1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright (c) 2013 by Delphix. All rights reserved.
  25  * Copyright 2013 Nexenta Systems, Inc.  All rights reserved.
  26  */
  27 
  28 /*
  29  * SPA: Storage Pool Allocator
  30  *
  31  * This file contains all the routines used when modifying on-disk SPA state.
  32  * This includes opening, importing, destroying, exporting a pool, and syncing a
  33  * pool.
  34  */
  35 
  36 #include <sys/zfs_context.h>
  37 #include <sys/fm/fs/zfs.h>
  38 #include <sys/spa_impl.h>
  39 #include <sys/zio.h>
  40 #include <sys/zio_checksum.h>
  41 #include <sys/dmu.h>
  42 #include <sys/dmu_tx.h>
  43 #include <sys/zap.h>
  44 #include <sys/zil.h>
  45 #include <sys/ddt.h>
  46 #include <sys/vdev_impl.h>
  47 #include <sys/metaslab.h>
  48 #include <sys/metaslab_impl.h>
  49 #include <sys/uberblock_impl.h>
  50 #include <sys/txg.h>
  51 #include <sys/avl.h>
  52 #include <sys/dmu_traverse.h>
  53 #include <sys/dmu_objset.h>
  54 #include <sys/unique.h>
  55 #include <sys/dsl_pool.h>
  56 #include <sys/dsl_dataset.h>
  57 #include <sys/dsl_dir.h>
  58 #include <sys/dsl_prop.h>
  59 #include <sys/dsl_synctask.h>
  60 #include <sys/fs/zfs.h>
  61 #include <sys/arc.h>
  62 #include <sys/callb.h>
  63 #include <sys/systeminfo.h>
  64 #include <sys/spa_boot.h>
  65 #include <sys/zfs_ioctl.h>
  66 #include <sys/dsl_scan.h>
  67 #include <sys/zfeature.h>
  68 #include <sys/dsl_destroy.h>
  69 
  70 #ifdef  _KERNEL
  71 #include <sys/bootprops.h>
  72 #include <sys/callb.h>
  73 #include <sys/cpupart.h>
  74 #include <sys/pool.h>
  75 #include <sys/sysdc.h>
  76 #include <sys/zone.h>
  77 #endif  /* _KERNEL */
  78 
  79 #include "zfs_prop.h"
  80 #include "zfs_comutil.h"
  81 
  82 /*
  83  * The interval, in seconds, at which failed configuration cache file writes
  84  * should be retried.
  85  */
  86 static int zfs_ccw_retry_interval = 300;
  87 
  88 typedef enum zti_modes {
  89         ZTI_MODE_FIXED,                 /* value is # of threads (min 1) */
  90         ZTI_MODE_ONLINE_PERCENT,        /* value is % of online CPUs */
  91         ZTI_MODE_BATCH,                 /* cpu-intensive; value is ignored */
  92         ZTI_MODE_NULL,                  /* don't create a taskq */
  93         ZTI_NMODES
  94 } zti_modes_t;
  95 
  96 #define ZTI_P(n, q)     { ZTI_MODE_FIXED, (n), (q) }
  97 #define ZTI_PCT(n)      { ZTI_MODE_ONLINE_PERCENT, (n), 1 }
  98 #define ZTI_BATCH       { ZTI_MODE_BATCH, 0, 1 }
  99 #define ZTI_NULL        { ZTI_MODE_NULL, 0, 0 }
 100 
 101 #define ZTI_N(n)        ZTI_P(n, 1)
 102 #define ZTI_ONE         ZTI_N(1)
 103 
 104 typedef struct zio_taskq_info {
 105         zti_modes_t zti_mode;
 106         uint_t zti_value;
 107         uint_t zti_count;
 108 } zio_taskq_info_t;
 109 
 110 static const char *const zio_taskq_types[ZIO_TASKQ_TYPES] = {
 111         "issue", "issue_high", "intr", "intr_high"
 112 };
 113 
 114 /*
 115  * This table defines the taskq settings for each ZFS I/O type. When
 116  * initializing a pool, we use this table to create an appropriately sized
 117  * taskq. Some operations are low volume and therefore have a small, static
 118  * number of threads assigned to their taskqs using the ZTI_N(#) or ZTI_ONE
 119  * macros. Other operations process a large amount of data; the ZTI_BATCH
 120  * macro causes us to create a taskq oriented for throughput. Some operations
 121  * are so high frequency and short-lived that the taskq itself can become a a
 122  * point of lock contention. The ZTI_P(#, #) macro indicates that we need an
 123  * additional degree of parallelism specified by the number of threads per-
 124  * taskq and the number of taskqs; when dispatching an event in this case, the
 125  * particular taskq is chosen at random.
 126  *
 127  * The different taskq priorities are to handle the different contexts (issue
 128  * and interrupt) and then to reserve threads for ZIO_PRIORITY_NOW I/Os that
 129  * need to be handled with minimum delay.
 130  */
 131 const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
 132         /* ISSUE        ISSUE_HIGH      INTR            INTR_HIGH */
 133         { ZTI_ONE,      ZTI_NULL,       ZTI_ONE,        ZTI_NULL }, /* NULL */
 134         { ZTI_N(8),     ZTI_NULL,       ZTI_BATCH,      ZTI_NULL }, /* READ */
 135         { ZTI_BATCH,    ZTI_N(5),       ZTI_N(8),       ZTI_N(5) }, /* WRITE */
 136         { ZTI_P(12, 8), ZTI_NULL,       ZTI_ONE,        ZTI_NULL }, /* FREE */
 137         { ZTI_ONE,      ZTI_NULL,       ZTI_ONE,        ZTI_NULL }, /* CLAIM */
 138         { ZTI_ONE,      ZTI_NULL,       ZTI_ONE,        ZTI_NULL }, /* IOCTL */
 139 };
 140 
 141 static void spa_sync_version(void *arg, dmu_tx_t *tx);
 142 static void spa_sync_props(void *arg, dmu_tx_t *tx);
 143 static boolean_t spa_has_active_shared_spare(spa_t *spa);
 144 static int spa_load_impl(spa_t *spa, uint64_t, nvlist_t *config,
 145     spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
 146     char **ereport);
 147 static void spa_vdev_resilver_done(spa_t *spa);
 148 
 149 uint_t          zio_taskq_batch_pct = 100;      /* 1 thread per cpu in pset */
 150 id_t            zio_taskq_psrset_bind = PS_NONE;
 151 boolean_t       zio_taskq_sysdc = B_TRUE;       /* use SDC scheduling class */
 152 uint_t          zio_taskq_basedc = 80;          /* base duty cycle */
 153 
 154 boolean_t       spa_create_process = B_TRUE;    /* no process ==> no sysdc */
 155 extern int      zfs_sync_pass_deferred_free;
 156 
 157 /*
 158  * This (illegal) pool name is used when temporarily importing a spa_t in order
 159  * to get the vdev stats associated with the imported devices.
 160  */
 161 #define TRYIMPORT_NAME  "$import"
 162 
 163 /*
 164  * ==========================================================================
 165  * SPA properties routines
 166  * ==========================================================================
 167  */
 168 
 169 /*
 170  * Add a (source=src, propname=propval) list to an nvlist.
 171  */
 172 static void
 173 spa_prop_add_list(nvlist_t *nvl, zpool_prop_t prop, char *strval,
 174     uint64_t intval, zprop_source_t src)
 175 {
 176         const char *propname = zpool_prop_to_name(prop);
 177         nvlist_t *propval;
 178 
 179         VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 180         VERIFY(nvlist_add_uint64(propval, ZPROP_SOURCE, src) == 0);
 181 
 182         if (strval != NULL)
 183                 VERIFY(nvlist_add_string(propval, ZPROP_VALUE, strval) == 0);
 184         else
 185                 VERIFY(nvlist_add_uint64(propval, ZPROP_VALUE, intval) == 0);
 186 
 187         VERIFY(nvlist_add_nvlist(nvl, propname, propval) == 0);
 188         nvlist_free(propval);
 189 }
 190 
 191 /*
 192  * Get property values from the spa configuration.
 193  */
 194 static void
 195 spa_prop_get_config(spa_t *spa, nvlist_t **nvp)
 196 {
 197         vdev_t *rvd = spa->spa_root_vdev;
 198         dsl_pool_t *pool = spa->spa_dsl_pool;
 199         uint64_t size;
 200         uint64_t alloc;
 201         uint64_t space;
 202         uint64_t cap, version;
 203         zprop_source_t src = ZPROP_SRC_NONE;
 204         spa_config_dirent_t *dp;
 205 
 206         ASSERT(MUTEX_HELD(&spa->spa_props_lock));
 207 
 208         if (rvd != NULL) {
 209                 alloc = metaslab_class_get_alloc(spa_normal_class(spa));
 210                 size = metaslab_class_get_space(spa_normal_class(spa));
 211                 spa_prop_add_list(*nvp, ZPOOL_PROP_NAME, spa_name(spa), 0, src);
 212                 spa_prop_add_list(*nvp, ZPOOL_PROP_SIZE, NULL, size, src);
 213                 spa_prop_add_list(*nvp, ZPOOL_PROP_ALLOCATED, NULL, alloc, src);
 214                 spa_prop_add_list(*nvp, ZPOOL_PROP_FREE, NULL,
 215                     size - alloc, src);
 216 
 217                 space = 0;
 218                 for (int c = 0; c < rvd->vdev_children; c++) {
 219                         vdev_t *tvd = rvd->vdev_child[c];
 220                         space += tvd->vdev_max_asize - tvd->vdev_asize;
 221                 }
 222                 spa_prop_add_list(*nvp, ZPOOL_PROP_EXPANDSZ, NULL, space,
 223                     src);
 224 
 225                 spa_prop_add_list(*nvp, ZPOOL_PROP_READONLY, NULL,
 226                     (spa_mode(spa) == FREAD), src);
 227 
 228                 cap = (size == 0) ? 0 : (alloc * 100 / size);
 229                 spa_prop_add_list(*nvp, ZPOOL_PROP_CAPACITY, NULL, cap, src);
 230 
 231                 spa_prop_add_list(*nvp, ZPOOL_PROP_DEDUPRATIO, NULL,
 232                     ddt_get_pool_dedup_ratio(spa), src);
 233 
 234                 spa_prop_add_list(*nvp, ZPOOL_PROP_HEALTH, NULL,
 235                     rvd->vdev_state, src);
 236 
 237                 version = spa_version(spa);
 238                 if (version == zpool_prop_default_numeric(ZPOOL_PROP_VERSION))
 239                         src = ZPROP_SRC_DEFAULT;
 240                 else
 241                         src = ZPROP_SRC_LOCAL;
 242                 spa_prop_add_list(*nvp, ZPOOL_PROP_VERSION, NULL, version, src);
 243         }
 244 
 245         if (pool != NULL) {
 246                 dsl_dir_t *freedir = pool->dp_free_dir;
 247 
 248                 /*
 249                  * The $FREE directory was introduced in SPA_VERSION_DEADLISTS,
 250                  * when opening pools before this version freedir will be NULL.
 251                  */
 252                 if (freedir != NULL) {
 253                         spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING, NULL,
 254                             freedir->dd_phys->dd_used_bytes, src);
 255                 } else {
 256                         spa_prop_add_list(*nvp, ZPOOL_PROP_FREEING,
 257                             NULL, 0, src);
 258                 }
 259         }
 260 
 261         spa_prop_add_list(*nvp, ZPOOL_PROP_GUID, NULL, spa_guid(spa), src);
 262 
 263         if (spa->spa_comment != NULL) {
 264                 spa_prop_add_list(*nvp, ZPOOL_PROP_COMMENT, spa->spa_comment,
 265                     0, ZPROP_SRC_LOCAL);
 266         }
 267 
 268         if (spa->spa_root != NULL)
 269                 spa_prop_add_list(*nvp, ZPOOL_PROP_ALTROOT, spa->spa_root,
 270                     0, ZPROP_SRC_LOCAL);
 271 
 272         if ((dp = list_head(&spa->spa_config_list)) != NULL) {
 273                 if (dp->scd_path == NULL) {
 274                         spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
 275                             "none", 0, ZPROP_SRC_LOCAL);
 276                 } else if (strcmp(dp->scd_path, spa_config_path) != 0) {
 277                         spa_prop_add_list(*nvp, ZPOOL_PROP_CACHEFILE,
 278                             dp->scd_path, 0, ZPROP_SRC_LOCAL);
 279                 }
 280         }
 281 }
 282 
 283 /*
 284  * Get zpool property values.
 285  */
 286 int
 287 spa_prop_get(spa_t *spa, nvlist_t **nvp)
 288 {
 289         objset_t *mos = spa->spa_meta_objset;
 290         zap_cursor_t zc;
 291         zap_attribute_t za;
 292         int err;
 293 
 294         VERIFY(nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 295 
 296         mutex_enter(&spa->spa_props_lock);
 297 
 298         /*
 299          * Get properties from the spa config.
 300          */
 301         spa_prop_get_config(spa, nvp);
 302 
 303         /* If no pool property object, no more prop to get. */
 304         if (mos == NULL || spa->spa_pool_props_object == 0) {
 305                 mutex_exit(&spa->spa_props_lock);
 306                 return (0);
 307         }
 308 
 309         /*
 310          * Get properties from the MOS pool property object.
 311          */
 312         for (zap_cursor_init(&zc, mos, spa->spa_pool_props_object);
 313             (err = zap_cursor_retrieve(&zc, &za)) == 0;
 314             zap_cursor_advance(&zc)) {
 315                 uint64_t intval = 0;
 316                 char *strval = NULL;
 317                 zprop_source_t src = ZPROP_SRC_DEFAULT;
 318                 zpool_prop_t prop;
 319 
 320                 if ((prop = zpool_name_to_prop(za.za_name)) == ZPROP_INVAL)
 321                         continue;
 322 
 323                 switch (za.za_integer_length) {
 324                 case 8:
 325                         /* integer property */
 326                         if (za.za_first_integer !=
 327                             zpool_prop_default_numeric(prop))
 328                                 src = ZPROP_SRC_LOCAL;
 329 
 330                         if (prop == ZPOOL_PROP_BOOTFS) {
 331                                 dsl_pool_t *dp;
 332                                 dsl_dataset_t *ds = NULL;
 333 
 334                                 dp = spa_get_dsl(spa);
 335                                 dsl_pool_config_enter(dp, FTAG);
 336                                 if (err = dsl_dataset_hold_obj(dp,
 337                                     za.za_first_integer, FTAG, &ds)) {
 338                                         dsl_pool_config_exit(dp, FTAG);
 339                                         break;
 340                                 }
 341 
 342                                 strval = kmem_alloc(
 343                                     MAXNAMELEN + strlen(MOS_DIR_NAME) + 1,
 344                                     KM_SLEEP);
 345                                 dsl_dataset_name(ds, strval);
 346                                 dsl_dataset_rele(ds, FTAG);
 347                                 dsl_pool_config_exit(dp, FTAG);
 348                         } else {
 349                                 strval = NULL;
 350                                 intval = za.za_first_integer;
 351                         }
 352 
 353                         spa_prop_add_list(*nvp, prop, strval, intval, src);
 354 
 355                         if (strval != NULL)
 356                                 kmem_free(strval,
 357                                     MAXNAMELEN + strlen(MOS_DIR_NAME) + 1);
 358 
 359                         break;
 360 
 361                 case 1:
 362                         /* string property */
 363                         strval = kmem_alloc(za.za_num_integers, KM_SLEEP);
 364                         err = zap_lookup(mos, spa->spa_pool_props_object,
 365                             za.za_name, 1, za.za_num_integers, strval);
 366                         if (err) {
 367                                 kmem_free(strval, za.za_num_integers);
 368                                 break;
 369                         }
 370                         spa_prop_add_list(*nvp, prop, strval, 0, src);
 371                         kmem_free(strval, za.za_num_integers);
 372                         break;
 373 
 374                 default:
 375                         break;
 376                 }
 377         }
 378         zap_cursor_fini(&zc);
 379         mutex_exit(&spa->spa_props_lock);
 380 out:
 381         if (err && err != ENOENT) {
 382                 nvlist_free(*nvp);
 383                 *nvp = NULL;
 384                 return (err);
 385         }
 386 
 387         return (0);
 388 }
 389 
 390 /*
 391  * Validate the given pool properties nvlist and modify the list
 392  * for the property values to be set.
 393  */
 394 static int
 395 spa_prop_validate(spa_t *spa, nvlist_t *props)
 396 {
 397         nvpair_t *elem;
 398         int error = 0, reset_bootfs = 0;
 399         uint64_t objnum = 0;
 400         boolean_t has_feature = B_FALSE;
 401 
 402         elem = NULL;
 403         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
 404                 uint64_t intval;
 405                 char *strval, *slash, *check, *fname;
 406                 const char *propname = nvpair_name(elem);
 407                 zpool_prop_t prop = zpool_name_to_prop(propname);
 408 
 409                 switch (prop) {
 410                 case ZPROP_INVAL:
 411                         if (!zpool_prop_feature(propname)) {
 412                                 error = SET_ERROR(EINVAL);
 413                                 break;
 414                         }
 415 
 416                         /*
 417                          * Sanitize the input.
 418                          */
 419                         if (nvpair_type(elem) != DATA_TYPE_UINT64) {
 420                                 error = SET_ERROR(EINVAL);
 421                                 break;
 422                         }
 423 
 424                         if (nvpair_value_uint64(elem, &intval) != 0) {
 425                                 error = SET_ERROR(EINVAL);
 426                                 break;
 427                         }
 428 
 429                         if (intval != 0) {
 430                                 error = SET_ERROR(EINVAL);
 431                                 break;
 432                         }
 433 
 434                         fname = strchr(propname, '@') + 1;
 435                         if (zfeature_lookup_name(fname, NULL) != 0) {
 436                                 error = SET_ERROR(EINVAL);
 437                                 break;
 438                         }
 439 
 440                         has_feature = B_TRUE;
 441                         break;
 442 
 443                 case ZPOOL_PROP_VERSION:
 444                         error = nvpair_value_uint64(elem, &intval);
 445                         if (!error &&
 446                             (intval < spa_version(spa) ||
 447                             intval > SPA_VERSION_BEFORE_FEATURES ||
 448                             has_feature))
 449                                 error = SET_ERROR(EINVAL);
 450                         break;
 451 
 452                 case ZPOOL_PROP_DELEGATION:
 453                 case ZPOOL_PROP_AUTOREPLACE:
 454                 case ZPOOL_PROP_LISTSNAPS:
 455                 case ZPOOL_PROP_AUTOEXPAND:
 456                         error = nvpair_value_uint64(elem, &intval);
 457                         if (!error && intval > 1)
 458                                 error = SET_ERROR(EINVAL);
 459                         break;
 460 
 461                 case ZPOOL_PROP_BOOTFS:
 462                         /*
 463                          * If the pool version is less than SPA_VERSION_BOOTFS,
 464                          * or the pool is still being created (version == 0),
 465                          * the bootfs property cannot be set.
