1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
  25  * Copyright (c) 2012 by Delphix. All rights reserved.
  26  */
  27 
  28 #include <sys/zfs_context.h>
  29 #include <sys/fm/fs/zfs.h>
  30 #include <sys/spa.h>
  31 #include <sys/spa_impl.h>
  32 #include <sys/dmu.h>
  33 #include <sys/dmu_tx.h>
  34 #include <sys/vdev_impl.h>
  35 #include <sys/uberblock_impl.h>
  36 #include <sys/metaslab.h>
  37 #include <sys/metaslab_impl.h>
  38 #include <sys/space_map.h>
  39 #include <sys/zio.h>
  40 #include <sys/zap.h>
  41 #include <sys/fs/zfs.h>
  42 #include <sys/arc.h>
  43 #include <sys/zil.h>
  44 #include <sys/dsl_scan.h>
  45 
  46 /*
  47  * Virtual device management.
  48  */
  49 
  50 static vdev_ops_t *vdev_ops_table[] = {
  51         &vdev_root_ops,
  52         &vdev_raidz_ops,
  53         &vdev_mirror_ops,
  54         &vdev_replacing_ops,
  55         &vdev_spare_ops,
  56         &vdev_disk_ops,
  57         &vdev_file_ops,
  58         &vdev_missing_ops,
  59         &vdev_hole_ops,
  60         NULL
  61 };
  62 
  63 /* maximum scrub/resilver I/O queue per leaf vdev */
  64 int zfs_scrub_limit = 10;
  65 
  66 /*
  67  * Given a vdev type, return the appropriate ops vector.
  68  */
  69 static vdev_ops_t *
  70 vdev_getops(const char *type)
  71 {
  72         vdev_ops_t *ops, **opspp;
  73 
  74         for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++)
  75                 if (strcmp(ops->vdev_op_type, type) == 0)
  76                         break;
  77 
  78         return (ops);
  79 }
  80 
  81 /*
  82  * Default asize function: return the MAX of psize with the asize of
  83  * all children.  This is what's used by anything other than RAID-Z.
  84  */
  85 uint64_t
  86 vdev_default_asize(vdev_t *vd, uint64_t psize)
  87 {
  88         uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
  89         uint64_t csize;
  90 
  91         for (int c = 0; c < vd->vdev_children; c++) {
  92                 csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
  93                 asize = MAX(asize, csize);
  94         }
  95 
  96         return (asize);
  97 }
  98 
  99 /*
 100  * Get the minimum allocatable size. We define the allocatable size as
 101  * the vdev's asize rounded to the nearest metaslab. This allows us to
 102  * replace or attach devices which don't have the same physical size but
 103  * can still satisfy the same number of allocations.
 104  */
 105 uint64_t
 106 vdev_get_min_asize(vdev_t *vd)
 107 {
 108         vdev_t *pvd = vd->vdev_parent;
 109 
 110         /*
 111          * If our parent is NULL (inactive spare or cache) or is the root,
 112          * just return our own asize.
 113          */
 114         if (pvd == NULL)
 115                 return (vd->vdev_asize);
 116 
 117         /*
 118          * The top-level vdev just returns the allocatable size rounded
 119          * to the nearest metaslab.
 120          */
 121         if (vd == vd->vdev_top)
 122                 return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift));
 123 
 124         /*
 125          * The allocatable space for a raidz vdev is N * sizeof(smallest child),
 126          * so each child must provide at least 1/Nth of its asize.
 127          */
 128         if (pvd->vdev_ops == &vdev_raidz_ops)
 129                 return (pvd->vdev_min_asize / pvd->vdev_children);
 130 
 131         return (pvd->vdev_min_asize);
 132 }
 133 
 134 void
 135 vdev_set_min_asize(vdev_t *vd)
 136 {
 137         vd->vdev_min_asize = vdev_get_min_asize(vd);
 138 
 139         for (int c = 0; c < vd->vdev_children; c++)
 140                 vdev_set_min_asize(vd->vdev_child[c]);
 141 }
 142 
 143 vdev_t *
 144 vdev_lookup_top(spa_t *spa, uint64_t vdev)
 145 {
 146         vdev_t *rvd = spa->spa_root_vdev;
 147 
 148         ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
 149 
 150         if (vdev < rvd->vdev_children) {
 151                 ASSERT(rvd->vdev_child[vdev] != NULL);
 152                 return (rvd->vdev_child[vdev]);
 153         }
 154 
 155         return (NULL);
 156 }
 157 
 158 vdev_t *
 159 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
 160 {
 161         vdev_t *mvd;
 162 
 163         if (vd->vdev_guid == guid)
 164                 return (vd);
 165 
 166         for (int c = 0; c < vd->vdev_children; c++)
 167                 if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
 168                     NULL)
 169                         return (mvd);
 170 
 171         return (NULL);
 172 }
 173 
 174 void
 175 vdev_add_child(vdev_t *pvd, vdev_t *cvd)
 176 {
 177         size_t oldsize, newsize;
 178         uint64_t id = cvd->vdev_id;
 179         vdev_t **newchild;
 180 
 181         ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
 182         ASSERT(cvd->vdev_parent == NULL);
 183 
 184         cvd->vdev_parent = pvd;
 185 
 186         if (pvd == NULL)
 187                 return;
 188 
 189         ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL);
 190 
 191         oldsize = pvd->vdev_children * sizeof (vdev_t *);
 192         pvd->vdev_children = MAX(pvd->vdev_children, id + 1);
 193         newsize = pvd->vdev_children * sizeof (vdev_t *);
 194 
 195         newchild = kmem_zalloc(newsize, KM_SLEEP);
 196         if (pvd->vdev_child != NULL) {
 197                 bcopy(pvd->vdev_child, newchild, oldsize);
 198                 kmem_free(pvd->vdev_child, oldsize);
 199         }
 200 
 201         pvd->vdev_child = newchild;
 202         pvd->vdev_child[id] = cvd;
 203 
 204         cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd);
 205         ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL);
 206 
 207         /*
 208          * Walk up all ancestors to update guid sum.
 209          */
 210         for (; pvd != NULL; pvd = pvd->vdev_parent)
 211                 pvd->vdev_guid_sum += cvd->vdev_guid_sum;
 212 }
 213 
 214 void
 215 vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
 216 {
 217         int c;
 218         uint_t id = cvd->vdev_id;
 219 
 220         ASSERT(cvd->vdev_parent == pvd);
 221 
 222         if (pvd == NULL)
 223                 return;
 224 
 225         ASSERT(id < pvd->vdev_children);
 226         ASSERT(pvd->vdev_child[id] == cvd);
 227 
 228         pvd->vdev_child[id] = NULL;
 229         cvd->vdev_parent = NULL;
 230 
 231         for (c = 0; c < pvd->vdev_children; c++)
 232                 if (pvd->vdev_child[c])
 233                         break;
 234 
 235         if (c == pvd->vdev_children) {
 236                 kmem_free(pvd->vdev_child, c * sizeof (vdev_t *));
 237                 pvd->vdev_child = NULL;
 238                 pvd->vdev_children = 0;
 239         }
 240 
 241         /*
 242          * Walk up all ancestors to update guid sum.
 243          */
 244         for (; pvd != NULL; pvd = pvd->vdev_parent)
 245                 pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
 246 }
 247 
 248 /*
 249  * Remove any holes in the child array.
 250  */
 251 void
 252 vdev_compact_children(vdev_t *pvd)
 253 {
 254         vdev_t **newchild, *cvd;
 255         int oldc = pvd->vdev_children;
 256         int newc;
 257 
 258         ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
 259 
 260         for (int c = newc = 0; c < oldc; c++)
 261                 if (pvd->vdev_child[c])
 262                         newc++;
 263 
 264         newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
 265 
 266         for (int c = newc = 0; c < oldc; c++) {
 267                 if ((cvd = pvd->vdev_child[c]) != NULL) {
 268                         newchild[newc] = cvd;
 269                         cvd->vdev_id = newc++;
 270                 }
 271         }
 272 
 273         kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *));
 274         pvd->vdev_child = newchild;
 275         pvd->vdev_children = newc;
 276 }
 277 
 278 /*
 279  * Allocate and minimally initialize a vdev_t.
 280  */
 281 vdev_t *
 282 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
 283 {
 284         vdev_t *vd;
 285 
 286         vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
 287 
 288         if (spa->spa_root_vdev == NULL) {
 289                 ASSERT(ops == &vdev_root_ops);
 290                 spa->spa_root_vdev = vd;
 291                 spa->spa_load_guid = spa_generate_guid(NULL);
 292         }
 293 
 294         if (guid == 0 && ops != &vdev_hole_ops) {
 295                 if (spa->spa_root_vdev == vd) {
 296                         /*
 297                          * The root vdev's guid will also be the pool guid,
 298                          * which must be unique among all pools.
 299                          */
 300                         guid = spa_generate_guid(NULL);
 301                 } else {
 302                         /*
 303                          * Any other vdev's guid must be unique within the pool.
 304                          */
 305                         guid = spa_generate_guid(spa);
 306                 }
 307                 ASSERT(!spa_guid_exists(spa_guid(spa), guid));
 308         }
 309 
 310         vd->vdev_spa = spa;
 311         vd->vdev_id = id;
 312         vd->vdev_guid = guid;
 313         vd->vdev_guid_sum = guid;
 314         vd->vdev_ops = ops;
 315         vd->vdev_state = VDEV_STATE_CLOSED;
 316         vd->vdev_ishole = (ops == &vdev_hole_ops);
 317 
 318         mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
 319         mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
 320         mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
 321         for (int t = 0; t < DTL_TYPES; t++) {
 322                 space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0,
 323                     &vd->vdev_dtl_lock);
 324         }
 325         txg_list_create(&vd->vdev_ms_list,
 326             offsetof(struct metaslab, ms_txg_node));
 327         txg_list_create(&vd->vdev_dtl_list,
 328             offsetof(struct vdev, vdev_dtl_node));
 329         vd->vdev_stat.vs_timestamp = gethrtime();
 330         vdev_queue_init(vd);
 331         vdev_cache_init(vd);
 332 
 333         return (vd);
 334 }
 335 
 336 /*
 337  * Allocate a new vdev.  The 'alloctype' is used to control whether we are
 338  * creating a new vdev or loading an existing one - the behavior is slightly
 339  * different for each case.
 340  */
 341 int
 342 vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
 343     int alloctype)
 344 {
 345         vdev_ops_t *ops;
 346         char *type;
 347         uint64_t guid = 0, islog, nparity;
 348         vdev_t *vd;
 349 
 350         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
 351 
 352         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
 353                 return (EINVAL);
 354 
 355         if ((ops = vdev_getops(type)) == NULL)
 356                 return (EINVAL);
 357 
 358         /*
 359          * If this is a load, get the vdev guid from the nvlist.
 360          * Otherwise, vdev_alloc_common() will generate one for us.
 361          */
 362         if (alloctype == VDEV_ALLOC_LOAD) {
 363                 uint64_t label_id;
 364 
 365                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) ||
 366                     label_id != id)
 367                         return (EINVAL);
 368 
 369                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
 370                         return (EINVAL);
 371         } else if (alloctype == VDEV_ALLOC_SPARE) {
 372                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
 373                         return (EINVAL);
 374         } else if (alloctype == VDEV_ALLOC_L2CACHE) {
 375                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
 376                         return (EINVAL);
 377         } else if (alloctype == VDEV_ALLOC_ROOTPOOL) {
 378                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0)
 379                         return (EINVAL);
 380         }
 381 
 382         /*
 383          * The first allocated vdev must be of type 'root'.
 384          */
 385         if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL)
 386                 return (EINVAL);
 387 
 388         /*
 389          * Determine whether we're a log vdev.
 390          */
 391         islog = 0;
 392         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog);
 393         if (islog && spa_version(spa) < SPA_VERSION_SLOGS)
 394                 return (ENOTSUP);
 395 
 396         if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES)
 397                 return (ENOTSUP);
 398 
 399         /*
 400          * Set the nparity property for RAID-Z vdevs.
 401          */
 402         nparity = -1ULL;
 403         if (ops == &vdev_raidz_ops) {
 404                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
 405                     &nparity) == 0) {
 406                         if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY)
 407                                 return (EINVAL);
 408                         /*
 409                          * Previous versions could only support 1 or 2 parity
 410                          * device.
 411                          */
 412                         if (nparity > 1 &&
 413                             spa_version(spa) < SPA_VERSION_RAIDZ2)
 414                                 return (ENOTSUP);
 415                         if (nparity > 2 &&
 416                             spa_version(spa) < SPA_VERSION_RAIDZ3)
 417                                 return (ENOTSUP);
 418                 } else {
 419                         /*
 420                          * We require the parity to be specified for SPAs that
 421                          * support multiple parity levels.
 422                          */
 423                         if (spa_version(spa) >= SPA_VERSION_RAIDZ2)
 424                                 return (EINVAL);
 425                         /*
 426                          * Otherwise, we default to 1 parity device for RAID-Z.
 427                          */
 428                         nparity = 1;
 429                 }
 430         } else {
 431                 nparity = 0;
 432         }
 433         ASSERT(nparity != -1ULL);
 434 
 435         vd = vdev_alloc_common(spa, id, guid, ops);
 436 
 437         vd->vdev_islog = islog;
 438         vd->vdev_nparity = nparity;
 439 
 440         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0)
 441                 vd->vdev_path = spa_strdup(vd->vdev_path);
 442         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0)
 443                 vd->vdev_devid = spa_strdup(vd->vdev_devid);
 444         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH,
 445             &vd->vdev_physpath) == 0)
 446                 vd->vdev_physpath = spa_strdup(vd->vdev_physpath);
 447         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0)
 448                 vd->vdev_fru = spa_strdup(vd->vdev_fru);
 449 
 450         /*
 451          * Set the whole_disk property.  If it's not specified, leave the value
 452          * as -1.
