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 /*
  29  * This file contains all the routines used when modifying on-disk SPA state.
  30  * This includes opening, importing, destroying, exporting a pool, and syncing a
  31  * pool.
  32  */
  33 
  34 #include <sys/zfs_context.h>
  35 #include <sys/fm/fs/zfs.h>
  36 #include <sys/spa_impl.h>
  37 #include <sys/zio.h>
  38 #include <sys/zio_checksum.h>
  39 #include <sys/dmu.h>
  40 #include <sys/dmu_tx.h>
  41 #include <sys/zap.h>
  42 #include <sys/zil.h>
  43 #include <sys/ddt.h>
  44 #include <sys/vdev_impl.h>
  45 #include <sys/metaslab.h>
  46 #include <sys/metaslab_impl.h>
  47 #include <sys/uberblock_impl.h>
  48 #include <sys/txg.h>
  49 #include <sys/avl.h>
  50 #include <sys/dmu_traverse.h>
  51 #include <sys/dmu_objset.h>
  52 #include <sys/unique.h>
  53 #include <sys/dsl_pool.h>
  54 #include <sys/dsl_dataset.h>
  55 #include <sys/dsl_dir.h>
  56 #include <sys/dsl_prop.h>
  57 #include <sys/dsl_synctask.h>
  58 #include <sys/fs/zfs.h>
  59 #include <sys/arc.h>
  60 #include <sys/callb.h>
  61 #include <sys/systeminfo.h>
  62 #include <sys/spa_boot.h>
  63 #include <sys/zfs_ioctl.h>
  64 #include <sys/dsl_scan.h>
  65 #include <sys/zfeature.h>
  66 
  67 #ifdef  _KERNEL
  68 #include <sys/bootprops.h>
  69 #include <sys/callb.h>
  70 #include <sys/cpupart.h>
  71 #include <sys/pool.h>
  72 #include <sys/sysdc.h>
  73 #include <sys/zone.h>
  74 #endif  /* _KERNEL */
  75 
  76 #include "zfs_prop.h"
  77 #include "zfs_comutil.h"
  78 
  79 typedef enum zti_modes {
  80         zti_mode_fixed,                 /* value is # of threads (min 1) */
  81         zti_mode_online_percent,        /* value is % of online CPUs */
  82         zti_mode_batch,                 /* cpu-intensive; value is ignored */
  83         zti_mode_null,                  /* don't create a taskq */
  84         zti_nmodes
  85 } zti_modes_t;
  86 
  87 #define ZTI_FIX(n)      { zti_mode_fixed, (n) }
  88 #define ZTI_PCT(n)      { zti_mode_online_percent, (n) }
  89 #define ZTI_BATCH       { zti_mode_batch, 0 }
  90 #define ZTI_NULL        { zti_mode_null, 0 }
  91 
  92 #define ZTI_ONE         ZTI_FIX(1)
  93 
  94 typedef struct zio_taskq_info {
  95         enum zti_modes zti_mode;
  96         uint_t zti_value;
  97 } zio_taskq_info_t;
  98 
  99 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
 100         "issue", "issue_high", "intr", "intr_high"
 101 };
 102 
 103 /*
 104  * Define the taskq threads for the following I/O types:
 105  *      NULL, READ, WRITE, FREE, CLAIM, and IOCTL
 106  */
 107 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
 108         /* ISSUE        ISSUE_HIGH      INTR            INTR_HIGH */
 109         { ZTI_ONE,      ZTI_NULL,       ZTI_ONE,        ZTI_NULL },
 110         { ZTI_FIX(8),   ZTI_NULL,       ZTI_BATCH,      ZTI_NULL },
 111         { ZTI_BATCH,    ZTI_FIX(5),     ZTI_FIX(8),     ZTI_FIX(5) },
 112         { ZTI_FIX(100), ZTI_NULL,       ZTI_ONE,        ZTI_NULL },
 113         { ZTI_ONE,      ZTI_NULL,       ZTI_ONE,        ZTI_NULL },
 114         { ZTI_ONE,      ZTI_NULL,       ZTI_ONE,        ZTI_NULL },
 115 };
 116 
 117 static dsl_syncfunc_t spa_sync_version;
 118 static dsl_syncfunc_t spa_sync_props;
 119 static boolean_t spa_has_active_shared_spare(spa_t *spa);
 120 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
 121     spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
 122     char **ereport);
 123 static void spa_vdev_resilver_done(spa_t *spa);
 124 
 125 uint_t          zio_taskq_batch_pct = 100;      /* 1 thread per cpu in pset */
 126 id_t            zio_taskq_psrset_bind = PS_NONE;
 127 boolean_t       zio_taskq_sysdc = B_TRUE;       /* use SDC scheduling class */
 128 uint_t          zio_taskq_basedc = 80;          /* base duty cycle */
 129 
 130 boolean_t       spa_create_process = B_TRUE;    /* no process ==> no sysdc */
 131 
 132 /*
 133  * This (illegal) pool name is used when temporarily importing a spa_t in order
 134  * to get the vdev stats associated with the imported devices.
 135  */
 136 #define TRYIMPORT_NAME  "$import"
 137 
 138 /*
 139  * ==========================================================================
 140  * SPA properties routines
 141  * ==========================================================================
 142  */
 143 
 144 /*
 145  * Add a (source=src, propname=propval) list to an nvlist.
 146  */
 147 static void
 148 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
 149     uint64_t intval, zprop_source_t src)
 150 {
 151         const char *propname = zpool_prop_to_name(prop);
 152         nvlist_t *propval;
 153 
 154         VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 155         VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
 156 
 157         if (strval != NULL)
 158                 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
 159         else
 160                 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
 161 
 162         VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
 163         nvlist_free(propval);
 164 }
 165 
 166 /*
 167  * Get property values from the spa configuration.
 168  */
 169 static void
 170 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
 171 {
 172         vdev_t *rvd = spa->spa_root_vdev;
 173         dsl_pool_t *pool = spa->spa_dsl_pool;
 174         uint64_t size;
 175         uint64_t alloc;
 176         uint64_t space;
 177         uint64_t cap, version;
 178         zprop_source_t src = ZPROP_SRC_NONE;
 179         spa_config_dirent_t *dp;
 180 
 181         ASSERT(MUTEX_HELD(&spa->spa_props_lock));
 182 
 183         if (rvd != NULL) {
 184                 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
 185                 size = metaslab_class_get_space(spa_normal_class(spa));
 186                 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
 187                 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
 188                 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
 189                 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
 190                     size - alloc, src);
 191 
 192                 space = 0;
 193                 for (int c = 0; c < rvd->vdev_children; c++) {
 194                         vdev_t *tvd = rvd->vdev_child[c];
 195                         space += tvd->vdev_max_asize - tvd->vdev_asize;
 196                 }
 197                 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
 198                     src);
 199 
 200                 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
 201                     (spa_mode(spa) == FREAD), src);
 202 
 203                 cap = (size == 0) ? 0 : (alloc * 100 / size);
 204                 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
 205 
 206                 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
 207                     ddt_get_pool_dedup_ratio(spa), src);
 208 
 209                 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
 210                     rvd->vdev_state, src);
 211 
 212                 version = spa_version(spa);
 213                 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
 214                         src = ZPROP_SRC_DEFAULT;
 215                 else
 216                         src = ZPROP_SRC_LOCAL;
 217                 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
 218         }
 219 
 220         if (pool != NULL) {
 221                 dsl_dir_t *freedir = pool->dp_free_dir;
 222 
 223                 /*
 224                  * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
 225                  * when opening pools before this version freedir will be NULL.
 226                  */
 227                 if (freedir != NULL) {
 228                         spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
 229                             freedir->dd_phys->dd_used_bytes, src);
 230                 } else {
 231                         spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
 232                             NULL, 0, src);
 233                 }
 234         }
 235 
 236         spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
 237 
 238         if (spa->spa_comment != NULL) {
 239                 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
 240                     0, ZPROP_SRC_LOCAL);
 241         }
 242 
 243         if (spa->spa_root != NULL)
 244                 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
 245                     0, ZPROP_SRC_LOCAL);
 246 
 247         if ((dp = list_head(&spa->spa_config_list)) != NULL) {
 248                 if (dp->scd_path == NULL) {
 249                         spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
 250                             "none", 0, ZPROP_SRC_LOCAL);
 251                 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
 252                         spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
 253                             dp->scd_path, 0, ZPROP_SRC_LOCAL);
 254                 }
 255         }
 256 }
 257 
 258 /*
 259  * Get zpool property values.
 260  */
 261 int
 262 spa_prop_get(spa_t *spa, nvlist_t **nvp)
 263 {
 264         objset_t *mos = spa->spa_meta_objset;
 265         zap_cursor_t zc;
 266         zap_attribute_t za;
 267         int err;
 268 
 269         VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 270 
 271         mutex_enter(&spa->spa_props_lock);
 272 
 273         /*
 274          * Get properties from the spa config.
 275          */
 276         spa_prop_get_config(spa, nvp);
 277 
 278         /* If no pool property object, no more prop to get. */
 279         if (mos == NULL || spa->spa_pool_props_object == 0) {
 280                 mutex_exit(&spa->spa_props_lock);
 281                 return (0);
 282         }
 283 
 284         /*
 285          * Get properties from the MOS pool property object.
 286          */
 287         for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
 288             (err = zap_cursor_retrieve(&zc, &za)) == 0;
 289             zap_cursor_advance(&zc)) {
 290                 uint64_t intval = 0;
 291                 char *strval = NULL;
 292                 zprop_source_t src = ZPROP_SRC_DEFAULT;
 293                 zpool_prop_t prop;
 294 
 295                 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
 296                         continue;
 297 
 298                 switch (za.za_integer_length) {
 299                 case 8:
 300                         /* integer property */
 301                         if (za.za_first_integer !=
 302                             zpool_prop_default_numeric(prop))
 303                                 src = ZPROP_SRC_LOCAL;
 304 
 305                         if (prop == ZPOOL_PROP_BOOTFS) {
 306                                 dsl_pool_t *dp;
 307                                 dsl_dataset_t *ds = NULL;
 308 
 309                                 dp = spa_get_dsl(spa);
 310                                 rw_enter(&dp->dp_config_rwlock, RW_READER);
 311                                 if (err = dsl_dataset_hold_obj(dp,
 312                                     za.za_first_integer, FTAG, &ds)) {
 313                                         rw_exit(&dp->dp_config_rwlock);
 314                                         break;
 315                                 }
 316 
 317                                 strval = kmem_alloc(
 318                                     MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
 319                                     KM_SLEEP);
 320                                 dsl_dataset_name(ds, strval);
 321                                 dsl_dataset_rele(ds, FTAG);
 322                                 rw_exit(&dp->dp_config_rwlock);
 323                         } else {
 324                                 strval = NULL;
 325                                 intval = za.za_first_integer;
 326                         }
 327 
 328                         spa_prop_add_list(*nvp, prop, strval, intval, src);
 329 
 330                         if (strval != NULL)
 331                                 kmem_free(strval,
 332                                     MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
 333 
 334                         break;
 335 
 336                 case 1:
 337                         /* string property */
 338                         strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
 339                         err = zap_lookup(mos, spa->spa_pool_props_object,
 340                             za.za_name, 1, za.za_num_integers, strval);
 341                         if (err) {
 342                                 kmem_free(strval, za.za_num_integers);
 343                                 break;
 344                         }
 345                         spa_prop_add_list(*nvp, prop, strval, 0, src);
 346                         kmem_free(strval, za.za_num_integers);
 347                         break;
 348 
 349                 default:
 350                         break;
 351                 }
 352         }
 353         zap_cursor_fini(&zc);
 354         mutex_exit(&spa->spa_props_lock);
 355 out:
 356         if (err && err != ENOENT) {
 357                 nvlist_free(*nvp);
 358                 *nvp = NULL;
 359                 return (err);
 360         }
 361 
 362         return (0);
 363 }
 364 
 365 /*
 366  * Validate the given pool properties nvlist and modify the list
 367  * for the property values to be set.
 368  */
 369 static int
 370 spa_prop_validate(spa_t *spa, nvlist_t *props)
 371 {
 372         nvpair_t *elem;
 373         int error = 0, reset_bootfs = 0;
 374         uint64_t objnum;
 375         boolean_t has_feature = B_FALSE;
 376 
 377         elem = NULL;
 378         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
 379                 uint64_t intval;
 380                 char *strval, *slash, *check, *fname;
 381                 const char *propname = nvpair_name(elem);
 382                 zpool_prop_t prop = zpool_name_to_prop(propname);
 383 
 384                 switch (prop) {
 385                 case ZPROP_INVAL:
 386                         if (!zpool_prop_feature(propname)) {
 387                                 error = EINVAL;
 388                                 break;
 389                         }
 390 
 391                         /*
 392                          * Sanitize the input.
 393                          */
 394                         if (nvpair_type(elem) != DATA_TYPE_UINT64) {
 395                                 error = EINVAL;
 396                                 break;
 397                         }
 398 
 399                         if (nvpair_value_uint64(elem, &intval) != 0) {
 400                                 error = EINVAL;
 401                                 break;
 402                         }
 403 
 404                         if (intval != 0) {
 405                                 error = EINVAL;
 406                                 break;
 407                         }
 408 
 409                         fname = strchr(propname, '@') + 1;
 410                         if (zfeature_lookup_name(fname, NULL) != 0) {
 411                                 error = EINVAL;
 412                                 break;
 413                         }
 414 
 415                         has_feature = B_TRUE;
 416                         break;
 417 
 418                 case ZPOOL_PROP_VERSION:
 419                         error = nvpair_value_uint64(elem, &intval);
 420                         if (!error &&
 421                             (intval < spa_version(spa) ||
 422                             intval > SPA_VERSION_BEFORE_FEATURES ||
 423                             has_feature))
 424                                 error = EINVAL;
 425                         break;
 426 
 427                 case ZPOOL_PROP_DELEGATION:
 428                 case ZPOOL_PROP_AUTOREPLACE:
 429                 case ZPOOL_PROP_LISTSNAPS:
 430                 case ZPOOL_PROP_AUTOEXPAND:
 431                         error = nvpair_value_uint64(elem, &intval);
 432                         if (!error && intval > 1)
 433                                 error = EINVAL;
 434                         break;
 435 
 436                 case ZPOOL_PROP_BOOTFS:
 437                         /*
 438                          * If the pool version is less than SPA_VERSION_BOOTFS,
 439                          * or the pool is still being created (version == 0),
 440                          * the bootfs property cannot be set.
 441                          */
 442                         if (spa_version(spa) < SPA_VERSION_BOOTFS) {
 443                                 error = ENOTSUP;
 444                                 break;
 445                         }
 446 
 447                         /*
 448                          * Make sure the vdev config is bootable
 449                          */
 450                         if (!vdev_is_bootable(spa->spa_root_vdev)) {
 451                                 error = ENOTSUP;
 452                                 break;
 453                         }
 454 
 455                         reset_bootfs = 1;
 456 
 457                         error = nvpair_value_string(elem, &strval);
 458 
 459                         if (!error) {
 460                                 objset_t *os;
 461                                 uint64_t compress;
 462 
 463                                 if (strval == NULL || strval[0] == '\0') {
 464                                         objnum = zpool_prop_default_numeric(
 465                                             ZPOOL_PROP_BOOTFS);
 466                                         break;
 467                                 }
 468 
 469                                 if (error = dmu_objset_hold(strval, FTAG, &os))
 470                                         break;
 471 
 472                                 /* Must be ZPL and not gzip compressed. */
 473 
 474                                 if (dmu_objset_type(os) != DMU_OST_ZFS) {
 475                                         error = ENOTSUP;
 476                                 } else if ((error = dsl_prop_get_integer(strval,
 477                                     zfs_prop_to_name(ZFS_PROP_COMPRESSION),
 478                                     &compress, NULL)) == 0 &&
 479                                     !BOOTFS_COMPRESS_VALID(compress)) {
 480                                         error = ENOTSUP;
 481                                 } else {
 482                                         objnum = dmu_objset_id(os);
 483                                 }
 484                                 dmu_objset_rele(os, FTAG);
 485                         }
 486                         break;
 487 
 488                 case ZPOOL_PROP_FAILUREMODE:
 489                         error = nvpair_value_uint64(elem, &intval);
 490                         if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
 491                             intval > ZIO_FAILURE_MODE_PANIC))
 492                                 error = EINVAL;
 493 
 494                         /*
 495                          * This is a special case which only occurs when
 496                          * the pool has completely failed. This allows
 497                          * the user to change the in-core failmode property
 498                          * without syncing it out to disk (I/Os might
 499                          * currently be blocked). We do this by returning
 500                          * EIO to the caller (spa_prop_set) to trick it
 501                          * into thinking we encountered a property validation
 502                          * error.
 503                          */
 504                         if (!error && spa_suspended(spa)) {
 505                                 spa->spa_failmode = intval;
 506                                 error = EIO;
 507                         }
 508                         break;
 509 
 510                 case ZPOOL_PROP_CACHEFILE:
 511                         if ((error = nvpair_value_string(elem, &strval)) != 0)
 512                                 break;
 513 
 514                         if (strval[0] == '\0')
 515                                 break;
 516 
 517                         if (strcmp(strval, "none") == 0)
 518                                 break;
 519 
 520                         if (strval[0] != '/') {
 521                                 error = EINVAL;
 522                                 break;
 523                         }
 524 
 525                         slash = strrchr(strval, '/');
 526                         ASSERT(slash != NULL);
 527 
 528                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
 529                             strcmp(slash, "/..") == 0)
 530                                 error = EINVAL;
 531                         break;
 532 
 533                 case ZPOOL_PROP_COMMENT:
 534                         if ((error = nvpair_value_string(elem, &strval)) != 0)
 535                                 break;
 536                         for (check = strval; *check != '\0'; check++) {
 537                                 /*
 538                                  * The kernel doesn't have an easy isprint()
 539                                  * check.  For this kernel check, we merely
 540                                  * check ASCII apart from DEL.  Fix this if
 541                                  * there is an easy-to-use kernel isprint().
 542                                  */
 543                                 if (*check >= 0x7f) {
 544                                         error = EINVAL;
 545                                         break;
 546                                 }
 547                                 check++;
 548                         }
 549                         if (strlen(strval) > ZPROP_MAX_COMMENT)
 550                                 error = E2BIG;
 551                         break;
 552 
 553                 case ZPOOL_PROP_DEDUPDITTO:
 554                         if (spa_version(spa) < SPA_VERSION_DEDUP)
 555                                 error = ENOTSUP;
 556                         else
 557                                 error = nvpair_value_uint64(elem, &intval);
 558                         if (error == 0 &&
 559                             intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
 560                                 error = EINVAL;
 561                         break;
 562                 }
 563 
 564                 if (error)
 565                         break;
 566         }
 567 
 568         if (!error && reset_bootfs) {
 569                 error = nvlist_remove(props,
 570                     zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
 571 
 572                 if (!error) {
 573                         error = nvlist_add_uint64(props,
 574                             zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
 575                 }
 576         }
 577 
 578         return (error);
 579 }
 580 
 581 void
 582 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
 583 {
 584         char *cachefile;
 585         spa_config_dirent_t *dp;
 586 
 587         if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
 588             &cachefile) != 0)
 589                 return;
 590 
 591         dp = kmem_alloc(sizeof (spa_config_dirent_t),
 592             KM_SLEEP);
 593 
 594         if (cachefile[0] == '\0')
 595                 dp->scd_path = spa_strdup(spa_config_path);
 596         else if (strcmp(cachefile, "none") == 0)
 597                 dp->scd_path = NULL;
 598         else
 599                 dp->scd_path = spa_strdup(cachefile);
 600 
 601         list_insert_head(&spa->spa_config_list, dp);
 602         if (need_sync)
 603                 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
 604 }
 605 
 606 int
 607 spa_prop_set(spa_t *spa, nvlist_t *nvp)
 608 {
 609         int error;
 610         nvpair_t *elem = NULL;
 611         boolean_t need_sync = B_FALSE;
 612 
 613         if ((error = spa_prop_validate(spa, nvp)) != 0)
 614                 return (error);
 615 
 616         while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
 617                 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
 618 
 619                 if (prop == ZPOOL_PROP_CACHEFILE ||
 620                     prop == ZPOOL_PROP_ALTROOT ||
 621                     prop == ZPOOL_PROP_READONLY)
 622                         continue;
 623 
 624                 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
 625                         uint64_t ver;
 626 
 627                         if (prop == ZPOOL_PROP_VERSION) {
 628                                 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
 629                         } else {
 630                                 ASSERT(zpool_prop_feature(nvpair_name(elem)));
 631                                 ver = SPA_VERSION_FEATURES;
 632                                 need_sync = B_TRUE;
 633                         }
 634 
 635                         /* Save time if the version is already set. */
 636                         if (ver == spa_version(spa))
 637                                 continue;
 638 
 639                         /*
 640                          * In addition to the pool directory object, we might
 641                          * create the pool properties object, the features for
 642                          * read object, the features for write object, or the
 643                          * feature descriptions object.
 644                          */
 645                         error = dsl_sync_task_do(spa_get_dsl(spa), NULL,
 646                             spa_sync_version, spa, &ver, 6);
 647                         if (error)
 648                                 return (error);
 649                         continue;
 650                 }
 651 
 652                 need_sync = B_TRUE;
 653                 break;
 654         }
 655 
 656         if (need_sync) {
 657                 return (dsl_sync_task_do(spa_get_dsl(spa), NULL, spa_sync_props,
 658                     spa, nvp, 6));
 659         }
 660 
 661         return (0);
 662 }
 663 
 664 /*
 665  * If the bootfs property value is dsobj, clear it.
 666  */
 667 void
 668 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
 669 {
 670         if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
 671                 VERIFY(zap_remove(spa->spa_meta_objset,
 672                     spa->spa_pool_props_object,
 673                     zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
 674                 spa->spa_bootfs = 0;
 675         }
 676 }
 677 
 678 /*
 679  * Change the GUID for the pool.  This is done so that we can later
 680  * re-import a pool built from a clone of our own vdevs.  We will modify
 681  * the root vdev's guid, our own pool guid, and then mark all of our
 682  * vdevs dirty.  Note that we must make sure that all our vdevs are
 683  * online when we do this, or else any vdevs that weren't present
 684  * would be orphaned from our pool.  We are also going to issue a
 685  * sysevent to update any watchers.
 686  */
 687 int
 688 spa_change_guid(spa_t *spa)
 689 {
 690         uint64_t        oldguid, newguid;
 691         uint64_t        txg;
 692 
 693         if (!(spa_mode_global & FWRITE))
 694                 return (EROFS);
 695 
 696         txg = spa_vdev_enter(spa);
 697 
 698         if (spa->spa_root_vdev->vdev_state != VDEV_STATE_HEALTHY)
 699                 return (spa_vdev_exit(spa, NULL, txg, ENXIO));
 700 
 701         oldguid = spa_guid(spa);
 702         newguid = spa_generate_guid(NULL);
 703         ASSERT3U(oldguid, !=, newguid);
 704 
 705         spa->spa_root_vdev->vdev_guid = newguid;
 706         spa->spa_root_vdev->vdev_guid_sum += (newguid - oldguid);
 707 
 708         vdev_config_dirty(spa->spa_root_vdev);
 709 
 710         spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
 711 
 712         return (spa_vdev_exit(spa, NULL, txg, 0));
 713 }
 714 
 715 /*
 716  * ==========================================================================
 717  * SPA state manipulation (open/create/destroy/import/export)
 718  * ==========================================================================
 719  */
 720 
 721 static int
 722 spa_error_entry_compare(const void *a, const void *b)
 723 {
 724         spa_error_entry_t *sa = (spa_error_entry_t *)a;
 725         spa_error_entry_t *sb = (spa_error_entry_t *)b;
 726         int ret;
 727 
 728         ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
 729             sizeof (zbookmark_t));
 730 
 731         if (ret < 0)
 732                 return (-1);
 733         else if (ret > 0)
 734                 return (1);
 735         else
 736                 return (0);
 737 }
 738 
 739 /*
 740  * Utility function which retrieves copies of the current logs and
 741  * re-initializes them in the process.