 466                          */
 467                         if (spa_version(spa) < SPA_VERSION_BOOTFS) {
 468                                 error = SET_ERROR(ENOTSUP);
 469                                 break;
 470                         }
 471 
 472                         /*
 473                          * Make sure the vdev config is bootable
 474                          */
 475                         if (!vdev_is_bootable(spa->spa_root_vdev)) {
 476                                 error = SET_ERROR(ENOTSUP);
 477                                 break;
 478                         }
 479 
 480                         reset_bootfs = 1;
 481 
 482                         error = nvpair_value_string(elem, &strval);
 483 
 484                         if (!error) {
 485                                 objset_t *os;
 486                                 uint64_t compress;
 487 
 488                                 if (strval == NULL || strval[0] == '\0') {
 489                                         objnum = zpool_prop_default_numeric(
 490                                             ZPOOL_PROP_BOOTFS);
 491                                         break;
 492                                 }
 493 
 494                                 if (error = dmu_objset_hold(strval, FTAG, &os))
 495                                         break;
 496 
 497                                 /* Must be ZPL and not gzip compressed. */
 498 
 499                                 if (dmu_objset_type(os) != DMU_OST_ZFS) {
 500                                         error = SET_ERROR(ENOTSUP);
 501                                 } else if ((error =
 502                                     dsl_prop_get_int_ds(dmu_objset_ds(os),
 503                                     zfs_prop_to_name(ZFS_PROP_COMPRESSION),
 504                                     &compress)) == 0 &&
 505                                     !BOOTFS_COMPRESS_VALID(compress)) {
 506                                         error = SET_ERROR(ENOTSUP);
 507                                 } else {
 508                                         objnum = dmu_objset_id(os);
 509                                 }
 510                                 dmu_objset_rele(os, FTAG);
 511                         }
 512                         break;
 513 
 514                 case ZPOOL_PROP_FAILUREMODE:
 515                         error = nvpair_value_uint64(elem, &intval);
 516                         if (!error && (intval < ZIO_FAILURE_MODE_WAIT ||
 517                             intval > ZIO_FAILURE_MODE_PANIC))
 518                                 error = SET_ERROR(EINVAL);
 519 
 520                         /*
 521                          * This is a special case which only occurs when
 522                          * the pool has completely failed. This allows
 523                          * the user to change the in-core failmode property
 524                          * without syncing it out to disk (I/Os might
 525                          * currently be blocked). We do this by returning
 526                          * EIO to the caller (spa_prop_set) to trick it
 527                          * into thinking we encountered a property validation
 528                          * error.
 529                          */
 530                         if (!error && spa_suspended(spa)) {
 531                                 spa->spa_failmode = intval;
 532                                 error = SET_ERROR(EIO);
 533                         }
 534                         break;
 535 
 536                 case ZPOOL_PROP_CACHEFILE:
 537                         if ((error = nvpair_value_string(elem, &strval)) != 0)
 538                                 break;
 539 
 540                         if (strval[0] == '\0')
 541                                 break;
 542 
 543                         if (strcmp(strval, "none") == 0)
 544                                 break;
 545 
 546                         if (strval[0] != '/') {
 547                                 error = SET_ERROR(EINVAL);
 548                                 break;
 549                         }
 550 
 551                         slash = strrchr(strval, '/');
 552                         ASSERT(slash != NULL);
 553 
 554                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
 555                             strcmp(slash, "/..") == 0)
 556                                 error = SET_ERROR(EINVAL);
 557                         break;
 558 
 559                 case ZPOOL_PROP_COMMENT:
 560                         if ((error = nvpair_value_string(elem, &strval)) != 0)
 561                                 break;
 562                         for (check = strval; *check != '\0'; check++) {
 563                                 /*
 564                                  * The kernel doesn't have an easy isprint()
 565                                  * check.  For this kernel check, we merely
 566                                  * check ASCII apart from DEL.  Fix this if
 567                                  * there is an easy-to-use kernel isprint().
 568                                  */
 569                                 if (*check >= 0x7f) {
 570                                         error = SET_ERROR(EINVAL);
 571                                         break;
 572                                 }
 573                                 check++;
 574                         }
 575                         if (strlen(strval) > ZPROP_MAX_COMMENT)
 576                                 error = E2BIG;
 577                         break;
 578 
 579                 case ZPOOL_PROP_DEDUPDITTO:
 580                         if (spa_version(spa) < SPA_VERSION_DEDUP)
 581                                 error = SET_ERROR(ENOTSUP);
 582                         else
 583                                 error = nvpair_value_uint64(elem, &intval);
 584                         if (error == 0 &&
 585                             intval != 0 && intval < ZIO_DEDUPDITTO_MIN)
 586                                 error = SET_ERROR(EINVAL);
 587                         break;
 588                 }
 589 
 590                 if (error)
 591                         break;
 592         }
 593 
 594         if (!error && reset_bootfs) {
 595                 error = nvlist_remove(props,
 596                     zpool_prop_to_name(ZPOOL_PROP_BOOTFS), DATA_TYPE_STRING);
 597 
 598                 if (!error) {
 599                         error = nvlist_add_uint64(props,
 600                             zpool_prop_to_name(ZPOOL_PROP_BOOTFS), objnum);
 601                 }
 602         }
 603 
 604         return (error);
 605 }
 606 
 607 void
 608 spa_configfile_set(spa_t *spa, nvlist_t *nvp, boolean_t need_sync)
 609 {
 610         char *cachefile;
 611         spa_config_dirent_t *dp;
 612 
 613         if (nvlist_lookup_string(nvp, zpool_prop_to_name(ZPOOL_PROP_CACHEFILE),
 614             &cachefile) != 0)
 615                 return;
 616 
 617         dp = kmem_alloc(sizeof (spa_config_dirent_t),
 618             KM_SLEEP);
 619 
 620         if (cachefile[0] == '\0')
 621                 dp->scd_path = spa_strdup(spa_config_path);
 622         else if (strcmp(cachefile, "none") == 0)
 623                 dp->scd_path = NULL;
 624         else
 625                 dp->scd_path = spa_strdup(cachefile);
 626 
 627         list_insert_head(&spa->spa_config_list, dp);
 628         if (need_sync)
 629                 spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
 630 }
 631 
 632 int
 633 spa_prop_set(spa_t *spa, nvlist_t *nvp)
 634 {
 635         int error;
 636         nvpair_t *elem = NULL;
 637         boolean_t need_sync = B_FALSE;
 638 
 639         if ((error = spa_prop_validate(spa, nvp)) != 0)
 640                 return (error);
 641 
 642         while ((elem = nvlist_next_nvpair(nvp, elem)) != NULL) {
 643                 zpool_prop_t prop = zpool_name_to_prop(nvpair_name(elem));
 644 
 645                 if (prop == ZPOOL_PROP_CACHEFILE ||
 646                     prop == ZPOOL_PROP_ALTROOT ||
 647                     prop == ZPOOL_PROP_READONLY)
 648                         continue;
 649 
 650                 if (prop == ZPOOL_PROP_VERSION || prop == ZPROP_INVAL) {
 651                         uint64_t ver;
 652 
 653                         if (prop == ZPOOL_PROP_VERSION) {
 654                                 VERIFY(nvpair_value_uint64(elem, &ver) == 0);
 655                         } else {
 656                                 ASSERT(zpool_prop_feature(nvpair_name(elem)));
 657                                 ver = SPA_VERSION_FEATURES;
 658                                 need_sync = B_TRUE;
 659                         }
 660 
 661                         /* Save time if the version is already set. */
 662                         if (ver == spa_version(spa))
 663                                 continue;
 664 
 665                         /*
 666                          * In addition to the pool directory object, we might
 667                          * create the pool properties object, the features for
 668                          * read object, the features for write object, or the
 669                          * feature descriptions object.
 670                          */
 671                         error = dsl_sync_task(spa->spa_name, NULL,
 672                             spa_sync_version, &ver, 6);
 673                         if (error)
 674                                 return (error);
 675                         continue;
 676                 }
 677 
 678                 need_sync = B_TRUE;
 679                 break;
 680         }
 681 
 682         if (need_sync) {
 683                 return (dsl_sync_task(spa->spa_name, NULL, spa_sync_props,
 684                     nvp, 6));
 685         }
 686 
 687         return (0);
 688 }
 689 
 690 /*
 691  * If the bootfs property value is dsobj, clear it.
 692  */
 693 void
 694 spa_prop_clear_bootfs(spa_t *spa, uint64_t dsobj, dmu_tx_t *tx)
 695 {
 696         if (spa->spa_bootfs == dsobj && spa->spa_pool_props_object != 0) {
 697                 VERIFY(zap_remove(spa->spa_meta_objset,
 698                     spa->spa_pool_props_object,
 699                     zpool_prop_to_name(ZPOOL_PROP_BOOTFS), tx) == 0);
 700                 spa->spa_bootfs = 0;
 701         }
 702 }
 703 
 704 /*ARGSUSED*/
 705 static int
 706 spa_change_guid_check(void *arg, dmu_tx_t *tx)
 707 {
 708         uint64_t *newguid = arg;
 709         spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 710         vdev_t *rvd = spa->spa_root_vdev;
 711         uint64_t vdev_state;
 712 
 713         spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
 714         vdev_state = rvd->vdev_state;
 715         spa_config_exit(spa, SCL_STATE, FTAG);
 716 
 717         if (vdev_state != VDEV_STATE_HEALTHY)
 718                 return (SET_ERROR(ENXIO));
 719 
 720         ASSERT3U(spa_guid(spa), !=, *newguid);
 721 
 722         return (0);
 723 }
 724 
 725 static void
 726 spa_change_guid_sync(void *arg, dmu_tx_t *tx)
 727 {
 728         uint64_t *newguid = arg;
 729         spa_t *spa = dmu_tx_pool(tx)->dp_spa;
 730         uint64_t oldguid;
 731         vdev_t *rvd = spa->spa_root_vdev;
 732 
 733         oldguid = spa_guid(spa);
 734 
 735         spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
 736         rvd->vdev_guid = *newguid;
 737         rvd->vdev_guid_sum += (*newguid - oldguid);
 738         vdev_config_dirty(rvd);
 739         spa_config_exit(spa, SCL_STATE, FTAG);
 740 
 741         spa_history_log_internal(spa, "guid change", tx, "old=%llu new=%llu",
 742             oldguid, *newguid);
 743 }
 744 
 745 /*
 746  * Change the GUID for the pool.  This is done so that we can later
 747  * re-import a pool built from a clone of our own vdevs.  We will modify
 748  * the root vdev's guid, our own pool guid, and then mark all of our
 749  * vdevs dirty.  Note that we must make sure that all our vdevs are
 750  * online when we do this, or else any vdevs that weren't present
 751  * would be orphaned from our pool.  We are also going to issue a
 752  * sysevent to update any watchers.
 753  */
 754 int
 755 spa_change_guid(spa_t *spa)
 756 {
 757         int error;
 758         uint64_t guid;
 759 
 760         mutex_enter(&spa->spa_vdev_top_lock);
 761         mutex_enter(&spa_namespace_lock);
 762         guid = spa_generate_guid(NULL);
 763 
 764         error = dsl_sync_task(spa->spa_name, spa_change_guid_check,
 765             spa_change_guid_sync, &guid, 5);
 766 
 767         if (error == 0) {
 768                 spa_config_sync(spa, B_FALSE, B_TRUE);
 769                 spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
 770         }
 771 
 772         mutex_exit(&spa_namespace_lock);
 773         mutex_exit(&spa->spa_vdev_top_lock);
 774 
 775         return (error);
 776 }
 777 
 778 /*
 779  * ==========================================================================
 780  * SPA state manipulation (open/create/destroy/import/export)
 781  * ==========================================================================
 782  */
 783 
 784 static int
 785 spa_error_entry_compare(const void *a, const void *b)
 786 {
 787         spa_error_entry_t *sa = (spa_error_entry_t *)a;
 788         spa_error_entry_t *sb = (spa_error_entry_t *)b;
 789         int ret;
 790 
 791         ret = bcmp(&sa->se_bookmark, &sb->se_bookmark,
 792             sizeof (zbookmark_t));
 793 
 794         if (ret < 0)
 795                 return (-1);
 796         else if (ret > 0)
 797                 return (1);
 798         else
 799                 return (0);
 800 }
 801 
 802 /*
 803  * Utility function which retrieves copies of the current logs and
 804  * re-initializes them in the process.
 805  */
 806 void
 807 spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub)
 808 {
 809         ASSERT(MUTEX_HELD(&spa->spa_errlist_lock));
 810 
 811         bcopy(&spa->spa_errlist_last, last, sizeof (avl_tree_t));
 812         bcopy(&spa->spa_errlist_scrub, scrub, sizeof (avl_tree_t));
 813 
 814         avl_create(&spa->spa_errlist_scrub,
 815             spa_error_entry_compare, sizeof (spa_error_entry_t),
 816             offsetof(spa_error_entry_t, se_avl));
 817         avl_create(&spa->spa_errlist_last,
 818             spa_error_entry_compare, sizeof (spa_error_entry_t),
 819             offsetof(spa_error_entry_t, se_avl));
 820 }
 821 
 822 static void
 823 spa_taskqs_init(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
 824 {
 825         const zio_taskq_info_t *ztip = &zio_taskqs[t][q];
 826         enum zti_modes mode = ztip->zti_mode;
 827         uint_t value = ztip->zti_value;
 828         uint_t count = ztip->zti_count;
 829         spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
 830         char name[32];
 831         uint_t flags = 0;
 832         boolean_t batch = B_FALSE;
 833 
 834         if (mode == ZTI_MODE_NULL) {
 835                 tqs->stqs_count = 0;
 836                 tqs->stqs_taskq = NULL;
 837                 return;
 838         }
 839 
 840         ASSERT3U(count, >, 0);
 841 
 842         tqs->stqs_count = count;
 843         tqs->stqs_taskq = kmem_alloc(count * sizeof (taskq_t *), KM_SLEEP);
 844 
 845         for (uint_t i = 0; i < count; i++) {
 846                 taskq_t *tq;
 847 
 848                 switch (mode) {
 849                 case ZTI_MODE_FIXED:
 850                         ASSERT3U(value, >=, 1);
 851                         value = MAX(value, 1);
 852                         break;
 853 
 854                 case ZTI_MODE_BATCH:
 855                         batch = B_TRUE;
 856                         flags |= TASKQ_THREADS_CPU_PCT;
 857                         value = zio_taskq_batch_pct;
 858                         break;
 859 
 860                 case ZTI_MODE_ONLINE_PERCENT:
 861                         flags |= TASKQ_THREADS_CPU_PCT;
 862                         break;
 863 
 864                 default:
 865                         panic("unrecognized mode for %s_%s taskq (%u:%u) in "
 866                             "spa_activate()",
 867                             zio_type_name[t], zio_taskq_types[q], mode, value);
 868                         break;
 869                 }
 870 
 871                 if (count > 1) {
 872                         (void) snprintf(name, sizeof (name), "%s_%s_%u",
 873                             zio_type_name[t], zio_taskq_types[q], i);
 874                 } else {
 875                         (void) snprintf(name, sizeof (name), "%s_%s",
 876                             zio_type_name[t], zio_taskq_types[q]);
 877                 }
 878 
 879                 if (zio_taskq_sysdc && spa->spa_proc != &p0) {
 880                         if (batch)
 881                                 flags |= TASKQ_DC_BATCH;
 882 
 883                         tq = taskq_create_sysdc(name, value, 50, INT_MAX,
 884                             spa->spa_proc, zio_taskq_basedc, flags);
 885                 } else {
 886                         tq = taskq_create_proc(name, value, maxclsyspri, 50,
 887                             INT_MAX, spa->spa_proc, flags);
 888                 }
 889 
 890                 tqs->stqs_taskq[i] = tq;
 891         }
 892 }
 893 
 894 static void
 895 spa_taskqs_fini(spa_t *spa, zio_type_t t, zio_taskq_type_t q)
 896 {
 897         spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
 898 
 899         if (tqs->stqs_taskq == NULL) {
 900                 ASSERT0(tqs->stqs_count);
 901                 return;
 902         }
 903 
 904         for (uint_t i = 0; i < tqs->stqs_count; i++) {
 905                 ASSERT3P(tqs->stqs_taskq[i], !=, NULL);
 906                 taskq_destroy(tqs->stqs_taskq[i]);
 907         }
 908 
 909         kmem_free(tqs->stqs_taskq, tqs->stqs_count * sizeof (taskq_t *));
 910         tqs->stqs_taskq = NULL;
 911 }
 912 
 913 /*
 914  * Dispatch a task to the appropriate taskq for the ZFS I/O type and priority.
 915  * Note that a type may have multiple discrete taskqs to avoid lock contention
 916  * on the taskq itself. In that case we choose which taskq at random by using
 917  * the low bits of gethrtime().
 918  */
 919 void
 920 spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
 921     task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent)
 922 {
 923         spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
 924         taskq_t *tq;
 925 
 926         ASSERT3P(tqs->stqs_taskq, !=, NULL);
 927         ASSERT3U(tqs->stqs_count, !=, 0);
 928 
 929         if (tqs->stqs_count == 1) {
 930                 tq = tqs->stqs_taskq[0];
 931         } else {
 932                 tq = tqs->stqs_taskq[gethrtime() % tqs->stqs_count];
 933         }
 934 
 935         taskq_dispatch_ent(tq, func, arg, flags, ent);
 936 }
 937 
 938 static void
 939 spa_create_zio_taskqs(spa_t *spa)
 940 {
 941         for (int t = 0; t < ZIO_TYPES; t++) {
 942                 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
 943                         spa_taskqs_init(spa, t, q);
 944                 }
 945         }
 946 }
 947 
 948 #ifdef _KERNEL
 949 static void
 950 spa_thread(void *arg)
 951 {
 952         callb_cpr_t cprinfo;
 953 
 954         spa_t *spa = arg;
 955         user_t *pu = PTOU(curproc);
 956 
 957         CALLB_CPR_INIT(&cprinfo, &spa->spa_proc_lock, callb_generic_cpr,
 958             spa->spa_name);
 959 
 960         ASSERT(curproc != &p0);
 961         (void) snprintf(pu->u_psargs, sizeof (pu->u_psargs),
 962             "zpool-%s", spa->spa_name);
 963         (void) strlcpy(pu->u_comm, pu->u_psargs, sizeof (pu->u_comm));
 964 
 965         /* bind this thread to the requested psrset */
 966         if (zio_taskq_psrset_bind != PS_NONE) {
 967                 pool_lock();
 968                 mutex_enter(&cpu_lock);
 969                 mutex_enter(&pidlock);
 970                 mutex_enter(&curproc->p_lock);
 971 
 972                 if (cpupart_bind_thread(curthread, zio_taskq_psrset_bind,
 973                     0, NULL, NULL) == 0)  {
 974                         curthread->t_bind_pset = zio_taskq_psrset_bind;
 975                 } else {
 976                         cmn_err(CE_WARN,
 977                             "Couldn't bind process for zfs pool \"%s\" to "
 978                             "pset %d\n", spa->spa_name, zio_taskq_psrset_bind);
 979                 }
 980 
 981                 mutex_exit(&curproc->p_lock);
 982                 mutex_exit(&pidlock);
 983                 mutex_exit(&cpu_lock);
 984                 pool_unlock();
 985         }
 986 
 987         if (zio_taskq_sysdc) {
 988                 sysdc_thread_enter(curthread, 100, 0);
 989         }
 990 
 991         spa->spa_proc = curproc;
 992         spa->spa_did = curthread->t_did;
 993 
 994         spa_create_zio_taskqs(spa);
 995 
 996         mutex_enter(&spa->spa_proc_lock);
 997         ASSERT(spa->spa_proc_state == SPA_PROC_CREATED);
 998 
 999         spa->spa_proc_state = SPA_PROC_ACTIVE;
1000         cv_broadcast(&spa->spa_proc_cv);
1001 
1002         CALLB_CPR_SAFE_BEGIN(&cprinfo);
1003         while (spa->spa_proc_state == SPA_PROC_ACTIVE)
1004                 cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1005         CALLB_CPR_SAFE_END(&cprinfo, &spa->spa_proc_lock);
1006 
1007         ASSERT(spa->spa_proc_state == SPA_PROC_DEACTIVATE);
1008         spa->spa_proc_state = SPA_PROC_GONE;
1009         spa->spa_proc = &p0;
1010         cv_broadcast(&spa->spa_proc_cv);
1011         CALLB_CPR_EXIT(&cprinfo);   /* drops spa_proc_lock */
1012 
1013         mutex_enter(&curproc->p_lock);
1014         lwp_exit();
1015 }
1016 #endif
1017 
1018 /*
1019  * Activate an uninitialized pool.