 453          */
 454         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
 455             &vd->vdev_wholedisk) != 0)
 456                 vd->vdev_wholedisk = -1ULL;
 457 
 458         /*
 459          * Look for the 'not present' flag.  This will only be set if the device
 460          * was not present at the time of import.
 461          */
 462         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
 463             &vd->vdev_not_present);
 464 
 465         /*
 466          * Get the alignment requirement.
 467          */
 468         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift);
 469 
 470         /*
 471          * Retrieve the vdev creation time.
 472          */
 473         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
 474             &vd->vdev_crtxg);
 475 
 476         /*
 477          * If we're a top-level vdev, try to load the allocation parameters.
 478          */
 479         if (parent && !parent->vdev_parent &&
 480             (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) {
 481                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
 482                     &vd->vdev_ms_array);
 483                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
 484                     &vd->vdev_ms_shift);
 485                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE,
 486                     &vd->vdev_asize);
 487                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING,
 488                     &vd->vdev_removing);
 489         }
 490 
 491         if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
 492                 ASSERT(alloctype == VDEV_ALLOC_LOAD ||
 493                     alloctype == VDEV_ALLOC_ADD ||
 494                     alloctype == VDEV_ALLOC_SPLIT ||
 495                     alloctype == VDEV_ALLOC_ROOTPOOL);
 496                 vd->vdev_mg = metaslab_group_create(islog ?
 497                     spa_log_class(spa) : spa_normal_class(spa), vd);
 498         }
 499 
 500         /*
 501          * If we're a leaf vdev, try to load the DTL object and other state.
 502          */
 503         if (vd->vdev_ops->vdev_op_leaf &&
 504             (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE ||
 505             alloctype == VDEV_ALLOC_ROOTPOOL)) {
 506                 if (alloctype == VDEV_ALLOC_LOAD) {
 507                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL,
 508                             &vd->vdev_dtl_smo.smo_object);
 509                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE,
 510                             &vd->vdev_unspare);
 511                 }
 512 
 513                 if (alloctype == VDEV_ALLOC_ROOTPOOL) {
 514                         uint64_t spare = 0;
 515 
 516                         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
 517                             &spare) == 0 && spare)
 518                                 spa_spare_add(vd);
 519                 }
 520 
 521                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
 522                     &vd->vdev_offline);
 523 
 524                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING,
 525                     &vd->vdev_resilvering);
 526 
 527                 /*
 528                  * When importing a pool, we want to ignore the persistent fault
 529                  * state, as the diagnosis made on another system may not be
 530                  * valid in the current context.  Local vdevs will
 531                  * remain in the faulted state.
 532                  */
 533                 if (spa_load_state(spa) == SPA_LOAD_OPEN) {
 534                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED,
 535                             &vd->vdev_faulted);
 536                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED,
 537                             &vd->vdev_degraded);
 538                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED,
 539                             &vd->vdev_removed);
 540 
 541                         if (vd->vdev_faulted || vd->vdev_degraded) {
 542                                 char *aux;
 543 
 544                                 vd->vdev_label_aux =
 545                                     VDEV_AUX_ERR_EXCEEDED;
 546                                 if (nvlist_lookup_string(nv,
 547                                     ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
 548                                     strcmp(aux, "external") == 0)
 549                                         vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
 550                         }
 551                 }
 552         }
 553 
 554         /*
 555          * Add ourselves to the parent's list of children.
 556          */
 557         vdev_add_child(parent, vd);
 558 
 559         *vdp = vd;
 560 
 561         return (0);
 562 }
 563 
 564 void
 565 vdev_free(vdev_t *vd)
 566 {
 567         spa_t *spa = vd->vdev_spa;
 568 
 569         /*
 570          * vdev_free() implies closing the vdev first.  This is simpler than
 571          * trying to ensure complicated semantics for all callers.
 572          */
 573         vdev_close(vd);
 574 
 575         ASSERT(!list_link_active(&vd->vdev_config_dirty_node));
 576         ASSERT(!list_link_active(&vd->vdev_state_dirty_node));
 577 
 578         /*
 579          * Free all children.
 580          */
 581         for (int c = 0; c < vd->vdev_children; c++)
 582                 vdev_free(vd->vdev_child[c]);
 583 
 584         ASSERT(vd->vdev_child == NULL);
 585         ASSERT(vd->vdev_guid_sum == vd->vdev_guid);
 586 
 587         /*
 588          * Discard allocation state.
 589          */
 590         if (vd->vdev_mg != NULL) {
 591                 vdev_metaslab_fini(vd);
 592                 metaslab_group_destroy(vd->vdev_mg);
 593         }
 594 
 595         ASSERT0(vd->vdev_stat.vs_space);
 596         ASSERT0(vd->vdev_stat.vs_dspace);
 597         ASSERT0(vd->vdev_stat.vs_alloc);
 598 
 599         /*
 600          * Remove this vdev from its parent's child list.
 601          */
 602         vdev_remove_child(vd->vdev_parent, vd);
 603 
 604         ASSERT(vd->vdev_parent == NULL);
 605 
 606         /*
 607          * Clean up vdev structure.
 608          */
 609         vdev_queue_fini(vd);
 610         vdev_cache_fini(vd);
 611 
 612         if (vd->vdev_path)
 613                 spa_strfree(vd->vdev_path);
 614         if (vd->vdev_devid)
 615                 spa_strfree(vd->vdev_devid);
 616         if (vd->vdev_physpath)
 617                 spa_strfree(vd->vdev_physpath);
 618         if (vd->vdev_fru)
 619                 spa_strfree(vd->vdev_fru);
 620 
 621         if (vd->vdev_isspare)
 622                 spa_spare_remove(vd);
 623         if (vd->vdev_isl2cache)
 624                 spa_l2cache_remove(vd);
 625 
 626         txg_list_destroy(&vd->vdev_ms_list);
 627         txg_list_destroy(&vd->vdev_dtl_list);
 628 
 629         mutex_enter(&vd->vdev_dtl_lock);
 630         for (int t = 0; t < DTL_TYPES; t++) {
 631                 space_map_unload(&vd->vdev_dtl[t]);
 632                 space_map_destroy(&vd->vdev_dtl[t]);
 633         }
 634         mutex_exit(&vd->vdev_dtl_lock);
 635 
 636         mutex_destroy(&vd->vdev_dtl_lock);
 637         mutex_destroy(&vd->vdev_stat_lock);
 638         mutex_destroy(&vd->vdev_probe_lock);
 639 
 640         if (vd == spa->spa_root_vdev)
 641                 spa->spa_root_vdev = NULL;
 642 
 643         kmem_free(vd, sizeof (vdev_t));
 644 }
 645 
 646 /*
 647  * Transfer top-level vdev state from svd to tvd.
 648  */
 649 static void
 650 vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
 651 {
 652         spa_t *spa = svd->vdev_spa;
 653         metaslab_t *msp;
 654         vdev_t *vd;
 655         int t;
 656 
 657         ASSERT(tvd == tvd->vdev_top);
 658 
 659         tvd->vdev_ms_array = svd->vdev_ms_array;
 660         tvd->vdev_ms_shift = svd->vdev_ms_shift;
 661         tvd->vdev_ms_count = svd->vdev_ms_count;
 662 
 663         svd->vdev_ms_array = 0;
 664         svd->vdev_ms_shift = 0;
 665         svd->vdev_ms_count = 0;
 666 
 667         if (tvd->vdev_mg)
 668                 ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
 669         tvd->vdev_mg = svd->vdev_mg;
 670         tvd->vdev_ms = svd->vdev_ms;
 671 
 672         svd->vdev_mg = NULL;
 673         svd->vdev_ms = NULL;
 674 
 675         if (tvd->vdev_mg != NULL)
 676                 tvd->vdev_mg->mg_vd = tvd;
 677 
 678         tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc;
 679         tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space;
 680         tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace;
 681 
 682         svd->vdev_stat.vs_alloc = 0;
 683         svd->vdev_stat.vs_space = 0;
 684         svd->vdev_stat.vs_dspace = 0;
 685 
 686         for (t = 0; t < TXG_SIZE; t++) {
 687                 while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL)
 688                         (void) txg_list_add(&tvd->vdev_ms_list, msp, t);
 689                 while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL)
 690                         (void) txg_list_add(&tvd->vdev_dtl_list, vd, t);
 691                 if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t))
 692                         (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t);
 693         }
 694 
 695         if (list_link_active(&svd->vdev_config_dirty_node)) {
 696                 vdev_config_clean(svd);
 697                 vdev_config_dirty(tvd);
 698         }
 699 
 700         if (list_link_active(&svd->vdev_state_dirty_node)) {
 701                 vdev_state_clean(svd);
 702                 vdev_state_dirty(tvd);
 703         }
 704 
 705         tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio;
 706         svd->vdev_deflate_ratio = 0;
 707 
 708         tvd->vdev_islog = svd->vdev_islog;
 709         svd->vdev_islog = 0;
 710 }
 711 
 712 static void
 713 vdev_top_update(vdev_t *tvd, vdev_t *vd)
 714 {
 715         if (vd == NULL)
 716                 return;
 717 
 718         vd->vdev_top = tvd;
 719 
 720         for (int c = 0; c < vd->vdev_children; c++)
 721                 vdev_top_update(tvd, vd->vdev_child[c]);
 722 }
 723 
 724 /*
 725  * Add a mirror/replacing vdev above an existing vdev.
 726  */
 727 vdev_t *
 728 vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops)
 729 {
 730         spa_t *spa = cvd->vdev_spa;
 731         vdev_t *pvd = cvd->vdev_parent;
 732         vdev_t *mvd;
 733 
 734         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
 735 
 736         mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops);
 737 
 738         mvd->vdev_asize = cvd->vdev_asize;
 739         mvd->vdev_min_asize = cvd->vdev_min_asize;
 740         mvd->vdev_max_asize = cvd->vdev_max_asize;
 741         mvd->vdev_ashift = cvd->vdev_ashift;
 742         mvd->vdev_state = cvd->vdev_state;
 743         mvd->vdev_crtxg = cvd->vdev_crtxg;
 744 
 745         vdev_remove_child(pvd, cvd);
 746         vdev_add_child(pvd, mvd);
 747         cvd->vdev_id = mvd->vdev_children;
 748         vdev_add_child(mvd, cvd);
 749         vdev_top_update(cvd->vdev_top, cvd->vdev_top);
 750 
 751         if (mvd == mvd->vdev_top)
 752                 vdev_top_transfer(cvd, mvd);
 753 
 754         return (mvd);
 755 }
 756 
 757 /*
 758  * Remove a 1-way mirror/replacing vdev from the tree.
 759  */
 760 void
 761 vdev_remove_parent(vdev_t *cvd)
 762 {
 763         vdev_t *mvd = cvd->vdev_parent;
 764         vdev_t *pvd = mvd->vdev_parent;
 765 
 766         ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
 767 
 768         ASSERT(mvd->vdev_children == 1);
 769         ASSERT(mvd->vdev_ops == &vdev_mirror_ops ||
 770             mvd->vdev_ops == &vdev_replacing_ops ||
 771             mvd->vdev_ops == &vdev_spare_ops);
 772         cvd->vdev_ashift = mvd->vdev_ashift;
 773 
 774         vdev_remove_child(mvd, cvd);
 775         vdev_remove_child(pvd, mvd);
 776 
 777         /*
 778          * If cvd will replace mvd as a top-level vdev, preserve mvd's guid.
 779          * Otherwise, we could have detached an offline device, and when we
 780          * go to import the pool we'll think we have two top-level vdevs,
 781          * instead of a different version of the same top-level vdev.
 782          */
 783         if (mvd->vdev_top == mvd) {
 784                 uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid;
 785                 cvd->vdev_orig_guid = cvd->vdev_guid;
 786                 cvd->vdev_guid += guid_delta;
 787                 cvd->vdev_guid_sum += guid_delta;
 788         }
 789         cvd->vdev_id = mvd->vdev_id;
 790         vdev_add_child(pvd, cvd);
 791         vdev_top_update(cvd->vdev_top, cvd->vdev_top);
 792 
 793         if (cvd == cvd->vdev_top)
 794                 vdev_top_transfer(mvd, cvd);
 795 
 796         ASSERT(mvd->vdev_children == 0);
 797         vdev_free(mvd);
 798 }
 799 
 800 int
 801 vdev_metaslab_init(vdev_t *vd, uint64_t txg)
 802 {
 803         spa_t *spa = vd->vdev_spa;
 804         objset_t *mos = spa->spa_meta_objset;
 805         uint64_t m;
 806         uint64_t oldc = vd->vdev_ms_count;
 807         uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift;
 808         metaslab_t **mspp;
 809         int error;
 810 
 811         ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER));
 812 
 813         /*
 814          * This vdev is not being allocated from yet or is a hole.
 815          */
 816         if (vd->vdev_ms_shift == 0)
 817                 return (0);
 818 
 819         ASSERT(!vd->vdev_ishole);
 820 
 821         /*
 822          * Compute the raidz-deflation ratio.  Note, we hard-code
 823          * in 128k (1 << 17) because it is the current "typical" blocksize.
 824          * Even if SPA_MAXBLOCKSIZE changes, this algorithm must never change,
 825          * or we will inconsistently account for existing bp's.