 742  */
 743 void
 744 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
 745 {
 746         ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
 747 
 748         bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
 749         bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
 750 
 751         avl_create(&spa->spa_errlist_scrub,
 752             spa_error_entry_compare, sizeof (spa_error_entry_t),
 753             offsetof(spa_error_entry_t, se_avl));
 754         avl_create(&spa->spa_errlist_last,
 755             spa_error_entry_compare, sizeof (spa_error_entry_t),
 756             offsetof(spa_error_entry_t, se_avl));
 757 }
 758 
 759 static taskq_t *
 760 spa_taskq_create(spa_t *spa, const char *name, enum zti_modes mode,
 761     uint_t value)
 762 {
 763         uint_t flags = 0;
 764         boolean_t batch = B_FALSE;
 765 
 766         switch (mode) {
 767         case zti_mode_null:
 768                 return (NULL);          /* no taskq needed */
 769 
 770         case zti_mode_fixed:
 771                 ASSERT3U(value, >=, 1);
 772                 value = MAX(value, 1);
 773                 break;
 774 
 775         case zti_mode_batch:
 776                 batch = B_TRUE;
 777                 flags |= TASKQ_THREADS_CPU_PCT;
 778                 value = zio_taskq_batch_pct;
 779                 break;
 780 
 781         case zti_mode_online_percent:
 782                 flags |= TASKQ_THREADS_CPU_PCT;
 783                 break;
 784 
 785         default:
 786                 panic("unrecognized mode for %s taskq (%u:%u) in "
 787                     "spa_activate()",
 788                     name, mode, value);
 789                 break;
 790         }
 791 
 792         if (zio_taskq_sysdc && spa->spa_proc != &p0) {
 793                 if (batch)
 794                         flags |= TASKQ_DC_BATCH;
 795 
 796                 return (taskq_create_sysdc(name, value, 50, INT_MAX,
 797                     spa->spa_proc, zio_taskq_basedc, flags));
 798         }
 799         return (taskq_create_proc(name, value, maxclsyspri, 50, INT_MAX,
 800             spa->spa_proc, flags));
 801 }
 802 
 803 static void
 804 spa_create_zio_taskqs(spa_t *spa)
 805 {
 806         for (int t = 0; t < ZIO_TYPES; t++) {
 807                 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
 808                         const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
 809                         enum zti_modes mode = ztip->zti_mode;
 810                         uint_t value = ztip->zti_value;
 811                         char name[32];
 812 
 813                         (void) snprintf(name, sizeof (name),
 814                             "%s_%s", zio_type_name[t], zio_taskq_types[q]);
 815 
 816                         spa->spa_zio_taskq[t][q] =
 817                             spa_taskq_create(spa, name, mode, value);
 818                 }
 819         }
 820 }
 821 
 822 #ifdef _KERNEL
 823 static void
 824 spa_thread(void *arg)
 825 {
 826         callb_cpr_t cprinfo;
 827 
 828         spa_t *spa = arg;
 829         user_t *pu = PTOU(curproc);
 830 
 831         CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
 832             spa->spa_name);
 833 
 834         ASSERT(curproc != &p0);
 835         (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
 836             "zpool-%s", spa->spa_name);
 837         (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
 838 
 839         /* bind this thread to the requested psrset */
 840         if (zio_taskq_psrset_bind != PS_NONE) {
 841                 pool_lock();
 842                 mutex_enter(&cpu_lock);
 843                 mutex_enter(&pidlock);
 844                 mutex_enter(&curproc->p_lock);
 845 
 846                 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
 847                     0, NULL, NULL) == 0)  {
 848                         curthread->t_bind_pset = zio_taskq_psrset_bind;
 849                 } else {
 850                         cmn_err(CE_WARN,
 851                             "Couldn't bind process for zfs pool \"%s\" to "
 852                             "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
 853                 }
 854 
 855                 mutex_exit(&curproc->p_lock);
 856                 mutex_exit(&pidlock);
 857                 mutex_exit(&cpu_lock);
 858                 pool_unlock();
 859         }
 860 
 861         if (zio_taskq_sysdc) {
 862                 sysdc_thread_enter(curthread, 100, 0);
 863         }
 864 
 865         spa->spa_proc = curproc;
 866         spa->spa_did = curthread->t_did;
 867 
 868         spa_create_zio_taskqs(spa);
 869 
 870         mutex_enter(&spa->spa_proc_lock);
 871         ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
 872 
 873         spa->spa_proc_state = SPA_PROC_ACTIVE;
 874         cv_broadcast(&spa->spa_proc_cv);
 875 
 876         CALLB_CPR_SAFE_BEGIN(&cprinfo);
 877         while (spa->spa_proc_state == SPA_PROC_ACTIVE)
 878                 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
 879         CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
 880 
 881         ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
 882         spa->spa_proc_state = SPA_PROC_GONE;
 883         spa->spa_proc = &p0;
 884         cv_broadcast(&spa->spa_proc_cv);
 885         CALLB_CPR_EXIT(&cprinfo);   /* drops spa_proc_lock */
 886 
 887         mutex_enter(&curproc->p_lock);
 888         lwp_exit();
 889 }
 890 #endif
 891 
 892 /*
 893  * Activate an uninitialized pool.
 894  */
 895 static void
 896 spa_activate(spa_t *spa, int mode)
 897 {
 898         ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
 899 
 900         spa->spa_state = POOL_STATE_ACTIVE;
 901         spa->spa_mode = mode;
 902 
 903         spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
 904         spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
 905 
 906         /* Try to create a covering process */
 907         mutex_enter(&spa->spa_proc_lock);
 908         ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
 909         ASSERT(spa->spa_proc == &p0);
 910         spa->spa_did = 0;
 911 
 912         /* Only create a process if we're going to be around a while. */
 913         if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
 914                 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
 915                     NULL, 0) == 0) {
 916                         spa->spa_proc_state = SPA_PROC_CREATED;
 917                         while (spa->spa_proc_state == SPA_PROC_CREATED) {
 918                                 cv_wait(&spa->spa_proc_cv,
 919                                     &spa->spa_proc_lock);
 920                         }
 921                         ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
 922                         ASSERT(spa->spa_proc != &p0);
 923                         ASSERT(spa->spa_did != 0);
 924                 } else {
 925 #ifdef _KERNEL
 926                         cmn_err(CE_WARN,
 927                             "Couldn't create process for zfs pool \"%s\"\n",
 928                             spa->spa_name);
 929 #endif
 930                 }
 931         }
 932         mutex_exit(&spa->spa_proc_lock);
 933 
 934         /* If we didn't create a process, we need to create our taskqs. */
 935         if (spa->spa_proc == &p0) {
 936                 spa_create_zio_taskqs(spa);
 937         }
 938 
 939         list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
 940             offsetof(vdev_t, vdev_config_dirty_node));
 941         list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
 942             offsetof(vdev_t, vdev_state_dirty_node));
 943 
 944         txg_list_create(&spa->spa_vdev_txg_list,
 945             offsetof(struct vdev, vdev_txg_node));
 946 
 947         avl_create(&spa->spa_errlist_scrub,
 948             spa_error_entry_compare, sizeof (spa_error_entry_t),
 949             offsetof(spa_error_entry_t, se_avl));
 950         avl_create(&spa->spa_errlist_last,
 951             spa_error_entry_compare, sizeof (spa_error_entry_t),
 952             offsetof(spa_error_entry_t, se_avl));
 953 }
 954 
 955 /*
 956  * Opposite of spa_activate().
 957  */
 958 static void
 959 spa_deactivate(spa_t *spa)
 960 {
 961         ASSERT(spa->spa_sync_on == B_FALSE);
 962         ASSERT(spa->spa_dsl_pool == NULL);
 963         ASSERT(spa->spa_root_vdev == NULL);
 964         ASSERT(spa->spa_async_zio_root == NULL);
 965         ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
 966 
 967         txg_list_destroy(&spa->spa_vdev_txg_list);
 968 
 969         list_destroy(&spa->spa_config_dirty_list);
 970         list_destroy(&spa->spa_state_dirty_list);
 971 
 972         for (int t = 0; t < ZIO_TYPES; t++) {
 973                 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
 974                         if (spa->spa_zio_taskq[t][q] != NULL)
 975                                 taskq_destroy(spa->spa_zio_taskq[t][q]);
 976                         spa->spa_zio_taskq[t][q] = NULL;
 977                 }
 978         }
 979 
 980         metaslab_class_destroy(spa->spa_normal_class);
 981         spa->spa_normal_class = NULL;
 982 
 983         metaslab_class_destroy(spa->spa_log_class);
 984         spa->spa_log_class = NULL;
 985 
 986         /*
 987          * If this was part of an import or the open otherwise failed, we may
 988          * still have errors left in the queues.  Empty them just in case.
 989          */
 990         spa_errlog_drain(spa);
 991 
 992         avl_destroy(&spa->spa_errlist_scrub);
 993         avl_destroy(&spa->spa_errlist_last);
 994 
 995         spa->spa_state = POOL_STATE_UNINITIALIZED;
 996 
 997         mutex_enter(&spa->spa_proc_lock);
 998         if (spa->spa_proc_state != SPA_PROC_NONE) {
 999                 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1000                 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1001                 cv_broadcast(&spa->spa_proc_cv);
1002                 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1003                         ASSERT(spa->spa_proc != &p0);
1004                         cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1005                 }
1006                 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1007                 spa->spa_proc_state = SPA_PROC_NONE;
1008         }
1009         ASSERT(spa->spa_proc == &p0);
1010         mutex_exit(&spa->spa_proc_lock);
1011 
1012         /*
1013          * We want to make sure spa_thread() has actually exited the ZFS
1014          * module, so that the module can't be unloaded out from underneath
1015          * it.
1016          */
1017         if (spa->spa_did != 0) {
1018                 thread_join(spa->spa_did);
1019                 spa->spa_did = 0;
1020         }
1021 }
1022 
1023 /*
1024  * Verify a pool configuration, and construct the vdev tree appropriately.  This
1025  * will create all the necessary vdevs in the appropriate layout, with each vdev
1026  * in the CLOSED state.  This will prep the pool before open/creation/import.
1027  * All vdev validation is done by the vdev_alloc() routine.
1028  */
1029 static int
1030 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1031     uint_t id, int atype)
1032 {
1033         nvlist_t **child;
1034         uint_t children;
1035         int error;
1036 
1037         if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1038                 return (error);
1039 
1040         if ((*vdp)->vdev_ops->vdev_op_leaf)
1041                 return (0);
1042 
1043         error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1044             &child, &children);
1045 
1046         if (error == ENOENT)
1047                 return (0);
1048 
1049         if (error) {
1050                 vdev_free(*vdp);
1051                 *vdp = NULL;
1052                 return (EINVAL);
1053         }
1054 
1055         for (int c = 0; c < children; c++) {
1056                 vdev_t *vd;
1057                 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1058                     atype)) != 0) {
1059                         vdev_free(*vdp);
1060                         *vdp = NULL;
1061                         return (error);
1062                 }
1063         }
1064 
1065         ASSERT(*vdp != NULL);
1066 
1067         return (0);
1068 }
1069 
1070 /*
1071  * Opposite of spa_load().
1072  */
1073 static void
1074 spa_unload(spa_t *spa)
1075 {
1076         int i;
1077 
1078         ASSERT(MUTEX_HELD(&spa_namespace_lock));
1079 
1080         /*
1081          * Stop async tasks.
1082          */
1083         spa_async_suspend(spa);
1084 
1085         /*
1086          * Stop syncing.
1087          */
1088         if (spa->spa_sync_on) {
1089                 txg_sync_stop(spa->spa_dsl_pool);
1090                 spa->spa_sync_on = B_FALSE;
1091         }
1092 
1093         /*
1094          * Wait for any outstanding async I/O to complete.
1095          */
1096         if (spa->spa_async_zio_root != NULL) {
1097                 (void) zio_wait(spa->spa_async_zio_root);
1098                 spa->spa_async_zio_root = NULL;
1099         }
1100 
1101         bpobj_close(&spa->spa_deferred_bpobj);
1102 
1103         /*
1104          * Close the dsl pool.
1105          */
1106         if (spa->spa_dsl_pool) {
1107                 dsl_pool_close(spa->spa_dsl_pool);
1108                 spa->spa_dsl_pool = NULL;
1109                 spa->spa_meta_objset = NULL;
1110         }
1111 
1112         ddt_unload(spa);
1113 
1114         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1115 
1116         /*
1117          * Drop and purge level 2 cache
1118          */
1119         spa_l2cache_drop(spa);
1120 
1121         /*
1122          * Close all vdevs.
1123          */
1124         if (spa->spa_root_vdev)
1125                 vdev_free(spa->spa_root_vdev);
1126         ASSERT(spa->spa_root_vdev == NULL);
1127 
1128         for (i = 0; i < spa->spa_spares.sav_count; i++)
1129                 vdev_free(spa->spa_spares.sav_vdevs[i]);
1130         if (spa->spa_spares.sav_vdevs) {
1131                 kmem_free(spa->spa_spares.sav_vdevs,
1132                     spa->spa_spares.sav_count * sizeof (void *));
1133                 spa->spa_spares.sav_vdevs = NULL;
1134         }
1135         if (spa->spa_spares.sav_config) {
1136                 nvlist_free(spa->spa_spares.sav_config);
1137                 spa->spa_spares.sav_config = NULL;
1138         }
1139         spa->spa_spares.sav_count = 0;
1140 
1141         for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1142                 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1143                 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1144         }
1145         if (spa->spa_l2cache.sav_vdevs) {
1146                 kmem_free(spa->spa_l2cache.sav_vdevs,
1147                     spa->spa_l2cache.sav_count * sizeof (void *));
1148                 spa->spa_l2cache.sav_vdevs = NULL;
1149         }
1150         if (spa->spa_l2cache.sav_config) {
1151                 nvlist_free(spa->spa_l2cache.sav_config);
1152                 spa->spa_l2cache.sav_config = NULL;
1153         }
1154         spa->spa_l2cache.sav_count = 0;
1155 
1156         spa->spa_async_suspended = 0;
1157 
1158         if (spa->spa_comment != NULL) {
1159                 spa_strfree(spa->spa_comment);
1160                 spa->spa_comment = NULL;
1161         }
1162 
1163         spa_config_exit(spa, SCL_ALL, FTAG);
1164 }
1165 
1166 /*
1167  * Load (or re-load) the current list of vdevs describing the active spares for
1168  * this pool.  When this is called, we have some form of basic information in
1169  * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1170  * then re-generate a more complete list including status information.
1171  */
1172 static void
1173 spa_load_spares(spa_t *spa)
1174 {
1175         nvlist_t **spares;
1176         uint_t nspares;
1177         int i;
1178         vdev_t *vd, *tvd;
1179 
1180         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1181 
1182         /*
1183          * First, close and free any existing spare vdevs.
1184          */
1185         for (i = 0; i < spa->spa_spares.sav_count; i++) {
1186                 vd = spa->spa_spares.sav_vdevs[i];
1187 
1188                 /* Undo the call to spa_activate() below */
1189                 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1190                     B_FALSE)) != NULL && tvd->vdev_isspare)
1191                         spa_spare_remove(tvd);
1192                 vdev_close(vd);
1193                 vdev_free(vd);
1194         }
1195 
1196         if (spa->spa_spares.sav_vdevs)
1197                 kmem_free(spa->spa_spares.sav_vdevs,
1198                     spa->spa_spares.sav_count * sizeof (void *));
1199 
1200         if (spa->spa_spares.sav_config == NULL)
1201                 nspares = 0;
1202         else
1203                 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1204                     ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1205 
1206         spa->spa_spares.sav_count = (int)nspares;
1207         spa->spa_spares.sav_vdevs = NULL;
1208 
1209         if (nspares == 0)
1210                 return;
1211 
1212         /*
1213          * Construct the array of vdevs, opening them to get status in the
1214          * process.   For each spare, there is potentially two different vdev_t
1215          * structures associated with it: one in the list of spares (used only
1216          * for basic validation purposes) and one in the active vdev
1217          * configuration (if it's spared in).  During this phase we open and
1218          * validate each vdev on the spare list.  If the vdev also exists in the
1219          * active configuration, then we also mark this vdev as an active spare.
1220          */
1221         spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1222             KM_SLEEP);
1223         for (i = 0; i < spa->spa_spares.sav_count; i++) {
1224                 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1225                     VDEV_ALLOC_SPARE) == 0);
1226                 ASSERT(vd != NULL);
1227 
1228                 spa->spa_spares.sav_vdevs[i] = vd;
1229 
1230                 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1231                     B_FALSE)) != NULL) {
1232                         if (!tvd->vdev_isspare)
1233                                 spa_spare_add(tvd);
1234 
1235                         /*
1236                          * We only mark the spare active if we were successfully
1237                          * able to load the vdev.  Otherwise, importing a pool
1238                          * with a bad active spare would result in strange
1239                          * behavior, because multiple pool would think the spare
1240                          * is actively in use.
1241                          *
1242                          * There is a vulnerability here to an equally bizarre
1243                          * circumstance, where a dead active spare is later
1244                          * brought back to life (onlined or otherwise).  Given
1245                          * the rarity of this scenario, and the extra complexity
1246                          * it adds, we ignore the possibility.
1247                          */
1248                         if (!vdev_is_dead(tvd))
1249                                 spa_spare_activate(tvd);
1250                 }
1251 
1252                 vd->vdev_top = vd;
1253                 vd->vdev_aux = &spa->spa_spares;
1254 
1255                 if (vdev_open(vd) != 0)
1256                         continue;
1257 
1258                 if (vdev_validate_aux(vd) == 0)
1259                         spa_spare_add(vd);
1260         }
1261 
1262         /*
1263          * Recompute the stashed list of spares, with status information
1264          * this time.
1265          */
1266         VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1267             DATA_TYPE_NVLIST_ARRAY) == 0);
1268 
1269         spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1270             KM_SLEEP);
1271         for (i = 0; i < spa->spa_spares.sav_count; i++)
1272                 spares[i] = vdev_config_generate(spa,
1273                     spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1274         VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1275             ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1276         for (i = 0; i < spa->spa_spares.sav_count; i++)
1277                 nvlist_free(spares[i]);
1278         kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1279 }
1280 
1281 /*
1282  * Load (or re-load) the current list of vdevs describing the active l2cache for
1283  * this pool.  When this is called, we have some form of basic information in
1284  * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1285  * then re-generate a more complete list including status information.
1286  * Devices which are already active have their details maintained, and are
1287  * not re-opened.
1288  */
1289 static void
1290 spa_load_l2cache(spa_t *spa)
1291 {
1292         nvlist_t **l2cache;
1293         uint_t nl2cache;
1294         int i, j, oldnvdevs;
1295         uint64_t guid;
1296         vdev_t *vd, **oldvdevs, **newvdevs;
1297         spa_aux_vdev_t *sav = &spa->spa_l2cache;
1298 
1299         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1300 
1301         if (sav->sav_config != NULL) {
1302                 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1303                     ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1304                 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1305         } else {
1306                 nl2cache = 0;
1307         }
1308 
1309         oldvdevs = sav->sav_vdevs;
1310         oldnvdevs = sav->sav_count;
1311         sav->sav_vdevs = NULL;
1312         sav->sav_count = 0;
1313 
1314         /*
1315          * Process new nvlist of vdevs.
1316          */
1317         for (i = 0; i < nl2cache; i++) {
1318                 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1319                     &guid) == 0);
1320 
1321                 newvdevs[i] = NULL;
1322                 for (j = 0; j < oldnvdevs; j++) {
1323                         vd = oldvdevs[j];
1324                         if (vd != NULL && guid == vd->vdev_guid) {
1325                                 /*
1326                                  * Retain previous vdev for add/remove ops.
1327                                  */
1328                                 newvdevs[i] = vd;
1329                                 oldvdevs[j] = NULL;
1330                                 break;
1331                         }
1332                 }
1333 
1334                 if (newvdevs[i] == NULL) {
1335                         /*
1336                          * Create new vdev
1337                          */
1338                         VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1339                             VDEV_ALLOC_L2CACHE) == 0);
1340                         ASSERT(vd != NULL);
1341                         newvdevs[i] = vd;
1342 
1343                         /*
1344                          * Commit this vdev as an l2cache device,
1345                          * even if it fails to open.
1346                          */
1347                         spa_l2cache_add(vd);
1348 
1349                         vd->vdev_top = vd;
1350                         vd->vdev_aux = sav;
1351 
1352                         spa_l2cache_activate(vd);
1353 
1354                         if (vdev_open(vd) != 0)
1355                                 continue;
1356 
1357                         (void) vdev_validate_aux(vd);
1358 
1359                         if (!vdev_is_dead(vd))
1360                                 l2arc_add_vdev(spa, vd);
1361                 }
1362         }
1363 
1364         /*
1365          * Purge vdevs that were dropped
1366          */
1367         for (i = 0; i < oldnvdevs; i++) {
1368                 uint64_t pool;
1369 
1370                 vd = oldvdevs[i];
1371                 if (vd != NULL) {
1372                         ASSERT(vd->vdev_isl2cache);
1373 
1374                         if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1375                             pool != 0ULL && l2arc_vdev_present(vd))
1376                                 l2arc_remove_vdev(vd);
1377                         vdev_clear_stats(vd);
1378                         vdev_free(vd);
1379                 }
1380         }
1381 
1382         if (oldvdevs)
1383                 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1384 
1385         if (sav->sav_config == NULL)
1386                 goto out;
1387 
1388         sav->sav_vdevs = newvdevs;
1389         sav->sav_count = (int)nl2cache;
1390 
1391         /*
1392          * Recompute the stashed list of l2cache devices, with status
1393          * information this time.
1394          */
1395         VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1396             DATA_TYPE_NVLIST_ARRAY) == 0);
1397 
1398         l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1399         for (i = 0; i < sav->sav_count; i++)
1400                 l2cache[i] = vdev_config_generate(spa,
1401                     sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1402         VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1403             ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1404 out:
1405         for (i = 0; i < sav->sav_count; i++)
1406                 nvlist_free(l2cache[i]);
1407         if (sav->sav_count)
1408                 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1409 }
1410 
1411 static int
1412 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1413 {
1414         dmu_buf_t *db;
1415         char *packed = NULL;
1416         size_t nvsize = 0;
1417         int error;
1418         *value = NULL;
1419 
1420         VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1421         nvsize = *(uint64_t *)db->db_data;
1422         dmu_buf_rele(db, FTAG);
1423 
1424         packed = kmem_alloc(nvsize, KM_SLEEP);
1425         error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1426             DMU_READ_PREFETCH);
1427         if (error == 0)
1428                 error = nvlist_unpack(packed, nvsize, value, 0);
1429         kmem_free(packed, nvsize);
1430 
1431         return (error);
1432 }
1433 
1434 /*
1435  * Checks to see if the given vdev could not be opened, in which case we post a
1436  * sysevent to notify the autoreplace code that the device has been removed.
1437  */
1438 static void
1439 spa_check_removed(vdev_t *vd)
1440 {
1441         for (int c = 0; c < vd->vdev_children; c++)
1442                 spa_check_removed(vd->vdev_child[c]);
1443 
1444         if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd)) {
1445                 zfs_post_autoreplace(vd->vdev_spa, vd);
1446                 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1447         }
1448 }
1449 
1450 /*
1451  * Validate the current config against the MOS config
1452  */
1453 static boolean_t
1454 spa_config_valid(spa_t *spa, nvlist_t *config)
1455 {
1456         vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1457         nvlist_t *nv;
1458 
1459         VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1460 
1461         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1462         VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1463 
1464         ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1465 
1466         /*
1467          * If we're doing a normal import, then build up any additional
1468          * diagnostic information about missing devices in this config.
1469          * We'll pass this up to the user for further processing.
1470          */
1471         if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1472                 nvlist_t **child, *nv;
1473                 uint64_t idx = 0;
1474 
1475                 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1476                     KM_SLEEP);
1477                 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1478 
1479                 for (int c = 0; c < rvd->vdev_children; c++) {
1480                         vdev_t *tvd = rvd->vdev_child[c];
1481                         vdev_t *mtvd  = mrvd->vdev_child[c];
1482 
1483                         if (tvd->vdev_ops == &vdev_missing_ops &&
1484                             mtvd->vdev_ops != &vdev_missing_ops &&
1485                             mtvd->vdev_islog)
1486                                 child[idx++] = vdev_config_generate(spa, mtvd,
1487                                     B_FALSE, 0);
1488                 }
1489 
1490                 if (idx) {
1491                         VERIFY(nvlist_add_nvlist_array(nv,
1492                             ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1493                         VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1494                             ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1495 
1496                         for (int i = 0; i < idx; i++)
1497                                 nvlist_free(child[i]);
1498                 }
1499                 nvlist_free(nv);
1500                 kmem_free(child, rvd->vdev_children * sizeof (char **));
1501         }
1502 
1503         /*
1504          * Compare the root vdev tree with the information we have
1505          * from the MOS config (mrvd). Check each top-level vdev
1506          * with the corresponding MOS config top-level (mtvd).
1507          */
1508         for (int c = 0; c < rvd->vdev_children; c++) {
1509                 vdev_t *tvd = rvd->vdev_child[c];
1510                 vdev_t *mtvd  = mrvd->vdev_child[c];
1511 
1512                 /*
1513                  * Resolve any "missing" vdevs in the current configuration.
1514                  * If we find that the MOS config has more accurate information
1515                  * about the top-level vdev then use that vdev instead.
1516                  */
1517                 if (tvd->vdev_ops == &vdev_missing_ops &&
1518                     mtvd->vdev_ops != &vdev_missing_ops) {
1519 
1520                         if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1521                                 continue;
1522 
1523                         /*
1524                          * Device specific actions.
