 2553 metaslab_group_passivate(mg); 2554 (void) spa_vdev_state_exit(spa, vd, 0); 2555 2556 error = spa_offline_log(spa); 2557 2558 spa_vdev_state_enter(spa, SCL_ALLOC); 2559 2560 /* 2561 * Check to see if the config has changed. 2562 */ 2563 if (error || generation != spa->spa_config_generation) { 2564 metaslab_group_activate(mg); 2565 if (error) 2566 return (spa_vdev_state_exit(spa, 2567 vd, error)); 2568 (void) spa_vdev_state_exit(spa, vd, 0); 2569 goto top; 2570 } 2571 ASSERT0(tvd->vdev_stat.vs_alloc); 2572 } 2573 2574 /* 2575 * Offline this device and reopen its top-level vdev. 2576 * If the top-level vdev is a log device then just offline 2577 * it. Otherwise, if this action results in the top-level 2578 * vdev becoming unusable, undo it and fail the request. 2579 */ 2580 vd->vdev_offline = B_TRUE; 2581 vdev_reopen(tvd); 2582 2583 if (!tvd->vdev_islog && vd->vdev_aux == NULL && 2584 vdev_is_dead(tvd)) { 2585 vd->vdev_offline = B_FALSE; 2586 vdev_reopen(tvd); 2587 return (spa_vdev_state_exit(spa, NULL, EBUSY)); 2588 } 2589 2590 /* 2591 * Add the device back into the metaslab rotor so that 2592 * once we online the device it's open for business. 2593 */ 2594 if (tvd->vdev_islog && mg != NULL) 2595 metaslab_group_activate(mg); 2596 } 2597 2598 vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 2599 2600 return (spa_vdev_state_exit(spa, vd, 0)); 2601} 2602 2603int 2604vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 2605{ 2606 int error; 2607 2608 mutex_enter(&spa->spa_vdev_top_lock); 2609 error = vdev_offline_locked(spa, guid, flags); 2610 mutex_exit(&spa->spa_vdev_top_lock); 2611 2612 return (error); 2613} 2614 2615/* 2616 * Clear the error counts associated with this vdev. Unlike vdev_online() and 2617 * vdev_offline(), we assume the spa config is locked. We also clear all 2618 * children. If 'vd' is NULL, then the user wants to clear all vdevs. 2619 */ 2620void 2621vdev_clear(spa_t *spa, vdev_t *vd) 2622{ 2623 vdev_t *rvd = spa->spa_root_vdev; 2624 2625 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 2626 2627 if (vd == NULL) 2628 vd = rvd; 2629 2630 vd->vdev_stat.vs_read_errors = 0; 2631 vd->vdev_stat.vs_write_errors = 0; 2632 vd->vdev_stat.vs_checksum_errors = 0; 2633 2634 for (int c = 0; c < vd->vdev_children; c++) 2635 vdev_clear(spa, vd->vdev_child[c]); 2636 2637 if (vd == rvd) { 2638 for (int c = 0; c < spa->spa_l2cache.sav_count; c++) 2639 vdev_clear(spa, spa->spa_l2cache.sav_vdevs[c]); 2640 2641 for (int c = 0; c < spa->spa_spares.sav_count; c++) 2642 vdev_clear(spa, spa->spa_spares.sav_vdevs[c]); 2643 } 2644 2645 /* 2646 * If we're in the FAULTED state or have experienced failed I/O, then 2647 * clear the persistent state and attempt to reopen the device. We 2648 * also mark the vdev config dirty, so that the new faulted state is 2649 * written out to disk. 2650 */ 2651 if (vd->vdev_faulted || vd->vdev_degraded || 2652 !vdev_readable(vd) || !vdev_writeable(vd)) { 2653 2654 /* 2655 * When reopening in reponse to a clear event, it may be due to 2656 * a fmadm repair request. In this case, if the device is 2657 * still broken, we want to still post the ereport again. 2658 */ 2659 vd->vdev_forcefault = B_TRUE; 2660 2661 vd->vdev_faulted = vd->vdev_degraded = 0ULL; 2662 vd->vdev_cant_read = B_FALSE; 2663 vd->vdev_cant_write = B_FALSE; 2664 2665 vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 2666 2667 vd->vdev_forcefault = B_FALSE; 2668 2669 if (vd != rvd && vdev_writeable(vd->vdev_top)) 2670 vdev_state_dirty(vd->vdev_top); 2671 2672 if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 2673 spa_async_request(spa, SPA_ASYNC_RESILVER); 2674 2675 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR); 2676 } 2677 2678 /* 2679 * When clearing a FMA-diagnosed fault, we always want to 2680 * unspare the device, as we assume that the original spare was 2681 * done in response to the FMA fault. 