 826          */
 827         vd->vdev_deflate_ratio = (1 << 17) /
 828             (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT);
 829 
 830         ASSERT(oldc <= newc);
 831 
 832         mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
 833 
 834         if (oldc != 0) {
 835                 bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
 836                 kmem_free(vd->vdev_ms, oldc * sizeof (*mspp));
 837         }
 838 
 839         vd->vdev_ms = mspp;
 840         vd->vdev_ms_count = newc;
 841 
 842         for (m = oldc; m < newc; m++) {
 843                 space_map_obj_t smo = { 0, 0, 0 };
 844                 if (txg == 0) {
 845                         uint64_t object = 0;
 846                         error = dmu_read(mos, vd->vdev_ms_array,
 847                             m * sizeof (uint64_t), sizeof (uint64_t), &object,
 848                             DMU_READ_PREFETCH);
 849                         if (error)
 850                                 return (error);
 851                         if (object != 0) {
 852                                 dmu_buf_t *db;
 853                                 error = dmu_bonus_hold(mos, object, FTAG, &db);
 854                                 if (error)
 855                                         return (error);
 856                                 ASSERT3U(db->db_size, >=, sizeof (smo));
 857                                 bcopy(db->db_data, &smo, sizeof (smo));
 858                                 ASSERT3U(smo.smo_object, ==, object);
 859                                 dmu_buf_rele(db, FTAG);
 860                         }
 861                 }
 862                 vd->vdev_ms[m] = metaslab_init(vd->vdev_mg, &smo,
 863                     m << vd->vdev_ms_shift, 1ULL << vd->vdev_ms_shift, txg);
 864         }
 865 
 866         if (txg == 0)
 867                 spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER);
 868 
 869         /*
 870          * If the vdev is being removed we don't activate
 871          * the metaslabs since we want to ensure that no new
 872          * allocations are performed on this device.
 873          */
 874         if (oldc == 0 && !vd->vdev_removing)
 875                 metaslab_group_activate(vd->vdev_mg);
 876 
 877         if (txg == 0)
 878                 spa_config_exit(spa, SCL_ALLOC, FTAG);
 879 
 880         return (0);
 881 }
 882 
 883 void
 884 vdev_metaslab_fini(vdev_t *vd)
 885 {
 886         uint64_t m;
 887         uint64_t count = vd->vdev_ms_count;
 888 
 889         if (vd->vdev_ms != NULL) {
 890                 metaslab_group_passivate(vd->vdev_mg);
 891                 for (m = 0; m < count; m++)
 892                         if (vd->vdev_ms[m] != NULL)
 893                                 metaslab_fini(vd->vdev_ms[m]);
 894                 kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *));
 895                 vd->vdev_ms = NULL;
 896         }
 897 }
 898 
 899 typedef struct vdev_probe_stats {
 900         boolean_t       vps_readable;
 901         boolean_t       vps_writeable;
 902         int             vps_flags;
 903 } vdev_probe_stats_t;
 904 
 905 static void
 906 vdev_probe_done(zio_t *zio)
 907 {
 908         spa_t *spa = zio->io_spa;
 909         vdev_t *vd = zio->io_vd;
 910         vdev_probe_stats_t *vps = zio->io_private;
 911 
 912         ASSERT(vd->vdev_probe_zio != NULL);
 913 
 914         if (zio->io_type == ZIO_TYPE_READ) {
 915                 if (zio->io_error == 0)
 916                         vps->vps_readable = 1;
 917                 if (zio->io_error == 0 && spa_writeable(spa)) {
 918                         zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd,
 919                             zio->io_offset, zio->io_size, zio->io_data,
 920                             ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
 921                             ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE));
 922                 } else {
 923                         zio_buf_free(zio->io_data, zio->io_size);
 924                 }
 925         } else if (zio->io_type == ZIO_TYPE_WRITE) {
 926                 if (zio->io_error == 0)
 927                         vps->vps_writeable = 1;
 928                 zio_buf_free(zio->io_data, zio->io_size);
 929         } else if (zio->io_type == ZIO_TYPE_NULL) {
 930                 zio_t *pio;
 931 
 932                 vd->vdev_cant_read |= !vps->vps_readable;
 933                 vd->vdev_cant_write |= !vps->vps_writeable;
 934 
 935                 if (vdev_readable(vd) &&
 936                     (vdev_writeable(vd) || !spa_writeable(spa))) {
 937                         zio->io_error = 0;
 938                 } else {
 939                         ASSERT(zio->io_error != 0);
 940                         zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
 941                             spa, vd, NULL, 0, 0);
 942                         zio->io_error = ENXIO;
 943                 }
 944 
 945                 mutex_enter(&vd->vdev_probe_lock);
 946                 ASSERT(vd->vdev_probe_zio == zio);
 947                 vd->vdev_probe_zio = NULL;
 948                 mutex_exit(&vd->vdev_probe_lock);
 949 
 950                 while ((pio = zio_walk_parents(zio)) != NULL)
 951                         if (!vdev_accessible(vd, pio))
 952                                 pio->io_error = ENXIO;
 953 
 954                 kmem_free(vps, sizeof (*vps));
 955         }
 956 }
 957 
 958 /*
 959  * Determine whether this device is accessible by reading and writing
 960  * to several known locations: the pad regions of each vdev label
 961  * but the first (which we leave alone in case it contains a VTOC).
 962  */
 963 zio_t *
 964 vdev_probe(vdev_t *vd, zio_t *zio)
 965 {
 966         spa_t *spa = vd->vdev_spa;
 967         vdev_probe_stats_t *vps = NULL;
 968         zio_t *pio;
 969 
 970         ASSERT(vd->vdev_ops->vdev_op_leaf);
 971 
 972         /*
 973          * Don't probe the probe.
 974          */
 975         if (zio && (zio->io_flags & ZIO_FLAG_PROBE))
 976                 return (NULL);
 977 
 978         /*
 979          * To prevent 'probe storms' when a device fails, we create
 980          * just one probe i/o at a time.  All zios that want to probe
 981          * this vdev will become parents of the probe io.
 982          */
 983         mutex_enter(&vd->vdev_probe_lock);
 984 
 985         if ((pio = vd->vdev_probe_zio) == NULL) {
 986                 vps = kmem_zalloc(sizeof (*vps), KM_SLEEP);
 987 
 988                 vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE |
 989                     ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE |
 990                     ZIO_FLAG_TRYHARD;
 991 
 992                 if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) {
 993                         /*
 994                          * vdev_cant_read and vdev_cant_write can only
 995                          * transition from TRUE to FALSE when we have the
 996                          * SCL_ZIO lock as writer; otherwise they can only
 997                          * transition from FALSE to TRUE.  This ensures that
 998                          * any zio looking at these values can assume that
 999                          * failures persist for the life of the I/O.  That's
1000                          * important because when a device has intermittent
1001                          * connectivity problems, we want to ensure that
1002                          * they're ascribed to the device (ENXIO) and not
1003                          * the zio (EIO).
1004                          *
1005                          * Since we hold SCL_ZIO as writer here, clear both
1006                          * values so the probe can reevaluate from first
1007                          * principles.
1008                          */
1009                         vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER;
1010                         vd->vdev_cant_read = B_FALSE;
1011                         vd->vdev_cant_write = B_FALSE;
1012                 }
1013 
1014                 vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd,
1015                     vdev_probe_done, vps,
1016                     vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE);
1017 
1018                 /*
1019                  * We can't change the vdev state in this context, so we
1020                  * kick off an async task to do it on our behalf.
1021                  */
1022                 if (zio != NULL) {
1023                         vd->vdev_probe_wanted = B_TRUE;
1024                         spa_async_request(spa, SPA_ASYNC_PROBE);
1025                 }
1026         }
1027 
1028         if (zio != NULL)
1029                 zio_add_child(zio, pio);
1030 
1031         mutex_exit(&vd->vdev_probe_lock);
1032 
1033         if (vps == NULL) {
1034                 ASSERT(zio != NULL);
1035                 return (NULL);
1036         }
1037 
1038         for (int l = 1; l < VDEV_LABELS; l++) {
1039                 zio_nowait(zio_read_phys(pio, vd,
1040                     vdev_label_offset(vd->vdev_psize, l,
1041                     offsetof(vdev_label_t, vl_pad2)),
1042                     VDEV_PAD_SIZE, zio_buf_alloc(VDEV_PAD_SIZE),
1043                     ZIO_CHECKSUM_OFF, vdev_probe_done, vps,
1044                     ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE));
1045         }
1046 
1047         if (zio == NULL)
1048                 return (pio);
1049 
1050         zio_nowait(pio);
1051         return (NULL);
1052 }
1053 
1054 static void
1055 vdev_open_child(void *arg)
1056 {
1057         vdev_t *vd = arg;
1058 
1059         vd->vdev_open_thread = curthread;
1060         vd->vdev_open_error = vdev_open(vd);
1061         vd->vdev_open_thread = NULL;
1062 }
1063 
1064 boolean_t
1065 vdev_uses_zvols(vdev_t *vd)
1066 {
1067         if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
1068             strlen(ZVOL_DIR)) == 0)
1069                 return (B_TRUE);
1070         for (int c = 0; c < vd->vdev_children; c++)
1071                 if (vdev_uses_zvols(vd->vdev_child[c]))
1072                         return (B_TRUE);
1073         return (B_FALSE);
1074 }
1075 
1076 void
1077 vdev_open_children(vdev_t *vd)
1078 {
1079         taskq_t *tq;
1080         int children = vd->vdev_children;
1081 
1082         /*
1083          * in order to handle pools on top of zvols, do the opens
1084          * in a single thread so that the same thread holds the
1085          * spa_namespace_lock
1086          */
1087         if (vdev_uses_zvols(vd)) {
1088                 for (int c = 0; c < children; c++)
1089                         vd->vdev_child[c]->vdev_open_error =
1090                             vdev_open(vd->vdev_child[c]);
1091                 return;
1092         }
1093         tq = taskq_create("vdev_open", children, minclsyspri,
1094             children, children, TASKQ_PREPOPULATE);
1095 
1096         for (int c = 0; c < children; c++)
1097                 VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
1098                     TQ_SLEEP) != NULL);
1099 
1100         taskq_destroy(tq);
1101 }
1102 
1103 /*
1104  * Prepare a virtual device for access.
1105  */
1106 int
1107 vdev_open(vdev_t *vd)
1108 {
1109         spa_t *spa = vd->vdev_spa;
1110         int error;
1111         uint64_t osize = 0;
1112         uint64_t max_osize = 0;
1113         uint64_t asize, max_asize, psize;
1114         uint64_t ashift = 0;
1115 
1116         ASSERT(vd->vdev_open_thread == curthread ||
1117             spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1118         ASSERT(vd->vdev_state == VDEV_STATE_CLOSED ||
1119             vd->vdev_state == VDEV_STATE_CANT_OPEN ||
1120             vd->vdev_state == VDEV_STATE_OFFLINE);
1121 
1122         vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1123         vd->vdev_cant_read = B_FALSE;
1124         vd->vdev_cant_write = B_FALSE;
1125         vd->vdev_min_asize = vdev_get_min_asize(vd);
1126 
1127         /*
1128          * If this vdev is not removed, check its fault status.  If it's
1129          * faulted, bail out of the open.
1130          */
1131         if (!vd->vdev_removed && vd->vdev_faulted) {
1132                 ASSERT(vd->vdev_children == 0);
1133                 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1134                     vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1135                 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1136                     vd->vdev_label_aux);
1137                 return (ENXIO);
1138         } else if (vd->vdev_offline) {
1139                 ASSERT(vd->vdev_children == 0);
1140                 vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE);
1141                 return (ENXIO);
1142         }
1143 
1144         error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift);
1145 
1146         /*
1147          * Reset the vdev_reopening flag so that we actually close
1148          * the vdev on error.
1149          */
1150         vd->vdev_reopening = B_FALSE;
1151         if (zio_injection_enabled && error == 0)
1152                 error = zio_handle_device_injection(vd, NULL, ENXIO);
1153 
1154         if (error) {
1155                 if (vd->vdev_removed &&
1156                     vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
1157                         vd->vdev_removed = B_FALSE;
1158 
1159                 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1160                     vd->vdev_stat.vs_aux);
1161                 return (error);
1162         }
1163 
1164         vd->vdev_removed = B_FALSE;
1165 
1166         /*
1167          * Recheck the faulted flag now that we have confirmed that
1168          * the vdev is accessible.  If we're faulted, bail.
1169          */
1170         if (vd->vdev_faulted) {
1171                 ASSERT(vd->vdev_children == 0);
1172                 ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED ||
1173                     vd->vdev_label_aux == VDEV_AUX_EXTERNAL);
1174                 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1175                     vd->vdev_label_aux);
1176                 return (ENXIO);
1177         }
1178 
1179         if (vd->vdev_degraded) {
1180                 ASSERT(vd->vdev_children == 0);
1181                 vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1182                     VDEV_AUX_ERR_EXCEEDED);
1183         } else {
1184                 vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0);
1185         }
1186 
1187         /*
1188          * For hole or missing vdevs we just return success.
1189          */
1190         if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
1191                 return (0);
1192 
1193         for (int c = 0; c < vd->vdev_children; c++) {
1194                 if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
1195                         vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
1196                             VDEV_AUX_NONE);
1197                         break;
1198                 }
1199         }
1200 
1201         osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t));
1202         max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t));
1203 
1204         if (vd->vdev_children == 0) {
1205                 if (osize < SPA_MINDEVSIZE) {
1206                         vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1207                             VDEV_AUX_TOO_SMALL);
1208                         return (EOVERFLOW);
1209                 }
1210                 psize = osize;
1211                 asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE);
1212                 max_asize = max_osize - (VDEV_LABEL_START_SIZE +
1213                     VDEV_LABEL_END_SIZE);
1214         } else {
1215                 if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE -
1216                     (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) {
1217                         vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1218                             VDEV_AUX_TOO_SMALL);
1219                         return (EOVERFLOW);
1220                 }
1221                 psize = 0;
1222                 asize = osize;
1223                 max_asize = max_osize;
1224         }
1225 
1226         vd->vdev_psize = psize;
1227 
1228         /*
1229          * Make sure the allocatable size hasn't shrunk.
1230          */
1231         if (asize < vd->vdev_min_asize) {
1232                 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1233                     VDEV_AUX_BAD_LABEL);
1234                 return (EINVAL);
1235         }
1236 
1237         if (vd->vdev_asize == 0) {
1238                 /*
1239                  * This is the first-ever open, so use the computed values.
1240                  * For testing purposes, a higher ashift can be requested.
1241                  */
1242                 vd->vdev_asize = asize;
1243                 vd->vdev_max_asize = max_asize;
1244                 vd->vdev_ashift = MAX(ashift, vd->vdev_ashift);
1245         } else {
1246                 /*
1247                  * Detect if the alignment requirement has increased.