1525                          */
1526                         if (mtvd->vdev_islog) {
1527                                 spa_set_log_state(spa, SPA_LOG_CLEAR);
1528                         } else {
1529                                 /*
1530                                  * XXX - once we have 'readonly' pool
1531                                  * support we should be able to handle
1532                                  * missing data devices by transitioning
1533                                  * the pool to readonly.
1534                                  */
1535                                 continue;
1536                         }
1537 
1538                         /*
1539                          * Swap the missing vdev with the data we were
1540                          * able to obtain from the MOS config.
1541                          */
1542                         vdev_remove_child(rvd, tvd);
1543                         vdev_remove_child(mrvd, mtvd);
1544 
1545                         vdev_add_child(rvd, mtvd);
1546                         vdev_add_child(mrvd, tvd);
1547 
1548                         spa_config_exit(spa, SCL_ALL, FTAG);
1549                         vdev_load(mtvd);
1550                         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1551 
1552                         vdev_reopen(rvd);
1553                 } else if (mtvd->vdev_islog) {
1554                         /*
1555                          * Load the slog device's state from the MOS config
1556                          * since it's possible that the label does not
1557                          * contain the most up-to-date information.
1558                          */
1559                         vdev_load_log_state(tvd, mtvd);
1560                         vdev_reopen(tvd);
1561                 }
1562         }
1563         vdev_free(mrvd);
1564         spa_config_exit(spa, SCL_ALL, FTAG);
1565 
1566         /*
1567          * Ensure we were able to validate the config.
1568          */
1569         return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1570 }
1571 
1572 /*
1573  * Check for missing log devices
1574  */
1575 static int
1576 spa_check_logs(spa_t *spa)
1577 {
1578         switch (spa->spa_log_state) {
1579         case SPA_LOG_MISSING:
1580                 /* need to recheck in case slog has been restored */
1581         case SPA_LOG_UNKNOWN:
1582                 if (dmu_objset_find(spa->spa_name, zil_check_log_chain, NULL,
1583                     DS_FIND_CHILDREN)) {
1584                         spa_set_log_state(spa, SPA_LOG_MISSING);
1585                         return (1);
1586                 }
1587                 break;
1588         }
1589         return (0);
1590 }
1591 
1592 static boolean_t
1593 spa_passivate_log(spa_t *spa)
1594 {
1595         vdev_t *rvd = spa->spa_root_vdev;
1596         boolean_t slog_found = B_FALSE;
1597 
1598         ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1599 
1600         if (!spa_has_slogs(spa))
1601                 return (B_FALSE);
1602 
1603         for (int c = 0; c < rvd->vdev_children; c++) {
1604                 vdev_t *tvd = rvd->vdev_child[c];
1605                 metaslab_group_t *mg = tvd->vdev_mg;
1606 
1607                 if (tvd->vdev_islog) {
1608                         metaslab_group_passivate(mg);
1609                         slog_found = B_TRUE;
1610                 }
1611         }
1612 
1613         return (slog_found);
1614 }
1615 
1616 static void
1617 spa_activate_log(spa_t *spa)
1618 {
1619         vdev_t *rvd = spa->spa_root_vdev;
1620 
1621         ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1622 
1623         for (int c = 0; c < rvd->vdev_children; c++) {
1624                 vdev_t *tvd = rvd->vdev_child[c];
1625                 metaslab_group_t *mg = tvd->vdev_mg;
1626 
1627                 if (tvd->vdev_islog)
1628                         metaslab_group_activate(mg);
1629         }
1630 }
1631 
1632 int
1633 spa_offline_log(spa_t *spa)
1634 {
1635         int error = 0;
1636 
1637         if ((error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1638             NULL, DS_FIND_CHILDREN)) == 0) {
1639 
1640                 /*
1641                  * We successfully offlined the log device, sync out the
1642                  * current txg so that the "stubby" block can be removed
1643                  * by zil_sync().
1644                  */
1645                 txg_wait_synced(spa->spa_dsl_pool, 0);
1646         }
1647         return (error);
1648 }
1649 
1650 static void
1651 spa_aux_check_removed(spa_aux_vdev_t *sav)
1652 {
1653         for (int i = 0; i < sav->sav_count; i++)
1654                 spa_check_removed(sav->sav_vdevs[i]);
1655 }
1656 
1657 void
1658 spa_claim_notify(zio_t *zio)
1659 {
1660         spa_t *spa = zio->io_spa;
1661 
1662         if (zio->io_error)
1663                 return;
1664 
1665         mutex_enter(&spa->spa_props_lock);       /* any mutex will do */
1666         if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1667                 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1668         mutex_exit(&spa->spa_props_lock);
1669 }
1670 
1671 typedef struct spa_load_error {
1672         uint64_t        sle_meta_count;
1673         uint64_t        sle_data_count;
1674 } spa_load_error_t;
1675 
1676 static void
1677 spa_load_verify_done(zio_t *zio)
1678 {
1679         blkptr_t *bp = zio->io_bp;
1680         spa_load_error_t *sle = zio->io_private;
1681         dmu_object_type_t type = BP_GET_TYPE(bp);
1682         int error = zio->io_error;
1683 
1684         if (error) {
1685                 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1686                     type != DMU_OT_INTENT_LOG)
1687                         atomic_add_64(&sle->sle_meta_count, 1);
1688                 else
1689                         atomic_add_64(&sle->sle_data_count, 1);
1690         }
1691         zio_data_buf_free(zio->io_data, zio->io_size);
1692 }
1693 
1694 /*ARGSUSED*/
1695 static int
1696 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1697     arc_buf_t *pbuf, const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1698 {
1699         if (bp != NULL) {
1700                 zio_t *rio = arg;
1701                 size_t size = BP_GET_PSIZE(bp);
1702                 void *data = zio_data_buf_alloc(size);
1703 
1704                 zio_nowait(zio_read(rio, spa, bp, data, size,
1705                     spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1706                     ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1707                     ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1708         }
1709         return (0);
1710 }
1711 
1712 static int
1713 spa_load_verify(spa_t *spa)
1714 {
1715         zio_t *rio;
1716         spa_load_error_t sle = { 0 };
1717         zpool_rewind_policy_t policy;
1718         boolean_t verify_ok = B_FALSE;
1719         int error;
1720 
1721         zpool_get_rewind_policy(spa->spa_config, &policy);
1722 
1723         if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1724                 return (0);
1725 
1726         rio = zio_root(spa, NULL, &sle,
1727             ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1728 
1729         error = traverse_pool(spa, spa->spa_verify_min_txg,
1730             TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1731 
1732         (void) zio_wait(rio);
1733 
1734         spa->spa_load_meta_errors = sle.sle_meta_count;
1735         spa->spa_load_data_errors = sle.sle_data_count;
1736 
1737         if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1738             sle.sle_data_count <= policy.zrp_maxdata) {
1739                 int64_t loss = 0;
1740 
1741                 verify_ok = B_TRUE;
1742                 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1743                 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1744 
1745                 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1746                 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1747                     ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1748                 VERIFY(nvlist_add_int64(spa->spa_load_info,
1749                     ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1750                 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1751                     ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1752         } else {
1753                 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1754         }
1755 
1756         if (error) {
1757                 if (error != ENXIO && error != EIO)
1758                         error = EIO;
1759                 return (error);
1760         }
1761 
1762         return (verify_ok ? 0 : EIO);
1763 }
1764 
1765 /*
1766  * Find a value in the pool props object.
1767  */
1768 static void
1769 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1770 {
1771         (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1772             zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1773 }
1774 
1775 /*
1776  * Find a value in the pool directory object.
1777  */
1778 static int
1779 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1780 {
1781         return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1782             name, sizeof (uint64_t), 1, val));
1783 }
1784 
1785 static int
1786 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1787 {
1788         vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1789         return (err);
1790 }
1791 
1792 /*
1793  * Fix up config after a partly-completed split.  This is done with the
1794  * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1795  * pool have that entry in their config, but only the splitting one contains
1796  * a list of all the guids of the vdevs that are being split off.
1797  *
1798  * This function determines what to do with that list: either rejoin
1799  * all the disks to the pool, or complete the splitting process.  To attempt
1800  * the rejoin, each disk that is offlined is marked online again, and
1801  * we do a reopen() call.  If the vdev label for every disk that was
1802  * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1803  * then we call vdev_split() on each disk, and complete the split.
1804  *
1805  * Otherwise we leave the config alone, with all the vdevs in place in
1806  * the original pool.
1807  */
1808 static void
1809 spa_try_repair(spa_t *spa, nvlist_t *config)
1810 {
1811         uint_t extracted;
1812         uint64_t *glist;
1813         uint_t i, gcount;
1814         nvlist_t *nvl;
1815         vdev_t **vd;
1816         boolean_t attempt_reopen;
1817 
1818         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1819                 return;
1820 
1821         /* check that the config is complete */
1822         if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1823             &glist, &gcount) != 0)
1824                 return;
1825 
1826         vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1827 
1828         /* attempt to online all the vdevs & validate */
1829         attempt_reopen = B_TRUE;
1830         for (i = 0; i < gcount; i++) {
1831                 if (glist[i] == 0)      /* vdev is hole */
1832                         continue;
1833 
1834                 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1835                 if (vd[i] == NULL) {
1836                         /*
1837                          * Don't bother attempting to reopen the disks;
1838                          * just do the split.
1839                          */
1840                         attempt_reopen = B_FALSE;
1841                 } else {
1842                         /* attempt to re-online it */
1843                         vd[i]->vdev_offline = B_FALSE;
1844                 }
1845         }
1846 
1847         if (attempt_reopen) {
1848                 vdev_reopen(spa->spa_root_vdev);
1849 
1850                 /* check each device to see what state it's in */
1851                 for (extracted = 0, i = 0; i < gcount; i++) {
1852                         if (vd[i] != NULL &&
1853                             vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1854                                 break;
1855                         ++extracted;
1856                 }
1857         }
1858 
1859         /*
1860          * If every disk has been moved to the new pool, or if we never
1861          * even attempted to look at them, then we split them off for
1862          * good.
1863          */
1864         if (!attempt_reopen || gcount == extracted) {
1865                 for (i = 0; i < gcount; i++)
1866                         if (vd[i] != NULL)
1867                                 vdev_split(vd[i]);
1868                 vdev_reopen(spa->spa_root_vdev);
1869         }
1870 
1871         kmem_free(vd, gcount * sizeof (vdev_t *));
1872 }
1873 
1874 static int
1875 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
1876     boolean_t mosconfig)
1877 {
1878         nvlist_t *config = spa->spa_config;
1879         char *ereport = FM_EREPORT_ZFS_POOL;
1880         char *comment;
1881         int error;
1882         uint64_t pool_guid;
1883         nvlist_t *nvl;
1884 
1885         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
1886                 return (EINVAL);
1887 
1888         ASSERT(spa->spa_comment == NULL);
1889         if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
1890                 spa->spa_comment = spa_strdup(comment);
1891 
1892         /*
1893          * Versioning wasn't explicitly added to the label until later, so if
1894          * it's not present treat it as the initial version.
1895          */
1896         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1897             &spa->spa_ubsync.ub_version) != 0)
1898                 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
1899 
1900         (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
1901             &spa->spa_config_txg);
1902 
1903         if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
1904             spa_guid_exists(pool_guid, 0)) {
1905                 error = EEXIST;
1906         } else {
1907                 spa->spa_config_guid = pool_guid;
1908 
1909                 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
1910                     &nvl) == 0) {
1911                         VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
1912                             KM_SLEEP) == 0);
1913                 }
1914 
1915                 nvlist_free(spa->spa_load_info);
1916                 spa->spa_load_info = fnvlist_alloc();
1917 
1918                 gethrestime(&spa->spa_loaded_ts);
1919                 error = spa_load_impl(spa, pool_guid, config, state, type,
1920                     mosconfig, &ereport);
1921         }
1922 
1923         spa->spa_minref = refcount_count(&spa->spa_refcount);
1924         if (error) {
1925                 if (error != EEXIST) {
1926                         spa->spa_loaded_ts.tv_sec = 0;
1927                         spa->spa_loaded_ts.tv_nsec = 0;
1928                 }
1929                 if (error != EBADF) {
1930                         zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
1931                 }
1932         }
1933         spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
1934         spa->spa_ena = 0;
1935 
1936         return (error);
1937 }
1938 
1939 /*
1940  * Load an existing storage pool, using the pool's builtin spa_config as a
1941  * source of configuration information.
1942  */
1943 static int
1944 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
1945     spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
1946     char **ereport)
1947 {
1948         int error = 0;
1949         nvlist_t *nvroot = NULL;
1950         nvlist_t *label;
1951         vdev_t *rvd;
1952         uberblock_t *ub = &spa->spa_uberblock;
1953         uint64_t children, config_cache_txg = spa->spa_config_txg;
1954         int orig_mode = spa->spa_mode;
1955         int parse;
1956         uint64_t obj;
1957         boolean_t missing_feat_write = B_FALSE;
1958 
1959         /*
1960          * If this is an untrusted config, access the pool in read-only mode.
1961          * This prevents things like resilvering recently removed devices.
1962          */
1963         if (!mosconfig)
1964                 spa->spa_mode = FREAD;
1965 
1966         ASSERT(MUTEX_HELD(&spa_namespace_lock));
1967 
1968         spa->spa_load_state = state;
1969 
1970         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
1971                 return (EINVAL);
1972 
1973         parse = (type == SPA_IMPORT_EXISTING ?
1974             VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
1975 
1976         /*
1977          * Create "The Godfather" zio to hold all async IOs
1978          */
1979         spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
1980             ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
1981 
1982         /*
1983          * Parse the configuration into a vdev tree.  We explicitly set the
1984          * value that will be returned by spa_version() since parsing the
1985          * configuration requires knowing the version number.
1986          */
1987         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1988         error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
1989         spa_config_exit(spa, SCL_ALL, FTAG);
1990 
1991         if (error != 0)
1992                 return (error);
1993 
1994         ASSERT(spa->spa_root_vdev == rvd);
1995 
1996         if (type != SPA_IMPORT_ASSEMBLE) {
1997                 ASSERT(spa_guid(spa) == pool_guid);
1998         }
1999 
2000         /*
2001          * Try to open all vdevs, loading each label in the process.
2002          */
2003         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2004         error = vdev_open(rvd);
2005         spa_config_exit(spa, SCL_ALL, FTAG);
2006         if (error != 0)
2007                 return (error);
2008 
2009         /*
2010          * We need to validate the vdev labels against the configuration that
2011          * we have in hand, which is dependent on the setting of mosconfig. If
2012          * mosconfig is true then we're validating the vdev labels based on
2013          * that config.  Otherwise, we're validating against the cached config
2014          * (zpool.cache) that was read when we loaded the zfs module, and then
2015          * later we will recursively call spa_load() and validate against
2016          * the vdev config.
2017          *
2018          * If we're assembling a new pool that's been split off from an
2019          * existing pool, the labels haven't yet been updated so we skip
2020          * validation for now.
2021          */
2022         if (type != SPA_IMPORT_ASSEMBLE) {
2023                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2024                 error = vdev_validate(rvd, mosconfig);
2025                 spa_config_exit(spa, SCL_ALL, FTAG);
2026 
2027                 if (error != 0)
2028                         return (error);
2029 
2030                 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2031                         return (ENXIO);
2032         }
2033 
2034         /*
2035          * Find the best uberblock.
2036          */
2037         vdev_uberblock_load(rvd, ub, &label);
2038 
2039         /*
2040          * If we weren't able to find a single valid uberblock, return failure.
2041          */
2042         if (ub->ub_txg == 0) {
2043                 nvlist_free(label);
2044                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2045         }
2046 
2047         /*
2048          * If the pool has an unsupported version we can't open it.
2049          */
2050         if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2051                 nvlist_free(label);
2052                 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2053         }
2054 
2055         if (ub->ub_version >= SPA_VERSION_FEATURES) {
2056                 nvlist_t *features;
2057 
2058                 /*
2059                  * If we weren't able to find what's necessary for reading the
2060                  * MOS in the label, return failure.
2061                  */
2062                 if (label == NULL || nvlist_lookup_nvlist(label,
2063                     ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2064                         nvlist_free(label);
2065                         return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2066                             ENXIO));
2067                 }
2068 
2069                 /*
2070                  * Update our in-core representation with the definitive values
2071                  * from the label.
2072                  */
2073                 nvlist_free(spa->spa_label_features);
2074                 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2075         }
2076 
2077         nvlist_free(label);
2078 
2079         /*
2080          * Look through entries in the label nvlist's features_for_read. If
2081          * there is a feature listed there which we don't understand then we
2082          * cannot open a pool.
2083          */
2084         if (ub->ub_version >= SPA_VERSION_FEATURES) {
2085                 nvlist_t *unsup_feat;
2086 
2087                 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2088                     0);
2089 
2090                 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2091                     NULL); nvp != NULL;
2092                     nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2093                         if (!zfeature_is_supported(nvpair_name(nvp))) {
2094                                 VERIFY(nvlist_add_string(unsup_feat,
2095                                     nvpair_name(nvp), "") == 0);
2096                         }
2097                 }
2098 
2099                 if (!nvlist_empty(unsup_feat)) {
2100                         VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2101                             ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2102                         nvlist_free(unsup_feat);
2103                         return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2104                             ENOTSUP));
2105                 }
2106 
2107                 nvlist_free(unsup_feat);
2108         }
2109 
2110         /*
2111          * If the vdev guid sum doesn't match the uberblock, we have an
2112          * incomplete configuration.  We first check to see if the pool
2113          * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2114          * If it is, defer the vdev_guid_sum check till later so we
2115          * can handle missing vdevs.
2116          */
2117         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2118             &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2119             rvd->vdev_guid_sum != ub->ub_guid_sum)
2120                 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2121 
2122         if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2123                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2124                 spa_try_repair(spa, config);
2125                 spa_config_exit(spa, SCL_ALL, FTAG);
2126                 nvlist_free(spa->spa_config_splitting);
2127                 spa->spa_config_splitting = NULL;
2128         }
2129 
2130         /*
2131          * Initialize internal SPA structures.
2132          */
2133         spa->spa_state = POOL_STATE_ACTIVE;
2134         spa->spa_ubsync = spa->spa_uberblock;
2135         spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2136             TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2137         spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2138             spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2139         spa->spa_claim_max_txg = spa->spa_first_txg;
2140         spa->spa_prev_software_version = ub->ub_software_version;
2141 
2142         error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2143         if (error)
2144                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2145         spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2146 
2147         if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2148                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2149 
2150         if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2151                 boolean_t missing_feat_read = B_FALSE;
2152                 nvlist_t *unsup_feat;
2153 
2154                 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2155                     &spa->spa_feat_for_read_obj) != 0) {
2156                         return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2157                 }
2158 
2159                 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2160                     &spa->spa_feat_for_write_obj) != 0) {
2161                         return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2162                 }
2163 
2164                 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2165                     &spa->spa_feat_desc_obj) != 0) {
2166                         return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2167                 }
2168 
2169                 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2170                     0);
2171 
2172                 if (!feature_is_supported(spa->spa_meta_objset,
2173                     spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2174                     unsup_feat))
2175                         missing_feat_read = B_TRUE;
2176 
2177                 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2178                         if (!feature_is_supported(spa->spa_meta_objset,
2179                             spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2180                             unsup_feat))
2181                                 missing_feat_write = B_TRUE;
2182                 }
2183 
2184                 if (!nvlist_empty(unsup_feat)) {
2185                         VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2186                             ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2187                 }
2188 
2189                 nvlist_free(unsup_feat);
2190 
2191                 if (!missing_feat_read) {
2192                         fnvlist_add_boolean(spa->spa_load_info,
2193                             ZPOOL_CONFIG_CAN_RDONLY);
2194                 }
2195 
2196                 /*
2197                  * If the state is SPA_LOAD_TRYIMPORT, our objective is
2198                  * twofold: to determine whether the pool is available for
2199                  * import in read-write mode and (if it is not) whether the
2200                  * pool is available for import in read-only mode. If the pool
2201                  * is available for import in read-write mode, it is displayed
2202                  * as available in userland; if it is not available for import
2203                  * in read-only mode, it is displayed as unavailable in
2204                  * userland. If the pool is available for import in read-only
2205                  * mode but not read-write mode, it is displayed as unavailable
2206                  * in userland with a special note that the pool is actually
2207                  * available for open in read-only mode.
2208                  *
2209                  * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2210                  * missing a feature for write, we must first determine whether
2211                  * the pool can be opened read-only before returning to
2212                  * userland in order to know whether to display the
2213                  * abovementioned note.
2214                  */
2215                 if (missing_feat_read || (missing_feat_write &&
2216                     spa_writeable(spa))) {
2217                         return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2218                             ENOTSUP));
2219                 }
2220         }
2221 
2222         spa->spa_is_initializing = B_TRUE;
2223         error = dsl_pool_open(spa->spa_dsl_pool);
2224         spa->spa_is_initializing = B_FALSE;
2225         if (error != 0)
2226                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2227 
2228         if (!mosconfig) {
2229                 uint64_t hostid;
2230                 nvlist_t *policy = NULL, *nvconfig;
2231 
2232                 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2233                         return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2234 
2235                 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2236                     ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2237                         char *hostname;
2238                         unsigned long myhostid = 0;
2239 
2240                         VERIFY(nvlist_lookup_string(nvconfig,
2241                             ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2242 
2243 #ifdef  _KERNEL
2244                         myhostid = zone_get_hostid(NULL);
2245 #else   /* _KERNEL */
2246                         /*
2247                          * We're emulating the system's hostid in userland, so
2248                          * we can't use zone_get_hostid().
2249                          */
2250                         (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2251 #endif  /* _KERNEL */
2252                         if (hostid != 0 && myhostid != 0 &&
2253                             hostid != myhostid) {
2254                                 nvlist_free(nvconfig);
2255                                 cmn_err(CE_WARN, "pool '%s' could not be "
2256                                     "loaded as it was last accessed by "
2257                                     "another system (host: %s hostid: 0x%lx). "
2258                                     "See: http://illumos.org/msg/ZFS-8000-EY",
2259                                     spa_name(spa), hostname,
2260                                     (unsigned long)hostid);
2261                                 return (EBADF);
2262                         }
2263                 }
2264                 if (nvlist_lookup_nvlist(spa->spa_config,
2265                     ZPOOL_REWIND_POLICY, &policy) == 0)
2266                         VERIFY(nvlist_add_nvlist(nvconfig,
2267                             ZPOOL_REWIND_POLICY, policy) == 0);
2268 
2269                 spa_config_set(spa, nvconfig);
2270                 spa_unload(spa);
2271                 spa_deactivate(spa);
2272                 spa_activate(spa, orig_mode);
2273 
2274                 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2275         }
2276 
2277         if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2278                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2279         error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2280         if (error != 0)
2281                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2282 
2283         /*
2284          * Load the bit that tells us to use the new accounting function
2285          * (raid-z deflation).  If we have an older pool, this will not
2286          * be present.
2287          */
2288         error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2289         if (error != 0 && error != ENOENT)
2290                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2291 
2292         error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2293             &spa->spa_creation_version);
2294         if (error != 0 && error != ENOENT)
2295                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2296 
2297         /*
2298          * Load the persistent error log.  If we have an older pool, this will
2299          * not be present.
2300          */
2301         error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2302         if (error != 0 && error != ENOENT)
2303                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2304 
2305         error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2306             &spa->spa_errlog_scrub);
2307         if (error != 0 && error != ENOENT)
2308                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2309 
2310         /*
2311          * Load the history object.  If we have an older pool, this
2312          * will not be present.
2313          */
2314         error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2315         if (error != 0 && error != ENOENT)
2316                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2317 
2318         /*
2319          * If we're assembling the pool from the split-off vdevs of
2320          * an existing pool, we don't want to attach the spares & cache
2321          * devices.
2322          */
2323 
2324         /*
2325          * Load any hot spares for this pool.
2326          */
2327         error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2328         if (error != 0 && error != ENOENT)
2329                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2330         if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2331                 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2332                 if (load_nvlist(spa, spa->spa_spares.sav_object,
2333                     &spa->spa_spares.sav_config) != 0)
2334                         return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2335 
2336                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2337                 spa_load_spares(spa);
2338                 spa_config_exit(spa, SCL_ALL, FTAG);
2339         } else if (error == 0) {
2340                 spa->spa_spares.sav_sync = B_TRUE;
2341         }
2342 
2343         /*
2344          * Load any level 2 ARC devices for this pool.