2682 */ 2683 if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 2684 vd->vdev_parent->vdev_ops == &vdev_spare_ops && 2685 vd->vdev_parent->vdev_child[0] == vd) 2686 vd->vdev_unspare = B_TRUE; 2687} 2688 2689boolean_t 2690vdev_is_dead(vdev_t *vd) 2691{ 2692 /* 2693 * Holes and missing devices are always considered "dead". 2694 * This simplifies the code since we don't have to check for 2695 * these types of devices in the various code paths. 2696 * Instead we rely on the fact that we skip over dead devices 2697 * before issuing I/O to them. 2698 */ 2699 return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole || 2700 vd->vdev_ops == &vdev_missing_ops); 2701} 2702 2703boolean_t 2704vdev_readable(vdev_t *vd) 2705{ 2706 return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 2707} 2708 2709boolean_t 2710vdev_writeable(vdev_t *vd) 2711{ 2712 return (!vdev_is_dead(vd) && !vd->vdev_cant_write); 2713} 2714 2715boolean_t 2716vdev_allocatable(vdev_t *vd) 2717{ 2718 uint64_t state = vd->vdev_state; 2719 2720 /* 2721 * We currently allow allocations from vdevs which may be in the 2722 * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 2723 * fails to reopen then we'll catch it later when we're holding 2724 * the proper locks. Note that we have to get the vdev state 2725 * in a local variable because although it changes atomically, 2726 * we're asking two separate questions about it. 2727 */ 2728 return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 2729 !vd->vdev_cant_write && !vd->vdev_ishole); 2730} 2731 2732boolean_t 2733vdev_accessible(vdev_t *vd, zio_t *zio) 2734{ 2735 ASSERT(zio->io_vd == vd); 2736 2737 if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 2738 return (B_FALSE); 2739 2740 if (zio->io_type == ZIO_TYPE_READ) 2741 return (!vd->vdev_cant_read); 2742 2743 if (zio->io_type == ZIO_TYPE_WRITE) 2744 return (!vd->vdev_cant_write); 2745 2746 return (B_TRUE); 2747} 2748 2749/* 2750 * Get statistics for the given vdev. 2751 */ 2752void 2753vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 2754{ 2755 spa_t *spa = vd->vdev_spa; 2756 vdev_t *rvd = spa->spa_root_vdev; 2757 2758 ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 2759 2760 mutex_enter(&vd->vdev_stat_lock); 2761 bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 2762 vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 2763 vs->vs_state = vd->vdev_state; 2764 vs->vs_rsize = vdev_get_min_asize(vd); 2765 if (vd->vdev_ops->vdev_op_leaf) 2766 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 2767 vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize; 2768 vs->vs_configured_ashift = vd->vdev_top != NULL 2769 ? vd->vdev_top->vdev_ashift : vd->vdev_ashift; 2770 vs->vs_logical_ashift = vd->vdev_logical_ashift; 2771 vs->vs_physical_ashift = vd->vdev_physical_ashift; 2772 if (vd->vdev_aux == NULL && vd == vd->vdev_top && !vd->vdev_ishole) { 2773 vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation; 2774 } 2775 2776 /* 2777 * If we're getting stats on the root vdev, aggregate the I/O counts 2778 * over all top-level vdevs (i.e. the direct children of the root). 