1248                  * We don't want to make the pool unavailable, just
1249                  * issue a warning instead.
1250                  */
1251                 if (ashift > vd->vdev_top->vdev_ashift &&
1252                     vd->vdev_ops->vdev_op_leaf) {
1253                         cmn_err(CE_WARN,
1254                             "Disk, '%s', has a block alignment that is "
1255                             "larger than the pool's alignment\n",
1256                             vd->vdev_path);
1257                 }
1258                 vd->vdev_max_asize = max_asize;
1259         }
1260 
1261         /*
1262          * If all children are healthy and the asize has increased,
1263          * then we've experienced dynamic LUN growth.  If automatic
1264          * expansion is enabled then use the additional space.
1265          */
1266         if (vd->vdev_state == VDEV_STATE_HEALTHY && asize > vd->vdev_asize &&
1267             (vd->vdev_expanding || spa->spa_autoexpand))
1268                 vd->vdev_asize = asize;
1269 
1270         vdev_set_min_asize(vd);
1271 
1272         /*
1273          * Ensure we can issue some IO before declaring the
1274          * vdev open for business.
1275          */
1276         if (vd->vdev_ops->vdev_op_leaf &&
1277             (error = zio_wait(vdev_probe(vd, NULL))) != 0) {
1278                 vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED,
1279                     VDEV_AUX_ERR_EXCEEDED);
1280                 return (error);
1281         }
1282 
1283         /*
1284          * If a leaf vdev has a DTL, and seems healthy, then kick off a
1285          * resilver.  But don't do this if we are doing a reopen for a scrub,
1286          * since this would just restart the scrub we are already doing.
1287          */
1288         if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen &&
1289             vdev_resilver_needed(vd, NULL, NULL))
1290                 spa_async_request(spa, SPA_ASYNC_RESILVER);
1291 
1292         return (0);
1293 }
1294 
1295 /*
1296  * Called once the vdevs are all opened, this routine validates the label
1297  * contents.  This needs to be done before vdev_load() so that we don't
1298  * inadvertently do repair I/Os to the wrong device.
1299  *
1300  * If 'strict' is false ignore the spa guid check. This is necessary because
1301  * if the machine crashed during a re-guid the new guid might have been written
1302  * to all of the vdev labels, but not the cached config. The strict check
1303  * will be performed when the pool is opened again using the mos config.
1304  *
1305  * This function will only return failure if one of the vdevs indicates that it
1306  * has since been destroyed or exported.  This is only possible if
1307  * /etc/zfs/zpool.cache was readonly at the time.  Otherwise, the vdev state
1308  * will be updated but the function will return 0.
1309  */
1310 int
1311 vdev_validate(vdev_t *vd, boolean_t strict)
1312 {
1313         spa_t *spa = vd->vdev_spa;
1314         nvlist_t *label;
1315         uint64_t guid = 0, top_guid;
1316         uint64_t state;
1317 
1318         for (int c = 0; c < vd->vdev_children; c++)
1319                 if (vdev_validate(vd->vdev_child[c], strict) != 0)
1320                         return (EBADF);
1321 
1322         /*
1323          * If the device has already failed, or was marked offline, don't do
1324          * any further validation.  Otherwise, label I/O will fail and we will
1325          * overwrite the previous state.
1326          */
1327         if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
1328                 uint64_t aux_guid = 0;
1329                 nvlist_t *nvl;
1330 
1331                 if ((label = vdev_label_read_config(vd, VDEV_BEST_LABEL)) ==
1332                     NULL) {
1333                         vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1334                             VDEV_AUX_BAD_LABEL);
1335                         return (0);
1336                 }
1337 
1338                 /*
1339                  * Determine if this vdev has been split off into another
1340                  * pool.  If so, then refuse to open it.
1341                  */
1342                 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
1343                     &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
1344                         vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1345                             VDEV_AUX_SPLIT_POOL);
1346                         nvlist_free(label);
1347                         return (0);
1348                 }
1349 
1350                 if (strict && (nvlist_lookup_uint64(label,
1351                     ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
1352                     guid != spa_guid(spa))) {
1353                         vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1354                             VDEV_AUX_CORRUPT_DATA);
1355                         nvlist_free(label);
1356                         return (0);
1357                 }
1358 
1359                 if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
1360                     != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
1361                     &aux_guid) != 0)
1362                         aux_guid = 0;
1363 
1364                 /*
1365                  * If this vdev just became a top-level vdev because its
1366                  * sibling was detached, it will have adopted the parent's
1367                  * vdev guid -- but the label may or may not be on disk yet.
1368                  * Fortunately, either version of the label will have the
1369                  * same top guid, so if we're a top-level vdev, we can
1370                  * safely compare to that instead.
1371                  *
1372                  * If we split this vdev off instead, then we also check the
1373                  * original pool's guid.  We don't want to consider the vdev
1374                  * corrupt if it is partway through a split operation.
1375                  */
1376                 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
1377                     &guid) != 0 ||
1378                     nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
1379                     &top_guid) != 0 ||
1380                     ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
1381                     (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
1382                         vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1383                             VDEV_AUX_CORRUPT_DATA);
1384                         nvlist_free(label);
1385                         return (0);
1386                 }
1387 
1388                 if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
1389                     &state) != 0) {
1390                         vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1391                             VDEV_AUX_CORRUPT_DATA);
1392                         nvlist_free(label);
1393                         return (0);
1394                 }
1395 
1396                 nvlist_free(label);
1397 
1398                 /*
1399                  * If this is a verbatim import, no need to check the
1400                  * state of the pool.
1401                  */
1402                 if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
1403                     spa_load_state(spa) == SPA_LOAD_OPEN &&
1404                     state != POOL_STATE_ACTIVE)
1405                         return (EBADF);
1406 
1407                 /*
1408                  * If we were able to open and validate a vdev that was
1409                  * previously marked permanently unavailable, clear that state
1410                  * now.
1411                  */
1412                 if (vd->vdev_not_present)
1413                         vd->vdev_not_present = 0;
1414         }
1415 
1416         return (0);
1417 }
1418 
1419 /*
1420  * Close a virtual device.
1421  */
1422 void
1423 vdev_close(vdev_t *vd)
1424 {
1425         spa_t *spa = vd->vdev_spa;
1426         vdev_t *pvd = vd->vdev_parent;
1427 
1428         ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1429 
1430         /*
1431          * If our parent is reopening, then we are as well, unless we are
1432          * going offline.
1433          */
1434         if (pvd != NULL && pvd->vdev_reopening)
1435                 vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline);
1436 
1437         vd->vdev_ops->vdev_op_close(vd);
1438 
1439         vdev_cache_purge(vd);
1440 
1441         /*
1442          * We record the previous state before we close it, so that if we are
1443          * doing a reopen(), we don't generate FMA ereports if we notice that
1444          * it's still faulted.
1445          */
1446         vd->vdev_prevstate = vd->vdev_state;
1447 
1448         if (vd->vdev_offline)
1449                 vd->vdev_state = VDEV_STATE_OFFLINE;
1450         else
1451                 vd->vdev_state = VDEV_STATE_CLOSED;
1452         vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
1453 }
1454 
1455 void
1456 vdev_hold(vdev_t *vd)
1457 {
1458         spa_t *spa = vd->vdev_spa;
1459 
1460         ASSERT(spa_is_root(spa));
1461         if (spa->spa_state == POOL_STATE_UNINITIALIZED)
1462                 return;
1463 
1464         for (int c = 0; c < vd->vdev_children; c++)
1465                 vdev_hold(vd->vdev_child[c]);
1466 
1467         if (vd->vdev_ops->vdev_op_leaf)
1468                 vd->vdev_ops->vdev_op_hold(vd);
1469 }
1470 
1471 void
1472 vdev_rele(vdev_t *vd)
1473 {
1474         spa_t *spa = vd->vdev_spa;
1475 
1476         ASSERT(spa_is_root(spa));
1477         for (int c = 0; c < vd->vdev_children; c++)
1478                 vdev_rele(vd->vdev_child[c]);
1479 
1480         if (vd->vdev_ops->vdev_op_leaf)
1481                 vd->vdev_ops->vdev_op_rele(vd);
1482 }
1483 
1484 /*
1485  * Reopen all interior vdevs and any unopened leaves.  We don't actually
1486  * reopen leaf vdevs which had previously been opened as they might deadlock
1487  * on the spa_config_lock.  Instead we only obtain the leaf's physical size.
1488  * If the leaf has never been opened then open it, as usual.
1489  */
1490 void
1491 vdev_reopen(vdev_t *vd)
1492 {
1493         spa_t *spa = vd->vdev_spa;
1494 
1495         ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1496 
1497         /* set the reopening flag unless we're taking the vdev offline */
1498         vd->vdev_reopening = !vd->vdev_offline;
1499         vdev_close(vd);
1500         (void) vdev_open(vd);
1501 
1502         /*
1503          * Call vdev_validate() here to make sure we have the same device.
1504          * Otherwise, a device with an invalid label could be successfully
1505          * opened in response to vdev_reopen().
1506          */
1507         if (vd->vdev_aux) {
1508                 (void) vdev_validate_aux(vd);
1509                 if (vdev_readable(vd) && vdev_writeable(vd) &&
1510                     vd->vdev_aux == &spa->spa_l2cache &&
1511                     !l2arc_vdev_present(vd))
1512                         l2arc_add_vdev(spa, vd);
1513         } else {
1514                 (void) vdev_validate(vd, B_TRUE);
1515         }
1516 
1517         /*
1518          * Reassess parent vdev's health.
1519          */
1520         vdev_propagate_state(vd);
1521 }
1522 
1523 int
1524 vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing)
1525 {
1526         int error;
1527 
1528         /*
1529          * Normally, partial opens (e.g. of a mirror) are allowed.
1530          * For a create, however, we want to fail the request if
1531          * there are any components we can't open.
1532          */
1533         error = vdev_open(vd);
1534 
1535         if (error || vd->vdev_state != VDEV_STATE_HEALTHY) {
1536                 vdev_close(vd);
1537                 return (error ? error : ENXIO);
1538         }
1539 
1540         /*
1541          * Recursively initialize all labels.
1542          */
1543         if ((error = vdev_label_init(vd, txg, isreplacing ?
1544             VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) {
1545                 vdev_close(vd);
1546                 return (error);
1547         }
1548 
1549         return (0);
1550 }
1551 
1552 void
1553 vdev_metaslab_set_size(vdev_t *vd)
1554 {
1555         /*
1556          * Aim for roughly 200 metaslabs per vdev.
1557          */
1558         vd->vdev_ms_shift = highbit(vd->vdev_asize / 200);
1559         vd->vdev_ms_shift = MAX(vd->vdev_ms_shift, SPA_MAXBLOCKSHIFT);
1560 }
1561 
1562 void
1563 vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
1564 {
1565         ASSERT(vd == vd->vdev_top);
1566         ASSERT(!vd->vdev_ishole);
1567         ASSERT(ISP2(flags));
1568         ASSERT(spa_writeable(vd->vdev_spa));
1569 
1570         if (flags & VDD_METASLAB)
1571                 (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
1572 
1573         if (flags & VDD_DTL)
1574                 (void) txg_list_add(&vd->vdev_dtl_list, arg, txg);
1575 
1576         (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg);
1577 }
1578 
1579 /*
1580  * DTLs.
1581  *
1582  * A vdev's DTL (dirty time log) is the set of transaction groups for which
1583  * the vdev has less than perfect replication.  There are four kinds of DTL:
1584  *
1585  * DTL_MISSING: txgs for which the vdev has no valid copies of the data
1586  *
1587  * DTL_PARTIAL: txgs for which data is available, but not fully replicated
1588  *
1589  * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon
1590  *      scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of
1591  *      txgs that was scrubbed.
1592  *
1593  * DTL_OUTAGE: txgs which cannot currently be read, whether due to
1594  *      persistent errors or just some device being offline.
1595  *      Unlike the other three, the DTL_OUTAGE map is not generally
1596  *      maintained; it's only computed when needed, typically to
1597  *      determine whether a device can be detached.
1598  *
1599  * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device
1600  * either has the data or it doesn't.
1601  *
1602  * For interior vdevs such as mirror and RAID-Z the picture is more complex.
1603  * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because
1604  * if any child is less than fully replicated, then so is its parent.
1605  * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs,
1606  * comprising only those txgs which appear in 'maxfaults' or more children;
1607  * those are the txgs we don't have enough replication to read.  For example,
1608  * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2);
1609  * thus, its DTL_MISSING consists of the set of txgs that appear in more than
1610  * two child DTL_MISSING maps.
1611  *
1612  * It should be clear from the above that to compute the DTLs and outage maps
1613  * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps.
1614  * Therefore, that is all we keep on disk.  When loading the pool, or after
1615  * a configuration change, we generate all other DTLs from first principles.
1616  */
1617 void
1618 vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1619 {
1620         space_map_t *sm = &vd->vdev_dtl[t];
1621 
1622         ASSERT(t < DTL_TYPES);
1623         ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1624         ASSERT(spa_writeable(vd->vdev_spa));
1625 
1626         mutex_enter(sm->sm_lock);
1627         if (!space_map_contains(sm, txg, size))
1628                 space_map_add(sm, txg, size);
1629         mutex_exit(sm->sm_lock);
1630 }
1631 
1632 boolean_t
1633 vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
1634 {
1635         space_map_t *sm = &vd->vdev_dtl[t];
1636         boolean_t dirty = B_FALSE;
1637 
1638         ASSERT(t < DTL_TYPES);
1639         ASSERT(vd != vd->vdev_spa->spa_root_vdev);
1640 
1641         mutex_enter(sm->sm_lock);
1642         if (sm->sm_space != 0)
1643                 dirty = space_map_contains(sm, txg, size);
1644         mutex_exit(sm->sm_lock);
1645 
1646         return (dirty);
1647 }
1648 
1649 boolean_t
1650 vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t)
1651 {
1652         space_map_t *sm = &vd->vdev_dtl[t];
1653         boolean_t empty;
1654 
1655         mutex_enter(sm->sm_lock);
1656         empty = (sm->sm_space == 0);
1657         mutex_exit(sm->sm_lock);
1658 
1659         return (empty);
1660 }
1661 
1662 /*
1663  * Reassess DTLs after a config change or scrub completion.