1020  */
1021 static void
1022 spa_activate(spa_t *spa, int mode)
1023 {
1024         ASSERT(spa->spa_state == POOL_STATE_UNINITIALIZED);
1025 
1026         spa->spa_state = POOL_STATE_ACTIVE;
1027         spa->spa_mode = mode;
1028 
1029         spa->spa_normal_class = metaslab_class_create(spa, zfs_metaslab_ops);
1030         spa->spa_log_class = metaslab_class_create(spa, zfs_metaslab_ops);
1031 
1032         /* Try to create a covering process */
1033         mutex_enter(&spa->spa_proc_lock);
1034         ASSERT(spa->spa_proc_state == SPA_PROC_NONE);
1035         ASSERT(spa->spa_proc == &p0);
1036         spa->spa_did = 0;
1037 
1038         /* Only create a process if we're going to be around a while. */
1039         if (spa_create_process && strcmp(spa->spa_name, TRYIMPORT_NAME) != 0) {
1040                 if (newproc(spa_thread, (caddr_t)spa, syscid, maxclsyspri,
1041                     NULL, 0) == 0) {
1042                         spa->spa_proc_state = SPA_PROC_CREATED;
1043                         while (spa->spa_proc_state == SPA_PROC_CREATED) {
1044                                 cv_wait(&spa->spa_proc_cv,
1045                                     &spa->spa_proc_lock);
1046                         }
1047                         ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1048                         ASSERT(spa->spa_proc != &p0);
1049                         ASSERT(spa->spa_did != 0);
1050                 } else {
1051 #ifdef _KERNEL
1052                         cmn_err(CE_WARN,
1053                             "Couldn't create process for zfs pool \"%s\"\n",
1054                             spa->spa_name);
1055 #endif
1056                 }
1057         }
1058         mutex_exit(&spa->spa_proc_lock);
1059 
1060         /* If we didn't create a process, we need to create our taskqs. */
1061         if (spa->spa_proc == &p0) {
1062                 spa_create_zio_taskqs(spa);
1063         }
1064 
1065         list_create(&spa->spa_config_dirty_list, sizeof (vdev_t),
1066             offsetof(vdev_t, vdev_config_dirty_node));
1067         list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
1068             offsetof(vdev_t, vdev_state_dirty_node));
1069 
1070         txg_list_create(&spa->spa_vdev_txg_list,
1071             offsetof(struct vdev, vdev_txg_node));
1072 
1073         avl_create(&spa->spa_errlist_scrub,
1074             spa_error_entry_compare, sizeof (spa_error_entry_t),
1075             offsetof(spa_error_entry_t, se_avl));
1076         avl_create(&spa->spa_errlist_last,
1077             spa_error_entry_compare, sizeof (spa_error_entry_t),
1078             offsetof(spa_error_entry_t, se_avl));
1079 }
1080 
1081 /*
1082  * Opposite of spa_activate().
1083  */
1084 static void
1085 spa_deactivate(spa_t *spa)
1086 {
1087         ASSERT(spa->spa_sync_on == B_FALSE);
1088         ASSERT(spa->spa_dsl_pool == NULL);
1089         ASSERT(spa->spa_root_vdev == NULL);
1090         ASSERT(spa->spa_async_zio_root == NULL);
1091         ASSERT(spa->spa_state != POOL_STATE_UNINITIALIZED);
1092 
1093         txg_list_destroy(&spa->spa_vdev_txg_list);
1094 
1095         list_destroy(&spa->spa_config_dirty_list);
1096         list_destroy(&spa->spa_state_dirty_list);
1097 
1098         for (int t = 0; t < ZIO_TYPES; t++) {
1099                 for (int q = 0; q < ZIO_TASKQ_TYPES; q++) {
1100                         spa_taskqs_fini(spa, t, q);
1101                 }
1102         }
1103 
1104         metaslab_class_destroy(spa->spa_normal_class);
1105         spa->spa_normal_class = NULL;
1106 
1107         metaslab_class_destroy(spa->spa_log_class);
1108         spa->spa_log_class = NULL;
1109 
1110         /*
1111          * If this was part of an import or the open otherwise failed, we may
1112          * still have errors left in the queues.  Empty them just in case.
1113          */
1114         spa_errlog_drain(spa);
1115 
1116         avl_destroy(&spa->spa_errlist_scrub);
1117         avl_destroy(&spa->spa_errlist_last);
1118 
1119         spa->spa_state = POOL_STATE_UNINITIALIZED;
1120 
1121         mutex_enter(&spa->spa_proc_lock);
1122         if (spa->spa_proc_state != SPA_PROC_NONE) {
1123                 ASSERT(spa->spa_proc_state == SPA_PROC_ACTIVE);
1124                 spa->spa_proc_state = SPA_PROC_DEACTIVATE;
1125                 cv_broadcast(&spa->spa_proc_cv);
1126                 while (spa->spa_proc_state == SPA_PROC_DEACTIVATE) {
1127                         ASSERT(spa->spa_proc != &p0);
1128                         cv_wait(&spa->spa_proc_cv, &spa->spa_proc_lock);
1129                 }
1130                 ASSERT(spa->spa_proc_state == SPA_PROC_GONE);
1131                 spa->spa_proc_state = SPA_PROC_NONE;
1132         }
1133         ASSERT(spa->spa_proc == &p0);
1134         mutex_exit(&spa->spa_proc_lock);
1135 
1136         /*
1137          * We want to make sure spa_thread() has actually exited the ZFS
1138          * module, so that the module can't be unloaded out from underneath
1139          * it.
1140          */
1141         if (spa->spa_did != 0) {
1142                 thread_join(spa->spa_did);
1143                 spa->spa_did = 0;
1144         }
1145 }
1146 
1147 /*
1148  * Verify a pool configuration, and construct the vdev tree appropriately.  This
1149  * will create all the necessary vdevs in the appropriate layout, with each vdev
1150  * in the CLOSED state.  This will prep the pool before open/creation/import.
1151  * All vdev validation is done by the vdev_alloc() routine.
1152  */
1153 static int
1154 spa_config_parse(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent,
1155     uint_t id, int atype)
1156 {
1157         nvlist_t **child;
1158         uint_t children;
1159         int error;
1160 
1161         if ((error = vdev_alloc(spa, vdp, nv, parent, id, atype)) != 0)
1162                 return (error);
1163 
1164         if ((*vdp)->vdev_ops->vdev_op_leaf)
1165                 return (0);
1166 
1167         error = nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1168             &child, &children);
1169 
1170         if (error == ENOENT)
1171                 return (0);
1172 
1173         if (error) {
1174                 vdev_free(*vdp);
1175                 *vdp = NULL;
1176                 return (SET_ERROR(EINVAL));
1177         }
1178 
1179         for (int c = 0; c < children; c++) {
1180                 vdev_t *vd;
1181                 if ((error = spa_config_parse(spa, &vd, child[c], *vdp, c,
1182                     atype)) != 0) {
1183                         vdev_free(*vdp);
1184                         *vdp = NULL;
1185                         return (error);
1186                 }
1187         }
1188 
1189         ASSERT(*vdp != NULL);
1190 
1191         return (0);
1192 }
1193 
1194 /*
1195  * Opposite of spa_load().
1196  */
1197 static void
1198 spa_unload(spa_t *spa)
1199 {
1200         int i;
1201 
1202         ASSERT(MUTEX_HELD(&spa_namespace_lock));
1203 
1204         /*
1205          * Stop async tasks.
1206          */
1207         spa_async_suspend(spa);
1208 
1209         /*
1210          * Stop syncing.
1211          */
1212         if (spa->spa_sync_on) {
1213                 txg_sync_stop(spa->spa_dsl_pool);
1214                 spa->spa_sync_on = B_FALSE;
1215         }
1216 
1217         /*
1218          * Wait for any outstanding async I/O to complete.
1219          */
1220         if (spa->spa_async_zio_root != NULL) {
1221                 (void) zio_wait(spa->spa_async_zio_root);
1222                 spa->spa_async_zio_root = NULL;
1223         }
1224 
1225         bpobj_close(&spa->spa_deferred_bpobj);
1226 
1227         /*
1228          * Close the dsl pool.
1229          */
1230         if (spa->spa_dsl_pool) {
1231                 dsl_pool_close(spa->spa_dsl_pool);
1232                 spa->spa_dsl_pool = NULL;
1233                 spa->spa_meta_objset = NULL;
1234         }
1235 
1236         ddt_unload(spa);
1237 
1238         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1239 
1240         /*
1241          * Drop and purge level 2 cache
1242          */
1243         spa_l2cache_drop(spa);
1244 
1245         /*
1246          * Close all vdevs.
1247          */
1248         if (spa->spa_root_vdev)
1249                 vdev_free(spa->spa_root_vdev);
1250         ASSERT(spa->spa_root_vdev == NULL);
1251 
1252         for (i = 0; i < spa->spa_spares.sav_count; i++)
1253                 vdev_free(spa->spa_spares.sav_vdevs[i]);
1254         if (spa->spa_spares.sav_vdevs) {
1255                 kmem_free(spa->spa_spares.sav_vdevs,
1256                     spa->spa_spares.sav_count * sizeof (void *));
1257                 spa->spa_spares.sav_vdevs = NULL;
1258         }
1259         if (spa->spa_spares.sav_config) {
1260                 nvlist_free(spa->spa_spares.sav_config);
1261                 spa->spa_spares.sav_config = NULL;
1262         }
1263         spa->spa_spares.sav_count = 0;
1264 
1265         for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
1266                 vdev_clear_stats(spa->spa_l2cache.sav_vdevs[i]);
1267                 vdev_free(spa->spa_l2cache.sav_vdevs[i]);
1268         }
1269         if (spa->spa_l2cache.sav_vdevs) {
1270                 kmem_free(spa->spa_l2cache.sav_vdevs,
1271                     spa->spa_l2cache.sav_count * sizeof (void *));
1272                 spa->spa_l2cache.sav_vdevs = NULL;
1273         }
1274         if (spa->spa_l2cache.sav_config) {
1275                 nvlist_free(spa->spa_l2cache.sav_config);
1276                 spa->spa_l2cache.sav_config = NULL;
1277         }
1278         spa->spa_l2cache.sav_count = 0;
1279 
1280         spa->spa_async_suspended = 0;
1281 
1282         if (spa->spa_comment != NULL) {
1283                 spa_strfree(spa->spa_comment);
1284                 spa->spa_comment = NULL;
1285         }
1286 
1287         spa_config_exit(spa, SCL_ALL, FTAG);
1288 }
1289 
1290 /*
1291  * Load (or re-load) the current list of vdevs describing the active spares for
1292  * this pool.  When this is called, we have some form of basic information in
1293  * 'spa_spares.sav_config'.  We parse this into vdevs, try to open them, and
1294  * then re-generate a more complete list including status information.
1295  */
1296 static void
1297 spa_load_spares(spa_t *spa)
1298 {
1299         nvlist_t **spares;
1300         uint_t nspares;
1301         int i;
1302         vdev_t *vd, *tvd;
1303 
1304         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1305 
1306         /*
1307          * First, close and free any existing spare vdevs.
1308          */
1309         for (i = 0; i < spa->spa_spares.sav_count; i++) {
1310                 vd = spa->spa_spares.sav_vdevs[i];
1311 
1312                 /* Undo the call to spa_activate() below */
1313                 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1314                     B_FALSE)) != NULL && tvd->vdev_isspare)
1315                         spa_spare_remove(tvd);
1316                 vdev_close(vd);
1317                 vdev_free(vd);
1318         }
1319 
1320         if (spa->spa_spares.sav_vdevs)
1321                 kmem_free(spa->spa_spares.sav_vdevs,
1322                     spa->spa_spares.sav_count * sizeof (void *));
1323 
1324         if (spa->spa_spares.sav_config == NULL)
1325                 nspares = 0;
1326         else
1327                 VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
1328                     ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
1329 
1330         spa->spa_spares.sav_count = (int)nspares;
1331         spa->spa_spares.sav_vdevs = NULL;
1332 
1333         if (nspares == 0)
1334                 return;
1335 
1336         /*
1337          * Construct the array of vdevs, opening them to get status in the
1338          * process.   For each spare, there is potentially two different vdev_t
1339          * structures associated with it: one in the list of spares (used only
1340          * for basic validation purposes) and one in the active vdev
1341          * configuration (if it's spared in).  During this phase we open and
1342          * validate each vdev on the spare list.  If the vdev also exists in the
1343          * active configuration, then we also mark this vdev as an active spare.
1344          */
1345         spa->spa_spares.sav_vdevs = kmem_alloc(nspares * sizeof (void *),
1346             KM_SLEEP);
1347         for (i = 0; i < spa->spa_spares.sav_count; i++) {
1348                 VERIFY(spa_config_parse(spa, &vd, spares[i], NULL, 0,
1349                     VDEV_ALLOC_SPARE) == 0);
1350                 ASSERT(vd != NULL);
1351 
1352                 spa->spa_spares.sav_vdevs[i] = vd;
1353 
1354                 if ((tvd = spa_lookup_by_guid(spa, vd->vdev_guid,
1355                     B_FALSE)) != NULL) {
1356                         if (!tvd->vdev_isspare)
1357                                 spa_spare_add(tvd);
1358 
1359                         /*
1360                          * We only mark the spare active if we were successfully
1361                          * able to load the vdev.  Otherwise, importing a pool
1362                          * with a bad active spare would result in strange
1363                          * behavior, because multiple pool would think the spare
1364                          * is actively in use.
1365                          *
1366                          * There is a vulnerability here to an equally bizarre
1367                          * circumstance, where a dead active spare is later
1368                          * brought back to life (onlined or otherwise).  Given
1369                          * the rarity of this scenario, and the extra complexity
1370                          * it adds, we ignore the possibility.
1371                          */
1372                         if (!vdev_is_dead(tvd))
1373                                 spa_spare_activate(tvd);
1374                 }
1375 
1376                 vd->vdev_top = vd;
1377                 vd->vdev_aux = &spa->spa_spares;
1378 
1379                 if (vdev_open(vd) != 0)
1380                         continue;
1381 
1382                 if (vdev_validate_aux(vd) == 0)
1383                         spa_spare_add(vd);
1384         }
1385 
1386         /*
1387          * Recompute the stashed list of spares, with status information
1388          * this time.
1389          */
1390         VERIFY(nvlist_remove(spa->spa_spares.sav_config, ZPOOL_CONFIG_SPARES,
1391             DATA_TYPE_NVLIST_ARRAY) == 0);
1392 
1393         spares = kmem_alloc(spa->spa_spares.sav_count * sizeof (void *),
1394             KM_SLEEP);
1395         for (i = 0; i < spa->spa_spares.sav_count; i++)
1396                 spares[i] = vdev_config_generate(spa,
1397                     spa->spa_spares.sav_vdevs[i], B_TRUE, VDEV_CONFIG_SPARE);
1398         VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
1399             ZPOOL_CONFIG_SPARES, spares, spa->spa_spares.sav_count) == 0);
1400         for (i = 0; i < spa->spa_spares.sav_count; i++)
1401                 nvlist_free(spares[i]);
1402         kmem_free(spares, spa->spa_spares.sav_count * sizeof (void *));
1403 }
1404 
1405 /*
1406  * Load (or re-load) the current list of vdevs describing the active l2cache for
1407  * this pool.  When this is called, we have some form of basic information in
1408  * 'spa_l2cache.sav_config'.  We parse this into vdevs, try to open them, and
1409  * then re-generate a more complete list including status information.
1410  * Devices which are already active have their details maintained, and are
1411  * not re-opened.
1412  */
1413 static void
1414 spa_load_l2cache(spa_t *spa)
1415 {
1416         nvlist_t **l2cache;
1417         uint_t nl2cache;
1418         int i, j, oldnvdevs;
1419         uint64_t guid;
1420         vdev_t *vd, **oldvdevs, **newvdevs;
1421         spa_aux_vdev_t *sav = &spa->spa_l2cache;
1422 
1423         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
1424 
1425         if (sav->sav_config != NULL) {
1426                 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config,
1427                     ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
1428                 newvdevs = kmem_alloc(nl2cache * sizeof (void *), KM_SLEEP);
1429         } else {
1430                 nl2cache = 0;
1431                 newvdevs = NULL;
1432         }
1433 
1434         oldvdevs = sav->sav_vdevs;
1435         oldnvdevs = sav->sav_count;
1436         sav->sav_vdevs = NULL;
1437         sav->sav_count = 0;
1438 
1439         /*
1440          * Process new nvlist of vdevs.
1441          */
1442         for (i = 0; i < nl2cache; i++) {
1443                 VERIFY(nvlist_lookup_uint64(l2cache[i], ZPOOL_CONFIG_GUID,
1444                     &guid) == 0);
1445 
1446                 newvdevs[i] = NULL;
1447                 for (j = 0; j < oldnvdevs; j++) {
1448                         vd = oldvdevs[j];
1449                         if (vd != NULL && guid == vd->vdev_guid) {
1450                                 /*
1451                                  * Retain previous vdev for add/remove ops.
1452                                  */
1453                                 newvdevs[i] = vd;
1454                                 oldvdevs[j] = NULL;
1455                                 break;
1456                         }
1457                 }
1458 
1459                 if (newvdevs[i] == NULL) {
1460                         /*
1461                          * Create new vdev
1462                          */
1463                         VERIFY(spa_config_parse(spa, &vd, l2cache[i], NULL, 0,
1464                             VDEV_ALLOC_L2CACHE) == 0);
1465                         ASSERT(vd != NULL);
1466                         newvdevs[i] = vd;
1467 
1468                         /*
1469                          * Commit this vdev as an l2cache device,
1470                          * even if it fails to open.
1471                          */
1472                         spa_l2cache_add(vd);
1473 
1474                         vd->vdev_top = vd;
1475                         vd->vdev_aux = sav;
1476 
1477                         spa_l2cache_activate(vd);
1478 
1479                         if (vdev_open(vd) != 0)
1480                                 continue;
1481 
1482                         (void) vdev_validate_aux(vd);
1483 
1484                         if (!vdev_is_dead(vd))
1485                                 l2arc_add_vdev(spa, vd);
1486                 }
1487         }
1488 
1489         /*
1490          * Purge vdevs that were dropped
1491          */
1492         for (i = 0; i < oldnvdevs; i++) {
1493                 uint64_t pool;
1494 
1495                 vd = oldvdevs[i];
1496                 if (vd != NULL) {
1497                         ASSERT(vd->vdev_isl2cache);
1498 
1499                         if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
1500                             pool != 0ULL && l2arc_vdev_present(vd))
1501                                 l2arc_remove_vdev(vd);
1502                         vdev_clear_stats(vd);
1503                         vdev_free(vd);
1504                 }
1505         }
1506 
1507         if (oldvdevs)
1508                 kmem_free(oldvdevs, oldnvdevs * sizeof (void *));
1509 
1510         if (sav->sav_config == NULL)
1511                 goto out;
1512 
1513         sav->sav_vdevs = newvdevs;
1514         sav->sav_count = (int)nl2cache;
1515 
1516         /*
1517          * Recompute the stashed list of l2cache devices, with status
1518          * information this time.