2779 */ 2780 if (vd == rvd) { 2781 for (int c = 0; c < rvd->vdev_children; c++) { 2782 vdev_t *cvd = rvd->vdev_child[c]; 2783 vdev_stat_t *cvs = &cvd->vdev_stat; 2784 2785 for (int t = 0; t < ZIO_TYPES; t++) { 2786 vs->vs_ops[t] += cvs->vs_ops[t]; 2787 vs->vs_bytes[t] += cvs->vs_bytes[t]; 2788 } 2789 cvs->vs_scan_removing = cvd->vdev_removing; 2790 } 2791 } 2792 mutex_exit(&vd->vdev_stat_lock); 2793} 2794 2795void 2796vdev_clear_stats(vdev_t *vd) 2797{ 2798 mutex_enter(&vd->vdev_stat_lock); 2799 vd->vdev_stat.vs_space = 0; 2800 vd->vdev_stat.vs_dspace = 0; 2801 vd->vdev_stat.vs_alloc = 0; 2802 mutex_exit(&vd->vdev_stat_lock); 2803} 2804 2805void 2806vdev_scan_stat_init(vdev_t *vd) 2807{ 2808 vdev_stat_t *vs = &vd->vdev_stat; 2809 2810 for (int c = 0; c < vd->vdev_children; c++) 2811 vdev_scan_stat_init(vd->vdev_child[c]); 2812 2813 mutex_enter(&vd->vdev_stat_lock); 2814 vs->vs_scan_processed = 0; 2815 mutex_exit(&vd->vdev_stat_lock); 2816} 2817 2818void 2819vdev_stat_update(zio_t *zio, uint64_t psize) 2820{ 2821 spa_t *spa = zio->io_spa; 2822 vdev_t *rvd = spa->spa_root_vdev; 2823 vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 2824 vdev_t *pvd; 2825 uint64_t txg = zio->io_txg; 2826 vdev_stat_t *vs = &vd->vdev_stat; 2827 zio_type_t type = zio->io_type; 2828 int flags = zio->io_flags; 2829 2830 /* 2831 * If this i/o is a gang leader, it didn't do any actual work. 2832 */ 2833 if (zio->io_gang_tree) 2834 return; 2835 2836 if (zio->io_error == 0) { 2837 /* 2838 * If this is a root i/o, don't count it -- we've already 2839 * counted the top-level vdevs, and vdev_get_stats() will 2840 * aggregate them when asked. This reduces contention on 2841 * the root vdev_stat_lock and implicitly handles blocks 2842 * that compress away to holes, for which there is no i/o. 2843 * (Holes never create vdev children, so all the counters 2844 * remain zero, which is what we want.) 2845 * 2846 * Note: this only applies to successful i/o (io_error == 0) 2847 * because unlike i/o counts, errors are not additive. 2848 * When reading a ditto block, for example, failure of 2849 * one top-level vdev does not imply a root-level error. 2850 */ 2851 if (vd == rvd) 2852 return; 2853 2854 ASSERT(vd == zio->io_vd); 2855 2856 if (flags & ZIO_FLAG_IO_BYPASS) 2857 return; 2858 2859 mutex_enter(&vd->vdev_stat_lock); 2860 2861 if (flags & ZIO_FLAG_IO_REPAIR) { 2862 if (flags & ZIO_FLAG_SCAN_THREAD) { 2863 dsl_scan_phys_t *scn_phys = 2864 &spa->spa_dsl_pool->dp_scan->scn_phys; 2865 uint64_t *processed = &scn_phys->scn_processed; 2866 2867 /* XXX cleanup? */ 2868 if (vd->vdev_ops->vdev_op_leaf) 2869 atomic_add_64(processed, psize); 2870 vs->vs_scan_processed += psize; 2871 } 2872 2873 if (flags & ZIO_FLAG_SELF_HEAL) 2874 vs->vs_self_healed += psize; 2875 } 2876 2877 vs->vs_ops[type]++; 2878 vs->vs_bytes[type] += psize; 2879 2880 mutex_exit(&vd->vdev_stat_lock); 2881 return; 2882 } 2883 2884 if (flags & ZIO_FLAG_SPECULATIVE) 2885 return; 2886 2887 /* 2888 * If this is an I/O error that is going to be retried, then ignore the 2889 * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 2890 * hard errors, when in reality they can happen for any number of 2891 * innocuous reasons (bus resets, MPxIO link failure, etc). 2892 */ 2893 if (zio->io_error == EIO && 2894 !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 2895 return; 2896 2897 /* 2898 * Intent logs writes won't propagate their error to the root 2899 * I/O so don't mark these types of failures as pool-level 2900 * errors. 2901 */ 2902 if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 2903 return; 2904 2905 mutex_enter(&vd->vdev_stat_lock); 2906 if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 2907 if (zio->io_error == ECKSUM) 2908 vs->vs_checksum_errors++; 2909 else 2910 vs->vs_read_errors++; 2911 } 2912 if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 2913 vs->vs_write_errors++; 2914 mutex_exit(&vd->vdev_stat_lock); 2915 2916 if (type == ZIO_TYPE_WRITE && txg != 0 && 2917 (!