1664  */
1665 void
1666 vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
1667 {
1668         spa_t *spa = vd->vdev_spa;
1669         avl_tree_t reftree;
1670         int minref;
1671 
1672         ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
1673 
1674         for (int c = 0; c < vd->vdev_children; c++)
1675                 vdev_dtl_reassess(vd->vdev_child[c], txg,
1676                     scrub_txg, scrub_done);
1677 
1678         if (vd == spa->spa_root_vdev || vd->vdev_ishole || vd->vdev_aux)
1679                 return;
1680 
1681         if (vd->vdev_ops->vdev_op_leaf) {
1682                 dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan;
1683 
1684                 mutex_enter(&vd->vdev_dtl_lock);
1685                 if (scrub_txg != 0 &&
1686                     (spa->spa_scrub_started ||
1687                     (scn && scn->scn_phys.scn_errors == 0))) {
1688                         /*
1689                          * We completed a scrub up to scrub_txg.  If we
1690                          * did it without rebooting, then the scrub dtl
1691                          * will be valid, so excise the old region and
1692                          * fold in the scrub dtl.  Otherwise, leave the
1693                          * dtl as-is if there was an error.
1694                          *
1695                          * There's little trick here: to excise the beginning
1696                          * of the DTL_MISSING map, we put it into a reference
1697                          * tree and then add a segment with refcnt -1 that
1698                          * covers the range [0, scrub_txg).  This means
1699                          * that each txg in that range has refcnt -1 or 0.
1700                          * We then add DTL_SCRUB with a refcnt of 2, so that
1701                          * entries in the range [0, scrub_txg) will have a
1702                          * positive refcnt -- either 1 or 2.  We then convert
1703                          * the reference tree into the new DTL_MISSING map.
1704                          */
1705                         space_map_ref_create(&reftree);
1706                         space_map_ref_add_map(&reftree,
1707                             &vd->vdev_dtl[DTL_MISSING], 1);
1708                         space_map_ref_add_seg(&reftree, 0, scrub_txg, -1);
1709                         space_map_ref_add_map(&reftree,
1710                             &vd->vdev_dtl[DTL_SCRUB], 2);
1711                         space_map_ref_generate_map(&reftree,
1712                             &vd->vdev_dtl[DTL_MISSING], 1);
1713                         space_map_ref_destroy(&reftree);
1714                 }
1715                 space_map_vacate(&vd->vdev_dtl[DTL_PARTIAL], NULL, NULL);
1716                 space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1717                     space_map_add, &vd->vdev_dtl[DTL_PARTIAL]);
1718                 if (scrub_done)
1719                         space_map_vacate(&vd->vdev_dtl[DTL_SCRUB], NULL, NULL);
1720                 space_map_vacate(&vd->vdev_dtl[DTL_OUTAGE], NULL, NULL);
1721                 if (!vdev_readable(vd))
1722                         space_map_add(&vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL);
1723                 else
1724                         space_map_walk(&vd->vdev_dtl[DTL_MISSING],
1725                             space_map_add, &vd->vdev_dtl[DTL_OUTAGE]);
1726                 mutex_exit(&vd->vdev_dtl_lock);
1727 
1728                 if (txg != 0)
1729                         vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg);
1730                 return;
1731         }
1732 
1733         mutex_enter(&vd->vdev_dtl_lock);
1734         for (int t = 0; t < DTL_TYPES; t++) {
1735                 /* account for child's outage in parent's missing map */
1736                 int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
1737                 if (t == DTL_SCRUB)
1738                         continue;                       /* leaf vdevs only */
1739                 if (t == DTL_PARTIAL)
1740                         minref = 1;                     /* i.e. non-zero */
1741                 else if (vd->vdev_nparity != 0)
1742                         minref = vd->vdev_nparity + 1;       /* RAID-Z */
1743                 else
1744                         minref = vd->vdev_children;  /* any kind of mirror */
1745                 space_map_ref_create(&reftree);
1746                 for (int c = 0; c < vd->vdev_children; c++) {
1747                         vdev_t *cvd = vd->vdev_child[c];
1748                         mutex_enter(&cvd->vdev_dtl_lock);
1749                         space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
1750                         mutex_exit(&cvd->vdev_dtl_lock);
1751                 }
1752                 space_map_ref_generate_map(&reftree, &vd->vdev_dtl[t], minref);
1753                 space_map_ref_destroy(&reftree);
1754         }
1755         mutex_exit(&vd->vdev_dtl_lock);
1756 }
1757 
1758 static int
1759 vdev_dtl_load(vdev_t *vd)
1760 {
1761         spa_t *spa = vd->vdev_spa;
1762         space_map_obj_t *smo = &vd->vdev_dtl_smo;
1763         objset_t *mos = spa->spa_meta_objset;
1764         dmu_buf_t *db;
1765         int error;
1766 
1767         ASSERT(vd->vdev_children == 0);
1768 
1769         if (smo->smo_object == 0)
1770                 return (0);
1771 
1772         ASSERT(!vd->vdev_ishole);
1773 
1774         if ((error = dmu_bonus_hold(mos, smo->smo_object, FTAG, &db)) != 0)
1775                 return (error);
1776 
1777         ASSERT3U(db->db_size, >=, sizeof (*smo));
1778         bcopy(db->db_data, smo, sizeof (*smo));
1779         dmu_buf_rele(db, FTAG);
1780 
1781         mutex_enter(&vd->vdev_dtl_lock);
1782         error = space_map_load(&vd->vdev_dtl[DTL_MISSING],
1783             NULL, SM_ALLOC, smo, mos);
1784         mutex_exit(&vd->vdev_dtl_lock);
1785 
1786         return (error);
1787 }
1788 
1789 void
1790 vdev_dtl_sync(vdev_t *vd, uint64_t txg)
1791 {
1792         spa_t *spa = vd->vdev_spa;
1793         space_map_obj_t *smo = &vd->vdev_dtl_smo;
1794         space_map_t *sm = &vd->vdev_dtl[DTL_MISSING];
1795         objset_t *mos = spa->spa_meta_objset;
1796         space_map_t smsync;
1797         kmutex_t smlock;
1798         dmu_buf_t *db;
1799         dmu_tx_t *tx;
1800 
1801         ASSERT(!vd->vdev_ishole);
1802 
1803         tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
1804 
1805         if (vd->vdev_detached) {
1806                 if (smo->smo_object != 0) {
1807                         int err = dmu_object_free(mos, smo->smo_object, tx);
1808                         ASSERT0(err);
1809                         smo->smo_object = 0;
1810                 }
1811                 dmu_tx_commit(tx);
1812                 return;
1813         }
1814 
1815         if (smo->smo_object == 0) {
1816                 ASSERT(smo->smo_objsize == 0);
1817                 ASSERT(smo->smo_alloc == 0);
1818                 smo->smo_object = dmu_object_alloc(mos,
1819                     DMU_OT_SPACE_MAP, 1 << SPACE_MAP_BLOCKSHIFT,
1820                     DMU_OT_SPACE_MAP_HEADER, sizeof (*smo), tx);
1821                 ASSERT(smo->smo_object != 0);
1822                 vdev_config_dirty(vd->vdev_top);
1823         }
1824 
1825         mutex_init(&smlock, NULL, MUTEX_DEFAULT, NULL);
1826 
1827         space_map_create(&smsync, sm->sm_start, sm->sm_size, sm->sm_shift,
1828             &smlock);
1829 
1830         mutex_enter(&smlock);
1831 
1832         mutex_enter(&vd->vdev_dtl_lock);
1833         space_map_walk(sm, space_map_add, &smsync);
1834         mutex_exit(&vd->vdev_dtl_lock);
1835 
1836         space_map_truncate(smo, mos, tx);
1837         space_map_sync(&smsync, SM_ALLOC, smo, mos, tx);
1838 
1839         space_map_destroy(&smsync);
1840 
1841         mutex_exit(&smlock);
1842         mutex_destroy(&smlock);
1843 
1844         VERIFY(0 == dmu_bonus_hold(mos, smo->smo_object, FTAG, &db));
1845         dmu_buf_will_dirty(db, tx);
1846         ASSERT3U(db->db_size, >=, sizeof (*smo));
1847         bcopy(smo, db->db_data, sizeof (*smo));
1848         dmu_buf_rele(db, FTAG);
1849 
1850         dmu_tx_commit(tx);
1851 }
1852 
1853 /*
1854  * Determine whether the specified vdev can be offlined/detached/removed
1855  * without losing data.
1856  */
1857 boolean_t
1858 vdev_dtl_required(vdev_t *vd)
1859 {
1860         spa_t *spa = vd->vdev_spa;
1861         vdev_t *tvd = vd->vdev_top;
1862         uint8_t cant_read = vd->vdev_cant_read;
1863         boolean_t required;
1864 
1865         ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
1866 
1867         if (vd == spa->spa_root_vdev || vd == tvd)
1868                 return (B_TRUE);
1869 
1870         /*
1871          * Temporarily mark the device as unreadable, and then determine
1872          * whether this results in any DTL outages in the top-level vdev.
1873          * If not, we can safely offline/detach/remove the device.
1874          */
1875         vd->vdev_cant_read = B_TRUE;
1876         vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1877         required = !vdev_dtl_empty(tvd, DTL_OUTAGE);
1878         vd->vdev_cant_read = cant_read;
1879         vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
1880 
1881         if (!required && zio_injection_enabled)
1882                 required = !!zio_handle_device_injection(vd, NULL, ECHILD);
1883 
1884         return (required);
1885 }
1886 
1887 /*
1888  * Determine if resilver is needed, and if so the txg range.
1889  */
1890 boolean_t
1891 vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
1892 {
1893         boolean_t needed = B_FALSE;
1894         uint64_t thismin = UINT64_MAX;
1895         uint64_t thismax = 0;
1896 
1897         if (vd->vdev_children == 0) {
1898                 mutex_enter(&vd->vdev_dtl_lock);
1899                 if (vd->vdev_dtl[DTL_MISSING].sm_space != 0 &&
1900                     vdev_writeable(vd)) {
1901                         space_seg_t *ss;
1902 
1903                         ss = avl_first(&vd->vdev_dtl[DTL_MISSING].sm_root);
1904                         thismin = ss->ss_start - 1;
1905                         ss = avl_last(&vd->vdev_dtl[DTL_MISSING].sm_root);
1906                         thismax = ss->ss_end;
1907                         needed = B_TRUE;
1908                 }
1909                 mutex_exit(&vd->vdev_dtl_lock);
1910         } else {
1911                 for (int c = 0; c < vd->vdev_children; c++) {
1912                         vdev_t *cvd = vd->vdev_child[c];
1913                         uint64_t cmin, cmax;
1914 
1915                         if (vdev_resilver_needed(cvd, &cmin, &cmax)) {
1916                                 thismin = MIN(thismin, cmin);
1917                                 thismax = MAX(thismax, cmax);
1918                                 needed = B_TRUE;
1919                         }
1920                 }
1921         }
1922 
1923         if (needed && minp) {
1924                 *minp = thismin;
1925                 *maxp = thismax;
1926         }
1927         return (needed);
1928 }
1929 
1930 void
1931 vdev_load(vdev_t *vd)
1932 {
1933         /*
1934          * Recursively load all children.
1935          */
1936         for (int c = 0; c < vd->vdev_children; c++)
1937                 vdev_load(vd->vdev_child[c]);
1938 
1939         /*
1940          * If this is a top-level vdev, initialize its metaslabs.
1941          */
1942         if (vd == vd->vdev_top && !vd->vdev_ishole &&
1943             (vd->vdev_ashift == 0 || vd->vdev_asize == 0 ||
1944             vdev_metaslab_init(vd, 0) != 0))
1945                 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1946                     VDEV_AUX_CORRUPT_DATA);
1947 
1948         /*
1949          * If this is a leaf vdev, load its DTL.
1950          */
1951         if (vd->vdev_ops->vdev_op_leaf && vdev_dtl_load(vd) != 0)
1952                 vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
1953                     VDEV_AUX_CORRUPT_DATA);
1954 }
1955 
1956 /*
1957  * The special vdev case is used for hot spares and l2cache devices.  Its
1958  * sole purpose it to set the vdev state for the associated vdev.  To do this,
1959  * we make sure that we can open the underlying device, then try to read the
1960  * label, and make sure that the label is sane and that it hasn't been
1961  * repurposed to another pool.
1962  */
1963 int
1964 vdev_validate_aux(vdev_t *vd)
1965 {
1966         nvlist_t *label;
1967         uint64_t guid, version;
1968         uint64_t state;
1969 
1970         if (!vdev_readable(vd))
1971                 return (0);
1972 
1973         if ((label = vdev_label_read_config(vd, VDEV_BEST_LABEL)) == NULL) {
1974                 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1975                     VDEV_AUX_CORRUPT_DATA);
1976                 return (-1);
1977         }
1978 
1979         if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 ||
1980             !SPA_VERSION_IS_SUPPORTED(version) ||
1981             nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 ||
1982             guid != vd->vdev_guid ||
1983             nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) {
1984                 vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
1985                     VDEV_AUX_CORRUPT_DATA);
1986                 nvlist_free(label);
1987                 return (-1);
1988         }
1989 
1990         /*
1991          * We don't actually check the pool state here.  If it's in fact in
1992          * use by another pool, we update this fact on the fly when requested.