2345          */
2346         error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2347             &spa->spa_l2cache.sav_object);
2348         if (error != 0 && error != ENOENT)
2349                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2350         if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2351                 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2352                 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2353                     &spa->spa_l2cache.sav_config) != 0)
2354                         return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2355 
2356                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2357                 spa_load_l2cache(spa);
2358                 spa_config_exit(spa, SCL_ALL, FTAG);
2359         } else if (error == 0) {
2360                 spa->spa_l2cache.sav_sync = B_TRUE;
2361         }
2362 
2363         spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2364 
2365         error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2366         if (error && error != ENOENT)
2367                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2368 
2369         if (error == 0) {
2370                 uint64_t autoreplace;
2371 
2372                 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2373                 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2374                 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2375                 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2376                 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2377                 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2378                     &spa->spa_dedup_ditto);
2379 
2380                 spa->spa_autoreplace = (autoreplace != 0);
2381         }
2382 
2383         /*
2384          * If the 'autoreplace' property is set, then post a resource notifying
2385          * the ZFS DE that it should not issue any faults for unopenable
2386          * devices.  We also iterate over the vdevs, and post a sysevent for any
2387          * unopenable vdevs so that the normal autoreplace handler can take
2388          * over.
2389          */
2390         if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2391                 spa_check_removed(spa->spa_root_vdev);
2392                 /*
2393                  * For the import case, this is done in spa_import(), because
2394                  * at this point we're using the spare definitions from
2395                  * the MOS config, not necessarily from the userland config.
2396                  */
2397                 if (state != SPA_LOAD_IMPORT) {
2398                         spa_aux_check_removed(&spa->spa_spares);
2399                         spa_aux_check_removed(&spa->spa_l2cache);
2400                 }
2401         }
2402 
2403         /*
2404          * Load the vdev state for all toplevel vdevs.
2405          */
2406         vdev_load(rvd);
2407 
2408         /*
2409          * Propagate the leaf DTLs we just loaded all the way up the tree.
2410          */
2411         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2412         vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2413         spa_config_exit(spa, SCL_ALL, FTAG);
2414 
2415         /*
2416          * Load the DDTs (dedup tables).
2417          */
2418         error = ddt_load(spa);
2419         if (error != 0)
2420                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2421 
2422         spa_update_dspace(spa);
2423 
2424         /*
2425          * Validate the config, using the MOS config to fill in any
2426          * information which might be missing.  If we fail to validate
2427          * the config then declare the pool unfit for use. If we're
2428          * assembling a pool from a split, the log is not transferred
2429          * over.
2430          */
2431         if (type != SPA_IMPORT_ASSEMBLE) {
2432                 nvlist_t *nvconfig;
2433 
2434                 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2435                         return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2436 
2437                 if (!spa_config_valid(spa, nvconfig)) {
2438                         nvlist_free(nvconfig);
2439                         return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2440                             ENXIO));
2441                 }
2442                 nvlist_free(nvconfig);
2443 
2444                 /*
2445                  * Now that we've validated the config, check the state of the
2446                  * root vdev.  If it can't be opened, it indicates one or
2447                  * more toplevel vdevs are faulted.
2448                  */
2449                 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2450                         return (ENXIO);
2451 
2452                 if (spa_check_logs(spa)) {
2453                         *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2454                         return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2455                 }
2456         }
2457 
2458         if (missing_feat_write) {
2459                 ASSERT(state == SPA_LOAD_TRYIMPORT);
2460 
2461                 /*
2462                  * At this point, we know that we can open the pool in
2463                  * read-only mode but not read-write mode. We now have enough
2464                  * information and can return to userland.
2465                  */
2466                 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2467         }
2468 
2469         /*
2470          * We've successfully opened the pool, verify that we're ready
2471          * to start pushing transactions.
2472          */
2473         if (state != SPA_LOAD_TRYIMPORT) {
2474                 if (error = spa_load_verify(spa))
2475                         return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2476                             error));
2477         }
2478 
2479         if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2480             spa->spa_load_max_txg == UINT64_MAX)) {
2481                 dmu_tx_t *tx;
2482                 int need_update = B_FALSE;
2483 
2484                 ASSERT(state != SPA_LOAD_TRYIMPORT);
2485 
2486                 /*
2487                  * Claim log blocks that haven't been committed yet.
2488                  * This must all happen in a single txg.
2489                  * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2490                  * invoked from zil_claim_log_block()'s i/o done callback.
2491                  * Price of rollback is that we abandon the log.
2492                  */
2493                 spa->spa_claiming = B_TRUE;
2494 
2495                 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2496                     spa_first_txg(spa));
2497                 (void) dmu_objset_find(spa_name(spa),
2498                     zil_claim, tx, DS_FIND_CHILDREN);
2499                 dmu_tx_commit(tx);
2500 
2501                 spa->spa_claiming = B_FALSE;
2502 
2503                 spa_set_log_state(spa, SPA_LOG_GOOD);
2504                 spa->spa_sync_on = B_TRUE;
2505                 txg_sync_start(spa->spa_dsl_pool);
2506 
2507                 /*
2508                  * Wait for all claims to sync.  We sync up to the highest
2509                  * claimed log block birth time so that claimed log blocks
2510                  * don't appear to be from the future.  spa_claim_max_txg
2511                  * will have been set for us by either zil_check_log_chain()
2512                  * (invoked from spa_check_logs()) or zil_claim() above.
2513                  */
2514                 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2515 
2516                 /*
2517                  * If the config cache is stale, or we have uninitialized
2518                  * metaslabs (see spa_vdev_add()), then update the config.
2519                  *
2520                  * If this is a verbatim import, trust the current
2521                  * in-core spa_config and update the disk labels.
2522                  */
2523                 if (config_cache_txg != spa->spa_config_txg ||
2524                     state == SPA_LOAD_IMPORT ||
2525                     state == SPA_LOAD_RECOVER ||
2526                     (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2527                         need_update = B_TRUE;
2528 
2529                 for (int c = 0; c < rvd->vdev_children; c++)
2530                         if (rvd->vdev_child[c]->vdev_ms_array == 0)
2531                                 need_update = B_TRUE;
2532 
2533                 /*
2534                  * Update the config cache asychronously in case we're the
2535                  * root pool, in which case the config cache isn't writable yet.
2536                  */
2537                 if (need_update)
2538                         spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2539 
2540                 /*
2541                  * Check all DTLs to see if anything needs resilvering.
2542                  */
2543                 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2544                     vdev_resilver_needed(rvd, NULL, NULL))
2545                         spa_async_request(spa, SPA_ASYNC_RESILVER);
2546 
2547                 /*
2548                  * Delete any inconsistent datasets.
2549                  */
2550                 (void) dmu_objset_find(spa_name(spa),
2551                     dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2552 
2553                 /*
2554                  * Clean up any stale temporary dataset userrefs.
2555                  */
2556                 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2557         }
2558 
2559         return (0);
2560 }
2561 
2562 static int
2563 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2564 {
2565         int mode = spa->spa_mode;
2566 
2567         spa_unload(spa);
2568         spa_deactivate(spa);
2569 
2570         spa->spa_load_max_txg--;
2571 
2572         spa_activate(spa, mode);
2573         spa_async_suspend(spa);
2574 
2575         return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2576 }
2577 
2578 /*
2579  * If spa_load() fails this function will try loading prior txg's. If
2580  * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2581  * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2582  * function will not rewind the pool and will return the same error as
2583  * spa_load().
2584  */
2585 static int
2586 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2587     uint64_t max_request, int rewind_flags)
2588 {
2589         nvlist_t *loadinfo = NULL;
2590         nvlist_t *config = NULL;
2591         int load_error, rewind_error;
2592         uint64_t safe_rewind_txg;
2593         uint64_t min_txg;
2594 
2595         if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2596                 spa->spa_load_max_txg = spa->spa_load_txg;
2597                 spa_set_log_state(spa, SPA_LOG_CLEAR);
2598         } else {
2599                 spa->spa_load_max_txg = max_request;
2600         }
2601 
2602         load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2603             mosconfig);
2604         if (load_error == 0)
2605                 return (0);
2606 
2607         if (spa->spa_root_vdev != NULL)
2608                 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2609 
2610         spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2611         spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2612 
2613         if (rewind_flags & ZPOOL_NEVER_REWIND) {
2614                 nvlist_free(config);
2615                 return (load_error);
2616         }
2617 
2618         if (state == SPA_LOAD_RECOVER) {
2619                 /* Price of rolling back is discarding txgs, including log */
2620                 spa_set_log_state(spa, SPA_LOG_CLEAR);
2621         } else {
2622                 /*
2623                  * If we aren't rolling back save the load info from our first
2624                  * import attempt so that we can restore it after attempting
2625                  * to rewind.
2626                  */
2627                 loadinfo = spa->spa_load_info;
2628                 spa->spa_load_info = fnvlist_alloc();
2629         }
2630 
2631         spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2632         safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2633         min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2634             TXG_INITIAL : safe_rewind_txg;
2635 
2636         /*
2637          * Continue as long as we're finding errors, we're still within
2638          * the acceptable rewind range, and we're still finding uberblocks
2639          */
2640         while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2641             spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2642                 if (spa->spa_load_max_txg < safe_rewind_txg)
2643                         spa->spa_extreme_rewind = B_TRUE;
2644                 rewind_error = spa_load_retry(spa, state, mosconfig);
2645         }
2646 
2647         spa->spa_extreme_rewind = B_FALSE;
2648         spa->spa_load_max_txg = UINT64_MAX;
2649 
2650         if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2651                 spa_config_set(spa, config);
2652 
2653         if (state == SPA_LOAD_RECOVER) {
2654                 ASSERT3P(loadinfo, ==, NULL);
2655                 return (rewind_error);
2656         } else {
2657                 /* Store the rewind info as part of the initial load info */
2658                 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2659                     spa->spa_load_info);
2660 
2661                 /* Restore the initial load info */
2662                 fnvlist_free(spa->spa_load_info);
2663                 spa->spa_load_info = loadinfo;
2664 
2665                 return (load_error);
2666         }
2667 }
2668 
2669 /*
2670  * Pool Open/Import
2671  *
2672  * The import case is identical to an open except that the configuration is sent
2673  * down from userland, instead of grabbed from the configuration cache.  For the
2674  * case of an open, the pool configuration will exist in the
2675  * POOL_STATE_UNINITIALIZED state.
2676  *
2677  * The stats information (gen/count/ustats) is used to gather vdev statistics at
2678  * the same time open the pool, without having to keep around the spa_t in some
2679  * ambiguous state.
2680  */
2681 static int
2682 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2683     nvlist_t **config)
2684 {
2685         spa_t *spa;
2686         spa_load_state_t state = SPA_LOAD_OPEN;
2687         int error;
2688         int locked = B_FALSE;
2689 
2690         *spapp = NULL;
2691 
2692         /*
2693          * As disgusting as this is, we need to support recursive calls to this
2694          * function because dsl_dir_open() is called during spa_load(), and ends
2695          * up calling spa_open() again.  The real fix is to figure out how to
2696          * avoid dsl_dir_open() calling this in the first place.
2697          */
2698         if (mutex_owner(&spa_namespace_lock) != curthread) {
2699                 mutex_enter(&spa_namespace_lock);
2700                 locked = B_TRUE;
2701         }
2702 
2703         if ((spa = spa_lookup(pool)) == NULL) {
2704                 if (locked)
2705                         mutex_exit(&spa_namespace_lock);
2706                 return (ENOENT);
2707         }
2708 
2709         if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2710                 zpool_rewind_policy_t policy;
2711 
2712                 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2713                     &policy);
2714                 if (policy.zrp_request & ZPOOL_DO_REWIND)
2715                         state = SPA_LOAD_RECOVER;
2716 
2717                 spa_activate(spa, spa_mode_global);
2718 
2719                 if (state != SPA_LOAD_RECOVER)
2720                         spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2721 
2722                 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2723                     policy.zrp_request);
2724 
2725                 if (error == EBADF) {
2726                         /*
2727                          * If vdev_validate() returns failure (indicated by
2728                          * EBADF), it indicates that one of the vdevs indicates
2729                          * that the pool has been exported or destroyed.  If
2730                          * this is the case, the config cache is out of sync and
2731                          * we should remove the pool from the namespace.
2732                          */
2733                         spa_unload(spa);
2734                         spa_deactivate(spa);
2735                         spa_config_sync(spa, B_TRUE, B_TRUE);
2736                         spa_remove(spa);
2737                         if (locked)
2738                                 mutex_exit(&spa_namespace_lock);
2739                         return (ENOENT);
2740                 }
2741 
2742                 if (error) {
2743                         /*
2744                          * We can't open the pool, but we still have useful
2745                          * information: the state of each vdev after the
2746                          * attempted vdev_open().  Return this to the user.
2747                          */
2748                         if (config != NULL && spa->spa_config) {
2749                                 VERIFY(nvlist_dup(spa->spa_config, config,
2750                                     KM_SLEEP) == 0);
2751                                 VERIFY(nvlist_add_nvlist(*config,
2752                                     ZPOOL_CONFIG_LOAD_INFO,
2753                                     spa->spa_load_info) == 0);
2754                         }
2755                         spa_unload(spa);
2756                         spa_deactivate(spa);
2757                         spa->spa_last_open_failed = error;
2758                         if (locked)
2759                                 mutex_exit(&spa_namespace_lock);
2760                         *spapp = NULL;
2761                         return (error);
2762                 }
2763         }
2764 
2765         spa_open_ref(spa, tag);
2766 
2767         if (config != NULL)
2768                 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2769 
2770         /*
2771          * If we've recovered the pool, pass back any information we
2772          * gathered while doing the load.
2773          */
2774         if (state == SPA_LOAD_RECOVER) {
2775                 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2776                     spa->spa_load_info) == 0);
2777         }
2778 
2779         if (locked) {
2780                 spa->spa_last_open_failed = 0;
2781                 spa->spa_last_ubsync_txg = 0;
2782                 spa->spa_load_txg = 0;
2783                 mutex_exit(&spa_namespace_lock);
2784         }
2785 
2786         *spapp = spa;
2787 
2788         return (0);
2789 }
2790 
2791 int
2792 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2793     nvlist_t **config)
2794 {
2795         return (spa_open_common(name, spapp, tag, policy, config));
2796 }
2797 
2798 int
2799 spa_open(const char *name, spa_t **spapp, void *tag)
2800 {
2801         return (spa_open_common(name, spapp, tag, NULL, NULL));
2802 }
2803 
2804 /*
2805  * Lookup the given spa_t, incrementing the inject count in the process,
2806  * preventing it from being exported or destroyed.
2807  */
2808 spa_t *
2809 spa_inject_addref(char *name)
2810 {
2811         spa_t *spa;
2812 
2813         mutex_enter(&spa_namespace_lock);
2814         if ((spa = spa_lookup(name)) == NULL) {
2815                 mutex_exit(&spa_namespace_lock);
2816                 return (NULL);
2817         }
2818         spa->spa_inject_ref++;
2819         mutex_exit(&spa_namespace_lock);
2820 
2821         return (spa);
2822 }
2823 
2824 void
2825 spa_inject_delref(spa_t *spa)
2826 {
2827         mutex_enter(&spa_namespace_lock);
2828         spa->spa_inject_ref--;
2829         mutex_exit(&spa_namespace_lock);
2830 }
2831 
2832 /*
2833  * Add spares device information to the nvlist.
2834  */
2835 static void
2836 spa_add_spares(spa_t *spa, nvlist_t *config)
2837 {
2838         nvlist_t **spares;
2839         uint_t i, nspares;
2840         nvlist_t *nvroot;
2841         uint64_t guid;
2842         vdev_stat_t *vs;
2843         uint_t vsc;
2844         uint64_t pool;
2845 
2846         ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2847 
2848         if (spa->spa_spares.sav_count == 0)
2849                 return;
2850 
2851         VERIFY(nvlist_lookup_nvlist(config,
2852             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2853         VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2854             ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2855         if (nspares != 0) {
2856                 VERIFY(nvlist_add_nvlist_array(nvroot,
2857                     ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2858                 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2859                     ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2860 
2861                 /*
2862                  * Go through and find any spares which have since been
2863                  * repurposed as an active spare.  If this is the case, update
2864                  * their status appropriately.
2865                  */
2866                 for (i = 0; i < nspares; i++) {
2867                         VERIFY(nvlist_lookup_uint64(spares[i],
2868                             ZPOOL_CONFIG_GUID, &guid) == 0);
2869                         if (spa_spare_exists(guid, &pool, NULL) &&
2870                             pool != 0ULL) {
2871                                 VERIFY(nvlist_lookup_uint64_array(
2872                                     spares[i], ZPOOL_CONFIG_VDEV_STATS,
2873                                     (uint64_t **)&vs, &vsc) == 0);
2874                                 vs->vs_state = VDEV_STATE_CANT_OPEN;
2875                                 vs->vs_aux = VDEV_AUX_SPARED;
2876                         }
2877                 }
2878         }
2879 }
2880 
2881 /*
2882  * Add l2cache device information to the nvlist, including vdev stats.
2883  */
2884 static void
2885 spa_add_l2cache(spa_t *spa, nvlist_t *config)
2886 {
2887         nvlist_t **l2cache;
2888         uint_t i, j, nl2cache;
2889         nvlist_t *nvroot;
2890         uint64_t guid;
2891         vdev_t *vd;
2892         vdev_stat_t *vs;
2893         uint_t vsc;
2894 
2895         ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2896 
2897         if (spa->spa_l2cache.sav_count == 0)
2898                 return;
2899 
2900         VERIFY(nvlist_lookup_nvlist(config,
2901             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2902         VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
2903             ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2904         if (nl2cache != 0) {
2905                 VERIFY(nvlist_add_nvlist_array(nvroot,
2906                     ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
2907                 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2908                     ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
2909 
2910                 /*
2911                  * Update level 2 cache device stats.
2912                  */
2913 
2914                 for (i = 0; i < nl2cache; i++) {
2915                         VERIFY(nvlist_lookup_uint64(l2cache[i],
2916                             ZPOOL_CONFIG_GUID, &guid) == 0);
2917 
2918                         vd = NULL;
2919                         for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
2920                                 if (guid ==
2921                                     spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
2922                                         vd = spa->spa_l2cache.sav_vdevs[j];
2923                                         break;
2924                                 }
2925                         }
2926                         ASSERT(vd != NULL);
2927 
2928                         VERIFY(nvlist_lookup_uint64_array(l2cache[i],
2929                             ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
2930                             == 0);
2931                         vdev_get_stats(vd, vs);
2932                 }
2933         }
2934 }
2935 
2936 static void
2937 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
2938 {
2939         nvlist_t *features;
2940         zap_cursor_t zc;
2941         zap_attribute_t za;
2942 
2943         ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2944         VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2945 
2946         if (spa->spa_feat_for_read_obj != 0) {
2947                 for (zap_cursor_init(&zc, spa->spa_meta_objset,
2948                     spa->spa_feat_for_read_obj);
2949                     zap_cursor_retrieve(&zc, &za) == 0;
2950                     zap_cursor_advance(&zc)) {
2951                         ASSERT(za.za_integer_length == sizeof (uint64_t) &&
2952                             za.za_num_integers == 1);
2953                         VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
2954                             za.za_first_integer));
2955                 }
2956                 zap_cursor_fini(&zc);
2957         }
2958 
2959         if (spa->spa_feat_for_write_obj != 0) {
2960                 for (zap_cursor_init(&zc, spa->spa_meta_objset,
2961                     spa->spa_feat_for_write_obj);
2962                     zap_cursor_retrieve(&zc, &za) == 0;
2963                     zap_cursor_advance(&zc)) {
2964                         ASSERT(za.za_integer_length == sizeof (uint64_t) &&
2965                             za.za_num_integers == 1);
2966                         VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
2967                             za.za_first_integer));
2968                 }
2969                 zap_cursor_fini(&zc);
2970         }
2971 
2972         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
2973             features) == 0);
2974         nvlist_free(features);
2975 }
2976 
2977 int
2978 spa_get_stats(const char *name, nvlist_t **config,
2979     char *altroot, size_t buflen)
2980 {
2981         int error;
2982         spa_t *spa;
2983 
2984         *config = NULL;
2985         error = spa_open_common(name, &spa, FTAG, NULL, config);
2986 
2987         if (spa != NULL) {
2988                 /*
2989                  * This still leaves a window of inconsistency where the spares
2990                  * or l2cache devices could change and the config would be
2991                  * self-inconsistent.
2992                  */
2993                 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
2994 
2995                 if (*config != NULL) {
2996                         uint64_t loadtimes[2];
2997 
2998                         loadtimes[0] = spa->spa_loaded_ts.tv_sec;
2999                         loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3000                         VERIFY(nvlist_add_uint64_array(*config,
3001                             ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3002 
3003                         VERIFY(nvlist_add_uint64(*config,
3004                             ZPOOL_CONFIG_ERRCOUNT,
3005                             spa_get_errlog_size(spa)) == 0);
3006 
3007                         if (spa_suspended(spa))
3008                                 VERIFY(nvlist_add_uint64(*config,
3009                                     ZPOOL_CONFIG_SUSPENDED,
3010                                     spa->spa_failmode) == 0);
3011 
3012                         spa_add_spares(spa, *config);
3013                         spa_add_l2cache(spa, *config);
3014                         spa_add_feature_stats(spa, *config);
3015                 }
3016         }
3017 
3018         /*
3019          * We want to get the alternate root even for faulted pools, so we cheat
3020          * and call spa_lookup() directly.
3021          */
3022         if (altroot) {
3023                 if (spa == NULL) {
3024                         mutex_enter(&spa_namespace_lock);
3025                         spa = spa_lookup(name);
3026                         if (spa)
3027                                 spa_altroot(spa, altroot, buflen);
3028                         else
3029                                 altroot[0] = '\0';
3030                         spa = NULL;
3031                         mutex_exit(&spa_namespace_lock);
3032                 } else {
3033                         spa_altroot(spa, altroot, buflen);
3034                 }
3035         }
3036 
3037         if (spa != NULL) {
3038                 spa_config_exit(spa, SCL_CONFIG, FTAG);
3039                 spa_close(spa, FTAG);
3040         }
3041 
3042         return (error);
3043 }
3044 
3045 /*
3046  * Validate that the auxiliary device array is well formed.  We must have an
3047  * array of nvlists, each which describes a valid leaf vdev.  If this is an
3048  * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3049  * specified, as long as they are well-formed.
3050  */
3051 static int
3052 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3053     spa_aux_vdev_t *sav, const char *config, uint64_t version,
3054     vdev_labeltype_t label)
3055 {
3056         nvlist_t **dev;
3057         uint_t i, ndev;
3058         vdev_t *vd;
3059         int error;
3060 
3061         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3062 
3063         /*
3064          * It's acceptable to have no devs specified.
3065          */
3066         if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3067                 return (0);
3068 
3069         if (ndev == 0)
3070                 return (EINVAL);
3071 
3072         /*
3073          * Make sure the pool is formatted with a version that supports this
3074          * device type.
3075          */
3076         if (spa_version(spa) < version)
3077                 return (ENOTSUP);
3078 
3079         /*
3080          * Set the pending device list so we correctly handle device in-use
3081          * checking.
3082          */
3083         sav->sav_pending = dev;
3084         sav->sav_npending = ndev;
3085 
3086         for (i = 0; i < ndev; i++) {
3087                 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3088                     mode)) != 0)
3089                         goto out;
3090 
3091                 if (!vd->vdev_ops->vdev_op_leaf) {
3092                         vdev_free(vd);
3093                         error = EINVAL;
3094                         goto out;
3095                 }
3096 
3097                 /*
3098                  * The L2ARC currently only supports disk devices in
3099                  * kernel context.  For user-level testing, we allow it.
3100                  */
3101 #ifdef _KERNEL
3102                 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3103                     strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3104                         error = ENOTBLK;
3105                         vdev_free(vd);
3106                         goto out;
3107                 }
3108 #endif
3109                 vd->vdev_top = vd;
3110 
3111                 if ((error = vdev_open(vd)) == 0 &&
3112                     (error = vdev_label_init(vd, crtxg, label)) == 0) {
3113                         VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3114                             vd->vdev_guid) == 0);
3115                 }
3116 
3117                 vdev_free(vd);
3118 
3119                 if (error &&
3120                     (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3121                         goto out;
3122                 else
3123                         error = 0;
3124         }
3125 
3126 out:
3127         sav->sav_pending = NULL;
3128         sav->sav_npending = 0;
3129         return (error);
3130 }
3131 
3132 static int
3133 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3134 {
3135         int error;
3136 
3137         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3138 
3139         if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3140             &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3141             VDEV_LABEL_SPARE)) != 0) {
3142                 return (error);
3143         }
3144 
3145         return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3146             &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3147             VDEV_LABEL_L2CACHE));
3148 }
3149 
3150 static void
3151 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3152     const char *config)
3153 {
3154         int i;
3155 
3156         if (sav->sav_config != NULL) {
3157                 nvlist_t **olddevs;
3158                 uint_t oldndevs;
3159                 nvlist_t **newdevs;
3160 
3161                 /*
3162                  * Generate new dev list by concatentating with the
3163                  * current dev list.