(flags & ZIO_FLAG_IO_REPAIR) || 2918 (flags & ZIO_FLAG_SCAN_THREAD) || 2919 spa->spa_claiming)) { 2920 /* 2921 * This is either a normal write (not a repair), or it's 2922 * a repair induced by the scrub thread, or it's a repair 2923 * made by zil_claim() during spa_load() in the first txg. 2924 * In the normal case, we commit the DTL change in the same 2925 * txg as the block was born. In the scrub-induced repair 2926 * case, we know that scrubs run in first-pass syncing context, 2927 * so we commit the DTL change in spa_syncing_txg(spa). 2928 * In the zil_claim() case, we commit in spa_first_txg(spa). 2929 * 2930 * We currently do not make DTL entries for failed spontaneous 2931 * self-healing writes triggered by normal (non-scrubbing) 2932 * reads, because we have no transactional context in which to 2933 * do so -- and it's not clear that it'd be desirable anyway. 2934 */ 2935 if (vd->vdev_ops->vdev_op_leaf) { 2936 uint64_t commit_txg = txg; 2937 if (flags & ZIO_FLAG_SCAN_THREAD) { 2938 ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2939 ASSERT(spa_sync_pass(spa) == 1); 2940 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 2941 commit_txg = spa_syncing_txg(spa); 2942 } else if (spa->spa_claiming) { 2943 ASSERT(flags & ZIO_FLAG_IO_REPAIR); 2944 commit_txg = spa_first_txg(spa); 2945 } 2946 ASSERT(commit_txg >= spa_syncing_txg(spa)); 2947 if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 2948 return; 2949 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 2950 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 2951 vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 2952 } 2953 if (vd != rvd) 2954 vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 2955 } 2956} 2957 2958/* 2959 * Update the in-core space usage stats for this vdev, its metaslab class, 2960 * and the root vdev. 2961 */ 2962void 2963vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 2964 int64_t space_delta) 2965{ 2966 int64_t dspace_delta = space_delta; 2967 spa_t *spa = vd->vdev_spa; 2968 vdev_t *rvd = spa->spa_root_vdev; 2969 metaslab_group_t *mg = vd->vdev_mg; 2970 metaslab_class_t *mc = mg ? mg->mg_class : NULL; 2971 2972 ASSERT(vd == vd->vdev_top); 2973 2974 /* 2975 * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 2976 * factor. We must calculate this here and not at the root vdev 2977 * because the root vdev's psize-to-asize is simply the max of its 2978 * childrens', thus not accurate enough for us. 2979 */ 2980 ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 2981 ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 2982 dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 2983 vd->vdev_deflate_ratio; 2984 2985 mutex_enter(&vd->vdev_stat_lock); 2986 vd->vdev_stat.vs_alloc += alloc_delta; 2987 vd->vdev_stat.vs_space += space_delta; 2988 vd->vdev_stat.vs_dspace += dspace_delta; 2989 mutex_exit(&vd->vdev_stat_lock); 2990 2991 if (mc == spa_normal_class(spa)) { 2992 mutex_enter(&rvd->vdev_stat_lock); 2993 rvd->vdev_stat.vs_alloc += alloc_delta; 2994 rvd->vdev_stat.vs_space += space_delta; 2995 rvd->vdev_stat.vs_dspace += dspace_delta; 2996 mutex_exit(&rvd->vdev_stat_lock); 2997 } 2998 2999 if (mc != NULL) { 3000 ASSERT(rvd == vd->vdev_parent); 3001 ASSERT(vd->vdev_ms_count != 0); 3002 3003 metaslab_class_space_update(mc, 3004 alloc_delta, defer_delta, space_delta, dspace_delta); 3005 } 3006} 3007 3008/* 3009 * Mark a top-level vdev's config as dirty, placing it on the dirty list 3010 * so that it will be written out next time the vdev configuration is synced. 