1993          */
1994         nvlist_free(label);
1995         return (0);
1996 }
1997 
1998 void
1999 vdev_remove(vdev_t *vd, uint64_t txg)
2000 {
2001         spa_t *spa = vd->vdev_spa;
2002         objset_t *mos = spa->spa_meta_objset;
2003         dmu_tx_t *tx;
2004 
2005         tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
2006 
2007         if (vd->vdev_dtl_smo.smo_object) {
2008                 ASSERT0(vd->vdev_dtl_smo.smo_alloc);
2009                 (void) dmu_object_free(mos, vd->vdev_dtl_smo.smo_object, tx);
2010                 vd->vdev_dtl_smo.smo_object = 0;
2011         }
2012 
2013         if (vd->vdev_ms != NULL) {
2014                 for (int m = 0; m < vd->vdev_ms_count; m++) {
2015                         metaslab_t *msp = vd->vdev_ms[m];
2016 
2017                         if (msp == NULL || msp->ms_smo.smo_object == 0)
2018                                 continue;
2019 
2020                         ASSERT0(msp->ms_smo.smo_alloc);
2021                         (void) dmu_object_free(mos, msp->ms_smo.smo_object, tx);
2022                         msp->ms_smo.smo_object = 0;
2023                 }
2024         }
2025 
2026         if (vd->vdev_ms_array) {
2027                 (void) dmu_object_free(mos, vd->vdev_ms_array, tx);
2028                 vd->vdev_ms_array = 0;
2029                 vd->vdev_ms_shift = 0;
2030         }
2031         dmu_tx_commit(tx);
2032 }
2033 
2034 void
2035 vdev_sync_done(vdev_t *vd, uint64_t txg)
2036 {
2037         metaslab_t *msp;
2038         boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg));
2039 
2040         ASSERT(!vd->vdev_ishole);
2041 
2042         while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
2043                 metaslab_sync_done(msp, txg);
2044 
2045         if (reassess)
2046                 metaslab_sync_reassess(vd->vdev_mg);
2047 }
2048 
2049 void
2050 vdev_sync(vdev_t *vd, uint64_t txg)
2051 {
2052         spa_t *spa = vd->vdev_spa;
2053         vdev_t *lvd;
2054         metaslab_t *msp;
2055         dmu_tx_t *tx;
2056 
2057         ASSERT(!vd->vdev_ishole);
2058 
2059         if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0) {
2060                 ASSERT(vd == vd->vdev_top);
2061                 tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg);
2062                 vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset,
2063                     DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx);
2064                 ASSERT(vd->vdev_ms_array != 0);
2065                 vdev_config_dirty(vd);
2066                 dmu_tx_commit(tx);
2067         }
2068 
2069         /*
2070          * Remove the metadata associated with this vdev once it's empty.
2071          */
2072         if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing)
2073                 vdev_remove(vd, txg);
2074 
2075         while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) {
2076                 metaslab_sync(msp, txg);
2077                 (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg));
2078         }
2079 
2080         while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL)
2081                 vdev_dtl_sync(lvd, txg);
2082 
2083         (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg));
2084 }
2085 
2086 uint64_t
2087 vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
2088 {
2089         return (vd->vdev_ops->vdev_op_asize(vd, psize));
2090 }
2091 
2092 /*
2093  * Mark the given vdev faulted.  A faulted vdev behaves as if the device could
2094  * not be opened, and no I/O is attempted.
2095  */
2096 int
2097 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2098 {
2099         vdev_t *vd, *tvd;
2100 
2101         spa_vdev_state_enter(spa, SCL_NONE);
2102 
2103         if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2104                 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2105 
2106         if (!vd->vdev_ops->vdev_op_leaf)
2107                 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2108 
2109         tvd = vd->vdev_top;
2110 
2111         /*
2112          * We don't directly use the aux state here, but if we do a
2113          * vdev_reopen(), we need this value to be present to remember why we
2114          * were faulted.
2115          */
2116         vd->vdev_label_aux = aux;
2117 
2118         /*
2119          * Faulted state takes precedence over degraded.
2120          */
2121         vd->vdev_delayed_close = B_FALSE;
2122         vd->vdev_faulted = 1ULL;
2123         vd->vdev_degraded = 0ULL;
2124         vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux);
2125 
2126         /*
2127          * If this device has the only valid copy of the data, then
2128          * back off and simply mark the vdev as degraded instead.
2129          */
2130         if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
2131                 vd->vdev_degraded = 1ULL;
2132                 vd->vdev_faulted = 0ULL;
2133 
2134                 /*
2135                  * If we reopen the device and it's not dead, only then do we
2136                  * mark it degraded.
2137                  */
2138                 vdev_reopen(tvd);
2139 
2140                 if (vdev_readable(vd))
2141                         vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
2142         }
2143 
2144         return (spa_vdev_state_exit(spa, vd, 0));
2145 }
2146 
2147 /*
2148  * Mark the given vdev degraded.  A degraded vdev is purely an indication to the
2149  * user that something is wrong.  The vdev continues to operate as normal as far
2150  * as I/O is concerned.
2151  */
2152 int
2153 vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux)
2154 {
2155         vdev_t *vd;
2156 
2157         spa_vdev_state_enter(spa, SCL_NONE);
2158 
2159         if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2160                 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2161 
2162         if (!vd->vdev_ops->vdev_op_leaf)
2163                 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2164 
2165         /*
2166          * If the vdev is already faulted, then don't do anything.
2167          */
2168         if (vd->vdev_faulted || vd->vdev_degraded)
2169                 return (spa_vdev_state_exit(spa, NULL, 0));
2170 
2171         vd->vdev_degraded = 1ULL;
2172         if (!vdev_is_dead(vd))
2173                 vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED,
2174                     aux);
2175 
2176         return (spa_vdev_state_exit(spa, vd, 0));
2177 }
2178 
2179 /*
2180  * Online the given vdev.  If 'unspare' is set, it implies two things.  First,
2181  * any attached spare device should be detached when the device finishes
2182  * resilvering.  Second, the online should be treated like a 'test' online case,
2183  * so no FMA events are generated if the device fails to open.
2184  */
2185 int
2186 vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate)
2187 {
2188         vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev;
2189 
2190         spa_vdev_state_enter(spa, SCL_NONE);
2191 
2192         if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2193                 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2194 
2195         if (!vd->vdev_ops->vdev_op_leaf)
2196                 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2197 
2198         tvd = vd->vdev_top;
2199         vd->vdev_offline = B_FALSE;
2200         vd->vdev_tmpoffline = B_FALSE;
2201         vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE);
2202         vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT);
2203 
2204         /* XXX - L2ARC 1.0 does not support expansion */
2205         if (!vd->vdev_aux) {
2206                 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2207                         pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND);
2208         }
2209 
2210         vdev_reopen(tvd);
2211         vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE;
2212 
2213         if (!vd->vdev_aux) {
2214                 for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2215                         pvd->vdev_expanding = B_FALSE;
2216         }
2217 
2218         if (newstate)
2219                 *newstate = vd->vdev_state;
2220         if ((flags & ZFS_ONLINE_UNSPARE) &&
2221             !vdev_is_dead(vd) && vd->vdev_parent &&
2222             vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2223             vd->vdev_parent->vdev_child[0] == vd)
2224                 vd->vdev_unspare = B_TRUE;
2225 
2226         if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) {
2227 
2228                 /* XXX - L2ARC 1.0 does not support expansion */
2229                 if (vd->vdev_aux)
2230                         return (spa_vdev_state_exit(spa, vd, ENOTSUP));
2231                 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2232         }
2233         return (spa_vdev_state_exit(spa, vd, 0));
2234 }
2235 
2236 static int
2237 vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags)
2238 {
2239         vdev_t *vd, *tvd;
2240         int error = 0;
2241         uint64_t generation;
2242         metaslab_group_t *mg;
2243 
2244 top:
2245         spa_vdev_state_enter(spa, SCL_ALLOC);
2246 
2247         if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
2248                 return (spa_vdev_state_exit(spa, NULL, ENODEV));
2249 
2250         if (!vd->vdev_ops->vdev_op_leaf)
2251                 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
2252 
2253         tvd = vd->vdev_top;
2254         mg = tvd->vdev_mg;
2255         generation = spa->spa_config_generation + 1;
2256 
2257         /*
2258          * If the device isn't already offline, try to offline it.
2259          */
2260         if (!vd->vdev_offline) {
2261                 /*
2262                  * If this device has the only valid copy of some data,
2263                  * don't allow it to be offlined. Log devices are always
2264                  * expendable.
2265                  */
2266                 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2267                     vdev_dtl_required(vd))
2268                         return (spa_vdev_state_exit(spa, NULL, EBUSY));
2269 
2270                 /*
2271                  * If the top-level is a slog and it has had allocations
2272                  * then proceed.  We check that the vdev's metaslab group
2273                  * is not NULL since it's possible that we may have just
2274                  * added this vdev but not yet initialized its metaslabs.
2275                  */
2276                 if (tvd->vdev_islog && mg != NULL) {
2277                         /*
2278                          * Prevent any future allocations.
2279                          */
2280                         metaslab_group_passivate(mg);
2281                         (void) spa_vdev_state_exit(spa, vd, 0);
2282 
2283                         error = spa_offline_log(spa);
2284 
2285                         spa_vdev_state_enter(spa, SCL_ALLOC);
2286 
2287                         /*
2288                          * Check to see if the config has changed.
2289                          */
2290                         if (error || generation != spa->spa_config_generation) {
2291                                 metaslab_group_activate(mg);
2292                                 if (error)
2293                                         return (spa_vdev_state_exit(spa,
2294                                             vd, error));
2295                                 (void) spa_vdev_state_exit(spa, vd, 0);
2296                                 goto top;
2297                         }
2298                         ASSERT0(tvd->vdev_stat.vs_alloc);
2299                 }
2300 
2301                 /*
2302                  * Offline this device and reopen its top-level vdev.
2303                  * If the top-level vdev is a log device then just offline
2304                  * it. Otherwise, if this action results in the top-level
2305                  * vdev becoming unusable, undo it and fail the request.
2306                  */
2307                 vd->vdev_offline = B_TRUE;
2308                 vdev_reopen(tvd);
2309 
2310                 if (!tvd->vdev_islog && vd->vdev_aux == NULL &&
2311                     vdev_is_dead(tvd)) {
2312                         vd->vdev_offline = B_FALSE;
2313                         vdev_reopen(tvd);
2314                         return (spa_vdev_state_exit(spa, NULL, EBUSY));
2315                 }
2316 
2317                 /*
2318                  * Add the device back into the metaslab rotor so that
2319                  * once we online the device it's open for business.
2320                  */
2321                 if (tvd->vdev_islog && mg != NULL)
2322                         metaslab_group_activate(mg);
2323         }
2324 
2325         vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY);
2326 
2327         return (spa_vdev_state_exit(spa, vd, 0));
2328 }
2329 
2330 int
2331 vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags)
2332 {
2333         int error;
2334 
2335         mutex_enter(&spa->spa_vdev_top_lock);
2336         error = vdev_offline_locked(spa, guid, flags);
2337         mutex_exit(&spa->spa_vdev_top_lock);
2338 
2339         return (error);
2340 }
2341 
2342 /*
2343  * Clear the error counts associated with this vdev.  Unlike vdev_online() and
2344  * vdev_offline(), we assume the spa config is locked.  We also clear all
2345  * children.  If 'vd' is NULL, then the user wants to clear all vdevs.
2346  */
2347 void
2348 vdev_clear(spa_t *spa, vdev_t *vd)
2349 {
2350         vdev_t *rvd = spa->spa_root_vdev;
2351 
2352         ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
2353 
2354         if (vd == NULL)
2355                 vd = rvd;
2356 
2357         vd->vdev_stat.vs_read_errors = 0;
2358         vd->vdev_stat.vs_write_errors = 0;
2359         vd->vdev_stat.vs_checksum_errors = 0;
2360 
2361         for (int c = 0; c < vd->vdev_children; c++)
2362                 vdev_clear(spa, vd->vdev_child[c]);
2363 
2364         /*
2365          * If we're in the FAULTED state or have experienced failed I/O, then
2366          * clear the persistent state and attempt to reopen the device.  We
2367          * also mark the vdev config dirty, so that the new faulted state is
2368          * written out to disk.
2369          */
2370         if (vd->vdev_faulted || vd->vdev_degraded ||
2371             !vdev_readable(vd) || !vdev_writeable(vd)) {
2372 
2373                 /*
2374                  * When reopening in reponse to a clear event, it may be due to
2375                  * a fmadm repair request.  In this case, if the device is
2376                  * still broken, we want to still post the ereport again.
2377                  */
2378                 vd->vdev_forcefault = B_TRUE;
2379 
2380                 vd->vdev_faulted = vd->vdev_degraded = 0ULL;
2381                 vd->vdev_cant_read = B_FALSE;
2382                 vd->vdev_cant_write = B_FALSE;
2383 
2384                 vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
2385 
2386                 vd->vdev_forcefault = B_FALSE;
2387 
2388                 if (vd != rvd && vdev_writeable(vd->vdev_top))
2389                         vdev_state_dirty(vd->vdev_top);
2390 
2391                 if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
2392                         spa_async_request(spa, SPA_ASYNC_RESILVER);
2393 
2394                 spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
2395         }
2396 
2397         /*
2398          * When clearing a FMA-diagnosed fault, we always want to
2399          * unspare the device, as we assume that the original spare was
2400          * done in response to the FMA fault.
2401          */
2402         if (!vdev_is_dead(vd) && vd->vdev_parent != NULL &&
2403             vd->vdev_parent->vdev_ops == &vdev_spare_ops &&
2404             vd->vdev_parent->vdev_child[0] == vd)
2405                 vd->vdev_unspare = B_TRUE;
2406 }
2407 
2408 boolean_t
2409 vdev_is_dead(vdev_t *vd)
2410 {
2411         /*
2412          * Holes and missing devices are always considered "dead".
2413          * This simplifies the code since we don't have to check for
2414          * these types of devices in the various code paths.
2415          * Instead we rely on the fact that we skip over dead devices
2416          * before issuing I/O to them.