3164                  */
3165                 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3166                     &olddevs, &oldndevs) == 0);
3167 
3168                 newdevs = kmem_alloc(sizeof (void *) *
3169                     (ndevs + oldndevs), KM_SLEEP);
3170                 for (i = 0; i < oldndevs; i++)
3171                         VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3172                             KM_SLEEP) == 0);
3173                 for (i = 0; i < ndevs; i++)
3174                         VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3175                             KM_SLEEP) == 0);
3176 
3177                 VERIFY(nvlist_remove(sav->sav_config, config,
3178                     DATA_TYPE_NVLIST_ARRAY) == 0);
3179 
3180                 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3181                     config, newdevs, ndevs + oldndevs) == 0);
3182                 for (i = 0; i < oldndevs + ndevs; i++)
3183                         nvlist_free(newdevs[i]);
3184                 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3185         } else {
3186                 /*
3187                  * Generate a new dev list.
3188                  */
3189                 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3190                     KM_SLEEP) == 0);
3191                 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3192                     devs, ndevs) == 0);
3193         }
3194 }
3195 
3196 /*
3197  * Stop and drop level 2 ARC devices
3198  */
3199 void
3200 spa_l2cache_drop(spa_t *spa)
3201 {
3202         vdev_t *vd;
3203         int i;
3204         spa_aux_vdev_t *sav = &spa->spa_l2cache;
3205 
3206         for (i = 0; i < sav->sav_count; i++) {
3207                 uint64_t pool;
3208 
3209                 vd = sav->sav_vdevs[i];
3210                 ASSERT(vd != NULL);
3211 
3212                 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3213                     pool != 0ULL && l2arc_vdev_present(vd))
3214                         l2arc_remove_vdev(vd);
3215         }
3216 }
3217 
3218 /*
3219  * Pool Creation
3220  */
3221 int
3222 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3223     const char *history_str, nvlist_t *zplprops)
3224 {
3225         spa_t *spa;
3226         char *altroot = NULL;
3227         vdev_t *rvd;
3228         dsl_pool_t *dp;
3229         dmu_tx_t *tx;
3230         int error = 0;
3231         uint64_t txg = TXG_INITIAL;
3232         nvlist_t **spares, **l2cache;
3233         uint_t nspares, nl2cache;
3234         uint64_t version, obj;
3235         boolean_t has_features;
3236 
3237         /*
3238          * If this pool already exists, return failure.
3239          */
3240         mutex_enter(&spa_namespace_lock);
3241         if (spa_lookup(pool) != NULL) {
3242                 mutex_exit(&spa_namespace_lock);
3243                 return (EEXIST);
3244         }
3245 
3246         /*
3247          * Allocate a new spa_t structure.
3248          */
3249         (void) nvlist_lookup_string(props,
3250             zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3251         spa = spa_add(pool, NULL, altroot);
3252         spa_activate(spa, spa_mode_global);
3253 
3254         if (props && (error = spa_prop_validate(spa, props))) {
3255                 spa_deactivate(spa);
3256                 spa_remove(spa);
3257                 mutex_exit(&spa_namespace_lock);
3258                 return (error);
3259         }
3260 
3261         has_features = B_FALSE;
3262         for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3263             elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3264                 if (zpool_prop_feature(nvpair_name(elem)))
3265                         has_features = B_TRUE;
3266         }
3267 
3268         if (has_features || nvlist_lookup_uint64(props,
3269             zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3270                 version = SPA_VERSION;
3271         }
3272         ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3273 
3274         spa->spa_first_txg = txg;
3275         spa->spa_uberblock.ub_txg = txg - 1;
3276         spa->spa_uberblock.ub_version = version;
3277         spa->spa_ubsync = spa->spa_uberblock;
3278 
3279         /*
3280          * Create "The Godfather" zio to hold all async IOs
3281          */
3282         spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3283             ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3284 
3285         /*
3286          * Create the root vdev.
3287          */
3288         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3289 
3290         error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3291 
3292         ASSERT(error != 0 || rvd != NULL);
3293         ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3294 
3295         if (error == 0 && !zfs_allocatable_devs(nvroot))
3296                 error = EINVAL;
3297 
3298         if (error == 0 &&
3299             (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3300             (error = spa_validate_aux(spa, nvroot, txg,
3301             VDEV_ALLOC_ADD)) == 0) {
3302                 for (int c = 0; c < rvd->vdev_children; c++) {
3303                         vdev_metaslab_set_size(rvd->vdev_child[c]);
3304                         vdev_expand(rvd->vdev_child[c], txg);
3305                 }
3306         }
3307 
3308         spa_config_exit(spa, SCL_ALL, FTAG);
3309 
3310         if (error != 0) {
3311                 spa_unload(spa);
3312                 spa_deactivate(spa);
3313                 spa_remove(spa);
3314                 mutex_exit(&spa_namespace_lock);
3315                 return (error);
3316         }
3317 
3318         /*
3319          * Get the list of spares, if specified.
3320          */
3321         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3322             &spares, &nspares) == 0) {
3323                 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3324                     KM_SLEEP) == 0);
3325                 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3326                     ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3327                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3328                 spa_load_spares(spa);
3329                 spa_config_exit(spa, SCL_ALL, FTAG);
3330                 spa->spa_spares.sav_sync = B_TRUE;
3331         }
3332 
3333         /*
3334          * Get the list of level 2 cache devices, if specified.
3335          */
3336         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3337             &l2cache, &nl2cache) == 0) {
3338                 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3339                     NV_UNIQUE_NAME, KM_SLEEP) == 0);
3340                 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3341                     ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3342                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3343                 spa_load_l2cache(spa);
3344                 spa_config_exit(spa, SCL_ALL, FTAG);
3345                 spa->spa_l2cache.sav_sync = B_TRUE;
3346         }
3347 
3348         spa->spa_is_initializing = B_TRUE;
3349         spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3350         spa->spa_meta_objset = dp->dp_meta_objset;
3351         spa->spa_is_initializing = B_FALSE;
3352 
3353         /*
3354          * Create DDTs (dedup tables).
3355          */
3356         ddt_create(spa);
3357 
3358         spa_update_dspace(spa);
3359 
3360         tx = dmu_tx_create_assigned(dp, txg);
3361 
3362         /*
3363          * Create the pool config object.
3364          */
3365         spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3366             DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3367             DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3368 
3369         if (zap_add(spa->spa_meta_objset,
3370             DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3371             sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3372                 cmn_err(CE_PANIC, "failed to add pool config");
3373         }
3374 
3375         if (spa_version(spa) >= SPA_VERSION_FEATURES)
3376                 spa_feature_create_zap_objects(spa, tx);
3377 
3378         if (zap_add(spa->spa_meta_objset,
3379             DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3380             sizeof (uint64_t), 1, &version, tx) != 0) {
3381                 cmn_err(CE_PANIC, "failed to add pool version");
3382         }
3383 
3384         /* Newly created pools with the right version are always deflated. */
3385         if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3386                 spa->spa_deflate = TRUE;
3387                 if (zap_add(spa->spa_meta_objset,
3388                     DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3389                     sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3390                         cmn_err(CE_PANIC, "failed to add deflate");
3391                 }
3392         }
3393 
3394         /*
3395          * Create the deferred-free bpobj.  Turn off compression
3396          * because sync-to-convergence takes longer if the blocksize
3397          * keeps changing.
3398          */
3399         obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3400         dmu_object_set_compress(spa->spa_meta_objset, obj,
3401             ZIO_COMPRESS_OFF, tx);
3402         if (zap_add(spa->spa_meta_objset,
3403             DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3404             sizeof (uint64_t), 1, &obj, tx) != 0) {
3405                 cmn_err(CE_PANIC, "failed to add bpobj");
3406         }
3407         VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3408             spa->spa_meta_objset, obj));
3409 
3410         /*
3411          * Create the pool's history object.
3412          */
3413         if (version >= SPA_VERSION_ZPOOL_HISTORY)
3414                 spa_history_create_obj(spa, tx);
3415 
3416         /*
3417          * Set pool properties.
3418          */
3419         spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3420         spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3421         spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3422         spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3423 
3424         if (props != NULL) {
3425                 spa_configfile_set(spa, props, B_FALSE);
3426                 spa_sync_props(spa, props, tx);
3427         }
3428 
3429         dmu_tx_commit(tx);
3430 
3431         spa->spa_sync_on = B_TRUE;
3432         txg_sync_start(spa->spa_dsl_pool);
3433 
3434         /*
3435          * We explicitly wait for the first transaction to complete so that our
3436          * bean counters are appropriately updated.
3437          */
3438         txg_wait_synced(spa->spa_dsl_pool, txg);
3439 
3440         spa_config_sync(spa, B_FALSE, B_TRUE);
3441 
3442         if (version >= SPA_VERSION_ZPOOL_HISTORY && history_str != NULL)
3443                 (void) spa_history_log(spa, history_str, LOG_CMD_POOL_CREATE);
3444         spa_history_log_version(spa, LOG_POOL_CREATE);
3445 
3446         spa->spa_minref = refcount_count(&spa->spa_refcount);
3447 
3448         mutex_exit(&spa_namespace_lock);
3449 
3450         return (0);
3451 }
3452 
3453 #ifdef _KERNEL
3454 /*
3455  * Get the root pool information from the root disk, then import the root pool
3456  * during the system boot up time.
3457  */
3458 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3459 
3460 static nvlist_t *
3461 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3462 {
3463         nvlist_t *config;
3464         nvlist_t *nvtop, *nvroot;
3465         uint64_t pgid;
3466 
3467         if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3468                 return (NULL);
3469 
3470         /*
3471          * Add this top-level vdev to the child array.
3472          */
3473         VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3474             &nvtop) == 0);
3475         VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3476             &pgid) == 0);
3477         VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3478 
3479         /*
3480          * Put this pool's top-level vdevs into a root vdev.
3481          */
3482         VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3483         VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3484             VDEV_TYPE_ROOT) == 0);
3485         VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3486         VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3487         VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3488             &nvtop, 1) == 0);
3489 
3490         /*
3491          * Replace the existing vdev_tree with the new root vdev in
3492          * this pool's configuration (remove the old, add the new).
3493          */
3494         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3495         nvlist_free(nvroot);
3496         return (config);
3497 }
3498 
3499 /*
3500  * Walk the vdev tree and see if we can find a device with "better"
3501  * configuration. A configuration is "better" if the label on that
3502  * device has a more recent txg.
3503  */
3504 static void
3505 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3506 {
3507         for (int c = 0; c < vd->vdev_children; c++)
3508                 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3509 
3510         if (vd->vdev_ops->vdev_op_leaf) {
3511                 nvlist_t *label;
3512                 uint64_t label_txg;
3513 
3514                 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3515                     &label) != 0)
3516                         return;
3517 
3518                 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3519                     &label_txg) == 0);
3520 
3521                 /*
3522                  * Do we have a better boot device?
3523                  */
3524                 if (label_txg > *txg) {
3525                         *txg = label_txg;
3526                         *avd = vd;
3527                 }
3528                 nvlist_free(label);
3529         }
3530 }
3531 
3532 /*
3533  * Import a root pool.
3534  *
3535  * For x86. devpath_list will consist of devid and/or physpath name of
3536  * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3537  * The GRUB "findroot" command will return the vdev we should boot.
3538  *
3539  * For Sparc, devpath_list consists the physpath name of the booting device
3540  * no matter the rootpool is a single device pool or a mirrored pool.
3541  * e.g.
3542  *      "/pci@1f,0/ide@d/disk@0,0:a"
3543  */
3544 int
3545 spa_import_rootpool(char *devpath, char *devid)
3546 {
3547         spa_t *spa;
3548         vdev_t *rvd, *bvd, *avd = NULL;
3549         nvlist_t *config, *nvtop;
3550         uint64_t guid, txg;
3551         char *pname;
3552         int error;
3553 
3554         /*
3555          * Read the label from the boot device and generate a configuration.
3556          */
3557         config = spa_generate_rootconf(devpath, devid, &guid);
3558 #if defined(_OBP) && defined(_KERNEL)
3559         if (config == NULL) {
3560                 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3561                         /* iscsi boot */
3562                         get_iscsi_bootpath_phy(devpath);
3563                         config = spa_generate_rootconf(devpath, devid, &guid);
3564                 }
3565         }
3566 #endif
3567         if (config == NULL) {
3568                 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3569                     devpath);
3570                 return (EIO);
3571         }
3572 
3573         VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3574             &pname) == 0);
3575         VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3576 
3577         mutex_enter(&spa_namespace_lock);
3578         if ((spa = spa_lookup(pname)) != NULL) {
3579                 /*
3580                  * Remove the existing root pool from the namespace so that we
3581                  * can replace it with the correct config we just read in.
3582                  */
3583                 spa_remove(spa);
3584         }
3585 
3586         spa = spa_add(pname, config, NULL);
3587         spa->spa_is_root = B_TRUE;
3588         spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3589 
3590         /*
3591          * Build up a vdev tree based on the boot device's label config.
3592          */
3593         VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3594             &nvtop) == 0);
3595         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3596         error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3597             VDEV_ALLOC_ROOTPOOL);
3598         spa_config_exit(spa, SCL_ALL, FTAG);
3599         if (error) {
3600                 mutex_exit(&spa_namespace_lock);
3601                 nvlist_free(config);
3602                 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3603                     pname);
3604                 return (error);
3605         }
3606 
3607         /*
3608          * Get the boot vdev.
3609          */
3610         if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3611                 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3612                     (u_longlong_t)guid);
3613                 error = ENOENT;
3614                 goto out;
3615         }
3616 
3617         /*
3618          * Determine if there is a better boot device.
3619          */
3620         avd = bvd;
3621         spa_alt_rootvdev(rvd, &avd, &txg);
3622         if (avd != bvd) {
3623                 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3624                     "try booting from '%s'", avd->vdev_path);
3625                 error = EINVAL;
3626                 goto out;
3627         }
3628 
3629         /*
3630          * If the boot device is part of a spare vdev then ensure that
3631          * we're booting off the active spare.
3632          */
3633         if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3634             !bvd->vdev_isspare) {
3635                 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3636                     "try booting from '%s'",
3637                     bvd->vdev_parent->
3638                     vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3639                 error = EINVAL;
3640                 goto out;
3641         }
3642 
3643         error = 0;
3644         spa_history_log_version(spa, LOG_POOL_IMPORT);
3645 out:
3646         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3647         vdev_free(rvd);
3648         spa_config_exit(spa, SCL_ALL, FTAG);
3649         mutex_exit(&spa_namespace_lock);
3650 
3651         nvlist_free(config);
3652         return (error);
3653 }
3654 
3655 #endif
3656 
3657 /*
3658  * Import a non-root pool into the system.
3659  */
3660 int
3661 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3662 {
3663         spa_t *spa;
3664         char *altroot = NULL;
3665         spa_load_state_t state = SPA_LOAD_IMPORT;
3666         zpool_rewind_policy_t policy;
3667         uint64_t mode = spa_mode_global;
3668         uint64_t readonly = B_FALSE;
3669         int error;
3670         nvlist_t *nvroot;
3671         nvlist_t **spares, **l2cache;
3672         uint_t nspares, nl2cache;
3673 
3674         /*
3675          * If a pool with this name exists, return failure.
3676          */
3677         mutex_enter(&spa_namespace_lock);
3678         if (spa_lookup(pool) != NULL) {
3679                 mutex_exit(&spa_namespace_lock);
3680                 return (EEXIST);
3681         }
3682 
3683         /*
3684          * Create and initialize the spa structure.
3685          */
3686         (void) nvlist_lookup_string(props,
3687             zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3688         (void) nvlist_lookup_uint64(props,
3689             zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3690         if (readonly)
3691                 mode = FREAD;
3692         spa = spa_add(pool, config, altroot);
3693         spa->spa_import_flags = flags;
3694 
3695         /*
3696          * Verbatim import - Take a pool and insert it into the namespace
3697          * as if it had been loaded at boot.
3698          */
3699         if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3700                 if (props != NULL)
3701                         spa_configfile_set(spa, props, B_FALSE);
3702 
3703                 spa_config_sync(spa, B_FALSE, B_TRUE);
3704 
3705                 mutex_exit(&spa_namespace_lock);
3706                 spa_history_log_version(spa, LOG_POOL_IMPORT);
3707 
3708                 return (0);
3709         }
3710 
3711         spa_activate(spa, mode);
3712 
3713         /*
3714          * Don't start async tasks until we know everything is healthy.
3715          */
3716         spa_async_suspend(spa);
3717 
3718         zpool_get_rewind_policy(config, &policy);
3719         if (policy.zrp_request & ZPOOL_DO_REWIND)
3720                 state = SPA_LOAD_RECOVER;
3721 
3722         /*
3723          * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3724          * because the user-supplied config is actually the one to trust when
3725          * doing an import.
3726          */
3727         if (state != SPA_LOAD_RECOVER)
3728                 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3729 
3730         error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3731             policy.zrp_request);
3732 
3733         /*
3734          * Propagate anything learned while loading the pool and pass it
3735          * back to caller (i.e. rewind info, missing devices, etc).
3736          */
3737         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3738             spa->spa_load_info) == 0);
3739 
3740         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3741         /*
3742          * Toss any existing sparelist, as it doesn't have any validity
3743          * anymore, and conflicts with spa_has_spare().
3744          */
3745         if (spa->spa_spares.sav_config) {
3746                 nvlist_free(spa->spa_spares.sav_config);
3747                 spa->spa_spares.sav_config = NULL;
3748                 spa_load_spares(spa);
3749         }
3750         if (spa->spa_l2cache.sav_config) {
3751                 nvlist_free(spa->spa_l2cache.sav_config);
3752                 spa->spa_l2cache.sav_config = NULL;
3753                 spa_load_l2cache(spa);
3754         }
3755 
3756         VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3757             &nvroot) == 0);
3758         if (error == 0)
3759                 error = spa_validate_aux(spa, nvroot, -1ULL,
3760                     VDEV_ALLOC_SPARE);
3761         if (error == 0)
3762                 error = spa_validate_aux(spa, nvroot, -1ULL,
3763                     VDEV_ALLOC_L2CACHE);
3764         spa_config_exit(spa, SCL_ALL, FTAG);
3765 
3766         if (props != NULL)
3767                 spa_configfile_set(spa, props, B_FALSE);
3768 
3769         if (error != 0 || (props && spa_writeable(spa) &&
3770             (error = spa_prop_set(spa, props)))) {
3771                 spa_unload(spa);
3772                 spa_deactivate(spa);
3773                 spa_remove(spa);
3774                 mutex_exit(&spa_namespace_lock);
3775                 return (error);
3776         }
3777 
3778         spa_async_resume(spa);
3779 
3780         /*
3781          * Override any spares and level 2 cache devices as specified by
3782          * the user, as these may have correct device names/devids, etc.
3783          */
3784         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3785             &spares, &nspares) == 0) {
3786                 if (spa->spa_spares.sav_config)
3787                         VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3788                             ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3789                 else
3790                         VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3791                             NV_UNIQUE_NAME, KM_SLEEP) == 0);
3792                 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3793                     ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3794                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3795                 spa_load_spares(spa);
3796                 spa_config_exit(spa, SCL_ALL, FTAG);
3797                 spa->spa_spares.sav_sync = B_TRUE;
3798         }
3799         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3800             &l2cache, &nl2cache) == 0) {
3801                 if (spa->spa_l2cache.sav_config)
3802                         VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3803                             ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3804                 else
3805                         VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3806                             NV_UNIQUE_NAME, KM_SLEEP) == 0);
3807                 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3808                     ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3809                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3810                 spa_load_l2cache(spa);
3811                 spa_config_exit(spa, SCL_ALL, FTAG);
3812                 spa->spa_l2cache.sav_sync = B_TRUE;
3813         }
3814 
3815         /*
3816          * Check for any removed devices.
3817          */
3818         if (spa->spa_autoreplace) {
3819                 spa_aux_check_removed(&spa->spa_spares);
3820                 spa_aux_check_removed(&spa->spa_l2cache);
3821         }
3822 
3823         if (spa_writeable(spa)) {
3824                 /*
3825                  * Update the config cache to include the newly-imported pool.
3826                  */
3827                 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3828         }
3829 
3830         /*
3831          * It's possible that the pool was expanded while it was exported.
3832          * We kick off an async task to handle this for us.
3833          */
3834         spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3835 
3836         mutex_exit(&spa_namespace_lock);
3837         spa_history_log_version(spa, LOG_POOL_IMPORT);
3838 
3839         return (0);
3840 }
3841 
3842 nvlist_t *
3843 spa_tryimport(nvlist_t *tryconfig)
3844 {
3845         nvlist_t *config = NULL;
3846         char *poolname;
3847         spa_t *spa;
3848         uint64_t state;
3849         int error;
3850 
3851         if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3852                 return (NULL);
3853 
3854         if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3855                 return (NULL);
3856 
3857         /*
3858          * Create and initialize the spa structure.
3859          */
3860         mutex_enter(&spa_namespace_lock);
3861         spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3862         spa_activate(spa, FREAD);
3863 
3864         /*
3865          * Pass off the heavy lifting to spa_load().
3866          * Pass TRUE for mosconfig because the user-supplied config
3867          * is actually the one to trust when doing an import.
3868          */
3869         error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
3870 
3871         /*
3872          * If 'tryconfig' was at least parsable, return the current config.
3873          */
3874         if (spa->spa_root_vdev != NULL) {
3875                 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
3876                 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
3877                     poolname) == 0);
3878                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
3879                     state) == 0);
3880                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
3881                     spa->spa_uberblock.ub_timestamp) == 0);
3882                 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3883                     spa->spa_load_info) == 0);
3884 
3885                 /*
3886                  * If the bootfs property exists on this pool then we
3887                  * copy it out so that external consumers can tell which
3888                  * pools are bootable.
3889                  */
3890                 if ((!error || error == EEXIST) && spa->spa_bootfs) {
3891                         char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3892 
3893                         /*
3894                          * We have to play games with the name since the
3895                          * pool was opened as TRYIMPORT_NAME.
3896                          */
3897                         if (dsl_dsobj_to_dsname(spa_name(spa),
3898                             spa->spa_bootfs, tmpname) == 0) {
3899                                 char *cp;
3900                                 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
3901 
3902                                 cp = strchr(tmpname, '/');
3903                                 if (cp == NULL) {
3904                                         (void) strlcpy(dsname, tmpname,
3905                                             MAXPATHLEN);
3906                                 } else {
3907                                         (void) snprintf(dsname, MAXPATHLEN,
3908                                             "%s/%s", poolname, ++cp);
3909                                 }
3910                                 VERIFY(nvlist_add_string(config,
3911                                     ZPOOL_CONFIG_BOOTFS, dsname) == 0);
3912                                 kmem_free(dsname, MAXPATHLEN);
3913                         }
3914                         kmem_free(tmpname, MAXPATHLEN);
3915                 }
3916 
3917                 /*
3918                  * Add the list of hot spares and level 2 cache devices.
3919                  */
3920                 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3921                 spa_add_spares(spa, config);
3922                 spa_add_l2cache(spa, config);
3923                 spa_config_exit(spa, SCL_CONFIG, FTAG);
3924         }
3925 
3926         spa_unload(spa);
3927         spa_deactivate(spa);
3928         spa_remove(spa);
3929         mutex_exit(&spa_namespace_lock);
3930 
3931         return (config);
3932 }
3933 
3934 /*
3935  * Pool export/destroy
3936  *
3937  * The act of destroying or exporting a pool is very simple.  We make sure there
3938  * is no more pending I/O and any references to the pool are gone.  Then, we
3939  * update the pool state and sync all the labels to disk, removing the
3940  * configuration from the cache afterwards. If the 'hardforce' flag is set, then
3941  * we don't sync the labels or remove the configuration cache.
3942  */
3943 static int
3944 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
3945     boolean_t force, boolean_t hardforce)
3946 {
3947         spa_t *spa;
3948 
3949         if (oldconfig)
3950                 *oldconfig = NULL;
3951 
3952         if (!(spa_mode_global & FWRITE))
3953                 return (EROFS);
3954 
3955         mutex_enter(&spa_namespace_lock);
3956         if ((spa = spa_lookup(pool)) == NULL) {
3957                 mutex_exit(&spa_namespace_lock);
3958                 return (ENOENT);
3959         }
3960 
3961         /*
3962          * Put a hold on the pool, drop the namespace lock, stop async tasks,
3963          * reacquire the namespace lock, and see if we can export.
3964          */
3965         spa_open_ref(spa, FTAG);
3966         mutex_exit(&spa_namespace_lock);
3967         spa_async_suspend(spa);
3968         mutex_enter(&spa_namespace_lock);
3969         spa_close(spa, FTAG);
3970 
3971         /*
3972          * The pool will be in core if it's openable,
3973          * in which case we can modify its state.
3974          */
3975         if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
3976                 /*
3977                  * Objsets may be open only because they're dirty, so we
3978                  * have to force it to sync before checking spa_refcnt.
3979                  */
3980                 txg_wait_synced(spa->spa_dsl_pool, 0);
3981 
3982                 /*
3983                  * A pool cannot be exported or destroyed if there are active
3984                  * references.  If we are resetting a pool, allow references by
3985                  * fault injection handlers.