3011 * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 3012 */ 3013void 3014vdev_config_dirty(vdev_t *vd) 3015{ 3016 spa_t *spa = vd->vdev_spa; 3017 vdev_t *rvd = spa->spa_root_vdev; 3018 int c; 3019 3020 ASSERT(spa_writeable(spa)); 3021 3022 /* 3023 * If this is an aux vdev (as with l2cache and spare devices), then we 3024 * update the vdev config manually and set the sync flag. 3025 */ 3026 if (vd->vdev_aux != NULL) { 3027 spa_aux_vdev_t *sav = vd->vdev_aux; 3028 nvlist_t **aux; 3029 uint_t naux; 3030 3031 for (c = 0; c < sav->sav_count; c++) { 3032 if (sav->sav_vdevs[c] == vd) 3033 break; 3034 } 3035 3036 if (c == sav->sav_count) { 3037 /* 3038 * We're being removed. There's nothing more to do. 3039 */ 3040 ASSERT(sav->sav_sync == B_TRUE); 3041 return; 3042 } 3043 3044 sav->sav_sync = B_TRUE; 3045 3046 if (nvlist_lookup_nvlist_array(sav->sav_config, 3047 ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 3048 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 3049 ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 3050 } 3051 3052 ASSERT(c < naux); 3053 3054 /* 3055 * Setting the nvlist in the middle if the array is a little 3056 * sketchy, but it will work. 3057 */ 3058 nvlist_free(aux[c]); 3059 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 3060 3061 return; 3062 } 3063 3064 /* 3065 * The dirty list is protected by the SCL_CONFIG lock. The caller 3066 * must either hold SCL_CONFIG as writer, or must be the sync thread 3067 * (which holds SCL_CONFIG as reader). There's only one sync thread, 3068 * so this is sufficient to ensure mutual exclusion. 3069 */ 3070 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 3071 (dsl_pool_sync_context(spa_get_dsl(spa)) && 3072 spa_config_held(spa, SCL_CONFIG, RW_READER))); 3073 3074 if (vd == rvd) { 3075 for (c = 0; c < rvd->vdev_children; c++) 3076 vdev_config_dirty(rvd->vdev_child[c]); 3077 } else { 3078 ASSERT(vd == vd->vdev_top); 3079 3080 if (!list_link_active(&vd->vdev_config_dirty_node) && 3081 !vd->vdev_ishole) 3082 list_insert_head(&spa->spa_config_dirty_list, vd); 3083 } 3084} 3085 3086void 3087vdev_config_clean(vdev_t *vd) 3088{ 3089 spa_t *spa = vd->vdev_spa; 3090 3091 ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 3092 (dsl_pool_sync_context(spa_get_dsl(spa)) && 3093 spa_config_held(spa, SCL_CONFIG, RW_READER))); 3094 3095 ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 3096 list_remove(&spa->spa_config_dirty_list, vd); 3097} 3098 3099/* 3100 * Mark a top-level vdev's state as dirty, so that the next pass of 3101 * spa_sync() can convert this into vdev_config_dirty(). We distinguish 3102 * the state changes from larger config changes because they require 3103 * much less locking, and are often needed for administrative actions. 3104 */ 3105void 3106vdev_state_dirty(vdev_t *vd) 3107{ 3108 spa_t *spa = vd->vdev_spa; 3109 3110 ASSERT(spa_writeable(spa)); 3111 ASSERT(vd == vd->vdev_top); 3112 3113 /* 3114 * The state list is protected by the SCL_STATE lock. The caller 3115 * must either hold SCL_STATE as writer, or must be the sync thread 3116 * (which holds SCL_STATE as reader). There's only one sync thread, 3117 * so this is sufficient to ensure mutual exclusion. 3118 */ 3119 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3120 (dsl_pool_sync_context(spa_get_dsl(spa)) && 3121 spa_config_held(spa, SCL_STATE, RW_READER))); 3122 3123 if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole) 3124 list_insert_head(&spa->spa_state_dirty_list, vd); 3125} 3126 3127void 3128vdev_state_clean(vdev_t *vd) 3129{ 3130 spa_t *spa = vd->vdev_spa; 3131 3132 ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3133 (dsl_pool_sync_context(spa_get_dsl(spa)) && 3134 spa_config_held(spa, SCL_STATE, RW_READER))); 3135 3136 ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 3137 list_remove(&spa->spa_state_dirty_list, vd); 3138} 3139 3140/* 3141 * Propagate vdev state up from children to parent. 