2417          */
2418         return (vd->vdev_state < VDEV_STATE_DEGRADED || vd->vdev_ishole ||
2419             vd->vdev_ops == &vdev_missing_ops);
2420 }
2421 
2422 boolean_t
2423 vdev_readable(vdev_t *vd)
2424 {
2425         return (!vdev_is_dead(vd) && !vd->vdev_cant_read);
2426 }
2427 
2428 boolean_t
2429 vdev_writeable(vdev_t *vd)
2430 {
2431         return (!vdev_is_dead(vd) && !vd->vdev_cant_write);
2432 }
2433 
2434 boolean_t
2435 vdev_allocatable(vdev_t *vd)
2436 {
2437         uint64_t state = vd->vdev_state;
2438 
2439         /*
2440          * We currently allow allocations from vdevs which may be in the
2441          * process of reopening (i.e. VDEV_STATE_CLOSED). If the device
2442          * fails to reopen then we'll catch it later when we're holding
2443          * the proper locks.  Note that we have to get the vdev state
2444          * in a local variable because although it changes atomically,
2445          * we're asking two separate questions about it.
2446          */
2447         return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) &&
2448             !vd->vdev_cant_write && !vd->vdev_ishole);
2449 }
2450 
2451 boolean_t
2452 vdev_accessible(vdev_t *vd, zio_t *zio)
2453 {
2454         ASSERT(zio->io_vd == vd);
2455 
2456         if (vdev_is_dead(vd) || vd->vdev_remove_wanted)
2457                 return (B_FALSE);
2458 
2459         if (zio->io_type == ZIO_TYPE_READ)
2460                 return (!vd->vdev_cant_read);
2461 
2462         if (zio->io_type == ZIO_TYPE_WRITE)
2463                 return (!vd->vdev_cant_write);
2464 
2465         return (B_TRUE);
2466 }
2467 
2468 /*
2469  * Get statistics for the given vdev.
2470  */
2471 void
2472 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
2473 {
2474         vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
2475 
2476         mutex_enter(&vd->vdev_stat_lock);
2477         bcopy(&vd->vdev_stat, vs, sizeof (*vs));
2478         vs->vs_timestamp = gethrtime() - vs->vs_timestamp;
2479         vs->vs_state = vd->vdev_state;
2480         vs->vs_rsize = vdev_get_min_asize(vd);
2481         if (vd->vdev_ops->vdev_op_leaf)
2482                 vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
2483         vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
2484         mutex_exit(&vd->vdev_stat_lock);
2485 
2486         /*
2487          * If we're getting stats on the root vdev, aggregate the I/O counts
2488          * over all top-level vdevs (i.e. the direct children of the root).
2489          */
2490         if (vd == rvd) {
2491                 for (int c = 0; c < rvd->vdev_children; c++) {
2492                         vdev_t *cvd = rvd->vdev_child[c];
2493                         vdev_stat_t *cvs = &cvd->vdev_stat;
2494 
2495                         mutex_enter(&vd->vdev_stat_lock);
2496                         for (int t = 0; t < ZIO_TYPES; t++) {
2497                                 vs->vs_ops[t] += cvs->vs_ops[t];
2498                                 vs->vs_bytes[t] += cvs->vs_bytes[t];
2499                         }
2500                         cvs->vs_scan_removing = cvd->vdev_removing;
2501                         mutex_exit(&vd->vdev_stat_lock);
2502                 }
2503         }
2504 }
2505 
2506 void
2507 vdev_clear_stats(vdev_t *vd)
2508 {
2509         mutex_enter(&vd->vdev_stat_lock);
2510         vd->vdev_stat.vs_space = 0;
2511         vd->vdev_stat.vs_dspace = 0;
2512         vd->vdev_stat.vs_alloc = 0;
2513         mutex_exit(&vd->vdev_stat_lock);
2514 }
2515 
2516 void
2517 vdev_scan_stat_init(vdev_t *vd)
2518 {
2519         vdev_stat_t *vs = &vd->vdev_stat;
2520 
2521         for (int c = 0; c < vd->vdev_children; c++)
2522                 vdev_scan_stat_init(vd->vdev_child[c]);
2523 
2524         mutex_enter(&vd->vdev_stat_lock);
2525         vs->vs_scan_processed = 0;
2526         mutex_exit(&vd->vdev_stat_lock);
2527 }
2528 
2529 void
2530 vdev_stat_update(zio_t *zio, uint64_t psize)
2531 {
2532         spa_t *spa = zio->io_spa;
2533         vdev_t *rvd = spa->spa_root_vdev;
2534         vdev_t *vd = zio->io_vd ? zio->io_vd : rvd;
2535         vdev_t *pvd;
2536         uint64_t txg = zio->io_txg;
2537         vdev_stat_t *vs = &vd->vdev_stat;
2538         zio_type_t type = zio->io_type;
2539         int flags = zio->io_flags;
2540 
2541         /*
2542          * If this i/o is a gang leader, it didn't do any actual work.
2543          */
2544         if (zio->io_gang_tree)
2545                 return;
2546 
2547         if (zio->io_error == 0) {
2548                 /*
2549                  * If this is a root i/o, don't count it -- we've already
2550                  * counted the top-level vdevs, and vdev_get_stats() will
2551                  * aggregate them when asked.  This reduces contention on
2552                  * the root vdev_stat_lock and implicitly handles blocks
2553                  * that compress away to holes, for which there is no i/o.
2554                  * (Holes never create vdev children, so all the counters
2555                  * remain zero, which is what we want.)
2556                  *
2557                  * Note: this only applies to successful i/o (io_error == 0)
2558                  * because unlike i/o counts, errors are not additive.
2559                  * When reading a ditto block, for example, failure of
2560                  * one top-level vdev does not imply a root-level error.
2561                  */
2562                 if (vd == rvd)
2563                         return;
2564 
2565                 ASSERT(vd == zio->io_vd);
2566 
2567                 if (flags & ZIO_FLAG_IO_BYPASS)
2568                         return;
2569 
2570                 mutex_enter(&vd->vdev_stat_lock);
2571 
2572                 if (flags & ZIO_FLAG_IO_REPAIR) {
2573                         if (flags & ZIO_FLAG_SCAN_THREAD) {
2574                                 dsl_scan_phys_t *scn_phys =
2575                                     &spa->spa_dsl_pool->dp_scan->scn_phys;
2576                                 uint64_t *processed = &scn_phys->scn_processed;
2577 
2578                                 /* XXX cleanup? */
2579                                 if (vd->vdev_ops->vdev_op_leaf)
2580                                         atomic_add_64(processed, psize);
2581                                 vs->vs_scan_processed += psize;
2582                         }
2583 
2584                         if (flags & ZIO_FLAG_SELF_HEAL)
2585                                 vs->vs_self_healed += psize;
2586                 }
2587 
2588                 vs->vs_ops[type]++;
2589                 vs->vs_bytes[type] += psize;
2590 
2591                 mutex_exit(&vd->vdev_stat_lock);
2592                 return;
2593         }
2594 
2595         if (flags & ZIO_FLAG_SPECULATIVE)
2596                 return;
2597 
2598         /*
2599          * If this is an I/O error that is going to be retried, then ignore the
2600          * error.  Otherwise, the user may interpret B_FAILFAST I/O errors as
2601          * hard errors, when in reality they can happen for any number of
2602          * innocuous reasons (bus resets, MPxIO link failure, etc).
2603          */
2604         if (zio->io_error == EIO &&
2605             !(zio->io_flags & ZIO_FLAG_IO_RETRY))
2606                 return;
2607 
2608         /*
2609          * Intent logs writes won't propagate their error to the root
2610          * I/O so don't mark these types of failures as pool-level
2611          * errors.
2612          */
2613         if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
2614                 return;
2615 
2616         mutex_enter(&vd->vdev_stat_lock);
2617         if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) {
2618                 if (zio->io_error == ECKSUM)
2619                         vs->vs_checksum_errors++;
2620                 else
2621                         vs->vs_read_errors++;
2622         }
2623         if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd))
2624                 vs->vs_write_errors++;
2625         mutex_exit(&vd->vdev_stat_lock);
2626 
2627         if (type == ZIO_TYPE_WRITE && txg != 0 &&
2628             (!(flags & ZIO_FLAG_IO_REPAIR) ||
2629             (flags & ZIO_FLAG_SCAN_THREAD) ||
2630             spa->spa_claiming)) {
2631                 /*
2632                  * This is either a normal write (not a repair), or it's
2633                  * a repair induced by the scrub thread, or it's a repair
2634                  * made by zil_claim() during spa_load() in the first txg.
2635                  * In the normal case, we commit the DTL change in the same
2636                  * txg as the block was born.  In the scrub-induced repair
2637                  * case, we know that scrubs run in first-pass syncing context,
2638                  * so we commit the DTL change in spa_syncing_txg(spa).
2639                  * In the zil_claim() case, we commit in spa_first_txg(spa).
2640                  *
2641                  * We currently do not make DTL entries for failed spontaneous
2642                  * self-healing writes triggered by normal (non-scrubbing)
2643                  * reads, because we have no transactional context in which to
2644                  * do so -- and it's not clear that it'd be desirable anyway.
2645                  */
2646                 if (vd->vdev_ops->vdev_op_leaf) {
2647                         uint64_t commit_txg = txg;
2648                         if (flags & ZIO_FLAG_SCAN_THREAD) {
2649                                 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2650                                 ASSERT(spa_sync_pass(spa) == 1);
2651                                 vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
2652                                 commit_txg = spa_syncing_txg(spa);
2653                         } else if (spa->spa_claiming) {
2654                                 ASSERT(flags & ZIO_FLAG_IO_REPAIR);
2655                                 commit_txg = spa_first_txg(spa);
2656                         }
2657                         ASSERT(commit_txg >= spa_syncing_txg(spa));
2658                         if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1))
2659                                 return;
2660                         for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent)
2661                                 vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1);
2662                         vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg);
2663                 }
2664                 if (vd != rvd)
2665                         vdev_dtl_dirty(vd, DTL_MISSING, txg, 1);
2666         }
2667 }
2668 
2669 /*
2670  * Update the in-core space usage stats for this vdev, its metaslab class,
2671  * and the root vdev.
2672  */
2673 void
2674 vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta,
2675     int64_t space_delta)
2676 {
2677         int64_t dspace_delta = space_delta;
2678         spa_t *spa = vd->vdev_spa;
2679         vdev_t *rvd = spa->spa_root_vdev;
2680         metaslab_group_t *mg = vd->vdev_mg;
2681         metaslab_class_t *mc = mg ? mg->mg_class : NULL;
2682 
2683         ASSERT(vd == vd->vdev_top);
2684 
2685         /*
2686          * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion
2687          * factor.  We must calculate this here and not at the root vdev
2688          * because the root vdev's psize-to-asize is simply the max of its
2689          * childrens', thus not accurate enough for us.
2690          */
2691         ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0);
2692         ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache);
2693         dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) *
2694             vd->vdev_deflate_ratio;
2695 
2696         mutex_enter(&vd->vdev_stat_lock);
2697         vd->vdev_stat.vs_alloc += alloc_delta;
2698         vd->vdev_stat.vs_space += space_delta;
2699         vd->vdev_stat.vs_dspace += dspace_delta;
2700         mutex_exit(&vd->vdev_stat_lock);
2701 
2702         if (mc == spa_normal_class(spa)) {
2703                 mutex_enter(&rvd->vdev_stat_lock);
2704                 rvd->vdev_stat.vs_alloc += alloc_delta;
2705                 rvd->vdev_stat.vs_space += space_delta;
2706                 rvd->vdev_stat.vs_dspace += dspace_delta;
2707                 mutex_exit(&rvd->vdev_stat_lock);
2708         }
2709 
2710         if (mc != NULL) {
2711                 ASSERT(rvd == vd->vdev_parent);
2712                 ASSERT(vd->vdev_ms_count != 0);
2713 
2714                 metaslab_class_space_update(mc,
2715                     alloc_delta, defer_delta, space_delta, dspace_delta);
2716         }
2717 }
2718 
2719 /*
2720  * Mark a top-level vdev's config as dirty, placing it on the dirty list
2721  * so that it will be written out next time the vdev configuration is synced.
2722  * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs.
2723  */
2724 void
2725 vdev_config_dirty(vdev_t *vd)
2726 {
2727         spa_t *spa = vd->vdev_spa;
2728         vdev_t *rvd = spa->spa_root_vdev;
2729         int c;
2730 
2731         ASSERT(spa_writeable(spa));
2732 
2733         /*
2734          * If this is an aux vdev (as with l2cache and spare devices), then we
2735          * update the vdev config manually and set the sync flag.
2736          */
2737         if (vd->vdev_aux != NULL) {
2738                 spa_aux_vdev_t *sav = vd->vdev_aux;
2739                 nvlist_t **aux;
2740                 uint_t naux;
2741 
2742                 for (c = 0; c < sav->sav_count; c++) {
2743                         if (sav->sav_vdevs[c] == vd)
2744                                 break;
2745                 }
2746 
2747                 if (c == sav->sav_count) {
2748                         /*
2749                          * We're being removed.  There's nothing more to do.
2750                          */
2751                         ASSERT(sav->sav_sync == B_TRUE);
2752                         return;
2753                 }
2754 
2755                 sav->sav_sync = B_TRUE;
2756 
2757                 if (nvlist_lookup_nvlist_array(sav->sav_config,
2758                     ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) {
2759                         VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
2760                             ZPOOL_CONFIG_SPARES, &aux, &naux) == 0);
2761                 }
2762 
2763                 ASSERT(c < naux);
2764 
2765                 /*
2766                  * Setting the nvlist in the middle if the array is a little
2767                  * sketchy, but it will work.
2768                  */
2769                 nvlist_free(aux[c]);
2770                 aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0);
2771 
2772                 return;
2773         }
2774 
2775         /*
2776          * The dirty list is protected by the SCL_CONFIG lock.  The caller
2777          * must either hold SCL_CONFIG as writer, or must be the sync thread
2778          * (which holds SCL_CONFIG as reader).  There's only one sync thread,
2779          * so this is sufficient to ensure mutual exclusion.