3986                  */
3987                 if (!spa_refcount_zero(spa) ||
3988                     (spa->spa_inject_ref != 0 &&
3989                     new_state != POOL_STATE_UNINITIALIZED)) {
3990                         spa_async_resume(spa);
3991                         mutex_exit(&spa_namespace_lock);
3992                         return (EBUSY);
3993                 }
3994 
3995                 /*
3996                  * A pool cannot be exported if it has an active shared spare.
3997                  * This is to prevent other pools stealing the active spare
3998                  * from an exported pool. At user's own will, such pool can
3999                  * be forcedly exported.
4000                  */
4001                 if (!force && new_state == POOL_STATE_EXPORTED &&
4002                     spa_has_active_shared_spare(spa)) {
4003                         spa_async_resume(spa);
4004                         mutex_exit(&spa_namespace_lock);
4005                         return (EXDEV);
4006                 }
4007 
4008                 /*
4009                  * We want this to be reflected on every label,
4010                  * so mark them all dirty.  spa_unload() will do the
4011                  * final sync that pushes these changes out.
4012                  */
4013                 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4014                         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4015                         spa->spa_state = new_state;
4016                         spa->spa_final_txg = spa_last_synced_txg(spa) +
4017                             TXG_DEFER_SIZE + 1;
4018                         vdev_config_dirty(spa->spa_root_vdev);
4019                         spa_config_exit(spa, SCL_ALL, FTAG);
4020                 }
4021         }
4022 
4023         spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4024 
4025         if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4026                 spa_unload(spa);
4027                 spa_deactivate(spa);
4028         }
4029 
4030         if (oldconfig && spa->spa_config)
4031                 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4032 
4033         if (new_state != POOL_STATE_UNINITIALIZED) {
4034                 if (!hardforce)
4035                         spa_config_sync(spa, B_TRUE, B_TRUE);
4036                 spa_remove(spa);
4037         }
4038         mutex_exit(&spa_namespace_lock);
4039 
4040         return (0);
4041 }
4042 
4043 /*
4044  * Destroy a storage pool.
4045  */
4046 int
4047 spa_destroy(char *pool)
4048 {
4049         return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4050             B_FALSE, B_FALSE));
4051 }
4052 
4053 /*
4054  * Export a storage pool.
4055  */
4056 int
4057 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4058     boolean_t hardforce)
4059 {
4060         return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4061             force, hardforce));
4062 }
4063 
4064 /*
4065  * Similar to spa_export(), this unloads the spa_t without actually removing it
4066  * from the namespace in any way.
4067  */
4068 int
4069 spa_reset(char *pool)
4070 {
4071         return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4072             B_FALSE, B_FALSE));
4073 }
4074 
4075 /*
4076  * ==========================================================================
4077  * Device manipulation
4078  * ==========================================================================
4079  */
4080 
4081 /*
4082  * Add a device to a storage pool.
4083  */
4084 int
4085 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4086 {
4087         uint64_t txg, id;
4088         int error;
4089         vdev_t *rvd = spa->spa_root_vdev;
4090         vdev_t *vd, *tvd;
4091         nvlist_t **spares, **l2cache;
4092         uint_t nspares, nl2cache;
4093 
4094         ASSERT(spa_writeable(spa));
4095 
4096         txg = spa_vdev_enter(spa);
4097 
4098         if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4099             VDEV_ALLOC_ADD)) != 0)
4100                 return (spa_vdev_exit(spa, NULL, txg, error));
4101 
4102         spa->spa_pending_vdev = vd;  /* spa_vdev_exit() will clear this */
4103 
4104         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4105             &nspares) != 0)
4106                 nspares = 0;
4107 
4108         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4109             &nl2cache) != 0)
4110                 nl2cache = 0;
4111 
4112         if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4113                 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4114 
4115         if (vd->vdev_children != 0 &&
4116             (error = vdev_create(vd, txg, B_FALSE)) != 0)
4117                 return (spa_vdev_exit(spa, vd, txg, error));
4118 
4119         /*
4120          * We must validate the spares and l2cache devices after checking the
4121          * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4122          */
4123         if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4124                 return (spa_vdev_exit(spa, vd, txg, error));
4125 
4126         /*
4127          * Transfer each new top-level vdev from vd to rvd.
4128          */
4129         for (int c = 0; c < vd->vdev_children; c++) {
4130 
4131                 /*
4132                  * Set the vdev id to the first hole, if one exists.
4133                  */
4134                 for (id = 0; id < rvd->vdev_children; id++) {
4135                         if (rvd->vdev_child[id]->vdev_ishole) {
4136                                 vdev_free(rvd->vdev_child[id]);
4137                                 break;
4138                         }
4139                 }
4140                 tvd = vd->vdev_child[c];
4141                 vdev_remove_child(vd, tvd);
4142                 tvd->vdev_id = id;
4143                 vdev_add_child(rvd, tvd);
4144                 vdev_config_dirty(tvd);
4145         }
4146 
4147         if (nspares != 0) {
4148                 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4149                     ZPOOL_CONFIG_SPARES);
4150                 spa_load_spares(spa);
4151                 spa->spa_spares.sav_sync = B_TRUE;
4152         }
4153 
4154         if (nl2cache != 0) {
4155                 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4156                     ZPOOL_CONFIG_L2CACHE);
4157                 spa_load_l2cache(spa);
4158                 spa->spa_l2cache.sav_sync = B_TRUE;
4159         }
4160 
4161         /*
4162          * We have to be careful when adding new vdevs to an existing pool.
4163          * If other threads start allocating from these vdevs before we
4164          * sync the config cache, and we lose power, then upon reboot we may
4165          * fail to open the pool because there are DVAs that the config cache
4166          * can't translate.  Therefore, we first add the vdevs without
4167          * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4168          * and then let spa_config_update() initialize the new metaslabs.
4169          *
4170          * spa_load() checks for added-but-not-initialized vdevs, so that
4171          * if we lose power at any point in this sequence, the remaining
4172          * steps will be completed the next time we load the pool.
4173          */
4174         (void) spa_vdev_exit(spa, vd, txg, 0);
4175 
4176         mutex_enter(&spa_namespace_lock);
4177         spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4178         mutex_exit(&spa_namespace_lock);
4179 
4180         return (0);
4181 }
4182 
4183 /*
4184  * Attach a device to a mirror.  The arguments are the path to any device
4185  * in the mirror, and the nvroot for the new device.  If the path specifies
4186  * a device that is not mirrored, we automatically insert the mirror vdev.
4187  *
4188  * If 'replacing' is specified, the new device is intended to replace the
4189  * existing device; in this case the two devices are made into their own
4190  * mirror using the 'replacing' vdev, which is functionally identical to
4191  * the mirror vdev (it actually reuses all the same ops) but has a few
4192  * extra rules: you can't attach to it after it's been created, and upon
4193  * completion of resilvering, the first disk (the one being replaced)
4194  * is automatically detached.
4195  */
4196 int
4197 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4198 {
4199         uint64_t txg, dtl_max_txg;
4200         vdev_t *rvd = spa->spa_root_vdev;
4201         vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4202         vdev_ops_t *pvops;
4203         char *oldvdpath, *newvdpath;
4204         int newvd_isspare;
4205         int error;
4206 
4207         ASSERT(spa_writeable(spa));
4208 
4209         txg = spa_vdev_enter(spa);
4210 
4211         oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4212 
4213         if (oldvd == NULL)
4214                 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4215 
4216         if (!oldvd->vdev_ops->vdev_op_leaf)
4217                 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4218 
4219         pvd = oldvd->vdev_parent;
4220 
4221         if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4222             VDEV_ALLOC_ATTACH)) != 0)
4223                 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4224 
4225         if (newrootvd->vdev_children != 1)
4226                 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4227 
4228         newvd = newrootvd->vdev_child[0];
4229 
4230         if (!newvd->vdev_ops->vdev_op_leaf)
4231                 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4232 
4233         if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4234                 return (spa_vdev_exit(spa, newrootvd, txg, error));
4235 
4236         /*
4237          * Spares can't replace logs
4238          */
4239         if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4240                 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4241 
4242         if (!replacing) {
4243                 /*
4244                  * For attach, the only allowable parent is a mirror or the root
4245                  * vdev.
4246                  */
4247                 if (pvd->vdev_ops != &vdev_mirror_ops &&
4248                     pvd->vdev_ops != &vdev_root_ops)
4249                         return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4250 
4251                 pvops = &vdev_mirror_ops;
4252         } else {
4253                 /*
4254                  * Active hot spares can only be replaced by inactive hot
4255                  * spares.
4256                  */
4257                 if (pvd->vdev_ops == &vdev_spare_ops &&
4258                     oldvd->vdev_isspare &&
4259                     !spa_has_spare(spa, newvd->vdev_guid))
4260                         return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4261 
4262                 /*
4263                  * If the source is a hot spare, and the parent isn't already a
4264                  * spare, then we want to create a new hot spare.  Otherwise, we
4265                  * want to create a replacing vdev.  The user is not allowed to
4266                  * attach to a spared vdev child unless the 'isspare' state is
4267                  * the same (spare replaces spare, non-spare replaces
4268                  * non-spare).
4269                  */
4270                 if (pvd->vdev_ops == &vdev_replacing_ops &&
4271                     spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4272                         return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4273                 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4274                     newvd->vdev_isspare != oldvd->vdev_isspare) {
4275                         return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4276                 }
4277 
4278                 if (newvd->vdev_isspare)
4279                         pvops = &vdev_spare_ops;
4280                 else
4281                         pvops = &vdev_replacing_ops;
4282         }
4283 
4284         /*
4285          * Make sure the new device is big enough.
4286          */
4287         if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4288                 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4289 
4290         /*
4291          * The new device cannot have a higher alignment requirement
4292          * than the top-level vdev.
4293          */
4294         if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4295                 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4296 
4297         /*
4298          * If this is an in-place replacement, update oldvd's path and devid
4299          * to make it distinguishable from newvd, and unopenable from now on.
4300          */
4301         if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4302                 spa_strfree(oldvd->vdev_path);
4303                 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4304                     KM_SLEEP);
4305                 (void) sprintf(oldvd->vdev_path, "%s/%s",
4306                     newvd->vdev_path, "old");
4307                 if (oldvd->vdev_devid != NULL) {
4308                         spa_strfree(oldvd->vdev_devid);
4309                         oldvd->vdev_devid = NULL;
4310                 }
4311         }
4312 
4313         /* mark the device being resilvered */
4314         newvd->vdev_resilvering = B_TRUE;
4315 
4316         /*
4317          * If the parent is not a mirror, or if we're replacing, insert the new
4318          * mirror/replacing/spare vdev above oldvd.
4319          */
4320         if (pvd->vdev_ops != pvops)
4321                 pvd = vdev_add_parent(oldvd, pvops);
4322 
4323         ASSERT(pvd->vdev_top->vdev_parent == rvd);
4324         ASSERT(pvd->vdev_ops == pvops);
4325         ASSERT(oldvd->vdev_parent == pvd);
4326 
4327         /*
4328          * Extract the new device from its root and add it to pvd.
4329          */
4330         vdev_remove_child(newrootvd, newvd);
4331         newvd->vdev_id = pvd->vdev_children;
4332         newvd->vdev_crtxg = oldvd->vdev_crtxg;
4333         vdev_add_child(pvd, newvd);
4334 
4335         tvd = newvd->vdev_top;
4336         ASSERT(pvd->vdev_top == tvd);
4337         ASSERT(tvd->vdev_parent == rvd);
4338 
4339         vdev_config_dirty(tvd);
4340 
4341         /*
4342          * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4343          * for any dmu_sync-ed blocks.  It will propagate upward when
4344          * spa_vdev_exit() calls vdev_dtl_reassess().
4345          */
4346         dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4347 
4348         vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4349             dtl_max_txg - TXG_INITIAL);
4350 
4351         if (newvd->vdev_isspare) {
4352                 spa_spare_activate(newvd);
4353                 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4354         }
4355 
4356         oldvdpath = spa_strdup(oldvd->vdev_path);
4357         newvdpath = spa_strdup(newvd->vdev_path);
4358         newvd_isspare = newvd->vdev_isspare;
4359 
4360         /*
4361          * Mark newvd's DTL dirty in this txg.
4362          */
4363         vdev_dirty(tvd, VDD_DTL, newvd, txg);
4364 
4365         /*
4366          * Restart the resilver
4367          */
4368         dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4369 
4370         /*
4371          * Commit the config
4372          */
4373         (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4374 
4375         spa_history_log_internal(LOG_POOL_VDEV_ATTACH, spa, NULL,
4376             "%s vdev=%s %s vdev=%s",
4377             replacing && newvd_isspare ? "spare in" :
4378             replacing ? "replace" : "attach", newvdpath,
4379             replacing ? "for" : "to", oldvdpath);
4380 
4381         spa_strfree(oldvdpath);
4382         spa_strfree(newvdpath);
4383 
4384         if (spa->spa_bootfs)
4385                 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4386 
4387         return (0);
4388 }
4389 
4390 /*
4391  * Detach a device from a mirror or replacing vdev.
4392  * If 'replace_done' is specified, only detach if the parent
4393  * is a replacing vdev.
4394  */
4395 int
4396 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4397 {
4398         uint64_t txg;
4399         int error;
4400         vdev_t *rvd = spa->spa_root_vdev;
4401         vdev_t *vd, *pvd, *cvd, *tvd;
4402         boolean_t unspare = B_FALSE;
4403         uint64_t unspare_guid;
4404         char *vdpath;
4405 
4406         ASSERT(spa_writeable(spa));
4407 
4408         txg = spa_vdev_enter(spa);
4409 
4410         vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4411 
4412         if (vd == NULL)
4413                 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4414 
4415         if (!vd->vdev_ops->vdev_op_leaf)
4416                 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4417 
4418         pvd = vd->vdev_parent;
4419 
4420         /*
4421          * If the parent/child relationship is not as expected, don't do it.
4422          * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4423          * vdev that's replacing B with C.  The user's intent in replacing
4424          * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4425          * the replace by detaching C, the expected behavior is to end up
4426          * M(A,B).  But suppose that right after deciding to detach C,
4427          * the replacement of B completes.  We would have M(A,C), and then
4428          * ask to detach C, which would leave us with just A -- not what
4429          * the user wanted.  To prevent this, we make sure that the
4430          * parent/child relationship hasn't changed -- in this example,
4431          * that C's parent is still the replacing vdev R.
4432          */
4433         if (pvd->vdev_guid != pguid && pguid != 0)
4434                 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4435 
4436         /*
4437          * Only 'replacing' or 'spare' vdevs can be replaced.
4438          */
4439         if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4440             pvd->vdev_ops != &vdev_spare_ops)
4441                 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4442 
4443         ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4444             spa_version(spa) >= SPA_VERSION_SPARES);
4445 
4446         /*
4447          * Only mirror, replacing, and spare vdevs support detach.
4448          */
4449         if (pvd->vdev_ops != &vdev_replacing_ops &&
4450             pvd->vdev_ops != &vdev_mirror_ops &&
4451             pvd->vdev_ops != &vdev_spare_ops)
4452                 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4453 
4454         /*
4455          * If this device has the only valid copy of some data,
4456          * we cannot safely detach it.
4457          */
4458         if (vdev_dtl_required(vd))
4459                 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4460 
4461         ASSERT(pvd->vdev_children >= 2);
4462 
4463         /*
4464          * If we are detaching the second disk from a replacing vdev, then
4465          * check to see if we changed the original vdev's path to have "/old"
4466          * at the end in spa_vdev_attach().  If so, undo that change now.
4467          */
4468         if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4469             vd->vdev_path != NULL) {
4470                 size_t len = strlen(vd->vdev_path);
4471 
4472                 for (int c = 0; c < pvd->vdev_children; c++) {
4473                         cvd = pvd->vdev_child[c];
4474 
4475                         if (cvd == vd || cvd->vdev_path == NULL)
4476                                 continue;
4477 
4478                         if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4479                             strcmp(cvd->vdev_path + len, "/old") == 0) {
4480                                 spa_strfree(cvd->vdev_path);
4481                                 cvd->vdev_path = spa_strdup(vd->vdev_path);
4482                                 break;
4483                         }
4484                 }
4485         }
4486 
4487         /*
4488          * If we are detaching the original disk from a spare, then it implies
4489          * that the spare should become a real disk, and be removed from the
4490          * active spare list for the pool.
4491          */
4492         if (pvd->vdev_ops == &vdev_spare_ops &&
4493             vd->vdev_id == 0 &&
4494             pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4495                 unspare = B_TRUE;
4496 
4497         /*
4498          * Erase the disk labels so the disk can be used for other things.
4499          * This must be done after all other error cases are handled,
4500          * but before we disembowel vd (so we can still do I/O to it).
4501          * But if we can't do it, don't treat the error as fatal --
4502          * it may be that the unwritability of the disk is the reason
4503          * it's being detached!
4504          */
4505         error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4506 
4507         /*
4508          * Remove vd from its parent and compact the parent's children.
4509          */
4510         vdev_remove_child(pvd, vd);
4511         vdev_compact_children(pvd);
4512 
4513         /*
4514          * Remember one of the remaining children so we can get tvd below.
4515          */
4516         cvd = pvd->vdev_child[pvd->vdev_children - 1];
4517 
4518         /*
4519          * If we need to remove the remaining child from the list of hot spares,
4520          * do it now, marking the vdev as no longer a spare in the process.
4521          * We must do this before vdev_remove_parent(), because that can
4522          * change the GUID if it creates a new toplevel GUID.  For a similar
4523          * reason, we must remove the spare now, in the same txg as the detach;
4524          * otherwise someone could attach a new sibling, change the GUID, and
4525          * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4526          */
4527         if (unspare) {
4528                 ASSERT(cvd->vdev_isspare);
4529                 spa_spare_remove(cvd);
4530                 unspare_guid = cvd->vdev_guid;
4531                 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4532                 cvd->vdev_unspare = B_TRUE;
4533         }
4534 
4535         /*
4536          * If the parent mirror/replacing vdev only has one child,
4537          * the parent is no longer needed.  Remove it from the tree.
4538          */
4539         if (pvd->vdev_children == 1) {
4540                 if (pvd->vdev_ops == &vdev_spare_ops)
4541                         cvd->vdev_unspare = B_FALSE;
4542                 vdev_remove_parent(cvd);
4543                 cvd->vdev_resilvering = B_FALSE;
4544         }
4545 
4546 
4547         /*
4548          * We don't set tvd until now because the parent we just removed
4549          * may have been the previous top-level vdev.
4550          */
4551         tvd = cvd->vdev_top;
4552         ASSERT(tvd->vdev_parent == rvd);
4553 
4554         /*
4555          * Reevaluate the parent vdev state.
4556          */
4557         vdev_propagate_state(cvd);
4558 
4559         /*
4560          * If the 'autoexpand' property is set on the pool then automatically
4561          * try to expand the size of the pool. For example if the device we
4562          * just detached was smaller than the others, it may be possible to
4563          * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4564          * first so that we can obtain the updated sizes of the leaf vdevs.
4565          */
4566         if (spa->spa_autoexpand) {
4567                 vdev_reopen(tvd);
4568                 vdev_expand(tvd, txg);
4569         }
4570 
4571         vdev_config_dirty(tvd);
4572 
4573         /*
4574          * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4575          * vd->vdev_detached is set and free vd's DTL object in syncing context.
4576          * But first make sure we're not on any *other* txg's DTL list, to
4577          * prevent vd from being accessed after it's freed.
4578          */
4579         vdpath = spa_strdup(vd->vdev_path);
4580         for (int t = 0; t < TXG_SIZE; t++)
4581                 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4582         vd->vdev_detached = B_TRUE;
4583         vdev_dirty(tvd, VDD_DTL, vd, txg);
4584 
4585         spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4586 
4587         /* hang on to the spa before we release the lock */
4588         spa_open_ref(spa, FTAG);
4589 
4590         error = spa_vdev_exit(spa, vd, txg, 0);
4591 
4592         spa_history_log_internal(LOG_POOL_VDEV_DETACH, spa, NULL,
4593             "vdev=%s", vdpath);
4594         spa_strfree(vdpath);
4595 
4596         /*
4597          * If this was the removal of the original device in a hot spare vdev,
4598          * then we want to go through and remove the device from the hot spare
4599          * list of every other pool.
4600          */
4601         if (unspare) {
4602                 spa_t *altspa = NULL;
4603 
4604                 mutex_enter(&spa_namespace_lock);
4605                 while ((altspa = spa_next(altspa)) != NULL) {
4606                         if (altspa->spa_state != POOL_STATE_ACTIVE ||
4607                             altspa == spa)
4608                                 continue;
4609 
4610                         spa_open_ref(altspa, FTAG);
4611                         mutex_exit(&spa_namespace_lock);
4612                         (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4613                         mutex_enter(&spa_namespace_lock);
4614                         spa_close(altspa, FTAG);
4615                 }
4616                 mutex_exit(&spa_namespace_lock);
4617 
4618                 /* search the rest of the vdevs for spares to remove */
4619                 spa_vdev_resilver_done(spa);
4620         }
4621 
4622         /* all done with the spa; OK to release */
4623         mutex_enter(&spa_namespace_lock);
4624         spa_close(spa, FTAG);
4625         mutex_exit(&spa_namespace_lock);
4626 
4627         return (error);
4628 }
4629 
4630 /*
4631  * Split a set of devices from their mirrors, and create a new pool from them.
4632  */
4633 int
4634 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4635     nvlist_t *props, boolean_t exp)
4636 {
4637         int error = 0;
4638         uint64_t txg, *glist;
4639         spa_t *newspa;
4640         uint_t c, children, lastlog;
4641         nvlist_t **child, *nvl, *tmp;
4642         dmu_tx_t *tx;
4643         char *altroot = NULL;
4644         vdev_t *rvd, **vml = NULL;                      /* vdev modify list */
4645         boolean_t activate_slog;
4646 
4647         ASSERT(spa_writeable(spa));
4648 
4649         txg = spa_vdev_enter(spa);
4650 
4651         /* clear the log and flush everything up to now */
4652         activate_slog = spa_passivate_log(spa);
4653         (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4654         error = spa_offline_log(spa);
4655         txg = spa_vdev_config_enter(spa);
4656 
4657         if (activate_slog)
4658                 spa_activate_log(spa);
4659 
4660         if (error != 0)
4661                 return (spa_vdev_exit(spa, NULL, txg, error));
4662 
4663         /* check new spa name before going any further */
4664         if (spa_lookup(newname) != NULL)
4665                 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4666 
4667         /*
4668          * scan through all the children to ensure they're all mirrors
4669          */
4670         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4671             nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4672             &children) != 0)
4673                 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4674 
4675         /* first, check to ensure we've got the right child count */
4676         rvd = spa->spa_root_vdev;
4677         lastlog = 0;
4678         for (c = 0; c < rvd->vdev_children; c++) {
4679                 vdev_t *vd = rvd->vdev_child[c];
4680 
4681                 /* don't count the holes & logs as children */
4682                 if (vd->vdev_islog || vd->vdev_ishole) {
4683                         if (lastlog == 0)
4684                                 lastlog = c;
4685                         continue;
4686                 }
4687 
4688                 lastlog = 0;
4689         }
4690         if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4691                 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4692 
4693         /* next, ensure no spare or cache devices are part of the split */
4694         if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4695             nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4696                 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4697 
4698         vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4699         glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4700 
4701         /* then, loop over each vdev and validate it */
4702         for (c = 0; c < children; c++) {
4703                 uint64_t is_hole = 0;
4704 
4705                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4706                     &is_hole);
4707 
4708                 if (is_hole != 0) {
4709                         if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4710                             spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4711                                 continue;
4712                         } else {
4713                                 error = EINVAL;
4714                                 break;
4715                         }
4716                 }
4717 
4718                 /* which disk is going to be split? */
4719                 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4720                     &glist[c]) != 0) {
4721                         error = EINVAL;
4722                         break;
4723                 }
4724 
4725                 /* look it up in the spa */
4726                 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4727                 if (vml[c] == NULL) {
4728                         error = ENODEV;
4729                         break;
4730                 }
4731 
4732                 /* make sure there's nothing stopping the split */
4733                 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4734                     vml[c]->vdev_islog ||
4735                     vml[c]->vdev_ishole ||
4736                     vml[c]->vdev_isspare ||
4737                     vml[c]->vdev_isl2cache ||
4738                     !vdev_writeable(vml[c]) ||
4739                     vml[c]->vdev_children != 0 ||
4740                     vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4741                     c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4742                         error = EINVAL;
4743                         break;
4744                 }
4745 
4746                 if (vdev_dtl_required(vml[c])) {
4747                         error = EBUSY;
4748                         break;
4749                 }
4750 
4751                 /* we need certain info from the top level */
4752                 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4753                     vml[c]->vdev_top->vdev_ms_array) == 0);
4754                 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4755                     vml[c]->vdev_top->vdev_ms_shift) == 0);
4756                 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4757                     vml[c]->vdev_top->vdev_asize) == 0);
4758                 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4759                     vml[c]->vdev_top->vdev_ashift) == 0);
4760         }
4761 
4762         if (error != 0) {
4763                 kmem_free(vml, children * sizeof (vdev_t *));
4764                 kmem_free(glist, children * sizeof (uint64_t));
4765                 return (spa_vdev_exit(spa, NULL, txg, error));
4766         }
4767 
4768         /* stop writers from using the disks */
4769         for (c = 0; c < children; c++) {
4770                 if (vml[c] != NULL)
4771                         vml[c]->vdev_offline = B_TRUE;
4772         }
4773         vdev_reopen(spa->spa_root_vdev);
4774 
4775         /*
4776          * Temporarily record the splitting vdevs in the spa config.  This
4777          * will disappear once the config is regenerated.