1519          */
1520         VERIFY(nvlist_remove(sav->sav_config, ZPOOL_CONFIG_L2CACHE,
1521             DATA_TYPE_NVLIST_ARRAY) == 0);
1522 
1523         l2cache = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
1524         for (i = 0; i < sav->sav_count; i++)
1525                 l2cache[i] = vdev_config_generate(spa,
1526                     sav->sav_vdevs[i], B_TRUE, VDEV_CONFIG_L2CACHE);
1527         VERIFY(nvlist_add_nvlist_array(sav->sav_config,
1528             ZPOOL_CONFIG_L2CACHE, l2cache, sav->sav_count) == 0);
1529 out:
1530         for (i = 0; i < sav->sav_count; i++)
1531                 nvlist_free(l2cache[i]);
1532         if (sav->sav_count)
1533                 kmem_free(l2cache, sav->sav_count * sizeof (void *));
1534 }
1535 
1536 static int
1537 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
1538 {
1539         dmu_buf_t *db;
1540         char *packed = NULL;
1541         size_t nvsize = 0;
1542         int error;
1543         *value = NULL;
1544 
1545         VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
1546         nvsize = *(uint64_t *)db->db_data;
1547         dmu_buf_rele(db, FTAG);
1548 
1549         packed = kmem_alloc(nvsize, KM_SLEEP);
1550         error = dmu_read(spa->spa_meta_objset, obj, 0, nvsize, packed,
1551             DMU_READ_PREFETCH);
1552         if (error == 0)
1553                 error = nvlist_unpack(packed, nvsize, value, 0);
1554         kmem_free(packed, nvsize);
1555 
1556         return (error);
1557 }
1558 
1559 /*
1560  * Checks to see if the given vdev could not be opened, in which case we post a
1561  * sysevent to notify the autoreplace code that the device has been removed.
1562  */
1563 static void
1564 spa_check_removed(vdev_t *vd)
1565 {
1566         for (int c = 0; c < vd->vdev_children; c++)
1567                 spa_check_removed(vd->vdev_child[c]);
1568 
1569         if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
1570             !vd->vdev_ishole) {
1571                 zfs_post_autoreplace(vd->vdev_spa, vd);
1572                 spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
1573         }
1574 }
1575 
1576 /*
1577  * Validate the current config against the MOS config
1578  */
1579 static boolean_t
1580 spa_config_valid(spa_t *spa, nvlist_t *config)
1581 {
1582         vdev_t *mrvd, *rvd = spa->spa_root_vdev;
1583         nvlist_t *nv;
1584 
1585         VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nv) == 0);
1586 
1587         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1588         VERIFY(spa_config_parse(spa, &mrvd, nv, NULL, 0, VDEV_ALLOC_LOAD) == 0);
1589 
1590         ASSERT3U(rvd->vdev_children, ==, mrvd->vdev_children);
1591 
1592         /*
1593          * If we're doing a normal import, then build up any additional
1594          * diagnostic information about missing devices in this config.
1595          * We'll pass this up to the user for further processing.
1596          */
1597         if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG)) {
1598                 nvlist_t **child, *nv;
1599                 uint64_t idx = 0;
1600 
1601                 child = kmem_alloc(rvd->vdev_children * sizeof (nvlist_t **),
1602                     KM_SLEEP);
1603                 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1604 
1605                 for (int c = 0; c < rvd->vdev_children; c++) {
1606                         vdev_t *tvd = rvd->vdev_child[c];
1607                         vdev_t *mtvd  = mrvd->vdev_child[c];
1608 
1609                         if (tvd->vdev_ops == &vdev_missing_ops &&
1610                             mtvd->vdev_ops != &vdev_missing_ops &&
1611                             mtvd->vdev_islog)
1612                                 child[idx++] = vdev_config_generate(spa, mtvd,
1613                                     B_FALSE, 0);
1614                 }
1615 
1616                 if (idx) {
1617                         VERIFY(nvlist_add_nvlist_array(nv,
1618                             ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
1619                         VERIFY(nvlist_add_nvlist(spa->spa_load_info,
1620                             ZPOOL_CONFIG_MISSING_DEVICES, nv) == 0);
1621 
1622                         for (int i = 0; i < idx; i++)
1623                                 nvlist_free(child[i]);
1624                 }
1625                 nvlist_free(nv);
1626                 kmem_free(child, rvd->vdev_children * sizeof (char **));
1627         }
1628 
1629         /*
1630          * Compare the root vdev tree with the information we have
1631          * from the MOS config (mrvd). Check each top-level vdev
1632          * with the corresponding MOS config top-level (mtvd).
1633          */
1634         for (int c = 0; c < rvd->vdev_children; c++) {
1635                 vdev_t *tvd = rvd->vdev_child[c];
1636                 vdev_t *mtvd  = mrvd->vdev_child[c];
1637 
1638                 /*
1639                  * Resolve any "missing" vdevs in the current configuration.
1640                  * If we find that the MOS config has more accurate information
1641                  * about the top-level vdev then use that vdev instead.
1642                  */
1643                 if (tvd->vdev_ops == &vdev_missing_ops &&
1644                     mtvd->vdev_ops != &vdev_missing_ops) {
1645 
1646                         if (!(spa->spa_import_flags & ZFS_IMPORT_MISSING_LOG))
1647                                 continue;
1648 
1649                         /*
1650                          * Device specific actions.
1651                          */
1652                         if (mtvd->vdev_islog) {
1653                                 spa_set_log_state(spa, SPA_LOG_CLEAR);
1654                         } else {
1655                                 /*
1656                                  * XXX - once we have 'readonly' pool
1657                                  * support we should be able to handle
1658                                  * missing data devices by transitioning
1659                                  * the pool to readonly.
1660                                  */
1661                                 continue;
1662                         }
1663 
1664                         /*
1665                          * Swap the missing vdev with the data we were
1666                          * able to obtain from the MOS config.
1667                          */
1668                         vdev_remove_child(rvd, tvd);
1669                         vdev_remove_child(mrvd, mtvd);
1670 
1671                         vdev_add_child(rvd, mtvd);
1672                         vdev_add_child(mrvd, tvd);
1673 
1674                         spa_config_exit(spa, SCL_ALL, FTAG);
1675                         vdev_load(mtvd);
1676                         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
1677 
1678                         vdev_reopen(rvd);
1679                 } else if (mtvd->vdev_islog) {
1680                         /*
1681                          * Load the slog device's state from the MOS config
1682                          * since it's possible that the label does not
1683                          * contain the most up-to-date information.
1684                          */
1685                         vdev_load_log_state(tvd, mtvd);
1686                         vdev_reopen(tvd);
1687                 }
1688         }
1689         vdev_free(mrvd);
1690         spa_config_exit(spa, SCL_ALL, FTAG);
1691 
1692         /*
1693          * Ensure we were able to validate the config.
1694          */
1695         return (rvd->vdev_guid_sum == spa->spa_uberblock.ub_guid_sum);
1696 }
1697 
1698 /*
1699  * Check for missing log devices
1700  */
1701 static boolean_t
1702 spa_check_logs(spa_t *spa)
1703 {
1704         boolean_t rv = B_FALSE;
1705 
1706         switch (spa->spa_log_state) {
1707         case SPA_LOG_MISSING:
1708                 /* need to recheck in case slog has been restored */
1709         case SPA_LOG_UNKNOWN:
1710                 rv = (dmu_objset_find(spa->spa_name, zil_check_log_chain,
1711                     NULL, DS_FIND_CHILDREN) != 0);
1712                 if (rv)
1713                         spa_set_log_state(spa, SPA_LOG_MISSING);
1714                 break;
1715         }
1716         return (rv);
1717 }
1718 
1719 static boolean_t
1720 spa_passivate_log(spa_t *spa)
1721 {
1722         vdev_t *rvd = spa->spa_root_vdev;
1723         boolean_t slog_found = B_FALSE;
1724 
1725         ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1726 
1727         if (!spa_has_slogs(spa))
1728                 return (B_FALSE);
1729 
1730         for (int c = 0; c < rvd->vdev_children; c++) {
1731                 vdev_t *tvd = rvd->vdev_child[c];
1732                 metaslab_group_t *mg = tvd->vdev_mg;
1733 
1734                 if (tvd->vdev_islog) {
1735                         metaslab_group_passivate(mg);
1736                         slog_found = B_TRUE;
1737                 }
1738         }
1739 
1740         return (slog_found);
1741 }
1742 
1743 static void
1744 spa_activate_log(spa_t *spa)
1745 {
1746         vdev_t *rvd = spa->spa_root_vdev;
1747 
1748         ASSERT(spa_config_held(spa, SCL_ALLOC, RW_WRITER));
1749 
1750         for (int c = 0; c < rvd->vdev_children; c++) {
1751                 vdev_t *tvd = rvd->vdev_child[c];
1752                 metaslab_group_t *mg = tvd->vdev_mg;
1753 
1754                 if (tvd->vdev_islog)
1755                         metaslab_group_activate(mg);
1756         }
1757 }
1758 
1759 int
1760 spa_offline_log(spa_t *spa)
1761 {
1762         int error;
1763 
1764         error = dmu_objset_find(spa_name(spa), zil_vdev_offline,
1765             NULL, DS_FIND_CHILDREN);
1766         if (error == 0) {
1767                 /*
1768                  * We successfully offlined the log device, sync out the
1769                  * current txg so that the "stubby" block can be removed
1770                  * by zil_sync().
1771                  */
1772                 txg_wait_synced(spa->spa_dsl_pool, 0);
1773         }
1774         return (error);
1775 }
1776 
1777 static void
1778 spa_aux_check_removed(spa_aux_vdev_t *sav)
1779 {
1780         for (int i = 0; i < sav->sav_count; i++)
1781                 spa_check_removed(sav->sav_vdevs[i]);
1782 }
1783 
1784 void
1785 spa_claim_notify(zio_t *zio)
1786 {
1787         spa_t *spa = zio->io_spa;
1788 
1789         if (zio->io_error)
1790                 return;
1791 
1792         mutex_enter(&spa->spa_props_lock);       /* any mutex will do */
1793         if (spa->spa_claim_max_txg < zio->io_bp->blk_birth)
1794                 spa->spa_claim_max_txg = zio->io_bp->blk_birth;
1795         mutex_exit(&spa->spa_props_lock);
1796 }
1797 
1798 typedef struct spa_load_error {
1799         uint64_t        sle_meta_count;
1800         uint64_t        sle_data_count;
1801 } spa_load_error_t;
1802 
1803 static void
1804 spa_load_verify_done(zio_t *zio)
1805 {
1806         blkptr_t *bp = zio->io_bp;
1807         spa_load_error_t *sle = zio->io_private;
1808         dmu_object_type_t type = BP_GET_TYPE(bp);
1809         int error = zio->io_error;
1810 
1811         if (error) {
1812                 if ((BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type)) &&
1813                     type != DMU_OT_INTENT_LOG)
1814                         atomic_add_64(&sle->sle_meta_count, 1);
1815                 else
1816                         atomic_add_64(&sle->sle_data_count, 1);
1817         }
1818         zio_data_buf_free(zio->io_data, zio->io_size);
1819 }
1820 
1821 /*ARGSUSED*/
1822 static int
1823 spa_load_verify_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
1824     const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1825 {
1826         if (bp != NULL) {
1827                 zio_t *rio = arg;
1828                 size_t size = BP_GET_PSIZE(bp);
1829                 void *data = zio_data_buf_alloc(size);
1830 
1831                 zio_nowait(zio_read(rio, spa, bp, data, size,
1832                     spa_load_verify_done, rio->io_private, ZIO_PRIORITY_SCRUB,
1833                     ZIO_FLAG_SPECULATIVE | ZIO_FLAG_CANFAIL |
1834                     ZIO_FLAG_SCRUB | ZIO_FLAG_RAW, zb));
1835         }
1836         return (0);
1837 }
1838 
1839 static int
1840 spa_load_verify(spa_t *spa)
1841 {
1842         zio_t *rio;
1843         spa_load_error_t sle = { 0 };
1844         zpool_rewind_policy_t policy;
1845         boolean_t verify_ok = B_FALSE;
1846         int error;
1847 
1848         zpool_get_rewind_policy(spa->spa_config, &policy);
1849 
1850         if (policy.zrp_request & ZPOOL_NEVER_REWIND)
1851                 return (0);
1852 
1853         rio = zio_root(spa, NULL, &sle,
1854             ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE);
1855 
1856         error = traverse_pool(spa, spa->spa_verify_min_txg,
1857             TRAVERSE_PRE | TRAVERSE_PREFETCH, spa_load_verify_cb, rio);
1858 
1859         (void) zio_wait(rio);
1860 
1861         spa->spa_load_meta_errors = sle.sle_meta_count;
1862         spa->spa_load_data_errors = sle.sle_data_count;
1863 
1864         if (!error && sle.sle_meta_count <= policy.zrp_maxmeta &&
1865             sle.sle_data_count <= policy.zrp_maxdata) {
1866                 int64_t loss = 0;
1867 
1868                 verify_ok = B_TRUE;
1869                 spa->spa_load_txg = spa->spa_uberblock.ub_txg;
1870                 spa->spa_load_txg_ts = spa->spa_uberblock.ub_timestamp;
1871 
1872                 loss = spa->spa_last_ubsync_txg_ts - spa->spa_load_txg_ts;
1873                 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1874                     ZPOOL_CONFIG_LOAD_TIME, spa->spa_load_txg_ts) == 0);
1875                 VERIFY(nvlist_add_int64(spa->spa_load_info,
1876                     ZPOOL_CONFIG_REWIND_TIME, loss) == 0);
1877                 VERIFY(nvlist_add_uint64(spa->spa_load_info,
1878                     ZPOOL_CONFIG_LOAD_DATA_ERRORS, sle.sle_data_count) == 0);
1879         } else {
1880                 spa->spa_load_max_txg = spa->spa_uberblock.ub_txg;
1881         }
1882 
1883         if (error) {
1884                 if (error != ENXIO && error != EIO)
1885                         error = SET_ERROR(EIO);
1886                 return (error);
1887         }
1888 
1889         return (verify_ok ? 0 : EIO);
1890 }
1891 
1892 /*
1893  * Find a value in the pool props object.
1894  */
1895 static void
1896 spa_prop_find(spa_t *spa, zpool_prop_t prop, uint64_t *val)
1897 {
1898         (void) zap_lookup(spa->spa_meta_objset, spa->spa_pool_props_object,
1899             zpool_prop_to_name(prop), sizeof (uint64_t), 1, val);
1900 }
1901 
1902 /*
1903  * Find a value in the pool directory object.
1904  */
1905 static int
1906 spa_dir_prop(spa_t *spa, const char *name, uint64_t *val)
1907 {
1908         return (zap_lookup(spa->spa_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
1909             name, sizeof (uint64_t), 1, val));
1910 }
1911 
1912 static int
1913 spa_vdev_err(vdev_t *vdev, vdev_aux_t aux, int err)
1914 {
1915         vdev_set_state(vdev, B_TRUE, VDEV_STATE_CANT_OPEN, aux);
1916         return (err);
1917 }
1918 
1919 /*
1920  * Fix up config after a partly-completed split.  This is done with the
1921  * ZPOOL_CONFIG_SPLIT nvlist.  Both the splitting pool and the split-off
1922  * pool have that entry in their config, but only the splitting one contains
1923  * a list of all the guids of the vdevs that are being split off.
1924  *
1925  * This function determines what to do with that list: either rejoin
1926  * all the disks to the pool, or complete the splitting process.  To attempt
1927  * the rejoin, each disk that is offlined is marked online again, and
1928  * we do a reopen() call.  If the vdev label for every disk that was
1929  * marked online indicates it was successfully split off (VDEV_AUX_SPLIT_POOL)
1930  * then we call vdev_split() on each disk, and complete the split.
1931  *
1932  * Otherwise we leave the config alone, with all the vdevs in place in
1933  * the original pool.
1934  */
1935 static void
1936 spa_try_repair(spa_t *spa, nvlist_t *config)
1937 {
1938         uint_t extracted;
1939         uint64_t *glist;
1940         uint_t i, gcount;
1941         nvlist_t *nvl;
1942         vdev_t **vd;
1943         boolean_t attempt_reopen;
1944 
1945         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT, &nvl) != 0)
1946                 return;
1947 
1948         /* check that the config is complete */
1949         if (nvlist_lookup_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
1950             &glist, &gcount) != 0)
1951                 return;
1952 
1953         vd = kmem_zalloc(gcount * sizeof (vdev_t *), KM_SLEEP);
1954 
1955         /* attempt to online all the vdevs & validate */
1956         attempt_reopen = B_TRUE;
1957         for (i = 0; i < gcount; i++) {
1958                 if (glist[i] == 0)      /* vdev is hole */
1959                         continue;
1960 
1961                 vd[i] = spa_lookup_by_guid(spa, glist[i], B_FALSE);
1962                 if (vd[i] == NULL) {
1963                         /*
1964                          * Don't bother attempting to reopen the disks;
1965                          * just do the split.
1966                          */
1967                         attempt_reopen = B_FALSE;
1968                 } else {
1969                         /* attempt to re-online it */
1970                         vd[i]->vdev_offline = B_FALSE;
1971                 }
1972         }
1973 
1974         if (attempt_reopen) {
1975                 vdev_reopen(spa->spa_root_vdev);
1976 
1977                 /* check each device to see what state it's in */
1978                 for (extracted = 0, i = 0; i < gcount; i++) {
1979                         if (vd[i] != NULL &&
1980                             vd[i]->vdev_stat.vs_aux != VDEV_AUX_SPLIT_POOL)
1981                                 break;
1982                         ++extracted;
1983                 }
1984         }
1985 
1986         /*
1987          * If every disk has been moved to the new pool, or if we never
1988          * even attempted to look at them, then we split them off for
1989          * good.
1990          */
1991         if (!attempt_reopen || gcount == extracted) {
1992                 for (i = 0; i < gcount; i++)
1993                         if (vd[i] != NULL)
1994                                 vdev_split(vd[i]);
1995                 vdev_reopen(spa->spa_root_vdev);
1996         }
1997 
1998         kmem_free(vd, gcount * sizeof (vdev_t *));
1999 }
2000 
2001 static int
2002 spa_load(spa_t *spa, spa_load_state_t state, spa_import_type_t type,
2003     boolean_t mosconfig)
2004 {
2005         nvlist_t *config = spa->spa_config;
2006         char *ereport = FM_EREPORT_ZFS_POOL;
2007         char *comment;
2008         int error;
2009         uint64_t pool_guid;
2010         nvlist_t *nvl;
2011 
2012         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid))
2013                 return (SET_ERROR(EINVAL));
2014 
2015         ASSERT(spa->spa_comment == NULL);
2016         if (nvlist_lookup_string(config, ZPOOL_CONFIG_COMMENT, &comment) == 0)
2017                 spa->spa_comment = spa_strdup(comment);
2018 
2019         /*
2020          * Versioning wasn't explicitly added to the label until later, so if
2021          * it's not present treat it as the initial version.
2022          */
2023         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
2024             &spa->spa_ubsync.ub_version) != 0)
2025                 spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
2026 
2027         (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
2028             &spa->spa_config_txg);
2029 
2030         if ((state == SPA_LOAD_IMPORT || state == SPA_LOAD_TRYIMPORT) &&
2031             spa_guid_exists(pool_guid, 0)) {
2032                 error = SET_ERROR(EEXIST);
2033         } else {
2034                 spa->spa_config_guid = pool_guid;
2035 
2036                 if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_SPLIT,
2037                     &nvl) == 0) {
2038                         VERIFY(nvlist_dup(nvl, &spa->spa_config_splitting,
2039                             KM_SLEEP) == 0);
2040                 }
2041 
2042                 nvlist_free(spa->spa_load_info);
2043                 spa->spa_load_info = fnvlist_alloc();
2044 
2045                 gethrestime(&spa->spa_loaded_ts);
2046                 error = spa_load_impl(spa, pool_guid, config, state, type,
2047                     mosconfig, &ereport);
2048         }
2049 
2050         spa->spa_minref = refcount_count(&spa->spa_refcount);
2051         if (error) {
2052                 if (error != EEXIST) {
2053                         spa->spa_loaded_ts.tv_sec = 0;
2054                         spa->spa_loaded_ts.tv_nsec = 0;
2055                 }
2056                 if (error != EBADF) {
2057                         zfs_ereport_post(ereport, spa, NULL, NULL, 0, 0);
2058                 }
2059         }
2060         spa->spa_load_state = error ? SPA_LOAD_ERROR : SPA_LOAD_NONE;
2061         spa->spa_ena = 0;
2062 
2063         return (error);
2064 }
2065 
2066 /*
2067  * Load an existing storage pool, using the pool's builtin spa_config as a
2068  * source of configuration information.