3142 */ 3143void 3144vdev_propagate_state(vdev_t *vd) 3145{ 3146 spa_t *spa = vd->vdev_spa; 3147 vdev_t *rvd = spa->spa_root_vdev; 3148 int degraded = 0, faulted = 0; 3149 int corrupted = 0; 3150 vdev_t *child; 3151 3152 if (vd->vdev_children > 0) { 3153 for (int c = 0; c < vd->vdev_children; c++) { 3154 child = vd->vdev_child[c]; 3155 3156 /* 3157 * Don't factor holes into the decision. 3158 */ 3159 if (child->vdev_ishole) 3160 continue; 3161 3162 if (!vdev_readable(child) || 3163 (!vdev_writeable(child) && spa_writeable(spa))) { 3164 /* 3165 * Root special: if there is a top-level log 3166 * device, treat the root vdev as if it were 3167 * degraded. 3168 */ 3169 if (child->vdev_islog && vd == rvd) 3170 degraded++; 3171 else 3172 faulted++; 3173 } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 3174 degraded++; 3175 } 3176 3177 if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 3178 corrupted++; 3179 } 3180 3181 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 3182 3183 /* 3184 * Root special: if there is a top-level vdev that cannot be 3185 * opened due to corrupted metadata, then propagate the root 3186 * vdev's aux state as 'corrupt' rather than 'insufficient 3187 * replicas'. 3188 */ 3189 if (corrupted && vd == rvd && 3190 rvd->vdev_state == VDEV_STATE_CANT_OPEN) 3191 vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 3192 VDEV_AUX_CORRUPT_DATA); 3193 } 3194 3195 if (vd->vdev_parent) 3196 vdev_propagate_state(vd->vdev_parent); 3197} 3198 3199/* 3200 * Set a vdev's state. If this is during an open, we don't update the parent 3201 * state, because we're in the process of opening children depth-first. 3202 * Otherwise, we propagate the change to the parent. 3203 * 3204 * If this routine places a device in a faulted state, an appropriate ereport is 3205 * generated. 3206 */ 3207void 3208vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 3209{ 3210 uint64_t save_state; 3211 spa_t *spa = vd->vdev_spa; 3212 3213 if (state == vd->vdev_state) { 3214 vd->vdev_stat.vs_aux = aux; 3215 return; 3216 } 3217 3218 save_state = vd->vdev_state; 3219 3220 vd->vdev_state = state; 3221 vd->vdev_stat.vs_aux = aux; 3222 3223 /* 3224 * If we are setting the vdev state to anything but an open state, then 3225 * always close the underlying device unless the device has requested 3226 * a delayed close (i.e. we're about to remove or fault the device). 3227 * Otherwise, we keep accessible but invalid devices open forever. 3228 * We don't call vdev_close() itself, because that implies some extra 3229 * checks (offline, etc) that we don't want here. This is limited to 3230 * leaf devices, because otherwise closing the device will affect other 3231 * children. 3232 */ 3233 if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 3234 vd->vdev_ops->vdev_op_leaf) 3235 vd->vdev_ops->vdev_op_close(vd); 3236 3237 /* 3238 * If we have brought this vdev back into service, we need 3239 * to notify fmd so that it can gracefully repair any outstanding 3240 * cases due to a missing device. We do this in all cases, even those 3241 * that probably don't correlate to a repaired fault. This is sure to 3242 * catch all cases, and we let the zfs-retire agent sort it out. If 3243 * this is a transient state it's OK, as the retire agent will 3244 * double-check the state of the vdev before repairing it. 3245 */ 3246 if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 3247 vd->vdev_prevstate != state) 3248 zfs_post_state_change(spa, vd); 3249 3250 if (vd->vdev_removed && 3251 state == VDEV_STATE_CANT_OPEN && 3252 (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 3253 /* 3254 * If the previous state is set to VDEV_STATE_REMOVED, then this 3255 * device was previously marked removed and someone attempted to 3256 * reopen it. If this failed due to a nonexistent device, then 3257 * keep the device in the REMOVED state. We also let this be if 3258 * it is one of our special test online cases, which is only 3259 * attempting to online the device and shouldn't generate an FMA 3260 * fault. 