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