2069  */
2070 static int
2071 spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
2072     spa_load_state_t state, spa_import_type_t type, boolean_t mosconfig,
2073     char **ereport)
2074 {
2075         int error = 0;
2076         nvlist_t *nvroot = NULL;
2077         nvlist_t *label;
2078         vdev_t *rvd;
2079         uberblock_t *ub = &spa->spa_uberblock;
2080         uint64_t children, config_cache_txg = spa->spa_config_txg;
2081         int orig_mode = spa->spa_mode;
2082         int parse;
2083         uint64_t obj;
2084         boolean_t missing_feat_write = B_FALSE;
2085 
2086         /*
2087          * If this is an untrusted config, access the pool in read-only mode.
2088          * This prevents things like resilvering recently removed devices.
2089          */
2090         if (!mosconfig)
2091                 spa->spa_mode = FREAD;
2092 
2093         ASSERT(MUTEX_HELD(&spa_namespace_lock));
2094 
2095         spa->spa_load_state = state;
2096 
2097         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvroot))
2098                 return (SET_ERROR(EINVAL));
2099 
2100         parse = (type == SPA_IMPORT_EXISTING ?
2101             VDEV_ALLOC_LOAD : VDEV_ALLOC_SPLIT);
2102 
2103         /*
2104          * Create "The Godfather" zio to hold all async IOs
2105          */
2106         spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
2107             ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
2108 
2109         /*
2110          * Parse the configuration into a vdev tree.  We explicitly set the
2111          * value that will be returned by spa_version() since parsing the
2112          * configuration requires knowing the version number.
2113          */
2114         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2115         error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, parse);
2116         spa_config_exit(spa, SCL_ALL, FTAG);
2117 
2118         if (error != 0)
2119                 return (error);
2120 
2121         ASSERT(spa->spa_root_vdev == rvd);
2122 
2123         if (type != SPA_IMPORT_ASSEMBLE) {
2124                 ASSERT(spa_guid(spa) == pool_guid);
2125         }
2126 
2127         /*
2128          * Try to open all vdevs, loading each label in the process.
2129          */
2130         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2131         error = vdev_open(rvd);
2132         spa_config_exit(spa, SCL_ALL, FTAG);
2133         if (error != 0)
2134                 return (error);
2135 
2136         /*
2137          * We need to validate the vdev labels against the configuration that
2138          * we have in hand, which is dependent on the setting of mosconfig. If
2139          * mosconfig is true then we're validating the vdev labels based on
2140          * that config.  Otherwise, we're validating against the cached config
2141          * (zpool.cache) that was read when we loaded the zfs module, and then
2142          * later we will recursively call spa_load() and validate against
2143          * the vdev config.
2144          *
2145          * If we're assembling a new pool that's been split off from an
2146          * existing pool, the labels haven't yet been updated so we skip
2147          * validation for now.
2148          */
2149         if (type != SPA_IMPORT_ASSEMBLE) {
2150                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2151                 error = vdev_validate(rvd, mosconfig);
2152                 spa_config_exit(spa, SCL_ALL, FTAG);
2153 
2154                 if (error != 0)
2155                         return (error);
2156 
2157                 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2158                         return (SET_ERROR(ENXIO));
2159         }
2160 
2161         /*
2162          * Find the best uberblock.
2163          */
2164         vdev_uberblock_load(rvd, ub, &label);
2165 
2166         /*
2167          * If we weren't able to find a single valid uberblock, return failure.
2168          */
2169         if (ub->ub_txg == 0) {
2170                 nvlist_free(label);
2171                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
2172         }
2173 
2174         /*
2175          * If the pool has an unsupported version we can't open it.
2176          */
2177         if (!SPA_VERSION_IS_SUPPORTED(ub->ub_version)) {
2178                 nvlist_free(label);
2179                 return (spa_vdev_err(rvd, VDEV_AUX_VERSION_NEWER, ENOTSUP));
2180         }
2181 
2182         if (ub->ub_version >= SPA_VERSION_FEATURES) {
2183                 nvlist_t *features;
2184 
2185                 /*
2186                  * If we weren't able to find what's necessary for reading the
2187                  * MOS in the label, return failure.
2188                  */
2189                 if (label == NULL || nvlist_lookup_nvlist(label,
2190                     ZPOOL_CONFIG_FEATURES_FOR_READ, &features) != 0) {
2191                         nvlist_free(label);
2192                         return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2193                             ENXIO));
2194                 }
2195 
2196                 /*
2197                  * Update our in-core representation with the definitive values
2198                  * from the label.
2199                  */
2200                 nvlist_free(spa->spa_label_features);
2201                 VERIFY(nvlist_dup(features, &spa->spa_label_features, 0) == 0);
2202         }
2203 
2204         nvlist_free(label);
2205 
2206         /*
2207          * Look through entries in the label nvlist's features_for_read. If
2208          * there is a feature listed there which we don't understand then we
2209          * cannot open a pool.
2210          */
2211         if (ub->ub_version >= SPA_VERSION_FEATURES) {
2212                 nvlist_t *unsup_feat;
2213 
2214                 VERIFY(nvlist_alloc(&unsup_feat, NV_UNIQUE_NAME, KM_SLEEP) ==
2215                     0);
2216 
2217                 for (nvpair_t *nvp = nvlist_next_nvpair(spa->spa_label_features,
2218                     NULL); nvp != NULL;
2219                     nvp = nvlist_next_nvpair(spa->spa_label_features, nvp)) {
2220                         if (!zfeature_is_supported(nvpair_name(nvp))) {
2221                                 VERIFY(nvlist_add_string(unsup_feat,
2222                                     nvpair_name(nvp), "") == 0);
2223                         }
2224                 }
2225 
2226                 if (!nvlist_empty(unsup_feat)) {
2227                         VERIFY(nvlist_add_nvlist(spa->spa_load_info,
2228                             ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat) == 0);
2229                         nvlist_free(unsup_feat);
2230                         return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2231                             ENOTSUP));
2232                 }
2233 
2234                 nvlist_free(unsup_feat);
2235         }
2236 
2237         /*
2238          * If the vdev guid sum doesn't match the uberblock, we have an
2239          * incomplete configuration.  We first check to see if the pool
2240          * is aware of the complete config (i.e ZPOOL_CONFIG_VDEV_CHILDREN).
2241          * If it is, defer the vdev_guid_sum check till later so we
2242          * can handle missing vdevs.
2243          */
2244         if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
2245             &children) != 0 && mosconfig && type != SPA_IMPORT_ASSEMBLE &&
2246             rvd->vdev_guid_sum != ub->ub_guid_sum)
2247                 return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM, ENXIO));
2248 
2249         if (type != SPA_IMPORT_ASSEMBLE && spa->spa_config_splitting) {
2250                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2251                 spa_try_repair(spa, config);
2252                 spa_config_exit(spa, SCL_ALL, FTAG);
2253                 nvlist_free(spa->spa_config_splitting);
2254                 spa->spa_config_splitting = NULL;
2255         }
2256 
2257         /*
2258          * Initialize internal SPA structures.
2259          */
2260         spa->spa_state = POOL_STATE_ACTIVE;
2261         spa->spa_ubsync = spa->spa_uberblock;
2262         spa->spa_verify_min_txg = spa->spa_extreme_rewind ?
2263             TXG_INITIAL - 1 : spa_last_synced_txg(spa) - TXG_DEFER_SIZE - 1;
2264         spa->spa_first_txg = spa->spa_last_ubsync_txg ?
2265             spa->spa_last_ubsync_txg : spa_last_synced_txg(spa) + 1;
2266         spa->spa_claim_max_txg = spa->spa_first_txg;
2267         spa->spa_prev_software_version = ub->ub_software_version;
2268 
2269         error = dsl_pool_init(spa, spa->spa_first_txg, &spa->spa_dsl_pool);
2270         if (error)
2271                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2272         spa->spa_meta_objset = spa->spa_dsl_pool->dp_meta_objset;
2273 
2274         if (spa_dir_prop(spa, DMU_POOL_CONFIG, &spa->spa_config_object) != 0)
2275                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2276 
2277         if (spa_version(spa) >= SPA_VERSION_FEATURES) {
2278                 boolean_t missing_feat_read = B_FALSE;
2279                 nvlist_t *unsup_feat, *enabled_feat;
2280 
2281                 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_READ,
2282                     &spa->spa_feat_for_read_obj) != 0) {
2283                         return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2284                 }
2285 
2286                 if (spa_dir_prop(spa, DMU_POOL_FEATURES_FOR_WRITE,
2287                     &spa->spa_feat_for_write_obj) != 0) {
2288                         return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2289                 }
2290 
2291                 if (spa_dir_prop(spa, DMU_POOL_FEATURE_DESCRIPTIONS,
2292                     &spa->spa_feat_desc_obj) != 0) {
2293                         return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2294                 }
2295 
2296                 enabled_feat = fnvlist_alloc();
2297                 unsup_feat = fnvlist_alloc();
2298 
2299                 if (!feature_is_supported(spa->spa_meta_objset,
2300                     spa->spa_feat_for_read_obj, spa->spa_feat_desc_obj,
2301                     unsup_feat, enabled_feat))
2302                         missing_feat_read = B_TRUE;
2303 
2304                 if (spa_writeable(spa) || state == SPA_LOAD_TRYIMPORT) {
2305                         if (!feature_is_supported(spa->spa_meta_objset,
2306                             spa->spa_feat_for_write_obj, spa->spa_feat_desc_obj,
2307                             unsup_feat, enabled_feat)) {
2308                                 missing_feat_write = B_TRUE;
2309                         }
2310                 }
2311 
2312                 fnvlist_add_nvlist(spa->spa_load_info,
2313                     ZPOOL_CONFIG_ENABLED_FEAT, enabled_feat);
2314 
2315                 if (!nvlist_empty(unsup_feat)) {
2316                         fnvlist_add_nvlist(spa->spa_load_info,
2317                             ZPOOL_CONFIG_UNSUP_FEAT, unsup_feat);
2318                 }
2319 
2320                 fnvlist_free(enabled_feat);
2321                 fnvlist_free(unsup_feat);
2322 
2323                 if (!missing_feat_read) {
2324                         fnvlist_add_boolean(spa->spa_load_info,
2325                             ZPOOL_CONFIG_CAN_RDONLY);
2326                 }
2327 
2328                 /*
2329                  * If the state is SPA_LOAD_TRYIMPORT, our objective is
2330                  * twofold: to determine whether the pool is available for
2331                  * import in read-write mode and (if it is not) whether the
2332                  * pool is available for import in read-only mode. If the pool
2333                  * is available for import in read-write mode, it is displayed
2334                  * as available in userland; if it is not available for import
2335                  * in read-only mode, it is displayed as unavailable in
2336                  * userland. If the pool is available for import in read-only
2337                  * mode but not read-write mode, it is displayed as unavailable
2338                  * in userland with a special note that the pool is actually
2339                  * available for open in read-only mode.
2340                  *
2341                  * As a result, if the state is SPA_LOAD_TRYIMPORT and we are
2342                  * missing a feature for write, we must first determine whether
2343                  * the pool can be opened read-only before returning to
2344                  * userland in order to know whether to display the
2345                  * abovementioned note.
2346                  */
2347                 if (missing_feat_read || (missing_feat_write &&
2348                     spa_writeable(spa))) {
2349                         return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT,
2350                             ENOTSUP));
2351                 }
2352         }
2353 
2354         spa->spa_is_initializing = B_TRUE;
2355         error = dsl_pool_open(spa->spa_dsl_pool);
2356         spa->spa_is_initializing = B_FALSE;
2357         if (error != 0)
2358                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2359 
2360         if (!mosconfig) {
2361                 uint64_t hostid;
2362                 nvlist_t *policy = NULL, *nvconfig;
2363 
2364                 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2365                         return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2366 
2367                 if (!spa_is_root(spa) && nvlist_lookup_uint64(nvconfig,
2368                     ZPOOL_CONFIG_HOSTID, &hostid) == 0) {
2369                         char *hostname;
2370                         unsigned long myhostid = 0;
2371 
2372                         VERIFY(nvlist_lookup_string(nvconfig,
2373                             ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
2374 
2375 #ifdef  _KERNEL
2376                         myhostid = zone_get_hostid(NULL);
2377 #else   /* _KERNEL */
2378                         /*
2379                          * We're emulating the system's hostid in userland, so
2380                          * we can't use zone_get_hostid().
2381                          */
2382                         (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
2383 #endif  /* _KERNEL */
2384                         if (hostid != 0 && myhostid != 0 &&
2385                             hostid != myhostid) {
2386                                 nvlist_free(nvconfig);
2387                                 cmn_err(CE_WARN, "pool '%s' could not be "
2388                                     "loaded as it was last accessed by "
2389                                     "another system (host: %s hostid: 0x%lx). "
2390                                     "See: http://illumos.org/msg/ZFS-8000-EY",
2391                                     spa_name(spa), hostname,
2392                                     (unsigned long)hostid);
2393                                 return (SET_ERROR(EBADF));
2394                         }
2395                 }
2396                 if (nvlist_lookup_nvlist(spa->spa_config,
2397                     ZPOOL_REWIND_POLICY, &policy) == 0)
2398                         VERIFY(nvlist_add_nvlist(nvconfig,
2399                             ZPOOL_REWIND_POLICY, policy) == 0);
2400 
2401                 spa_config_set(spa, nvconfig);
2402                 spa_unload(spa);
2403                 spa_deactivate(spa);
2404                 spa_activate(spa, orig_mode);
2405 
2406                 return (spa_load(spa, state, SPA_IMPORT_EXISTING, B_TRUE));
2407         }
2408 
2409         if (spa_dir_prop(spa, DMU_POOL_SYNC_BPOBJ, &obj) != 0)
2410                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2411         error = bpobj_open(&spa->spa_deferred_bpobj, spa->spa_meta_objset, obj);
2412         if (error != 0)
2413                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2414 
2415         /*
2416          * Load the bit that tells us to use the new accounting function
2417          * (raid-z deflation).  If we have an older pool, this will not
2418          * be present.
2419          */
2420         error = spa_dir_prop(spa, DMU_POOL_DEFLATE, &spa->spa_deflate);
2421         if (error != 0 && error != ENOENT)
2422                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2423 
2424         error = spa_dir_prop(spa, DMU_POOL_CREATION_VERSION,
2425             &spa->spa_creation_version);
2426         if (error != 0 && error != ENOENT)
2427                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2428 
2429         /*
2430          * Load the persistent error log.  If we have an older pool, this will
2431          * not be present.
2432          */
2433         error = spa_dir_prop(spa, DMU_POOL_ERRLOG_LAST, &spa->spa_errlog_last);
2434         if (error != 0 && error != ENOENT)
2435                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2436 
2437         error = spa_dir_prop(spa, DMU_POOL_ERRLOG_SCRUB,
2438             &spa->spa_errlog_scrub);
2439         if (error != 0 && error != ENOENT)
2440                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2441 
2442         /*
2443          * Load the history object.  If we have an older pool, this
2444          * will not be present.
2445          */
2446         error = spa_dir_prop(spa, DMU_POOL_HISTORY, &spa->spa_history);
2447         if (error != 0 && error != ENOENT)
2448                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2449 
2450         /*
2451          * If we're assembling the pool from the split-off vdevs of
2452          * an existing pool, we don't want to attach the spares & cache
2453          * devices.
2454          */
2455 
2456         /*
2457          * Load any hot spares for this pool.
2458          */
2459         error = spa_dir_prop(spa, DMU_POOL_SPARES, &spa->spa_spares.sav_object);
2460         if (error != 0 && error != ENOENT)
2461                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2462         if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2463                 ASSERT(spa_version(spa) >= SPA_VERSION_SPARES);
2464                 if (load_nvlist(spa, spa->spa_spares.sav_object,
2465                     &spa->spa_spares.sav_config) != 0)
2466                         return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2467 
2468                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2469                 spa_load_spares(spa);
2470                 spa_config_exit(spa, SCL_ALL, FTAG);
2471         } else if (error == 0) {
2472                 spa->spa_spares.sav_sync = B_TRUE;
2473         }
2474 
2475         /*
2476          * Load any level 2 ARC devices for this pool.
2477          */
2478         error = spa_dir_prop(spa, DMU_POOL_L2CACHE,
2479             &spa->spa_l2cache.sav_object);
2480         if (error != 0 && error != ENOENT)
2481                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2482         if (error == 0 && type != SPA_IMPORT_ASSEMBLE) {
2483                 ASSERT(spa_version(spa) >= SPA_VERSION_L2CACHE);
2484                 if (load_nvlist(spa, spa->spa_l2cache.sav_object,
2485                     &spa->spa_l2cache.sav_config) != 0)
2486                         return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2487 
2488                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2489                 spa_load_l2cache(spa);
2490                 spa_config_exit(spa, SCL_ALL, FTAG);
2491         } else if (error == 0) {
2492                 spa->spa_l2cache.sav_sync = B_TRUE;
2493         }
2494 
2495         spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
2496 
2497         error = spa_dir_prop(spa, DMU_POOL_PROPS, &spa->spa_pool_props_object);
2498         if (error && error != ENOENT)
2499                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2500 
2501         if (error == 0) {
2502                 uint64_t autoreplace;
2503 
2504                 spa_prop_find(spa, ZPOOL_PROP_BOOTFS, &spa->spa_bootfs);
2505                 spa_prop_find(spa, ZPOOL_PROP_AUTOREPLACE, &autoreplace);
2506                 spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
2507                 spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
2508                 spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
2509                 spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
2510                     &spa->spa_dedup_ditto);
2511 
2512                 spa->spa_autoreplace = (autoreplace != 0);
2513         }
2514 
2515         /*
2516          * If the 'autoreplace' property is set, then post a resource notifying
2517          * the ZFS DE that it should not issue any faults for unopenable
2518          * devices.  We also iterate over the vdevs, and post a sysevent for any
2519          * unopenable vdevs so that the normal autoreplace handler can take
2520          * over.
2521          */
2522         if (spa->spa_autoreplace && state != SPA_LOAD_TRYIMPORT) {
2523                 spa_check_removed(spa->spa_root_vdev);
2524                 /*
2525                  * For the import case, this is done in spa_import(), because
2526                  * at this point we're using the spare definitions from
2527                  * the MOS config, not necessarily from the userland config.
2528                  */
2529                 if (state != SPA_LOAD_IMPORT) {
2530                         spa_aux_check_removed(&spa->spa_spares);
2531                         spa_aux_check_removed(&spa->spa_l2cache);
2532                 }
2533         }
2534 
2535         /*
2536          * Load the vdev state for all toplevel vdevs.
2537          */
2538         vdev_load(rvd);
2539 
2540         /*
2541          * Propagate the leaf DTLs we just loaded all the way up the tree.
2542          */
2543         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
2544         vdev_dtl_reassess(rvd, 0, 0, B_FALSE);
2545         spa_config_exit(spa, SCL_ALL, FTAG);
2546 
2547         /*
2548          * Load the DDTs (dedup tables).
2549          */
2550         error = ddt_load(spa);
2551         if (error != 0)
2552                 return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2553 
2554         spa_update_dspace(spa);
2555 
2556         /*
2557          * Validate the config, using the MOS config to fill in any
2558          * information which might be missing.  If we fail to validate
2559          * the config then declare the pool unfit for use. If we're
2560          * assembling a pool from a split, the log is not transferred
2561          * over.