3261 */ 3262 vd->vdev_state = VDEV_STATE_REMOVED; 3263 vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 3264 } else if (state == VDEV_STATE_REMOVED) { 3265 vd->vdev_removed = B_TRUE; 3266 } else if (state == VDEV_STATE_CANT_OPEN) { 3267 /* 3268 * If we fail to open a vdev during an import or recovery, we 3269 * mark it as "not available", which signifies that it was 3270 * never there to begin with. Failure to open such a device 3271 * is not considered an error. 3272 */ 3273 if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 3274 spa_load_state(spa) == SPA_LOAD_RECOVER) && 3275 vd->vdev_ops->vdev_op_leaf) 3276 vd->vdev_not_present = 1; 3277 3278 /* 3279 * Post the appropriate ereport. If the 'prevstate' field is 3280 * set to something other than VDEV_STATE_UNKNOWN, it indicates 3281 * that this is part of a vdev_reopen(). In this case, we don't 3282 * want to post the ereport if the device was already in the 3283 * CANT_OPEN state beforehand. 3284 * 3285 * If the 'checkremove' flag is set, then this is an attempt to 3286 * online the device in response to an insertion event. If we 3287 * hit this case, then we have detected an insertion event for a 3288 * faulted or offline device that wasn't in the removed state. 3289 * In this scenario, we don't post an ereport because we are 3290 * about to replace the device, or attempt an online with 3291 * vdev_forcefault, which will generate the fault for us. 3292 */ 3293 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 3294 !vd->vdev_not_present && !vd->vdev_checkremove && 3295 vd != spa->spa_root_vdev) { 3296 const char *class; 3297 3298 switch (aux) { 3299 case VDEV_AUX_OPEN_FAILED: 3300 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 3301 break; 3302 case VDEV_AUX_CORRUPT_DATA: 3303 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 3304 break; 3305 case VDEV_AUX_NO_REPLICAS: 3306 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 3307 break; 3308 case VDEV_AUX_BAD_GUID_SUM: 3309 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 3310 break; 3311 case VDEV_AUX_TOO_SMALL: 3312 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 3313 break; 3314 case VDEV_AUX_BAD_LABEL: 3315 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 3316 break; 3317 default: 3318 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 3319 } 3320 3321 zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 3322 } 3323 3324 /* Erase any notion of persistent removed state */ 3325 vd->vdev_removed = B_FALSE; 3326 } else { 3327 vd->vdev_removed = B_FALSE; 3328 } 3329 3330 if (!isopen && vd->vdev_parent) 3331 vdev_propagate_state(vd->vdev_parent); 3332} 3333 3334/* 3335 * Check the vdev configuration to ensure that it's capable of supporting 3336 * a root pool. 3337 * 3338 * On Solaris, we do not support RAID-Z or partial configuration. In 3339 * addition, only a single top-level vdev is allowed and none of the 3340 * leaves can be wholedisks. 