4778          */
4779         VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4780         VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4781             glist, children) == 0);
4782         kmem_free(glist, children * sizeof (uint64_t));
4783 
4784         mutex_enter(&spa->spa_props_lock);
4785         VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4786             nvl) == 0);
4787         mutex_exit(&spa->spa_props_lock);
4788         spa->spa_config_splitting = nvl;
4789         vdev_config_dirty(spa->spa_root_vdev);
4790 
4791         /* configure and create the new pool */
4792         VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4793         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4794             exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4795         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4796             spa_version(spa)) == 0);
4797         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4798             spa->spa_config_txg) == 0);
4799         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4800             spa_generate_guid(NULL)) == 0);
4801         (void) nvlist_lookup_string(props,
4802             zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4803 
4804         /* add the new pool to the namespace */
4805         newspa = spa_add(newname, config, altroot);
4806         newspa->spa_config_txg = spa->spa_config_txg;
4807         spa_set_log_state(newspa, SPA_LOG_CLEAR);
4808 
4809         /* release the spa config lock, retaining the namespace lock */
4810         spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4811 
4812         if (zio_injection_enabled)
4813                 zio_handle_panic_injection(spa, FTAG, 1);
4814 
4815         spa_activate(newspa, spa_mode_global);
4816         spa_async_suspend(newspa);
4817 
4818         /* create the new pool from the disks of the original pool */
4819         error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4820         if (error)
4821                 goto out;
4822 
4823         /* if that worked, generate a real config for the new pool */
4824         if (newspa->spa_root_vdev != NULL) {
4825                 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4826                     NV_UNIQUE_NAME, KM_SLEEP) == 0);
4827                 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4828                     ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4829                 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4830                     B_TRUE));
4831         }
4832 
4833         /* set the props */
4834         if (props != NULL) {
4835                 spa_configfile_set(newspa, props, B_FALSE);
4836                 error = spa_prop_set(newspa, props);
4837                 if (error)
4838                         goto out;
4839         }
4840 
4841         /* flush everything */
4842         txg = spa_vdev_config_enter(newspa);
4843         vdev_config_dirty(newspa->spa_root_vdev);
4844         (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4845 
4846         if (zio_injection_enabled)
4847                 zio_handle_panic_injection(spa, FTAG, 2);
4848 
4849         spa_async_resume(newspa);
4850 
4851         /* finally, update the original pool's config */
4852         txg = spa_vdev_config_enter(spa);
4853         tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4854         error = dmu_tx_assign(tx, TXG_WAIT);
4855         if (error != 0)
4856                 dmu_tx_abort(tx);
4857         for (c = 0; c < children; c++) {
4858                 if (vml[c] != NULL) {
4859                         vdev_split(vml[c]);
4860                         if (error == 0)
4861                                 spa_history_log_internal(LOG_POOL_VDEV_DETACH,
4862                                     spa, tx, "vdev=%s",
4863                                     vml[c]->vdev_path);
4864                         vdev_free(vml[c]);
4865                 }
4866         }
4867         vdev_config_dirty(spa->spa_root_vdev);
4868         spa->spa_config_splitting = NULL;
4869         nvlist_free(nvl);
4870         if (error == 0)
4871                 dmu_tx_commit(tx);
4872         (void) spa_vdev_exit(spa, NULL, txg, 0);
4873 
4874         if (zio_injection_enabled)
4875                 zio_handle_panic_injection(spa, FTAG, 3);
4876 
4877         /* split is complete; log a history record */
4878         spa_history_log_internal(LOG_POOL_SPLIT, newspa, NULL,
4879             "split new pool %s from pool %s", newname, spa_name(spa));
4880 
4881         kmem_free(vml, children * sizeof (vdev_t *));
4882 
4883         /* if we're not going to mount the filesystems in userland, export */
4884         if (exp)
4885                 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
4886                     B_FALSE, B_FALSE);
4887 
4888         return (error);
4889 
4890 out:
4891         spa_unload(newspa);
4892         spa_deactivate(newspa);
4893         spa_remove(newspa);
4894 
4895         txg = spa_vdev_config_enter(spa);
4896 
4897         /* re-online all offlined disks */
4898         for (c = 0; c < children; c++) {
4899                 if (vml[c] != NULL)
4900                         vml[c]->vdev_offline = B_FALSE;
4901         }
4902         vdev_reopen(spa->spa_root_vdev);
4903 
4904         nvlist_free(spa->spa_config_splitting);
4905         spa->spa_config_splitting = NULL;
4906         (void) spa_vdev_exit(spa, NULL, txg, error);
4907 
4908         kmem_free(vml, children * sizeof (vdev_t *));
4909         return (error);
4910 }
4911 
4912 static nvlist_t *
4913 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
4914 {
4915         for (int i = 0; i < count; i++) {
4916                 uint64_t guid;
4917 
4918                 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
4919                     &guid) == 0);
4920 
4921                 if (guid == target_guid)
4922                         return (nvpp[i]);
4923         }
4924 
4925         return (NULL);
4926 }
4927 
4928 static void
4929 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
4930         nvlist_t *dev_to_remove)
4931 {
4932         nvlist_t **newdev = NULL;
4933 
4934         if (count > 1)
4935                 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
4936 
4937         for (int i = 0, j = 0; i < count; i++) {
4938                 if (dev[i] == dev_to_remove)
4939                         continue;
4940                 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
4941         }
4942 
4943         VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
4944         VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
4945 
4946         for (int i = 0; i < count - 1; i++)
4947                 nvlist_free(newdev[i]);
4948 
4949         if (count > 1)
4950                 kmem_free(newdev, (count - 1) * sizeof (void *));
4951 }
4952 
4953 /*
4954  * Evacuate the device.
4955  */
4956 static int
4957 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
4958 {
4959         uint64_t txg;
4960         int error = 0;
4961 
4962         ASSERT(MUTEX_HELD(&spa_namespace_lock));
4963         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
4964         ASSERT(vd == vd->vdev_top);
4965 
4966         /*
4967          * Evacuate the device.  We don't hold the config lock as writer
4968          * since we need to do I/O but we do keep the
4969          * spa_namespace_lock held.  Once this completes the device
4970          * should no longer have any blocks allocated on it.
4971          */
4972         if (vd->vdev_islog) {
4973                 if (vd->vdev_stat.vs_alloc != 0)
4974                         error = spa_offline_log(spa);
4975         } else {
4976                 error = ENOTSUP;
4977         }
4978 
4979         if (error)
4980                 return (error);
4981 
4982         /*
4983          * The evacuation succeeded.  Remove any remaining MOS metadata
4984          * associated with this vdev, and wait for these changes to sync.
4985          */
4986         ASSERT3U(vd->vdev_stat.vs_alloc, ==, 0);
4987         txg = spa_vdev_config_enter(spa);
4988         vd->vdev_removing = B_TRUE;
4989         vdev_dirty(vd, 0, NULL, txg);
4990         vdev_config_dirty(vd);
4991         spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4992 
4993         return (0);
4994 }
4995 
4996 /*
4997  * Complete the removal by cleaning up the namespace.
4998  */
4999 static void
5000 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5001 {
5002         vdev_t *rvd = spa->spa_root_vdev;
5003         uint64_t id = vd->vdev_id;
5004         boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5005 
5006         ASSERT(MUTEX_HELD(&spa_namespace_lock));
5007         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5008         ASSERT(vd == vd->vdev_top);
5009 
5010         /*
5011          * Only remove any devices which are empty.
5012          */
5013         if (vd->vdev_stat.vs_alloc != 0)
5014                 return;
5015 
5016         (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5017 
5018         if (list_link_active(&vd->vdev_state_dirty_node))
5019                 vdev_state_clean(vd);
5020         if (list_link_active(&vd->vdev_config_dirty_node))
5021                 vdev_config_clean(vd);
5022 
5023         vdev_free(vd);
5024 
5025         if (last_vdev) {
5026                 vdev_compact_children(rvd);
5027         } else {
5028                 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5029                 vdev_add_child(rvd, vd);
5030         }
5031         vdev_config_dirty(rvd);
5032 
5033         /*
5034          * Reassess the health of our root vdev.
5035          */
5036         vdev_reopen(rvd);
5037 }
5038 
5039 /*
5040  * Remove a device from the pool -
5041  *
5042  * Removing a device from the vdev namespace requires several steps
5043  * and can take a significant amount of time.  As a result we use
5044  * the spa_vdev_config_[enter/exit] functions which allow us to
5045  * grab and release the spa_config_lock while still holding the namespace
5046  * lock.  During each step the configuration is synced out.
5047  */
5048 
5049 /*
5050  * Remove a device from the pool.  Currently, this supports removing only hot
5051  * spares, slogs, and level 2 ARC devices.
5052  */
5053 int
5054 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5055 {
5056         vdev_t *vd;
5057         metaslab_group_t *mg;
5058         nvlist_t **spares, **l2cache, *nv;
5059         uint64_t txg = 0;
5060         uint_t nspares, nl2cache;
5061         int error = 0;
5062         boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5063 
5064         ASSERT(spa_writeable(spa));
5065 
5066         if (!locked)
5067                 txg = spa_vdev_enter(spa);
5068 
5069         vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5070 
5071         if (spa->spa_spares.sav_vdevs != NULL &&
5072             nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5073             ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5074             (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5075                 /*
5076                  * Only remove the hot spare if it's not currently in use
5077                  * in this pool.
5078                  */
5079                 if (vd == NULL || unspare) {
5080                         spa_vdev_remove_aux(spa->spa_spares.sav_config,
5081                             ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5082                         spa_load_spares(spa);
5083                         spa->spa_spares.sav_sync = B_TRUE;
5084                 } else {
5085                         error = EBUSY;
5086                 }
5087         } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5088             nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5089             ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5090             (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5091                 /*
5092                  * Cache devices can always be removed.
5093                  */
5094                 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5095                     ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5096                 spa_load_l2cache(spa);
5097                 spa->spa_l2cache.sav_sync = B_TRUE;
5098         } else if (vd != NULL && vd->vdev_islog) {
5099                 ASSERT(!locked);
5100                 ASSERT(vd == vd->vdev_top);
5101 
5102                 /*
5103                  * XXX - Once we have bp-rewrite this should
5104                  * become the common case.
5105                  */
5106 
5107                 mg = vd->vdev_mg;
5108 
5109                 /*
5110                  * Stop allocating from this vdev.
5111                  */
5112                 metaslab_group_passivate(mg);
5113 
5114                 /*
5115                  * Wait for the youngest allocations and frees to sync,
5116                  * and then wait for the deferral of those frees to finish.
5117                  */
5118                 spa_vdev_config_exit(spa, NULL,
5119                     txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5120 
5121                 /*
5122                  * Attempt to evacuate the vdev.
5123                  */
5124                 error = spa_vdev_remove_evacuate(spa, vd);
5125 
5126                 txg = spa_vdev_config_enter(spa);
5127 
5128                 /*
5129                  * If we couldn't evacuate the vdev, unwind.
5130                  */
5131                 if (error) {
5132                         metaslab_group_activate(mg);
5133                         return (spa_vdev_exit(spa, NULL, txg, error));
5134                 }
5135 
5136                 /*
5137                  * Clean up the vdev namespace.
5138                  */
5139                 spa_vdev_remove_from_namespace(spa, vd);
5140 
5141         } else if (vd != NULL) {
5142                 /*
5143                  * Normal vdevs cannot be removed (yet).
5144                  */
5145                 error = ENOTSUP;
5146         } else {
5147                 /*
5148                  * There is no vdev of any kind with the specified guid.
5149                  */
5150                 error = ENOENT;
5151         }
5152 
5153         if (!locked)
5154                 return (spa_vdev_exit(spa, NULL, txg, error));
5155 
5156         return (error);
5157 }
5158 
5159 /*
5160  * Find any device that's done replacing, or a vdev marked 'unspare' that's
5161  * current spared, so we can detach it.
5162  */
5163 static vdev_t *
5164 spa_vdev_resilver_done_hunt(vdev_t *vd)
5165 {
5166         vdev_t *newvd, *oldvd;
5167 
5168         for (int c = 0; c < vd->vdev_children; c++) {
5169                 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5170                 if (oldvd != NULL)
5171                         return (oldvd);
5172         }
5173 
5174         /*
5175          * Check for a completed replacement.  We always consider the first
5176          * vdev in the list to be the oldest vdev, and the last one to be
5177          * the newest (see spa_vdev_attach() for how that works).  In
5178          * the case where the newest vdev is faulted, we will not automatically
5179          * remove it after a resilver completes.  This is OK as it will require
5180          * user intervention to determine which disk the admin wishes to keep.
5181          */
5182         if (vd->vdev_ops == &vdev_replacing_ops) {
5183                 ASSERT(vd->vdev_children > 1);
5184 
5185                 newvd = vd->vdev_child[vd->vdev_children - 1];
5186                 oldvd = vd->vdev_child[0];
5187 
5188                 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5189                     vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5190                     !vdev_dtl_required(oldvd))
5191                         return (oldvd);
5192         }
5193 
5194         /*
5195          * Check for a completed resilver with the 'unspare' flag set.
5196          */
5197         if (vd->vdev_ops == &vdev_spare_ops) {
5198                 vdev_t *first = vd->vdev_child[0];
5199                 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5200 
5201                 if (last->vdev_unspare) {
5202                         oldvd = first;
5203                         newvd = last;
5204                 } else if (first->vdev_unspare) {
5205                         oldvd = last;
5206                         newvd = first;
5207                 } else {
5208                         oldvd = NULL;
5209                 }
5210 
5211                 if (oldvd != NULL &&
5212                     vdev_dtl_empty(newvd, DTL_MISSING) &&
5213                     vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5214                     !vdev_dtl_required(oldvd))
5215                         return (oldvd);
5216 
5217                 /*
5218                  * If there are more than two spares attached to a disk,
5219                  * and those spares are not required, then we want to
5220                  * attempt to free them up now so that they can be used
5221                  * by other pools.  Once we're back down to a single
5222                  * disk+spare, we stop removing them.
5223                  */
5224                 if (vd->vdev_children > 2) {
5225                         newvd = vd->vdev_child[1];
5226 
5227                         if (newvd->vdev_isspare && last->vdev_isspare &&
5228                             vdev_dtl_empty(last, DTL_MISSING) &&
5229                             vdev_dtl_empty(last, DTL_OUTAGE) &&
5230                             !vdev_dtl_required(newvd))
5231                                 return (newvd);
5232                 }
5233         }
5234 
5235         return (NULL);
5236 }
5237 
5238 static void
5239 spa_vdev_resilver_done(spa_t *spa)
5240 {
5241         vdev_t *vd, *pvd, *ppvd;
5242         uint64_t guid, sguid, pguid, ppguid;
5243 
5244         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5245 
5246         while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5247                 pvd = vd->vdev_parent;
5248                 ppvd = pvd->vdev_parent;
5249                 guid = vd->vdev_guid;
5250                 pguid = pvd->vdev_guid;
5251                 ppguid = ppvd->vdev_guid;
5252                 sguid = 0;
5253                 /*
5254                  * If we have just finished replacing a hot spared device, then
5255                  * we need to detach the parent's first child (the original hot
5256                  * spare) as well.
5257                  */
5258                 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5259                     ppvd->vdev_children == 2) {
5260                         ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5261                         sguid = ppvd->vdev_child[1]->vdev_guid;
5262                 }
5263                 spa_config_exit(spa, SCL_ALL, FTAG);
5264                 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5265                         return;
5266                 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5267                         return;
5268                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5269         }
5270 
5271         spa_config_exit(spa, SCL_ALL, FTAG);
5272 }
5273 
5274 /*
5275  * Update the stored path or FRU for this vdev.
5276  */
5277 int
5278 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5279     boolean_t ispath)
5280 {
5281         vdev_t *vd;
5282         boolean_t sync = B_FALSE;
5283 
5284         ASSERT(spa_writeable(spa));
5285 
5286         spa_vdev_state_enter(spa, SCL_ALL);
5287 
5288         if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5289                 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5290 
5291         if (!vd->vdev_ops->vdev_op_leaf)
5292                 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5293 
5294         if (ispath) {
5295                 if (strcmp(value, vd->vdev_path) != 0) {
5296                         spa_strfree(vd->vdev_path);
5297                         vd->vdev_path = spa_strdup(value);
5298                         sync = B_TRUE;
5299                 }
5300         } else {
5301                 if (vd->vdev_fru == NULL) {
5302                         vd->vdev_fru = spa_strdup(value);
5303                         sync = B_TRUE;
5304                 } else if (strcmp(value, vd->vdev_fru) != 0) {
5305                         spa_strfree(vd->vdev_fru);
5306                         vd->vdev_fru = spa_strdup(value);
5307                         sync = B_TRUE;
5308                 }
5309         }
5310 
5311         return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5312 }
5313 
5314 int
5315 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5316 {
5317         return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5318 }
5319 
5320 int
5321 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5322 {
5323         return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5324 }
5325 
5326 /*
5327  * ==========================================================================
5328  * SPA Scanning
5329  * ==========================================================================
5330  */
5331 
5332 int
5333 spa_scan_stop(spa_t *spa)
5334 {
5335         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5336         if (dsl_scan_resilvering(spa->spa_dsl_pool))
5337                 return (EBUSY);
5338         return (dsl_scan_cancel(spa->spa_dsl_pool));
5339 }
5340 
5341 int
5342 spa_scan(spa_t *spa, pool_scan_func_t func)
5343 {
5344         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5345 
5346         if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5347                 return (ENOTSUP);
5348 
5349         /*
5350          * If a resilver was requested, but there is no DTL on a
5351          * writeable leaf device, we have nothing to do.
5352          */
5353         if (func == POOL_SCAN_RESILVER &&
5354             !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5355                 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5356                 return (0);
5357         }
5358 
5359         return (dsl_scan(spa->spa_dsl_pool, func));
5360 }
5361 
5362 /*
5363  * ==========================================================================
5364  * SPA async task processing
5365  * ==========================================================================
5366  */
5367 
5368 static void
5369 spa_async_remove(spa_t *spa, vdev_t *vd)
5370 {
5371         if (vd->vdev_remove_wanted) {
5372                 vd->vdev_remove_wanted = B_FALSE;
5373                 vd->vdev_delayed_close = B_FALSE;
5374                 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5375 
5376                 /*
5377                  * We want to clear the stats, but we don't want to do a full
5378                  * vdev_clear() as that will cause us to throw away
5379                  * degraded/faulted state as well as attempt to reopen the
5380                  * device, all of which is a waste.
5381                  */
5382                 vd->vdev_stat.vs_read_errors = 0;
5383                 vd->vdev_stat.vs_write_errors = 0;
5384                 vd->vdev_stat.vs_checksum_errors = 0;
5385 
5386                 vdev_state_dirty(vd->vdev_top);
5387         }
5388 
5389         for (int c = 0; c < vd->vdev_children; c++)
5390                 spa_async_remove(spa, vd->vdev_child[c]);
5391 }
5392 
5393 static void
5394 spa_async_probe(spa_t *spa, vdev_t *vd)
5395 {
5396         if (vd->vdev_probe_wanted) {
5397                 vd->vdev_probe_wanted = B_FALSE;
5398                 vdev_reopen(vd);        /* vdev_open() does the actual probe */
5399         }
5400 
5401         for (int c = 0; c < vd->vdev_children; c++)
5402                 spa_async_probe(spa, vd->vdev_child[c]);
5403 }
5404 
5405 static void
5406 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5407 {
5408         sysevent_id_t eid;
5409         nvlist_t *attr;
5410         char *physpath;
5411 
5412         if (!spa->spa_autoexpand)
5413                 return;
5414 
5415         for (int c = 0; c < vd->vdev_children; c++) {
5416                 vdev_t *cvd = vd->vdev_child[c];
5417                 spa_async_autoexpand(spa, cvd);
5418         }
5419 
5420         if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5421                 return;
5422 
5423         physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5424         (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5425 
5426         VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5427         VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5428 
5429         (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5430             ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5431 
5432         nvlist_free(attr);
5433         kmem_free(physpath, MAXPATHLEN);
5434 }
5435 
5436 static void
5437 spa_async_thread(spa_t *spa)
5438 {
5439         int tasks;
5440 
5441         ASSERT(spa->spa_sync_on);
5442 
5443         mutex_enter(&spa->spa_async_lock);
5444         tasks = spa->spa_async_tasks;
5445         spa->spa_async_tasks = 0;
5446         mutex_exit(&spa->spa_async_lock);
5447 
5448         /*
5449          * See if the config needs to be updated.
5450          */
5451         if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5452                 uint64_t old_space, new_space;
5453 
5454                 mutex_enter(&spa_namespace_lock);
5455                 old_space = metaslab_class_get_space(spa_normal_class(spa));
5456                 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5457                 new_space = metaslab_class_get_space(spa_normal_class(spa));
5458                 mutex_exit(&spa_namespace_lock);
5459 
5460                 /*
5461                  * If the pool grew as a result of the config update,
5462                  * then log an internal history event.
5463                  */
5464                 if (new_space != old_space) {
5465                         spa_history_log_internal(LOG_POOL_VDEV_ONLINE,
5466                             spa, NULL,
5467                             "pool '%s' size: %llu(+%llu)",
5468                             spa_name(spa), new_space, new_space - old_space);
5469                 }
5470         }
5471 
5472         /*
5473          * See if any devices need to be marked REMOVED.
5474          */
5475         if (tasks & SPA_ASYNC_REMOVE) {
5476                 spa_vdev_state_enter(spa, SCL_NONE);
5477                 spa_async_remove(spa, spa->spa_root_vdev);
5478                 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5479                         spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5480                 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5481                         spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5482                 (void) spa_vdev_state_exit(spa, NULL, 0);
5483         }
5484 
5485         if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5486                 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5487                 spa_async_autoexpand(spa, spa->spa_root_vdev);
5488                 spa_config_exit(spa, SCL_CONFIG, FTAG);
5489         }
5490 
5491         /*
5492          * See if any devices need to be probed.
5493          */
5494         if (tasks & SPA_ASYNC_PROBE) {
5495                 spa_vdev_state_enter(spa, SCL_NONE);
5496                 spa_async_probe(spa, spa->spa_root_vdev);
5497                 (void) spa_vdev_state_exit(spa, NULL, 0);
5498         }
5499 
5500         /*
5501          * If any devices are done replacing, detach them.
5502          */
5503         if (tasks & SPA_ASYNC_RESILVER_DONE)
5504                 spa_vdev_resilver_done(spa);
5505 
5506         /*
5507          * Kick off a resilver.
5508          */
5509         if (tasks & SPA_ASYNC_RESILVER)
5510                 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5511 
5512         /*
5513          * Let the world know that we're done.
5514          */
5515         mutex_enter(&spa->spa_async_lock);
5516         spa->spa_async_thread = NULL;
5517         cv_broadcast(&spa->spa_async_cv);
5518         mutex_exit(&spa->spa_async_lock);
5519         thread_exit();
5520 }
5521 
5522 void
5523 spa_async_suspend(spa_t *spa)
5524 {
5525         mutex_enter(&spa->spa_async_lock);
5526         spa->spa_async_suspended++;
5527         while (spa->spa_async_thread != NULL)
5528                 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5529         mutex_exit(&spa->spa_async_lock);
5530 }
5531 
5532 void
5533 spa_async_resume(spa_t *spa)
5534 {
5535         mutex_enter(&spa->spa_async_lock);
5536         ASSERT(spa->spa_async_suspended != 0);
5537         spa->spa_async_suspended--;
5538         mutex_exit(&spa->spa_async_lock);
5539 }
5540 
5541 static void
5542 spa_async_dispatch(spa_t *spa)
5543 {
5544         mutex_enter(&spa->spa_async_lock);
5545         if (spa->spa_async_tasks && !spa->spa_async_suspended &&
5546             spa->spa_async_thread == NULL &&
5547             rootdir != NULL && !vn_is_readonly(rootdir))
5548                 spa->spa_async_thread = thread_create(NULL, 0,
5549                     spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5550         mutex_exit(&spa->spa_async_lock);
5551 }
5552 
5553 void
5554 spa_async_request(spa_t *spa, int task)
5555 {
5556         zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5557         mutex_enter(&spa->spa_async_lock);
5558         spa->spa_async_tasks |= task;
5559         mutex_exit(&spa->spa_async_lock);
5560 }
5561 
5562 /*
5563  * ==========================================================================
5564  * SPA syncing routines
5565  * ==========================================================================
5566  */
5567 
5568 static int
5569 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5570 {
5571         bpobj_t *bpo = arg;
5572         bpobj_enqueue(bpo, bp, tx);
5573         return (0);
5574 }
5575 
5576 static int
5577 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5578 {
5579         zio_t *zio = arg;
5580 
5581         zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5582             zio->io_flags));
5583         return (0);
5584 }
5585 
5586 static void
5587 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5588 {
5589         char *packed = NULL;
5590         size_t bufsize;
5591         size_t nvsize = 0;
5592         dmu_buf_t *db;
5593 
5594         VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5595 
5596         /*
5597          * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5598          * information.  This avoids the dbuf_will_dirty() path and
5599          * saves us a pre-read to get data we don't actually care about.