2562          */
2563         if (type != SPA_IMPORT_ASSEMBLE) {
2564                 nvlist_t *nvconfig;
2565 
2566                 if (load_nvlist(spa, spa->spa_config_object, &nvconfig) != 0)
2567                         return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, EIO));
2568 
2569                 if (!spa_config_valid(spa, nvconfig)) {
2570                         nvlist_free(nvconfig);
2571                         return (spa_vdev_err(rvd, VDEV_AUX_BAD_GUID_SUM,
2572                             ENXIO));
2573                 }
2574                 nvlist_free(nvconfig);
2575 
2576                 /*
2577                  * Now that we've validated the config, check the state of the
2578                  * root vdev.  If it can't be opened, it indicates one or
2579                  * more toplevel vdevs are faulted.
2580                  */
2581                 if (rvd->vdev_state <= VDEV_STATE_CANT_OPEN)
2582                         return (SET_ERROR(ENXIO));
2583 
2584                 if (spa_check_logs(spa)) {
2585                         *ereport = FM_EREPORT_ZFS_LOG_REPLAY;
2586                         return (spa_vdev_err(rvd, VDEV_AUX_BAD_LOG, ENXIO));
2587                 }
2588         }
2589 
2590         if (missing_feat_write) {
2591                 ASSERT(state == SPA_LOAD_TRYIMPORT);
2592 
2593                 /*
2594                  * At this point, we know that we can open the pool in
2595                  * read-only mode but not read-write mode. We now have enough
2596                  * information and can return to userland.
2597                  */
2598                 return (spa_vdev_err(rvd, VDEV_AUX_UNSUP_FEAT, ENOTSUP));
2599         }
2600 
2601         /*
2602          * We've successfully opened the pool, verify that we're ready
2603          * to start pushing transactions.
2604          */
2605         if (state != SPA_LOAD_TRYIMPORT) {
2606                 if (error = spa_load_verify(spa))
2607                         return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA,
2608                             error));
2609         }
2610 
2611         if (spa_writeable(spa) && (state == SPA_LOAD_RECOVER ||
2612             spa->spa_load_max_txg == UINT64_MAX)) {
2613                 dmu_tx_t *tx;
2614                 int need_update = B_FALSE;
2615 
2616                 ASSERT(state != SPA_LOAD_TRYIMPORT);
2617 
2618                 /*
2619                  * Claim log blocks that haven't been committed yet.
2620                  * This must all happen in a single txg.
2621                  * Note: spa_claim_max_txg is updated by spa_claim_notify(),
2622                  * invoked from zil_claim_log_block()'s i/o done callback.
2623                  * Price of rollback is that we abandon the log.
2624                  */
2625                 spa->spa_claiming = B_TRUE;
2626 
2627                 tx = dmu_tx_create_assigned(spa_get_dsl(spa),
2628                     spa_first_txg(spa));
2629                 (void) dmu_objset_find(spa_name(spa),
2630                     zil_claim, tx, DS_FIND_CHILDREN);
2631                 dmu_tx_commit(tx);
2632 
2633                 spa->spa_claiming = B_FALSE;
2634 
2635                 spa_set_log_state(spa, SPA_LOG_GOOD);
2636                 spa->spa_sync_on = B_TRUE;
2637                 txg_sync_start(spa->spa_dsl_pool);
2638 
2639                 /*
2640                  * Wait for all claims to sync.  We sync up to the highest
2641                  * claimed log block birth time so that claimed log blocks
2642                  * don't appear to be from the future.  spa_claim_max_txg
2643                  * will have been set for us by either zil_check_log_chain()
2644                  * (invoked from spa_check_logs()) or zil_claim() above.
2645                  */
2646                 txg_wait_synced(spa->spa_dsl_pool, spa->spa_claim_max_txg);
2647 
2648                 /*
2649                  * If the config cache is stale, or we have uninitialized
2650                  * metaslabs (see spa_vdev_add()), then update the config.
2651                  *
2652                  * If this is a verbatim import, trust the current
2653                  * in-core spa_config and update the disk labels.
2654                  */
2655                 if (config_cache_txg != spa->spa_config_txg ||
2656                     state == SPA_LOAD_IMPORT ||
2657                     state == SPA_LOAD_RECOVER ||
2658                     (spa->spa_import_flags & ZFS_IMPORT_VERBATIM))
2659                         need_update = B_TRUE;
2660 
2661                 for (int c = 0; c < rvd->vdev_children; c++)
2662                         if (rvd->vdev_child[c]->vdev_ms_array == 0)
2663                                 need_update = B_TRUE;
2664 
2665                 /*
2666                  * Update the config cache asychronously in case we're the
2667                  * root pool, in which case the config cache isn't writable yet.
2668                  */
2669                 if (need_update)
2670                         spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE);
2671 
2672                 /*
2673                  * Check all DTLs to see if anything needs resilvering.
2674                  */
2675                 if (!dsl_scan_resilvering(spa->spa_dsl_pool) &&
2676                     vdev_resilver_needed(rvd, NULL, NULL))
2677                         spa_async_request(spa, SPA_ASYNC_RESILVER);
2678 
2679                 /*
2680                  * Log the fact that we booted up (so that we can detect if
2681                  * we rebooted in the middle of an operation).
2682                  */
2683                 spa_history_log_version(spa, "open");
2684 
2685                 /*
2686                  * Delete any inconsistent datasets.
2687                  */
2688                 (void) dmu_objset_find(spa_name(spa),
2689                     dsl_destroy_inconsistent, NULL, DS_FIND_CHILDREN);
2690 
2691                 /*
2692                  * Clean up any stale temporary dataset userrefs.
2693                  */
2694                 dsl_pool_clean_tmp_userrefs(spa->spa_dsl_pool);
2695         }
2696 
2697         return (0);
2698 }
2699 
2700 static int
2701 spa_load_retry(spa_t *spa, spa_load_state_t state, int mosconfig)
2702 {
2703         int mode = spa->spa_mode;
2704 
2705         spa_unload(spa);
2706         spa_deactivate(spa);
2707 
2708         spa->spa_load_max_txg--;
2709 
2710         spa_activate(spa, mode);
2711         spa_async_suspend(spa);
2712 
2713         return (spa_load(spa, state, SPA_IMPORT_EXISTING, mosconfig));
2714 }
2715 
2716 /*
2717  * If spa_load() fails this function will try loading prior txg's. If
2718  * 'state' is SPA_LOAD_RECOVER and one of these loads succeeds the pool
2719  * will be rewound to that txg. If 'state' is not SPA_LOAD_RECOVER this
2720  * function will not rewind the pool and will return the same error as
2721  * spa_load().
2722  */
2723 static int
2724 spa_load_best(spa_t *spa, spa_load_state_t state, int mosconfig,
2725     uint64_t max_request, int rewind_flags)
2726 {
2727         nvlist_t *loadinfo = NULL;
2728         nvlist_t *config = NULL;
2729         int load_error, rewind_error;
2730         uint64_t safe_rewind_txg;
2731         uint64_t min_txg;
2732 
2733         if (spa->spa_load_txg && state == SPA_LOAD_RECOVER) {
2734                 spa->spa_load_max_txg = spa->spa_load_txg;
2735                 spa_set_log_state(spa, SPA_LOG_CLEAR);
2736         } else {
2737                 spa->spa_load_max_txg = max_request;
2738         }
2739 
2740         load_error = rewind_error = spa_load(spa, state, SPA_IMPORT_EXISTING,
2741             mosconfig);
2742         if (load_error == 0)
2743                 return (0);
2744 
2745         if (spa->spa_root_vdev != NULL)
2746                 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2747 
2748         spa->spa_last_ubsync_txg = spa->spa_uberblock.ub_txg;
2749         spa->spa_last_ubsync_txg_ts = spa->spa_uberblock.ub_timestamp;
2750 
2751         if (rewind_flags & ZPOOL_NEVER_REWIND) {
2752                 nvlist_free(config);
2753                 return (load_error);
2754         }
2755 
2756         if (state == SPA_LOAD_RECOVER) {
2757                 /* Price of rolling back is discarding txgs, including log */
2758                 spa_set_log_state(spa, SPA_LOG_CLEAR);
2759         } else {
2760                 /*
2761                  * If we aren't rolling back save the load info from our first
2762                  * import attempt so that we can restore it after attempting
2763                  * to rewind.
2764                  */
2765                 loadinfo = spa->spa_load_info;
2766                 spa->spa_load_info = fnvlist_alloc();
2767         }
2768 
2769         spa->spa_load_max_txg = spa->spa_last_ubsync_txg;
2770         safe_rewind_txg = spa->spa_last_ubsync_txg - TXG_DEFER_SIZE;
2771         min_txg = (rewind_flags & ZPOOL_EXTREME_REWIND) ?
2772             TXG_INITIAL : safe_rewind_txg;
2773 
2774         /*
2775          * Continue as long as we're finding errors, we're still within
2776          * the acceptable rewind range, and we're still finding uberblocks
2777          */
2778         while (rewind_error && spa->spa_uberblock.ub_txg >= min_txg &&
2779             spa->spa_uberblock.ub_txg <= spa->spa_load_max_txg) {
2780                 if (spa->spa_load_max_txg < safe_rewind_txg)
2781                         spa->spa_extreme_rewind = B_TRUE;
2782                 rewind_error = spa_load_retry(spa, state, mosconfig);
2783         }
2784 
2785         spa->spa_extreme_rewind = B_FALSE;
2786         spa->spa_load_max_txg = UINT64_MAX;
2787 
2788         if (config && (rewind_error || state != SPA_LOAD_RECOVER))
2789                 spa_config_set(spa, config);
2790 
2791         if (state == SPA_LOAD_RECOVER) {
2792                 ASSERT3P(loadinfo, ==, NULL);
2793                 return (rewind_error);
2794         } else {
2795                 /* Store the rewind info as part of the initial load info */
2796                 fnvlist_add_nvlist(loadinfo, ZPOOL_CONFIG_REWIND_INFO,
2797                     spa->spa_load_info);
2798 
2799                 /* Restore the initial load info */
2800                 fnvlist_free(spa->spa_load_info);
2801                 spa->spa_load_info = loadinfo;
2802 
2803                 return (load_error);
2804         }
2805 }
2806 
2807 /*
2808  * Pool Open/Import
2809  *
2810  * The import case is identical to an open except that the configuration is sent
2811  * down from userland, instead of grabbed from the configuration cache.  For the
2812  * case of an open, the pool configuration will exist in the
2813  * POOL_STATE_UNINITIALIZED state.
2814  *
2815  * The stats information (gen/count/ustats) is used to gather vdev statistics at
2816  * the same time open the pool, without having to keep around the spa_t in some
2817  * ambiguous state.
2818  */
2819 static int
2820 spa_open_common(const char *pool, spa_t **spapp, void *tag, nvlist_t *nvpolicy,
2821     nvlist_t **config)
2822 {
2823         spa_t *spa;
2824         spa_load_state_t state = SPA_LOAD_OPEN;
2825         int error;
2826         int locked = B_FALSE;
2827 
2828         *spapp = NULL;
2829 
2830         /*
2831          * As disgusting as this is, we need to support recursive calls to this
2832          * function because dsl_dir_open() is called during spa_load(), and ends
2833          * up calling spa_open() again.  The real fix is to figure out how to
2834          * avoid dsl_dir_open() calling this in the first place.
2835          */
2836         if (mutex_owner(&spa_namespace_lock) != curthread) {
2837                 mutex_enter(&spa_namespace_lock);
2838                 locked = B_TRUE;
2839         }
2840 
2841         if ((spa = spa_lookup(pool)) == NULL) {
2842                 if (locked)
2843                         mutex_exit(&spa_namespace_lock);
2844                 return (SET_ERROR(ENOENT));
2845         }
2846 
2847         if (spa->spa_state == POOL_STATE_UNINITIALIZED) {
2848                 zpool_rewind_policy_t policy;
2849 
2850                 zpool_get_rewind_policy(nvpolicy ? nvpolicy : spa->spa_config,
2851                     &policy);
2852                 if (policy.zrp_request & ZPOOL_DO_REWIND)
2853                         state = SPA_LOAD_RECOVER;
2854 
2855                 spa_activate(spa, spa_mode_global);
2856 
2857                 if (state != SPA_LOAD_RECOVER)
2858                         spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
2859 
2860                 error = spa_load_best(spa, state, B_FALSE, policy.zrp_txg,
2861                     policy.zrp_request);
2862 
2863                 if (error == EBADF) {
2864                         /*
2865                          * If vdev_validate() returns failure (indicated by
2866                          * EBADF), it indicates that one of the vdevs indicates
2867                          * that the pool has been exported or destroyed.  If
2868                          * this is the case, the config cache is out of sync and
2869                          * we should remove the pool from the namespace.
2870                          */
2871                         spa_unload(spa);
2872                         spa_deactivate(spa);
2873                         spa_config_sync(spa, B_TRUE, B_TRUE);
2874                         spa_remove(spa);
2875                         if (locked)
2876                                 mutex_exit(&spa_namespace_lock);
2877                         return (SET_ERROR(ENOENT));
2878                 }
2879 
2880                 if (error) {
2881                         /*
2882                          * We can't open the pool, but we still have useful
2883                          * information: the state of each vdev after the
2884                          * attempted vdev_open().  Return this to the user.
2885                          */
2886                         if (config != NULL && spa->spa_config) {
2887                                 VERIFY(nvlist_dup(spa->spa_config, config,
2888                                     KM_SLEEP) == 0);
2889                                 VERIFY(nvlist_add_nvlist(*config,
2890                                     ZPOOL_CONFIG_LOAD_INFO,
2891                                     spa->spa_load_info) == 0);
2892                         }
2893                         spa_unload(spa);
2894                         spa_deactivate(spa);
2895                         spa->spa_last_open_failed = error;
2896                         if (locked)
2897                                 mutex_exit(&spa_namespace_lock);
2898                         *spapp = NULL;
2899                         return (error);
2900                 }
2901         }
2902 
2903         spa_open_ref(spa, tag);
2904 
2905         if (config != NULL)
2906                 *config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
2907 
2908         /*
2909          * If we've recovered the pool, pass back any information we
2910          * gathered while doing the load.
2911          */
2912         if (state == SPA_LOAD_RECOVER) {
2913                 VERIFY(nvlist_add_nvlist(*config, ZPOOL_CONFIG_LOAD_INFO,
2914                     spa->spa_load_info) == 0);
2915         }
2916 
2917         if (locked) {
2918                 spa->spa_last_open_failed = 0;
2919                 spa->spa_last_ubsync_txg = 0;
2920                 spa->spa_load_txg = 0;
2921                 mutex_exit(&spa_namespace_lock);
2922         }
2923 
2924         *spapp = spa;
2925 
2926         return (0);
2927 }
2928 
2929 int
2930 spa_open_rewind(const char *name, spa_t **spapp, void *tag, nvlist_t *policy,
2931     nvlist_t **config)
2932 {
2933         return (spa_open_common(name, spapp, tag, policy, config));
2934 }
2935 
2936 int
2937 spa_open(const char *name, spa_t **spapp, void *tag)
2938 {
2939         return (spa_open_common(name, spapp, tag, NULL, NULL));
2940 }
2941 
2942 /*
2943  * Lookup the given spa_t, incrementing the inject count in the process,
2944  * preventing it from being exported or destroyed.
2945  */
2946 spa_t *
2947 spa_inject_addref(char *name)
2948 {
2949         spa_t *spa;
2950 
2951         mutex_enter(&spa_namespace_lock);
2952         if ((spa = spa_lookup(name)) == NULL) {
2953                 mutex_exit(&spa_namespace_lock);
2954                 return (NULL);
2955         }
2956         spa->spa_inject_ref++;
2957         mutex_exit(&spa_namespace_lock);
2958 
2959         return (spa);
2960 }
2961 
2962 void
2963 spa_inject_delref(spa_t *spa)
2964 {
2965         mutex_enter(&spa_namespace_lock);
2966         spa->spa_inject_ref--;
2967         mutex_exit(&spa_namespace_lock);
2968 }
2969 
2970 /*
2971  * Add spares device information to the nvlist.
2972  */
2973 static void
2974 spa_add_spares(spa_t *spa, nvlist_t *config)
2975 {
2976         nvlist_t **spares;
2977         uint_t i, nspares;
2978         nvlist_t *nvroot;
2979         uint64_t guid;
2980         vdev_stat_t *vs;
2981         uint_t vsc;
2982         uint64_t pool;
2983 
2984         ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
2985 
2986         if (spa->spa_spares.sav_count == 0)
2987                 return;
2988 
2989         VERIFY(nvlist_lookup_nvlist(config,
2990             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2991         VERIFY(nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
2992             ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2993         if (nspares != 0) {
2994                 VERIFY(nvlist_add_nvlist_array(nvroot,
2995                     ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
2996                 VERIFY(nvlist_lookup_nvlist_array(nvroot,
2997                     ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0);
2998 
2999                 /*
3000                  * Go through and find any spares which have since been
3001                  * repurposed as an active spare.  If this is the case, update
3002                  * their status appropriately.
3003                  */
3004                 for (i = 0; i < nspares; i++) {
3005                         VERIFY(nvlist_lookup_uint64(spares[i],
3006                             ZPOOL_CONFIG_GUID, &guid) == 0);
3007                         if (spa_spare_exists(guid, &pool, NULL) &&
3008                             pool != 0ULL) {
3009                                 VERIFY(nvlist_lookup_uint64_array(
3010                                     spares[i], ZPOOL_CONFIG_VDEV_STATS,
3011                                     (uint64_t **)&vs, &vsc) == 0);
3012                                 vs->vs_state = VDEV_STATE_CANT_OPEN;
3013                                 vs->vs_aux = VDEV_AUX_SPARED;
3014                         }
3015                 }
3016         }
3017 }
3018 
3019 /*
3020  * Add l2cache device information to the nvlist, including vdev stats.
3021  */
3022 static void
3023 spa_add_l2cache(spa_t *spa, nvlist_t *config)
3024 {
3025         nvlist_t **l2cache;
3026         uint_t i, j, nl2cache;
3027         nvlist_t *nvroot;
3028         uint64_t guid;
3029         vdev_t *vd;
3030         vdev_stat_t *vs;
3031         uint_t vsc;
3032 
3033         ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3034 
3035         if (spa->spa_l2cache.sav_count == 0)
3036                 return;
3037 
3038         VERIFY(nvlist_lookup_nvlist(config,
3039             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3040         VERIFY(nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
3041             ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3042         if (nl2cache != 0) {
3043                 VERIFY(nvlist_add_nvlist_array(nvroot,
3044                     ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3045                 VERIFY(nvlist_lookup_nvlist_array(nvroot,
3046                     ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0);
3047 
3048                 /*
3049                  * Update level 2 cache device stats.
3050                  */
3051 
3052                 for (i = 0; i < nl2cache; i++) {
3053                         VERIFY(nvlist_lookup_uint64(l2cache[i],
3054                             ZPOOL_CONFIG_GUID, &guid) == 0);
3055 
3056                         vd = NULL;
3057                         for (j = 0; j < spa->spa_l2cache.sav_count; j++) {
3058                                 if (guid ==
3059                                     spa->spa_l2cache.sav_vdevs[j]->vdev_guid) {
3060                                         vd = spa->spa_l2cache.sav_vdevs[j];
3061                                         break;
3062                                 }
3063                         }
3064                         ASSERT(vd != NULL);
3065 
3066                         VERIFY(nvlist_lookup_uint64_array(l2cache[i],
3067                             ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
3068                             == 0);
3069                         vdev_get_stats(vd, vs);
3070                 }
3071         }
3072 }
3073 
3074 static void
3075 spa_add_feature_stats(spa_t *spa, nvlist_t *config)
3076 {
3077         nvlist_t *features;
3078         zap_cursor_t zc;
3079         zap_attribute_t za;
3080 
3081         ASSERT(spa_config_held(spa, SCL_CONFIG, RW_READER));
3082         VERIFY(nvlist_alloc(&features, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3083 
3084         if (spa->spa_feat_for_read_obj != 0) {
3085                 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3086                     spa->spa_feat_for_read_obj);
3087                     zap_cursor_retrieve(&zc, &za) == 0;
3088                     zap_cursor_advance(&zc)) {
3089                         ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3090                             za.za_num_integers == 1);
3091                         VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3092                             za.za_first_integer));
3093                 }
3094                 zap_cursor_fini(&zc);
3095         }
3096 
3097         if (spa->spa_feat_for_write_obj != 0) {
3098                 for (zap_cursor_init(&zc, spa->spa_meta_objset,
3099                     spa->spa_feat_for_write_obj);
3100                     zap_cursor_retrieve(&zc, &za) == 0;
3101                     zap_cursor_advance(&zc)) {
3102                         ASSERT(za.za_integer_length == sizeof (uint64_t) &&
3103                             za.za_num_integers == 1);
3104                         VERIFY3U(0, ==, nvlist_add_uint64(features, za.za_name,
3105                             za.za_first_integer));
3106                 }
3107                 zap_cursor_fini(&zc);
3108         }
3109 
3110         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURE_STATS,
3111             features) == 0);
3112         nvlist_free(features);
3113 }
3114 
3115 int
3116 spa_get_stats(const char *name, nvlist_t **config,
3117     char *altroot, size_t buflen)
3118 {
3119         int error;
3120         spa_t *spa;
3121 
3122         *config = NULL;
3123         error = spa_open_common(name, &spa, FTAG, NULL, config);
3124 
3125         if (spa != NULL) {
3126                 /*
3127                  * This still leaves a window of inconsistency where the spares
3128                  * or l2cache devices could change and the config would be
3129                  * self-inconsistent.