3341 * 3342 * For FreeBSD, we can boot from any configuration. There is a 3343 * limitation that the boot filesystem must be either uncompressed or 3344 * compresses with lzjb compression but I'm not sure how to enforce 3345 * that here. 3346 */ 3347boolean_t 3348vdev_is_bootable(vdev_t *vd) 3349{ 3350#ifdef sun 3351 if (!vd->vdev_ops->vdev_op_leaf) { 3352 char *vdev_type = vd->vdev_ops->vdev_op_type; 3353 3354 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 3355 vd->vdev_children > 1) { 3356 return (B_FALSE); 3357 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 || 3358 strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) { 3359 return (B_FALSE); 3360 } 3361 } else if (vd->vdev_wholedisk == 1) { 3362 return (B_FALSE); 3363 } 3364 3365 for (int c = 0; c < vd->vdev_children; c++) { 3366 if (!vdev_is_bootable(vd->vdev_child[c])) 3367 return (B_FALSE); 3368 } 3369#endif /* sun */ 3370 return (B_TRUE); 3371} 3372 3373/* 3374 * Load the state from the original vdev tree (ovd) which 3375 * we've retrieved from the MOS config object. If the original 3376 * vdev was offline or faulted then we transfer that state to the 3377 * device in the current vdev tree (nvd). 3378 */ 3379void 3380vdev_load_log_state(vdev_t *nvd, vdev_t *ovd) 3381{ 3382 spa_t *spa = nvd->vdev_spa; 3383 3384 ASSERT(nvd->vdev_top->vdev_islog); 3385 ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 3386 ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid); 3387 3388 for (int c = 0; c < nvd->vdev_children; c++) 3389 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]); 3390 3391 if (nvd->vdev_ops->vdev_op_leaf) { 3392 /* 3393 * Restore the persistent vdev state 3394 */ 3395 nvd->vdev_offline = ovd->vdev_offline; 3396 nvd->vdev_faulted = ovd->vdev_faulted; 3397 nvd->vdev_degraded = ovd->vdev_degraded; 3398 nvd->vdev_removed = ovd->vdev_removed; 3399 } 3400} 3401 3402/* 3403 * Determine if a log device has valid content. If the vdev was 3404 * removed or faulted in the MOS config then we know that 3405 * the content on the log device has already been written to the pool. 3406 */ 3407boolean_t 3408vdev_log_state_valid(vdev_t *vd) 3409{ 3410 if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 3411 !vd->vdev_removed) 3412 return (B_TRUE); 3413 3414 for (int c = 0; c < vd->vdev_children; c++) 3415 if (vdev_log_state_valid(vd->vdev_child[c])) 3416 return (B_TRUE); 3417 3418 return (B_FALSE); 3419} 3420 3421/* 3422 * Expand a vdev if possible. 3423 */ 3424void 3425vdev_expand(vdev_t *vd, uint64_t txg) 3426{ 3427 ASSERT(vd->vdev_top == vd); 3428 ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3429 3430 if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) { 3431 VERIFY(vdev_metaslab_init(vd, txg) == 0); 3432 vdev_config_dirty(vd); 3433 } 3434} 3435 3436/* 3437 * Split a vdev. 3438 */ 3439void 3440vdev_split(vdev_t *vd) 3441{ 3442 vdev_t *cvd, *pvd = vd->vdev_parent; 3443 3444 vdev_remove_child(pvd, vd); 3445 vdev_compact_children(pvd); 3446 3447 cvd = pvd->vdev_child[0]; 3448 if (pvd->vdev_children == 1) { 3449 vdev_remove_parent(cvd); 3450 cvd->vdev_splitting = B_TRUE; 3451 } 3452 vdev_propagate_state(cvd); 3453} 3454 3455void 3456vdev_deadman(vdev_t *vd) 3457{ 3458 for (int c = 0; c < vd->vdev_children; c++) { 3459 vdev_t *cvd = vd->vdev_child[c]; 3460 3461 vdev_deadman(cvd); 3462 } 3463 3464 if (vd->vdev_ops->vdev_op_leaf) { 3465 vdev_queue_t *vq = &vd->vdev_queue; 3466 3467 mutex_enter(&vq->vq_lock); 3468 if (avl_numnodes(&vq->vq_active_tree) > 0) { 3469 spa_t *spa = vd->vdev_spa; 3470 zio_t *fio; 3471 uint64_t delta; 3472 3473 /* 3474 * Look at the head of all the pending queues, 3475 * if any I/O has been outstanding for longer than 3476 * the spa_deadman_synctime we panic the system. 3477 */ 3478 fio = avl_first(&vq->vq_active_tree); 3479 delta = gethrtime() - fio->io_timestamp; 3480 if (delta > spa_deadman_synctime(spa)) { 3481 zfs_dbgmsg("SLOW IO: zio timestamp %lluns, " 3482 "delta %lluns, last io %lluns", 3483 fio->io_timestamp, delta, 3484 vq->vq_io_complete_ts); 3485 fm_panic("I/O to pool '%s' appears to be " 3486 "hung on vdev guid %llu at '%s'.", 3487 spa_name(spa), 3488 (long long unsigned int) vd->vdev_guid, 3489 vd->vdev_path); 3490 } 3491 } 3492 mutex_exit(&vq->vq_lock); 3493 } 3494} 3495