5600          */
5601         bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5602         packed = kmem_alloc(bufsize, KM_SLEEP);
5603 
5604         VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5605             KM_SLEEP) == 0);
5606         bzero(packed + nvsize, bufsize - nvsize);
5607 
5608         dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5609 
5610         kmem_free(packed, bufsize);
5611 
5612         VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5613         dmu_buf_will_dirty(db, tx);
5614         *(uint64_t *)db->db_data = nvsize;
5615         dmu_buf_rele(db, FTAG);
5616 }
5617 
5618 static void
5619 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5620     const char *config, const char *entry)
5621 {
5622         nvlist_t *nvroot;
5623         nvlist_t **list;
5624         int i;
5625 
5626         if (!sav->sav_sync)
5627                 return;
5628 
5629         /*
5630          * Update the MOS nvlist describing the list of available devices.
5631          * spa_validate_aux() will have already made sure this nvlist is
5632          * valid and the vdevs are labeled appropriately.
5633          */
5634         if (sav->sav_object == 0) {
5635                 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5636                     DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5637                     sizeof (uint64_t), tx);
5638                 VERIFY(zap_update(spa->spa_meta_objset,
5639                     DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5640                     &sav->sav_object, tx) == 0);
5641         }
5642 
5643         VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5644         if (sav->sav_count == 0) {
5645                 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5646         } else {
5647                 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5648                 for (i = 0; i < sav->sav_count; i++)
5649                         list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5650                             B_FALSE, VDEV_CONFIG_L2CACHE);
5651                 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5652                     sav->sav_count) == 0);
5653                 for (i = 0; i < sav->sav_count; i++)
5654                         nvlist_free(list[i]);
5655                 kmem_free(list, sav->sav_count * sizeof (void *));
5656         }
5657 
5658         spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5659         nvlist_free(nvroot);
5660 
5661         sav->sav_sync = B_FALSE;
5662 }
5663 
5664 static void
5665 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5666 {
5667         nvlist_t *config;
5668 
5669         if (list_is_empty(&spa->spa_config_dirty_list))
5670                 return;
5671 
5672         spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5673 
5674         config = spa_config_generate(spa, spa->spa_root_vdev,
5675             dmu_tx_get_txg(tx), B_FALSE);
5676 
5677         spa_config_exit(spa, SCL_STATE, FTAG);
5678 
5679         if (spa->spa_config_syncing)
5680                 nvlist_free(spa->spa_config_syncing);
5681         spa->spa_config_syncing = config;
5682 
5683         spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5684 }
5685 
5686 static void
5687 spa_sync_version(void *arg1, void *arg2, dmu_tx_t *tx)
5688 {
5689         spa_t *spa = arg1;
5690         uint64_t version = *(uint64_t *)arg2;
5691 
5692         /*
5693          * Setting the version is special cased when first creating the pool.
5694          */
5695         ASSERT(tx->tx_txg != TXG_INITIAL);
5696 
5697         ASSERT(version <= SPA_VERSION);
5698         ASSERT(version >= spa_version(spa));
5699 
5700         spa->spa_uberblock.ub_version = version;
5701         vdev_config_dirty(spa->spa_root_vdev);
5702 }
5703 
5704 /*
5705  * Set zpool properties.
5706  */
5707 static void
5708 spa_sync_props(void *arg1, void *arg2, dmu_tx_t *tx)
5709 {
5710         spa_t *spa = arg1;
5711         objset_t *mos = spa->spa_meta_objset;
5712         nvlist_t *nvp = arg2;
5713         nvpair_t *elem = NULL;
5714 
5715         mutex_enter(&spa->spa_props_lock);
5716 
5717         while ((elem = nvlist_next_nvpair(nvp, elem))) {
5718                 uint64_t intval;
5719                 char *strval, *fname;
5720                 zpool_prop_t prop;
5721                 const char *propname;
5722                 zprop_type_t proptype;
5723                 zfeature_info_t *feature;
5724 
5725                 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5726                 case ZPROP_INVAL:
5727                         /*
5728                          * We checked this earlier in spa_prop_validate().
5729                          */
5730                         ASSERT(zpool_prop_feature(nvpair_name(elem)));
5731 
5732                         fname = strchr(nvpair_name(elem), '@') + 1;
5733                         VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5734 
5735                         spa_feature_enable(spa, feature, tx);
5736                         break;
5737 
5738                 case ZPOOL_PROP_VERSION:
5739                         VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5740                         /*
5741                          * The version is synced seperatly before other
5742                          * properties and should be correct by now.
5743                          */
5744                         ASSERT3U(spa_version(spa), >=, intval);
5745                         break;
5746 
5747                 case ZPOOL_PROP_ALTROOT:
5748                         /*
5749                          * 'altroot' is a non-persistent property. It should
5750                          * have been set temporarily at creation or import time.
5751                          */
5752                         ASSERT(spa->spa_root != NULL);
5753                         break;
5754 
5755                 case ZPOOL_PROP_READONLY:
5756                 case ZPOOL_PROP_CACHEFILE:
5757                         /*
5758                          * 'readonly' and 'cachefile' are also non-persisitent
5759                          * properties.
5760                          */
5761                         break;
5762                 case ZPOOL_PROP_COMMENT:
5763                         VERIFY(nvpair_value_string(elem, &strval) == 0);
5764                         if (spa->spa_comment != NULL)
5765                                 spa_strfree(spa->spa_comment);
5766                         spa->spa_comment = spa_strdup(strval);
5767                         /*
5768                          * We need to dirty the configuration on all the vdevs
5769                          * so that their labels get updated.  It's unnecessary
5770                          * to do this for pool creation since the vdev's
5771                          * configuratoin has already been dirtied.
5772                          */
5773                         if (tx->tx_txg != TXG_INITIAL)
5774                                 vdev_config_dirty(spa->spa_root_vdev);
5775                         break;
5776                 default:
5777                         /*
5778                          * Set pool property values in the poolprops mos object.
5779                          */
5780                         if (spa->spa_pool_props_object == 0) {
5781                                 spa->spa_pool_props_object =
5782                                     zap_create_link(mos, DMU_OT_POOL_PROPS,
5783                                     DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5784                                     tx);
5785                         }
5786 
5787                         /* normalize the property name */
5788                         propname = zpool_prop_to_name(prop);
5789                         proptype = zpool_prop_get_type(prop);
5790 
5791                         if (nvpair_type(elem) == DATA_TYPE_STRING) {
5792                                 ASSERT(proptype == PROP_TYPE_STRING);
5793                                 VERIFY(nvpair_value_string(elem, &strval) == 0);
5794                                 VERIFY(zap_update(mos,
5795                                     spa->spa_pool_props_object, propname,
5796                                     1, strlen(strval) + 1, strval, tx) == 0);
5797 
5798                         } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5799                                 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5800 
5801                                 if (proptype == PROP_TYPE_INDEX) {
5802                                         const char *unused;
5803                                         VERIFY(zpool_prop_index_to_string(
5804                                             prop, intval, &unused) == 0);
5805                                 }
5806                                 VERIFY(zap_update(mos,
5807                                     spa->spa_pool_props_object, propname,
5808                                     8, 1, &intval, tx) == 0);
5809                         } else {
5810                                 ASSERT(0); /* not allowed */
5811                         }
5812 
5813                         switch (prop) {
5814                         case ZPOOL_PROP_DELEGATION:
5815                                 spa->spa_delegation = intval;
5816                                 break;
5817                         case ZPOOL_PROP_BOOTFS:
5818                                 spa->spa_bootfs = intval;
5819                                 break;
5820                         case ZPOOL_PROP_FAILUREMODE:
5821                                 spa->spa_failmode = intval;
5822                                 break;
5823                         case ZPOOL_PROP_AUTOEXPAND:
5824                                 spa->spa_autoexpand = intval;
5825                                 if (tx->tx_txg != TXG_INITIAL)
5826                                         spa_async_request(spa,
5827                                             SPA_ASYNC_AUTOEXPAND);
5828                                 break;
5829                         case ZPOOL_PROP_DEDUPDITTO:
5830                                 spa->spa_dedup_ditto = intval;
5831                                 break;
5832                         default:
5833                                 break;
5834                         }
5835                 }
5836 
5837                 /* log internal history if this is not a zpool create */
5838                 if (spa_version(spa) >= SPA_VERSION_ZPOOL_HISTORY &&
5839                     tx->tx_txg != TXG_INITIAL) {
5840                         spa_history_log_internal(LOG_POOL_PROPSET,
5841                             spa, tx, "%s %lld %s",
5842                             nvpair_name(elem), intval, spa_name(spa));
5843                 }
5844         }
5845 
5846         mutex_exit(&spa->spa_props_lock);
5847 }
5848 
5849 /*
5850  * Perform one-time upgrade on-disk changes.  spa_version() does not
5851  * reflect the new version this txg, so there must be no changes this
5852  * txg to anything that the upgrade code depends on after it executes.
5853  * Therefore this must be called after dsl_pool_sync() does the sync
5854  * tasks.
5855  */
5856 static void
5857 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
5858 {
5859         dsl_pool_t *dp = spa->spa_dsl_pool;
5860 
5861         ASSERT(spa->spa_sync_pass == 1);
5862 
5863         if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
5864             spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
5865                 dsl_pool_create_origin(dp, tx);
5866 
5867                 /* Keeping the origin open increases spa_minref */
5868                 spa->spa_minref += 3;
5869         }
5870 
5871         if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
5872             spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
5873                 dsl_pool_upgrade_clones(dp, tx);
5874         }
5875 
5876         if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
5877             spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
5878                 dsl_pool_upgrade_dir_clones(dp, tx);
5879 
5880                 /* Keeping the freedir open increases spa_minref */
5881                 spa->spa_minref += 3;
5882         }
5883 
5884         if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
5885             spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
5886                 spa_feature_create_zap_objects(spa, tx);
5887         }
5888 }
5889 
5890 /*
5891  * Sync the specified transaction group.  New blocks may be dirtied as
5892  * part of the process, so we iterate until it converges.
5893  */
5894 void
5895 spa_sync(spa_t *spa, uint64_t txg)
5896 {
5897         dsl_pool_t *dp = spa->spa_dsl_pool;
5898         objset_t *mos = spa->spa_meta_objset;
5899         bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
5900         bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
5901         vdev_t *rvd = spa->spa_root_vdev;
5902         vdev_t *vd;
5903         dmu_tx_t *tx;
5904         int error;
5905 
5906         VERIFY(spa_writeable(spa));
5907 
5908         /*
5909          * Lock out configuration changes.
5910          */
5911         spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5912 
5913         spa->spa_syncing_txg = txg;
5914         spa->spa_sync_pass = 0;
5915 
5916         /*
5917          * If there are any pending vdev state changes, convert them
5918          * into config changes that go out with this transaction group.
5919          */
5920         spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5921         while (list_head(&spa->spa_state_dirty_list) != NULL) {
5922                 /*
5923                  * We need the write lock here because, for aux vdevs,
5924                  * calling vdev_config_dirty() modifies sav_config.
5925                  * This is ugly and will become unnecessary when we
5926                  * eliminate the aux vdev wart by integrating all vdevs
5927                  * into the root vdev tree.
5928                  */
5929                 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5930                 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
5931                 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
5932                         vdev_state_clean(vd);
5933                         vdev_config_dirty(vd);
5934                 }
5935                 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
5936                 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
5937         }
5938         spa_config_exit(spa, SCL_STATE, FTAG);
5939 
5940         tx = dmu_tx_create_assigned(dp, txg);
5941 
5942         /*
5943          * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
5944          * set spa_deflate if we have no raid-z vdevs.
5945          */
5946         if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
5947             spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
5948                 int i;
5949 
5950                 for (i = 0; i < rvd->vdev_children; i++) {
5951                         vd = rvd->vdev_child[i];
5952                         if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
5953                                 break;
5954                 }
5955                 if (i == rvd->vdev_children) {
5956                         spa->spa_deflate = TRUE;
5957                         VERIFY(0 == zap_add(spa->spa_meta_objset,
5958                             DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
5959                             sizeof (uint64_t), 1, &spa->spa_deflate, tx));
5960                 }
5961         }
5962 
5963         /*
5964          * If anything has changed in this txg, or if someone is waiting
5965          * for this txg to sync (eg, spa_vdev_remove()), push the
5966          * deferred frees from the previous txg.  If not, leave them
5967          * alone so that we don't generate work on an otherwise idle
5968          * system.
5969          */
5970         if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
5971             !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
5972             !txg_list_empty(&dp->dp_sync_tasks, txg) ||
5973             ((dsl_scan_active(dp->dp_scan) ||
5974             txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
5975                 zio_t *zio = zio_root(spa, NULL, NULL, 0);
5976                 VERIFY3U(bpobj_iterate(defer_bpo,
5977                     spa_free_sync_cb, zio, tx), ==, 0);
5978                 VERIFY3U(zio_wait(zio), ==, 0);
5979         }
5980 
5981         /*
5982          * Iterate to convergence.
5983          */
5984         do {
5985                 int pass = ++spa->spa_sync_pass;
5986 
5987                 spa_sync_config_object(spa, tx);
5988                 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
5989                     ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
5990                 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
5991                     ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
5992                 spa_errlog_sync(spa, txg);
5993                 dsl_pool_sync(dp, txg);
5994 
5995                 if (pass <= SYNC_PASS_DEFERRED_FREE) {
5996                         zio_t *zio = zio_root(spa, NULL, NULL, 0);
5997                         bplist_iterate(free_bpl, spa_free_sync_cb,
5998                             zio, tx);
5999                         VERIFY(zio_wait(zio) == 0);
6000                 } else {
6001                         bplist_iterate(free_bpl, bpobj_enqueue_cb,
6002                             defer_bpo, tx);
6003                 }
6004 
6005                 ddt_sync(spa, txg);
6006                 dsl_scan_sync(dp, tx);
6007 
6008                 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6009                         vdev_sync(vd, txg);
6010 
6011                 if (pass == 1)
6012                         spa_sync_upgrades(spa, tx);
6013 
6014         } while (dmu_objset_is_dirty(mos, txg));
6015 
6016         /*
6017          * Rewrite the vdev configuration (which includes the uberblock)
6018          * to commit the transaction group.
6019          *
6020          * If there are no dirty vdevs, we sync the uberblock to a few
6021          * random top-level vdevs that are known to be visible in the
6022          * config cache (see spa_vdev_add() for a complete description).
6023          * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6024          */
6025         for (;;) {
6026                 /*
6027                  * We hold SCL_STATE to prevent vdev open/close/etc.
6028                  * while we're attempting to write the vdev labels.
6029                  */
6030                 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6031 
6032                 if (list_is_empty(&spa->spa_config_dirty_list)) {
6033                         vdev_t *svd[SPA_DVAS_PER_BP];
6034                         int svdcount = 0;
6035                         int children = rvd->vdev_children;
6036                         int c0 = spa_get_random(children);
6037 
6038                         for (int c = 0; c < children; c++) {
6039                                 vd = rvd->vdev_child[(c0 + c) % children];
6040                                 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6041                                         continue;
6042                                 svd[svdcount++] = vd;
6043                                 if (svdcount == SPA_DVAS_PER_BP)
6044                                         break;
6045                         }
6046                         error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6047                         if (error != 0)
6048                                 error = vdev_config_sync(svd, svdcount, txg,
6049                                     B_TRUE);
6050                 } else {
6051                         error = vdev_config_sync(rvd->vdev_child,
6052                             rvd->vdev_children, txg, B_FALSE);
6053                         if (error != 0)
6054                                 error = vdev_config_sync(rvd->vdev_child,
6055                                     rvd->vdev_children, txg, B_TRUE);
6056                 }
6057 
6058                 spa_config_exit(spa, SCL_STATE, FTAG);
6059 
6060                 if (error == 0)
6061                         break;
6062                 zio_suspend(spa, NULL);
6063                 zio_resume_wait(spa);
6064         }
6065         dmu_tx_commit(tx);
6066 
6067         /*
6068          * Clear the dirty config list.
6069          */
6070         while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6071                 vdev_config_clean(vd);
6072 
6073         /*
6074          * Now that the new config has synced transactionally,
6075          * let it become visible to the config cache.
6076          */
6077         if (spa->spa_config_syncing != NULL) {
6078                 spa_config_set(spa, spa->spa_config_syncing);
6079                 spa->spa_config_txg = txg;
6080                 spa->spa_config_syncing = NULL;
6081         }
6082 
6083         spa->spa_ubsync = spa->spa_uberblock;
6084 
6085         dsl_pool_sync_done(dp, txg);
6086 
6087         /*
6088          * Update usable space statistics.
6089          */
6090         while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6091                 vdev_sync_done(vd, txg);
6092 
6093         spa_update_dspace(spa);
6094 
6095         /*
6096          * It had better be the case that we didn't dirty anything
6097          * since vdev_config_sync().
6098          */
6099         ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6100         ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6101         ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6102 
6103         spa->spa_sync_pass = 0;
6104 
6105         spa_config_exit(spa, SCL_CONFIG, FTAG);
6106 
6107         spa_handle_ignored_writes(spa);
6108 
6109         /*
6110          * If any async tasks have been requested, kick them off.
6111          */
6112         spa_async_dispatch(spa);
6113 }
6114 
6115 /*
6116  * Sync all pools.  We don't want to hold the namespace lock across these
6117  * operations, so we take a reference on the spa_t and drop the lock during the
6118  * sync.
6119  */
6120 void
6121 spa_sync_allpools(void)
6122 {
6123         spa_t *spa = NULL;
6124         mutex_enter(&spa_namespace_lock);
6125         while ((spa = spa_next(spa)) != NULL) {
6126                 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6127                     !spa_writeable(spa) || spa_suspended(spa))
6128                         continue;
6129                 spa_open_ref(spa, FTAG);
6130                 mutex_exit(&spa_namespace_lock);
6131                 txg_wait_synced(spa_get_dsl(spa), 0);
6132                 mutex_enter(&spa_namespace_lock);
6133                 spa_close(spa, FTAG);
6134         }
6135         mutex_exit(&spa_namespace_lock);
6136 }
6137 
6138 /*
6139  * ==========================================================================
6140  * Miscellaneous routines
6141  * ==========================================================================
6142  */
6143 
6144 /*
6145  * Remove all pools in the system.
6146  */
6147 void
6148 spa_evict_all(void)
6149 {
6150         spa_t *spa;
6151 
6152         /*
6153          * Remove all cached state.  All pools should be closed now,
6154          * so every spa in the AVL tree should be unreferenced.
6155          */
6156         mutex_enter(&spa_namespace_lock);
6157         while ((spa = spa_next(NULL)) != NULL) {
6158                 /*
6159                  * Stop async tasks.  The async thread may need to detach
6160                  * a device that's been replaced, which requires grabbing
6161                  * spa_namespace_lock, so we must drop it here.
6162                  */
6163                 spa_open_ref(spa, FTAG);
6164                 mutex_exit(&spa_namespace_lock);
6165                 spa_async_suspend(spa);
6166                 mutex_enter(&spa_namespace_lock);
6167                 spa_close(spa, FTAG);
6168 
6169                 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6170                         spa_unload(spa);
6171                         spa_deactivate(spa);
6172                 }
6173                 spa_remove(spa);
6174         }
6175         mutex_exit(&spa_namespace_lock);
6176 }
6177 
6178 vdev_t *
6179 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6180 {
6181         vdev_t *vd;
6182         int i;
6183 
6184         if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6185                 return (vd);
6186 
6187         if (aux) {
6188                 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6189                         vd = spa->spa_l2cache.sav_vdevs[i];
6190                         if (vd->vdev_guid == guid)
6191                                 return (vd);
6192                 }
6193 
6194                 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6195                         vd = spa->spa_spares.sav_vdevs[i];
6196                         if (vd->vdev_guid == guid)
6197                                 return (vd);
6198                 }
6199         }
6200 
6201         return (NULL);
6202 }
6203 
6204 void
6205 spa_upgrade(spa_t *spa, uint64_t version)
6206 {
6207         ASSERT(spa_writeable(spa));
6208 
6209         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6210 
6211         /*
6212          * This should only be called for a non-faulted pool, and since a
6213          * future version would result in an unopenable pool, this shouldn't be
6214          * possible.
6215          */
6216         ASSERT(spa->spa_uberblock.ub_version <= SPA_VERSION);
6217         ASSERT(version >= spa->spa_uberblock.ub_version);
6218 
6219         spa->spa_uberblock.ub_version = version;
6220         vdev_config_dirty(spa->spa_root_vdev);
6221 
6222         spa_config_exit(spa, SCL_ALL, FTAG);
6223 
6224         txg_wait_synced(spa_get_dsl(spa), 0);
6225 }
6226 
6227 boolean_t
6228 spa_has_spare(spa_t *spa, uint64_t guid)
6229 {
6230         int i;
6231         uint64_t spareguid;
6232         spa_aux_vdev_t *sav = &spa->spa_spares;
6233 
6234         for (i = 0; i < sav->sav_count; i++)
6235                 if (sav->sav_vdevs[i]->vdev_guid == guid)
6236                         return (B_TRUE);
6237 
6238         for (i = 0; i < sav->sav_npending; i++) {
6239                 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6240                     &spareguid) == 0 && spareguid == guid)
6241                         return (B_TRUE);
6242         }
6243 
6244         return (B_FALSE);
6245 }
6246 
6247 /*
6248  * Check if a pool has an active shared spare device.
6249  * Note: reference count of an active spare is 2, as a spare and as a replace
6250  */
6251 static boolean_t
6252 spa_has_active_shared_spare(spa_t *spa)
6253 {
6254         int i, refcnt;
6255         uint64_t pool;
6256         spa_aux_vdev_t *sav = &spa->spa_spares;
6257 
6258         for (i = 0; i < sav->sav_count; i++) {
6259                 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6260                     &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6261                     refcnt > 2)
6262                         return (B_TRUE);
6263         }
6264 
6265         return (B_FALSE);
6266 }
6267 
6268 /*
6269  * Post a sysevent corresponding to the given event.  The 'name' must be one of
6270  * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
6271  * filled in from the spa and (optionally) the vdev.  This doesn't do anything
6272  * in the userland libzpool, as we don't want consumers to misinterpret ztest
6273  * or zdb as real changes.
6274  */
6275 void
6276 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6277 {
6278 #ifdef _KERNEL
6279         sysevent_t              *ev;
6280         sysevent_attr_list_t    *attr = NULL;
6281         sysevent_value_t        value;
6282         sysevent_id_t           eid;
6283 
6284         ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6285             SE_SLEEP);
6286 
6287         value.value_type = SE_DATA_TYPE_STRING;
6288         value.value.sv_string = spa_name(spa);
6289         if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6290                 goto done;
6291 
6292         value.value_type = SE_DATA_TYPE_UINT64;
6293         value.value.sv_uint64 = spa_guid(spa);
6294         if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6295                 goto done;
6296 
6297         if (vd) {
6298                 value.value_type = SE_DATA_TYPE_UINT64;
6299                 value.value.sv_uint64 = vd->vdev_guid;
6300                 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6301                     SE_SLEEP) != 0)
6302                         goto done;
6303 
6304                 if (vd->vdev_path) {
6305                         value.value_type = SE_DATA_TYPE_STRING;
6306                         value.value.sv_string = vd->vdev_path;
6307                         if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6308                             &value, SE_SLEEP) != 0)
6309                                 goto done;
6310                 }
6311         }
6312 
6313         if (sysevent_attach_attributes(ev, attr) != 0)
6314                 goto done;
6315         attr = NULL;
6316 
6317         (void) log_sysevent(ev, SE_SLEEP, &eid);
6318 
6319 done:
6320         if (attr)
6321                 sysevent_free_attr(attr);
6322         sysevent_free(ev);
6323 #endif
6324 }