3130                  */
3131                 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
3132 
3133                 if (*config != NULL) {
3134                         uint64_t loadtimes[2];
3135 
3136                         loadtimes[0] = spa->spa_loaded_ts.tv_sec;
3137                         loadtimes[1] = spa->spa_loaded_ts.tv_nsec;
3138                         VERIFY(nvlist_add_uint64_array(*config,
3139                             ZPOOL_CONFIG_LOADED_TIME, loadtimes, 2) == 0);
3140 
3141                         VERIFY(nvlist_add_uint64(*config,
3142                             ZPOOL_CONFIG_ERRCOUNT,
3143                             spa_get_errlog_size(spa)) == 0);
3144 
3145                         if (spa_suspended(spa))
3146                                 VERIFY(nvlist_add_uint64(*config,
3147                                     ZPOOL_CONFIG_SUSPENDED,
3148                                     spa->spa_failmode) == 0);
3149 
3150                         spa_add_spares(spa, *config);
3151                         spa_add_l2cache(spa, *config);
3152                         spa_add_feature_stats(spa, *config);
3153                 }
3154         }
3155 
3156         /*
3157          * We want to get the alternate root even for faulted pools, so we cheat
3158          * and call spa_lookup() directly.
3159          */
3160         if (altroot) {
3161                 if (spa == NULL) {
3162                         mutex_enter(&spa_namespace_lock);
3163                         spa = spa_lookup(name);
3164                         if (spa)
3165                                 spa_altroot(spa, altroot, buflen);
3166                         else
3167                                 altroot[0] = '\0';
3168                         spa = NULL;
3169                         mutex_exit(&spa_namespace_lock);
3170                 } else {
3171                         spa_altroot(spa, altroot, buflen);
3172                 }
3173         }
3174 
3175         if (spa != NULL) {
3176                 spa_config_exit(spa, SCL_CONFIG, FTAG);
3177                 spa_close(spa, FTAG);
3178         }
3179 
3180         return (error);
3181 }
3182 
3183 /*
3184  * Validate that the auxiliary device array is well formed.  We must have an
3185  * array of nvlists, each which describes a valid leaf vdev.  If this is an
3186  * import (mode is VDEV_ALLOC_SPARE), then we allow corrupted spares to be
3187  * specified, as long as they are well-formed.
3188  */
3189 static int
3190 spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
3191     spa_aux_vdev_t *sav, const char *config, uint64_t version,
3192     vdev_labeltype_t label)
3193 {
3194         nvlist_t **dev;
3195         uint_t i, ndev;
3196         vdev_t *vd;
3197         int error;
3198 
3199         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3200 
3201         /*
3202          * It's acceptable to have no devs specified.
3203          */
3204         if (nvlist_lookup_nvlist_array(nvroot, config, &dev, &ndev) != 0)
3205                 return (0);
3206 
3207         if (ndev == 0)
3208                 return (SET_ERROR(EINVAL));
3209 
3210         /*
3211          * Make sure the pool is formatted with a version that supports this
3212          * device type.
3213          */
3214         if (spa_version(spa) < version)
3215                 return (SET_ERROR(ENOTSUP));
3216 
3217         /*
3218          * Set the pending device list so we correctly handle device in-use
3219          * checking.
3220          */
3221         sav->sav_pending = dev;
3222         sav->sav_npending = ndev;
3223 
3224         for (i = 0; i < ndev; i++) {
3225                 if ((error = spa_config_parse(spa, &vd, dev[i], NULL, 0,
3226                     mode)) != 0)
3227                         goto out;
3228 
3229                 if (!vd->vdev_ops->vdev_op_leaf) {
3230                         vdev_free(vd);
3231                         error = SET_ERROR(EINVAL);
3232                         goto out;
3233                 }
3234 
3235                 /*
3236                  * The L2ARC currently only supports disk devices in
3237                  * kernel context.  For user-level testing, we allow it.
3238                  */
3239 #ifdef _KERNEL
3240                 if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
3241                     strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
3242                         error = SET_ERROR(ENOTBLK);
3243                         vdev_free(vd);
3244                         goto out;
3245                 }
3246 #endif
3247                 vd->vdev_top = vd;
3248 
3249                 if ((error = vdev_open(vd)) == 0 &&
3250                     (error = vdev_label_init(vd, crtxg, label)) == 0) {
3251                         VERIFY(nvlist_add_uint64(dev[i], ZPOOL_CONFIG_GUID,
3252                             vd->vdev_guid) == 0);
3253                 }
3254 
3255                 vdev_free(vd);
3256 
3257                 if (error &&
3258                     (mode != VDEV_ALLOC_SPARE && mode != VDEV_ALLOC_L2CACHE))
3259                         goto out;
3260                 else
3261                         error = 0;
3262         }
3263 
3264 out:
3265         sav->sav_pending = NULL;
3266         sav->sav_npending = 0;
3267         return (error);
3268 }
3269 
3270 static int
3271 spa_validate_aux(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode)
3272 {
3273         int error;
3274 
3275         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
3276 
3277         if ((error = spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3278             &spa->spa_spares, ZPOOL_CONFIG_SPARES, SPA_VERSION_SPARES,
3279             VDEV_LABEL_SPARE)) != 0) {
3280                 return (error);
3281         }
3282 
3283         return (spa_validate_aux_devs(spa, nvroot, crtxg, mode,
3284             &spa->spa_l2cache, ZPOOL_CONFIG_L2CACHE, SPA_VERSION_L2CACHE,
3285             VDEV_LABEL_L2CACHE));
3286 }
3287 
3288 static void
3289 spa_set_aux_vdevs(spa_aux_vdev_t *sav, nvlist_t **devs, int ndevs,
3290     const char *config)
3291 {
3292         int i;
3293 
3294         if (sav->sav_config != NULL) {
3295                 nvlist_t **olddevs;
3296                 uint_t oldndevs;
3297                 nvlist_t **newdevs;
3298 
3299                 /*
3300                  * Generate new dev list by concatentating with the
3301                  * current dev list.
3302                  */
3303                 VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, config,
3304                     &olddevs, &oldndevs) == 0);
3305 
3306                 newdevs = kmem_alloc(sizeof (void *) *
3307                     (ndevs + oldndevs), KM_SLEEP);
3308                 for (i = 0; i < oldndevs; i++)
3309                         VERIFY(nvlist_dup(olddevs[i], &newdevs[i],
3310                             KM_SLEEP) == 0);
3311                 for (i = 0; i < ndevs; i++)
3312                         VERIFY(nvlist_dup(devs[i], &newdevs[i + oldndevs],
3313                             KM_SLEEP) == 0);
3314 
3315                 VERIFY(nvlist_remove(sav->sav_config, config,
3316                     DATA_TYPE_NVLIST_ARRAY) == 0);
3317 
3318                 VERIFY(nvlist_add_nvlist_array(sav->sav_config,
3319                     config, newdevs, ndevs + oldndevs) == 0);
3320                 for (i = 0; i < oldndevs + ndevs; i++)
3321                         nvlist_free(newdevs[i]);
3322                 kmem_free(newdevs, (oldndevs + ndevs) * sizeof (void *));
3323         } else {
3324                 /*
3325                  * Generate a new dev list.
3326                  */
3327                 VERIFY(nvlist_alloc(&sav->sav_config, NV_UNIQUE_NAME,
3328                     KM_SLEEP) == 0);
3329                 VERIFY(nvlist_add_nvlist_array(sav->sav_config, config,
3330                     devs, ndevs) == 0);
3331         }
3332 }
3333 
3334 /*
3335  * Stop and drop level 2 ARC devices
3336  */
3337 void
3338 spa_l2cache_drop(spa_t *spa)
3339 {
3340         vdev_t *vd;
3341         int i;
3342         spa_aux_vdev_t *sav = &spa->spa_l2cache;
3343 
3344         for (i = 0; i < sav->sav_count; i++) {
3345                 uint64_t pool;
3346 
3347                 vd = sav->sav_vdevs[i];
3348                 ASSERT(vd != NULL);
3349 
3350                 if (spa_l2cache_exists(vd->vdev_guid, &pool) &&
3351                     pool != 0ULL && l2arc_vdev_present(vd))
3352                         l2arc_remove_vdev(vd);
3353         }
3354 }
3355 
3356 /*
3357  * Pool Creation
3358  */
3359 int
3360 spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
3361     nvlist_t *zplprops)
3362 {
3363         spa_t *spa;
3364         char *altroot = NULL;
3365         vdev_t *rvd;
3366         dsl_pool_t *dp;
3367         dmu_tx_t *tx;
3368         int error = 0;
3369         uint64_t txg = TXG_INITIAL;
3370         nvlist_t **spares, **l2cache;
3371         uint_t nspares, nl2cache;
3372         uint64_t version, obj;
3373         boolean_t has_features;
3374 
3375         /*
3376          * If this pool already exists, return failure.
3377          */
3378         mutex_enter(&spa_namespace_lock);
3379         if (spa_lookup(pool) != NULL) {
3380                 mutex_exit(&spa_namespace_lock);
3381                 return (SET_ERROR(EEXIST));
3382         }
3383 
3384         /*
3385          * Allocate a new spa_t structure.
3386          */
3387         (void) nvlist_lookup_string(props,
3388             zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3389         spa = spa_add(pool, NULL, altroot);
3390         spa_activate(spa, spa_mode_global);
3391 
3392         if (props && (error = spa_prop_validate(spa, props))) {
3393                 spa_deactivate(spa);
3394                 spa_remove(spa);
3395                 mutex_exit(&spa_namespace_lock);
3396                 return (error);
3397         }
3398 
3399         has_features = B_FALSE;
3400         for (nvpair_t *elem = nvlist_next_nvpair(props, NULL);
3401             elem != NULL; elem = nvlist_next_nvpair(props, elem)) {
3402                 if (zpool_prop_feature(nvpair_name(elem)))
3403                         has_features = B_TRUE;
3404         }
3405 
3406         if (has_features || nvlist_lookup_uint64(props,
3407             zpool_prop_to_name(ZPOOL_PROP_VERSION), &version) != 0) {
3408                 version = SPA_VERSION;
3409         }
3410         ASSERT(SPA_VERSION_IS_SUPPORTED(version));
3411 
3412         spa->spa_first_txg = txg;
3413         spa->spa_uberblock.ub_txg = txg - 1;
3414         spa->spa_uberblock.ub_version = version;
3415         spa->spa_ubsync = spa->spa_uberblock;
3416 
3417         /*
3418          * Create "The Godfather" zio to hold all async IOs
3419          */
3420         spa->spa_async_zio_root = zio_root(spa, NULL, NULL,
3421             ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | ZIO_FLAG_GODFATHER);
3422 
3423         /*
3424          * Create the root vdev.
3425          */
3426         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3427 
3428         error = spa_config_parse(spa, &rvd, nvroot, NULL, 0, VDEV_ALLOC_ADD);
3429 
3430         ASSERT(error != 0 || rvd != NULL);
3431         ASSERT(error != 0 || spa->spa_root_vdev == rvd);
3432 
3433         if (error == 0 && !zfs_allocatable_devs(nvroot))
3434                 error = SET_ERROR(EINVAL);
3435 
3436         if (error == 0 &&
3437             (error = vdev_create(rvd, txg, B_FALSE)) == 0 &&
3438             (error = spa_validate_aux(spa, nvroot, txg,
3439             VDEV_ALLOC_ADD)) == 0) {
3440                 for (int c = 0; c < rvd->vdev_children; c++) {
3441                         vdev_metaslab_set_size(rvd->vdev_child[c]);
3442                         vdev_expand(rvd->vdev_child[c], txg);
3443                 }
3444         }
3445 
3446         spa_config_exit(spa, SCL_ALL, FTAG);
3447 
3448         if (error != 0) {
3449                 spa_unload(spa);
3450                 spa_deactivate(spa);
3451                 spa_remove(spa);
3452                 mutex_exit(&spa_namespace_lock);
3453                 return (error);
3454         }
3455 
3456         /*
3457          * Get the list of spares, if specified.
3458          */
3459         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3460             &spares, &nspares) == 0) {
3461                 VERIFY(nvlist_alloc(&spa->spa_spares.sav_config, NV_UNIQUE_NAME,
3462                     KM_SLEEP) == 0);
3463                 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3464                     ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3465                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3466                 spa_load_spares(spa);
3467                 spa_config_exit(spa, SCL_ALL, FTAG);
3468                 spa->spa_spares.sav_sync = B_TRUE;
3469         }
3470 
3471         /*
3472          * Get the list of level 2 cache devices, if specified.
3473          */
3474         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3475             &l2cache, &nl2cache) == 0) {
3476                 VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3477                     NV_UNIQUE_NAME, KM_SLEEP) == 0);
3478                 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3479                     ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3480                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3481                 spa_load_l2cache(spa);
3482                 spa_config_exit(spa, SCL_ALL, FTAG);
3483                 spa->spa_l2cache.sav_sync = B_TRUE;
3484         }
3485 
3486         spa->spa_is_initializing = B_TRUE;
3487         spa->spa_dsl_pool = dp = dsl_pool_create(spa, zplprops, txg);
3488         spa->spa_meta_objset = dp->dp_meta_objset;
3489         spa->spa_is_initializing = B_FALSE;
3490 
3491         /*
3492          * Create DDTs (dedup tables).
3493          */
3494         ddt_create(spa);
3495 
3496         spa_update_dspace(spa);
3497 
3498         tx = dmu_tx_create_assigned(dp, txg);
3499 
3500         /*
3501          * Create the pool config object.
3502          */
3503         spa->spa_config_object = dmu_object_alloc(spa->spa_meta_objset,
3504             DMU_OT_PACKED_NVLIST, SPA_CONFIG_BLOCKSIZE,
3505             DMU_OT_PACKED_NVLIST_SIZE, sizeof (uint64_t), tx);
3506 
3507         if (zap_add(spa->spa_meta_objset,
3508             DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CONFIG,
3509             sizeof (uint64_t), 1, &spa->spa_config_object, tx) != 0) {
3510                 cmn_err(CE_PANIC, "failed to add pool config");
3511         }
3512 
3513         if (spa_version(spa) >= SPA_VERSION_FEATURES)
3514                 spa_feature_create_zap_objects(spa, tx);
3515 
3516         if (zap_add(spa->spa_meta_objset,
3517             DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_CREATION_VERSION,
3518             sizeof (uint64_t), 1, &version, tx) != 0) {
3519                 cmn_err(CE_PANIC, "failed to add pool version");
3520         }
3521 
3522         /* Newly created pools with the right version are always deflated. */
3523         if (version >= SPA_VERSION_RAIDZ_DEFLATE) {
3524                 spa->spa_deflate = TRUE;
3525                 if (zap_add(spa->spa_meta_objset,
3526                     DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
3527                     sizeof (uint64_t), 1, &spa->spa_deflate, tx) != 0) {
3528                         cmn_err(CE_PANIC, "failed to add deflate");
3529                 }
3530         }
3531 
3532         /*
3533          * Create the deferred-free bpobj.  Turn off compression
3534          * because sync-to-convergence takes longer if the blocksize
3535          * keeps changing.
3536          */
3537         obj = bpobj_alloc(spa->spa_meta_objset, 1 << 14, tx);
3538         dmu_object_set_compress(spa->spa_meta_objset, obj,
3539             ZIO_COMPRESS_OFF, tx);
3540         if (zap_add(spa->spa_meta_objset,
3541             DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_SYNC_BPOBJ,
3542             sizeof (uint64_t), 1, &obj, tx) != 0) {
3543                 cmn_err(CE_PANIC, "failed to add bpobj");
3544         }
3545         VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
3546             spa->spa_meta_objset, obj));
3547 
3548         /*
3549          * Create the pool's history object.
3550          */
3551         if (version >= SPA_VERSION_ZPOOL_HISTORY)
3552                 spa_history_create_obj(spa, tx);
3553 
3554         /*
3555          * Set pool properties.
3556          */
3557         spa->spa_bootfs = zpool_prop_default_numeric(ZPOOL_PROP_BOOTFS);
3558         spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
3559         spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
3560         spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
3561 
3562         if (props != NULL) {
3563                 spa_configfile_set(spa, props, B_FALSE);
3564                 spa_sync_props(props, tx);
3565         }
3566 
3567         dmu_tx_commit(tx);
3568 
3569         spa->spa_sync_on = B_TRUE;
3570         txg_sync_start(spa->spa_dsl_pool);
3571 
3572         /*
3573          * We explicitly wait for the first transaction to complete so that our
3574          * bean counters are appropriately updated.
3575          */
3576         txg_wait_synced(spa->spa_dsl_pool, txg);
3577 
3578         spa_config_sync(spa, B_FALSE, B_TRUE);
3579 
3580         spa_history_log_version(spa, "create");
3581 
3582         spa->spa_minref = refcount_count(&spa->spa_refcount);
3583 
3584         mutex_exit(&spa_namespace_lock);
3585 
3586         return (0);
3587 }
3588 
3589 #ifdef _KERNEL
3590 /*
3591  * Get the root pool information from the root disk, then import the root pool
3592  * during the system boot up time.
3593  */
3594 extern int vdev_disk_read_rootlabel(char *, char *, nvlist_t **);
3595 
3596 static nvlist_t *
3597 spa_generate_rootconf(char *devpath, char *devid, uint64_t *guid)
3598 {
3599         nvlist_t *config;
3600         nvlist_t *nvtop, *nvroot;
3601         uint64_t pgid;
3602 
3603         if (vdev_disk_read_rootlabel(devpath, devid, &config) != 0)
3604                 return (NULL);
3605 
3606         /*
3607          * Add this top-level vdev to the child array.
3608          */
3609         VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3610             &nvtop) == 0);
3611         VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
3612             &pgid) == 0);
3613         VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, guid) == 0);
3614 
3615         /*
3616          * Put this pool's top-level vdevs into a root vdev.
3617          */
3618         VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
3619         VERIFY(nvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE,
3620             VDEV_TYPE_ROOT) == 0);
3621         VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL) == 0);
3622         VERIFY(nvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid) == 0);
3623         VERIFY(nvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3624             &nvtop, 1) == 0);
3625 
3626         /*
3627          * Replace the existing vdev_tree with the new root vdev in
3628          * this pool's configuration (remove the old, add the new).
3629          */
3630         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
3631         nvlist_free(nvroot);
3632         return (config);
3633 }
3634 
3635 /*
3636  * Walk the vdev tree and see if we can find a device with "better"
3637  * configuration. A configuration is "better" if the label on that
3638  * device has a more recent txg.