2780          */
2781         ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2782             (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2783             spa_config_held(spa, SCL_CONFIG, RW_READER)));
2784 
2785         if (vd == rvd) {
2786                 for (c = 0; c < rvd->vdev_children; c++)
2787                         vdev_config_dirty(rvd->vdev_child[c]);
2788         } else {
2789                 ASSERT(vd == vd->vdev_top);
2790 
2791                 if (!list_link_active(&vd->vdev_config_dirty_node) &&
2792                     !vd->vdev_ishole)
2793                         list_insert_head(&spa->spa_config_dirty_list, vd);
2794         }
2795 }
2796 
2797 void
2798 vdev_config_clean(vdev_t *vd)
2799 {
2800         spa_t *spa = vd->vdev_spa;
2801 
2802         ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) ||
2803             (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2804             spa_config_held(spa, SCL_CONFIG, RW_READER)));
2805 
2806         ASSERT(list_link_active(&vd->vdev_config_dirty_node));
2807         list_remove(&spa->spa_config_dirty_list, vd);
2808 }
2809 
2810 /*
2811  * Mark a top-level vdev's state as dirty, so that the next pass of
2812  * spa_sync() can convert this into vdev_config_dirty().  We distinguish
2813  * the state changes from larger config changes because they require
2814  * much less locking, and are often needed for administrative actions.
2815  */
2816 void
2817 vdev_state_dirty(vdev_t *vd)
2818 {
2819         spa_t *spa = vd->vdev_spa;
2820 
2821         ASSERT(spa_writeable(spa));
2822         ASSERT(vd == vd->vdev_top);
2823 
2824         /*
2825          * The state list is protected by the SCL_STATE lock.  The caller
2826          * must either hold SCL_STATE as writer, or must be the sync thread
2827          * (which holds SCL_STATE as reader).  There's only one sync thread,
2828          * so this is sufficient to ensure mutual exclusion.
2829          */
2830         ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2831             (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2832             spa_config_held(spa, SCL_STATE, RW_READER)));
2833 
2834         if (!list_link_active(&vd->vdev_state_dirty_node) && !vd->vdev_ishole)
2835                 list_insert_head(&spa->spa_state_dirty_list, vd);
2836 }
2837 
2838 void
2839 vdev_state_clean(vdev_t *vd)
2840 {
2841         spa_t *spa = vd->vdev_spa;
2842 
2843         ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) ||
2844             (dsl_pool_sync_context(spa_get_dsl(spa)) &&
2845             spa_config_held(spa, SCL_STATE, RW_READER)));
2846 
2847         ASSERT(list_link_active(&vd->vdev_state_dirty_node));
2848         list_remove(&spa->spa_state_dirty_list, vd);
2849 }
2850 
2851 /*
2852  * Propagate vdev state up from children to parent.
2853  */
2854 void
2855 vdev_propagate_state(vdev_t *vd)
2856 {
2857         spa_t *spa = vd->vdev_spa;
2858         vdev_t *rvd = spa->spa_root_vdev;
2859         int degraded = 0, faulted = 0;
2860         int corrupted = 0;
2861         vdev_t *child;
2862 
2863         if (vd->vdev_children > 0) {
2864                 for (int c = 0; c < vd->vdev_children; c++) {
2865                         child = vd->vdev_child[c];
2866 
2867                         /*
2868                          * Don't factor holes into the decision.
2869                          */
2870                         if (child->vdev_ishole)
2871                                 continue;
2872 
2873                         if (!vdev_readable(child) ||
2874                             (!vdev_writeable(child) && spa_writeable(spa))) {
2875                                 /*
2876                                  * Root special: if there is a top-level log
2877                                  * device, treat the root vdev as if it were
2878                                  * degraded.
2879                                  */
2880                                 if (child->vdev_islog && vd == rvd)
2881                                         degraded++;
2882                                 else
2883                                         faulted++;
2884                         } else if (child->vdev_state <= VDEV_STATE_DEGRADED) {
2885                                 degraded++;
2886                         }
2887 
2888                         if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA)
2889                                 corrupted++;
2890                 }
2891 
2892                 vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded);
2893 
2894                 /*
2895                  * Root special: if there is a top-level vdev that cannot be
2896                  * opened due to corrupted metadata, then propagate the root
2897                  * vdev's aux state as 'corrupt' rather than 'insufficient
2898                  * replicas'.
2899                  */
2900                 if (corrupted && vd == rvd &&
2901                     rvd->vdev_state == VDEV_STATE_CANT_OPEN)
2902                         vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN,
2903                             VDEV_AUX_CORRUPT_DATA);
2904         }
2905 
2906         if (vd->vdev_parent)
2907                 vdev_propagate_state(vd->vdev_parent);
2908 }
2909 
2910 /*
2911  * Set a vdev's state.  If this is during an open, we don't update the parent
2912  * state, because we're in the process of opening children depth-first.
2913  * Otherwise, we propagate the change to the parent.
2914  *
2915  * If this routine places a device in a faulted state, an appropriate ereport is
2916  * generated.
2917  */
2918 void
2919 vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
2920 {
2921         uint64_t save_state;
2922         spa_t *spa = vd->vdev_spa;
2923 
2924         if (state == vd->vdev_state) {
2925                 vd->vdev_stat.vs_aux = aux;
2926                 return;
2927         }
2928 
2929         save_state = vd->vdev_state;
2930 
2931         vd->vdev_state = state;
2932         vd->vdev_stat.vs_aux = aux;
2933 
2934         /*
2935          * If we are setting the vdev state to anything but an open state, then
2936          * always close the underlying device unless the device has requested
2937          * a delayed close (i.e. we're about to remove or fault the device).
2938          * Otherwise, we keep accessible but invalid devices open forever.
2939          * We don't call vdev_close() itself, because that implies some extra
2940          * checks (offline, etc) that we don't want here.  This is limited to
2941          * leaf devices, because otherwise closing the device will affect other
2942          * children.
2943          */
2944         if (!vd->vdev_delayed_close && vdev_is_dead(vd) &&
2945             vd->vdev_ops->vdev_op_leaf)
2946                 vd->vdev_ops->vdev_op_close(vd);
2947 
2948         /*
2949          * If we have brought this vdev back into service, we need
2950          * to notify fmd so that it can gracefully repair any outstanding
2951          * cases due to a missing device.  We do this in all cases, even those
2952          * that probably don't correlate to a repaired fault.  This is sure to
2953          * catch all cases, and we let the zfs-retire agent sort it out.  If
2954          * this is a transient state it's OK, as the retire agent will
2955          * double-check the state of the vdev before repairing it.
2956          */
2957         if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf &&
2958             vd->vdev_prevstate != state)
2959                 zfs_post_state_change(spa, vd);
2960 
2961         if (vd->vdev_removed &&
2962             state == VDEV_STATE_CANT_OPEN &&
2963             (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) {
2964                 /*
2965                  * If the previous state is set to VDEV_STATE_REMOVED, then this
2966                  * device was previously marked removed and someone attempted to
2967                  * reopen it.  If this failed due to a nonexistent device, then
2968                  * keep the device in the REMOVED state.  We also let this be if
2969                  * it is one of our special test online cases, which is only
2970                  * attempting to online the device and shouldn't generate an FMA
2971                  * fault.
2972                  */
2973                 vd->vdev_state = VDEV_STATE_REMOVED;
2974                 vd->vdev_stat.vs_aux = VDEV_AUX_NONE;
2975         } else if (state == VDEV_STATE_REMOVED) {
2976                 vd->vdev_removed = B_TRUE;
2977         } else if (state == VDEV_STATE_CANT_OPEN) {
2978                 /*
2979                  * If we fail to open a vdev during an import or recovery, we
2980                  * mark it as "not available", which signifies that it was
2981                  * never there to begin with.  Failure to open such a device
2982                  * is not considered an error.
2983                  */
2984                 if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
2985                     spa_load_state(spa) == SPA_LOAD_RECOVER) &&
2986                     vd->vdev_ops->vdev_op_leaf)
2987                         vd->vdev_not_present = 1;
2988 
2989                 /*
2990                  * Post the appropriate ereport.  If the 'prevstate' field is
2991                  * set to something other than VDEV_STATE_UNKNOWN, it indicates
2992                  * that this is part of a vdev_reopen().  In this case, we don't
2993                  * want to post the ereport if the device was already in the
2994                  * CANT_OPEN state beforehand.
2995                  *
2996                  * If the 'checkremove' flag is set, then this is an attempt to
2997                  * online the device in response to an insertion event.  If we
2998                  * hit this case, then we have detected an insertion event for a
2999                  * faulted or offline device that wasn't in the removed state.
3000                  * In this scenario, we don't post an ereport because we are
3001                  * about to replace the device, or attempt an online with
3002                  * vdev_forcefault, which will generate the fault for us.
3003                  */
3004                 if ((vd->vdev_prevstate != state || vd->vdev_forcefault) &&
3005                     !vd->vdev_not_present && !vd->vdev_checkremove &&
3006                     vd != spa->spa_root_vdev) {
3007                         const char *class;
3008 
3009                         switch (aux) {
3010                         case VDEV_AUX_OPEN_FAILED:
3011                                 class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED;
3012                                 break;
3013                         case VDEV_AUX_CORRUPT_DATA:
3014                                 class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA;
3015                                 break;
3016                         case VDEV_AUX_NO_REPLICAS:
3017                                 class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS;
3018                                 break;
3019                         case VDEV_AUX_BAD_GUID_SUM:
3020                                 class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM;
3021                                 break;
3022                         case VDEV_AUX_TOO_SMALL:
3023                                 class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL;
3024                                 break;
3025                         case VDEV_AUX_BAD_LABEL:
3026                                 class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL;
3027                                 break;
3028                         default:
3029                                 class = FM_EREPORT_ZFS_DEVICE_UNKNOWN;
3030                         }
3031 
3032                         zfs_ereport_post(class, spa, vd, NULL, save_state, 0);
3033                 }
3034 
3035                 /* Erase any notion of persistent removed state */
3036                 vd->vdev_removed = B_FALSE;
3037         } else {
3038                 vd->vdev_removed = B_FALSE;
3039         }
3040 
3041         if (!isopen && vd->vdev_parent)
3042                 vdev_propagate_state(vd->vdev_parent);
3043 }
3044 
3045 /*
3046  * Check the vdev configuration to ensure that it's capable of supporting
3047  * a root pool. Currently, we do not support RAID-Z or partial configuration.
3048  * In addition, only a single top-level vdev is allowed and none of the leaves
3049  * can be wholedisks.
3050  */
3051 boolean_t
3052 vdev_is_bootable(vdev_t *vd)
3053 {
3054         if (!vd->vdev_ops->vdev_op_leaf) {
3055                 char *vdev_type = vd->vdev_ops->vdev_op_type;
3056 
3057                 if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 &&
3058                     vd->vdev_children > 1) {
3059                         return (B_FALSE);
3060                 } else if (strcmp(vdev_type, VDEV_TYPE_RAIDZ) == 0 ||
3061                     strcmp(vdev_type, VDEV_TYPE_MISSING) == 0) {
3062                         return (B_FALSE);
3063                 }
3064         } else if (vd->vdev_wholedisk == 1) {
3065                 return (B_FALSE);
3066         }
3067 
3068         for (int c = 0; c < vd->vdev_children; c++) {
3069                 if (!vdev_is_bootable(vd->vdev_child[c]))
3070                         return (B_FALSE);
3071         }
3072         return (B_TRUE);
3073 }
3074 
3075 /*
3076  * Load the state from the original vdev tree (ovd) which
3077  * we've retrieved from the MOS config object. If the original
3078  * vdev was offline or faulted then we transfer that state to the
3079  * device in the current vdev tree (nvd).
3080  */
3081 void
3082 vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
3083 {
3084         spa_t *spa = nvd->vdev_spa;
3085 
3086         ASSERT(nvd->vdev_top->vdev_islog);
3087         ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3088         ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
3089 
3090         for (int c = 0; c < nvd->vdev_children; c++)
3091                 vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
3092 
3093         if (nvd->vdev_ops->vdev_op_leaf) {
3094                 /*
3095                  * Restore the persistent vdev state
3096                  */
3097                 nvd->vdev_offline = ovd->vdev_offline;
3098                 nvd->vdev_faulted = ovd->vdev_faulted;
3099                 nvd->vdev_degraded = ovd->vdev_degraded;
3100                 nvd->vdev_removed = ovd->vdev_removed;
3101         }
3102 }
3103 
3104 /*
3105  * Determine if a log device has valid content.  If the vdev was
3106  * removed or faulted in the MOS config then we know that
3107  * the content on the log device has already been written to the pool.
3108  */
3109 boolean_t
3110 vdev_log_state_valid(vdev_t *vd)
3111 {
3112         if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
3113             !vd->vdev_removed)
3114                 return (B_TRUE);
3115 
3116         for (int c = 0; c < vd->vdev_children; c++)
3117                 if (vdev_log_state_valid(vd->vdev_child[c]))
3118                         return (B_TRUE);
3119 
3120         return (B_FALSE);
3121 }
3122 
3123 /*
3124  * Expand a vdev if possible.
3125  */
3126 void
3127 vdev_expand(vdev_t *vd, uint64_t txg)
3128 {
3129         ASSERT(vd->vdev_top == vd);
3130         ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3131 
3132         if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
3133                 VERIFY(vdev_metaslab_init(vd, txg) == 0);
3134                 vdev_config_dirty(vd);
3135         }
3136 }
3137 
3138 /*
3139  * Split a vdev.
3140  */
3141 void
3142 vdev_split(vdev_t *vd)
3143 {
3144         vdev_t *cvd, *pvd = vd->vdev_parent;
3145 
3146         vdev_remove_child(pvd, vd);
3147         vdev_compact_children(pvd);
3148 
3149         cvd = pvd->vdev_child[0];
3150         if (pvd->vdev_children == 1) {
3151                 vdev_remove_parent(cvd);
3152                 cvd->vdev_splitting = B_TRUE;
3153         }
3154         vdev_propagate_state(cvd);
3155 }