3639  */
3640 static void
3641 spa_alt_rootvdev(vdev_t *vd, vdev_t **avd, uint64_t *txg)
3642 {
3643         for (int c = 0; c < vd->vdev_children; c++)
3644                 spa_alt_rootvdev(vd->vdev_child[c], avd, txg);
3645 
3646         if (vd->vdev_ops->vdev_op_leaf) {
3647                 nvlist_t *label;
3648                 uint64_t label_txg;
3649 
3650                 if (vdev_disk_read_rootlabel(vd->vdev_physpath, vd->vdev_devid,
3651                     &label) != 0)
3652                         return;
3653 
3654                 VERIFY(nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
3655                     &label_txg) == 0);
3656 
3657                 /*
3658                  * Do we have a better boot device?
3659                  */
3660                 if (label_txg > *txg) {
3661                         *txg = label_txg;
3662                         *avd = vd;
3663                 }
3664                 nvlist_free(label);
3665         }
3666 }
3667 
3668 /*
3669  * Import a root pool.
3670  *
3671  * For x86. devpath_list will consist of devid and/or physpath name of
3672  * the vdev (e.g. "id1,sd@SSEAGATE..." or "/pci@1f,0/ide@d/disk@0,0:a").
3673  * The GRUB "findroot" command will return the vdev we should boot.
3674  *
3675  * For Sparc, devpath_list consists the physpath name of the booting device
3676  * no matter the rootpool is a single device pool or a mirrored pool.
3677  * e.g.
3678  *      "/pci@1f,0/ide@d/disk@0,0:a"
3679  */
3680 int
3681 spa_import_rootpool(char *devpath, char *devid)
3682 {
3683         spa_t *spa;
3684         vdev_t *rvd, *bvd, *avd = NULL;
3685         nvlist_t *config, *nvtop;
3686         uint64_t guid, txg;
3687         char *pname;
3688         int error;
3689 
3690         /*
3691          * Read the label from the boot device and generate a configuration.
3692          */
3693         config = spa_generate_rootconf(devpath, devid, &guid);
3694 #if defined(_OBP) && defined(_KERNEL)
3695         if (config == NULL) {
3696                 if (strstr(devpath, "/iscsi/ssd") != NULL) {
3697                         /* iscsi boot */
3698                         get_iscsi_bootpath_phy(devpath);
3699                         config = spa_generate_rootconf(devpath, devid, &guid);
3700                 }
3701         }
3702 #endif
3703         if (config == NULL) {
3704                 cmn_err(CE_NOTE, "Cannot read the pool label from '%s'",
3705                     devpath);
3706                 return (SET_ERROR(EIO));
3707         }
3708 
3709         VERIFY(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
3710             &pname) == 0);
3711         VERIFY(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
3712 
3713         mutex_enter(&spa_namespace_lock);
3714         if ((spa = spa_lookup(pname)) != NULL) {
3715                 /*
3716                  * Remove the existing root pool from the namespace so that we
3717                  * can replace it with the correct config we just read in.
3718                  */
3719                 spa_remove(spa);
3720         }
3721 
3722         spa = spa_add(pname, config, NULL);
3723         spa->spa_is_root = B_TRUE;
3724         spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
3725 
3726         /*
3727          * Build up a vdev tree based on the boot device's label config.
3728          */
3729         VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3730             &nvtop) == 0);
3731         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3732         error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
3733             VDEV_ALLOC_ROOTPOOL);
3734         spa_config_exit(spa, SCL_ALL, FTAG);
3735         if (error) {
3736                 mutex_exit(&spa_namespace_lock);
3737                 nvlist_free(config);
3738                 cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
3739                     pname);
3740                 return (error);
3741         }
3742 
3743         /*
3744          * Get the boot vdev.
3745          */
3746         if ((bvd = vdev_lookup_by_guid(rvd, guid)) == NULL) {
3747                 cmn_err(CE_NOTE, "Can not find the boot vdev for guid %llu",
3748                     (u_longlong_t)guid);
3749                 error = SET_ERROR(ENOENT);
3750                 goto out;
3751         }
3752 
3753         /*
3754          * Determine if there is a better boot device.
3755          */
3756         avd = bvd;
3757         spa_alt_rootvdev(rvd, &avd, &txg);
3758         if (avd != bvd) {
3759                 cmn_err(CE_NOTE, "The boot device is 'degraded'. Please "
3760                     "try booting from '%s'", avd->vdev_path);
3761                 error = SET_ERROR(EINVAL);
3762                 goto out;
3763         }
3764 
3765         /*
3766          * If the boot device is part of a spare vdev then ensure that
3767          * we're booting off the active spare.
3768          */
3769         if (bvd->vdev_parent->vdev_ops == &vdev_spare_ops &&
3770             !bvd->vdev_isspare) {
3771                 cmn_err(CE_NOTE, "The boot device is currently spared. Please "
3772                     "try booting from '%s'",
3773                     bvd->vdev_parent->
3774                     vdev_child[bvd->vdev_parent->vdev_children - 1]->vdev_path);
3775                 error = SET_ERROR(EINVAL);
3776                 goto out;
3777         }
3778 
3779         error = 0;
3780 out:
3781         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3782         vdev_free(rvd);
3783         spa_config_exit(spa, SCL_ALL, FTAG);
3784         mutex_exit(&spa_namespace_lock);
3785 
3786         nvlist_free(config);
3787         return (error);
3788 }
3789 
3790 #endif
3791 
3792 /*
3793  * Import a non-root pool into the system.
3794  */
3795 int
3796 spa_import(const char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
3797 {
3798         spa_t *spa;
3799         char *altroot = NULL;
3800         spa_load_state_t state = SPA_LOAD_IMPORT;
3801         zpool_rewind_policy_t policy;
3802         uint64_t mode = spa_mode_global;
3803         uint64_t readonly = B_FALSE;
3804         int error;
3805         nvlist_t *nvroot;
3806         nvlist_t **spares, **l2cache;
3807         uint_t nspares, nl2cache;
3808 
3809         /*
3810          * If a pool with this name exists, return failure.
3811          */
3812         mutex_enter(&spa_namespace_lock);
3813         if (spa_lookup(pool) != NULL) {
3814                 mutex_exit(&spa_namespace_lock);
3815                 return (SET_ERROR(EEXIST));
3816         }
3817 
3818         /*
3819          * Create and initialize the spa structure.
3820          */
3821         (void) nvlist_lookup_string(props,
3822             zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
3823         (void) nvlist_lookup_uint64(props,
3824             zpool_prop_to_name(ZPOOL_PROP_READONLY), &readonly);
3825         if (readonly)
3826                 mode = FREAD;
3827         spa = spa_add(pool, config, altroot);
3828         spa->spa_import_flags = flags;
3829 
3830         /*
3831          * Verbatim import - Take a pool and insert it into the namespace
3832          * as if it had been loaded at boot.
3833          */
3834         if (spa->spa_import_flags & ZFS_IMPORT_VERBATIM) {
3835                 if (props != NULL)
3836                         spa_configfile_set(spa, props, B_FALSE);
3837 
3838                 spa_config_sync(spa, B_FALSE, B_TRUE);
3839 
3840                 mutex_exit(&spa_namespace_lock);
3841                 spa_history_log_version(spa, "import");
3842 
3843                 return (0);
3844         }
3845 
3846         spa_activate(spa, mode);
3847 
3848         /*
3849          * Don't start async tasks until we know everything is healthy.
3850          */
3851         spa_async_suspend(spa);
3852 
3853         zpool_get_rewind_policy(config, &policy);
3854         if (policy.zrp_request & ZPOOL_DO_REWIND)
3855                 state = SPA_LOAD_RECOVER;
3856 
3857         /*
3858          * Pass off the heavy lifting to spa_load().  Pass TRUE for mosconfig
3859          * because the user-supplied config is actually the one to trust when
3860          * doing an import.
3861          */
3862         if (state != SPA_LOAD_RECOVER)
3863                 spa->spa_last_ubsync_txg = spa->spa_load_txg = 0;
3864 
3865         error = spa_load_best(spa, state, B_TRUE, policy.zrp_txg,
3866             policy.zrp_request);
3867 
3868         /*
3869          * Propagate anything learned while loading the pool and pass it
3870          * back to caller (i.e. rewind info, missing devices, etc).
3871          */
3872         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
3873             spa->spa_load_info) == 0);
3874 
3875         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3876         /*
3877          * Toss any existing sparelist, as it doesn't have any validity
3878          * anymore, and conflicts with spa_has_spare().
3879          */
3880         if (spa->spa_spares.sav_config) {
3881                 nvlist_free(spa->spa_spares.sav_config);
3882                 spa->spa_spares.sav_config = NULL;
3883                 spa_load_spares(spa);
3884         }
3885         if (spa->spa_l2cache.sav_config) {
3886                 nvlist_free(spa->spa_l2cache.sav_config);
3887                 spa->spa_l2cache.sav_config = NULL;
3888                 spa_load_l2cache(spa);
3889         }
3890 
3891         VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3892             &nvroot) == 0);
3893         if (error == 0)
3894                 error = spa_validate_aux(spa, nvroot, -1ULL,
3895                     VDEV_ALLOC_SPARE);
3896         if (error == 0)
3897                 error = spa_validate_aux(spa, nvroot, -1ULL,
3898                     VDEV_ALLOC_L2CACHE);
3899         spa_config_exit(spa, SCL_ALL, FTAG);
3900 
3901         if (props != NULL)
3902                 spa_configfile_set(spa, props, B_FALSE);
3903 
3904         if (error != 0 || (props && spa_writeable(spa) &&
3905             (error = spa_prop_set(spa, props)))) {
3906                 spa_unload(spa);
3907                 spa_deactivate(spa);
3908                 spa_remove(spa);
3909                 mutex_exit(&spa_namespace_lock);
3910                 return (error);
3911         }
3912 
3913         spa_async_resume(spa);
3914 
3915         /*
3916          * Override any spares and level 2 cache devices as specified by
3917          * the user, as these may have correct device names/devids, etc.
3918          */
3919         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3920             &spares, &nspares) == 0) {
3921                 if (spa->spa_spares.sav_config)
3922                         VERIFY(nvlist_remove(spa->spa_spares.sav_config,
3923                             ZPOOL_CONFIG_SPARES, DATA_TYPE_NVLIST_ARRAY) == 0);
3924                 else
3925                         VERIFY(nvlist_alloc(&spa->spa_spares.sav_config,
3926                             NV_UNIQUE_NAME, KM_SLEEP) == 0);
3927                 VERIFY(nvlist_add_nvlist_array(spa->spa_spares.sav_config,
3928                     ZPOOL_CONFIG_SPARES, spares, nspares) == 0);
3929                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3930                 spa_load_spares(spa);
3931                 spa_config_exit(spa, SCL_ALL, FTAG);
3932                 spa->spa_spares.sav_sync = B_TRUE;
3933         }
3934         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
3935             &l2cache, &nl2cache) == 0) {
3936                 if (spa->spa_l2cache.sav_config)
3937                         VERIFY(nvlist_remove(spa->spa_l2cache.sav_config,
3938                             ZPOOL_CONFIG_L2CACHE, DATA_TYPE_NVLIST_ARRAY) == 0);
3939                 else
3940                         VERIFY(nvlist_alloc(&spa->spa_l2cache.sav_config,
3941                             NV_UNIQUE_NAME, KM_SLEEP) == 0);
3942                 VERIFY(nvlist_add_nvlist_array(spa->spa_l2cache.sav_config,
3943                     ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache) == 0);
3944                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
3945                 spa_load_l2cache(spa);
3946                 spa_config_exit(spa, SCL_ALL, FTAG);
3947                 spa->spa_l2cache.sav_sync = B_TRUE;
3948         }
3949 
3950         /*
3951          * Check for any removed devices.
3952          */
3953         if (spa->spa_autoreplace) {
3954                 spa_aux_check_removed(&spa->spa_spares);
3955                 spa_aux_check_removed(&spa->spa_l2cache);
3956         }
3957 
3958         if (spa_writeable(spa)) {
3959                 /*
3960                  * Update the config cache to include the newly-imported pool.
3961                  */
3962                 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
3963         }
3964 
3965         /*
3966          * It's possible that the pool was expanded while it was exported.
3967          * We kick off an async task to handle this for us.
3968          */
3969         spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
3970 
3971         mutex_exit(&spa_namespace_lock);
3972         spa_history_log_version(spa, "import");
3973 
3974         return (0);
3975 }
3976 
3977 nvlist_t *
3978 spa_tryimport(nvlist_t *tryconfig)
3979 {
3980         nvlist_t *config = NULL;
3981         char *poolname;
3982         spa_t *spa;
3983         uint64_t state;
3984         int error;
3985 
3986         if (nvlist_lookup_string(tryconfig, ZPOOL_CONFIG_POOL_NAME, &poolname))
3987                 return (NULL);
3988 
3989         if (nvlist_lookup_uint64(tryconfig, ZPOOL_CONFIG_POOL_STATE, &state))
3990                 return (NULL);
3991 
3992         /*
3993          * Create and initialize the spa structure.
3994          */
3995         mutex_enter(&spa_namespace_lock);
3996         spa = spa_add(TRYIMPORT_NAME, tryconfig, NULL);
3997         spa_activate(spa, FREAD);
3998 
3999         /*
4000          * Pass off the heavy lifting to spa_load().
4001          * Pass TRUE for mosconfig because the user-supplied config
4002          * is actually the one to trust when doing an import.
4003          */
4004         error = spa_load(spa, SPA_LOAD_TRYIMPORT, SPA_IMPORT_EXISTING, B_TRUE);
4005 
4006         /*
4007          * If 'tryconfig' was at least parsable, return the current config.
4008          */
4009         if (spa->spa_root_vdev != NULL) {
4010                 config = spa_config_generate(spa, NULL, -1ULL, B_TRUE);
4011                 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
4012                     poolname) == 0);
4013                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4014                     state) == 0);
4015                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
4016                     spa->spa_uberblock.ub_timestamp) == 0);
4017                 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_LOAD_INFO,
4018                     spa->spa_load_info) == 0);
4019 
4020                 /*
4021                  * If the bootfs property exists on this pool then we
4022                  * copy it out so that external consumers can tell which
4023                  * pools are bootable.
4024                  */
4025                 if ((!error || error == EEXIST) && spa->spa_bootfs) {
4026                         char *tmpname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4027 
4028                         /*
4029                          * We have to play games with the name since the
4030                          * pool was opened as TRYIMPORT_NAME.
4031                          */
4032                         if (dsl_dsobj_to_dsname(spa_name(spa),
4033                             spa->spa_bootfs, tmpname) == 0) {
4034                                 char *cp;
4035                                 char *dsname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4036 
4037                                 cp = strchr(tmpname, '/');
4038                                 if (cp == NULL) {
4039                                         (void) strlcpy(dsname, tmpname,
4040                                             MAXPATHLEN);
4041                                 } else {
4042                                         (void) snprintf(dsname, MAXPATHLEN,
4043                                             "%s/%s", poolname, ++cp);
4044                                 }
4045                                 VERIFY(nvlist_add_string(config,
4046                                     ZPOOL_CONFIG_BOOTFS, dsname) == 0);
4047                                 kmem_free(dsname, MAXPATHLEN);
4048                         }
4049                         kmem_free(tmpname, MAXPATHLEN);
4050                 }
4051 
4052                 /*
4053                  * Add the list of hot spares and level 2 cache devices.
4054                  */
4055                 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
4056                 spa_add_spares(spa, config);
4057                 spa_add_l2cache(spa, config);
4058                 spa_config_exit(spa, SCL_CONFIG, FTAG);
4059         }
4060 
4061         spa_unload(spa);
4062         spa_deactivate(spa);
4063         spa_remove(spa);
4064         mutex_exit(&spa_namespace_lock);
4065 
4066         return (config);
4067 }
4068 
4069 /*
4070  * Pool export/destroy
4071  *
4072  * The act of destroying or exporting a pool is very simple.  We make sure there
4073  * is no more pending I/O and any references to the pool are gone.  Then, we
4074  * update the pool state and sync all the labels to disk, removing the
4075  * configuration from the cache afterwards. If the 'hardforce' flag is set, then
4076  * we don't sync the labels or remove the configuration cache.
4077  */
4078 static int
4079 spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
4080     boolean_t force, boolean_t hardforce)
4081 {
4082         spa_t *spa;
4083 
4084         if (oldconfig)
4085                 *oldconfig = NULL;
4086 
4087         if (!(spa_mode_global & FWRITE))
4088                 return (SET_ERROR(EROFS));
4089 
4090         mutex_enter(&spa_namespace_lock);
4091         if ((spa = spa_lookup(pool)) == NULL) {
4092                 mutex_exit(&spa_namespace_lock);
4093                 return (SET_ERROR(ENOENT));
4094         }
4095 
4096         /*
4097          * Put a hold on the pool, drop the namespace lock, stop async tasks,
4098          * reacquire the namespace lock, and see if we can export.
4099          */
4100         spa_open_ref(spa, FTAG);
4101         mutex_exit(&spa_namespace_lock);
4102         spa_async_suspend(spa);
4103         mutex_enter(&spa_namespace_lock);
4104         spa_close(spa, FTAG);
4105 
4106         /*
4107          * The pool will be in core if it's openable,
4108          * in which case we can modify its state.
4109          */
4110         if (spa->spa_state != POOL_STATE_UNINITIALIZED && spa->spa_sync_on) {
4111                 /*
4112                  * Objsets may be open only because they're dirty, so we
4113                  * have to force it to sync before checking spa_refcnt.
4114                  */
4115                 txg_wait_synced(spa->spa_dsl_pool, 0);
4116 
4117                 /*
4118                  * A pool cannot be exported or destroyed if there are active
4119                  * references.  If we are resetting a pool, allow references by
4120                  * fault injection handlers.
4121                  */
4122                 if (!spa_refcount_zero(spa) ||
4123                     (spa->spa_inject_ref != 0 &&
4124                     new_state != POOL_STATE_UNINITIALIZED)) {
4125                         spa_async_resume(spa);
4126                         mutex_exit(&spa_namespace_lock);
4127                         return (SET_ERROR(EBUSY));
4128                 }
4129 
4130                 /*
4131                  * A pool cannot be exported if it has an active shared spare.
4132                  * This is to prevent other pools stealing the active spare
4133                  * from an exported pool. At user's own will, such pool can
4134                  * be forcedly exported.
4135                  */
4136                 if (!force && new_state == POOL_STATE_EXPORTED &&
4137                     spa_has_active_shared_spare(spa)) {
4138                         spa_async_resume(spa);
4139                         mutex_exit(&spa_namespace_lock);
4140                         return (SET_ERROR(EXDEV));
4141                 }
4142 
4143                 /*
4144                  * We want this to be reflected on every label,
4145                  * so mark them all dirty.  spa_unload() will do the
4146                  * final sync that pushes these changes out.
4147                  */
4148                 if (new_state != POOL_STATE_UNINITIALIZED && !hardforce) {
4149                         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
4150                         spa->spa_state = new_state;
4151                         spa->spa_final_txg = spa_last_synced_txg(spa) +
4152                             TXG_DEFER_SIZE + 1;
4153                         vdev_config_dirty(spa->spa_root_vdev);
4154                         spa_config_exit(spa, SCL_ALL, FTAG);
4155                 }
4156         }
4157 
4158         spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
4159 
4160         if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
4161                 spa_unload(spa);
4162                 spa_deactivate(spa);
4163         }
4164 
4165         if (oldconfig && spa->spa_config)
4166                 VERIFY(nvlist_dup(spa->spa_config, oldconfig, 0) == 0);
4167 
4168         if (new_state != POOL_STATE_UNINITIALIZED) {
4169                 if (!hardforce)
4170                         spa_config_sync(spa, B_TRUE, B_TRUE);
4171                 spa_remove(spa);
4172         }
4173         mutex_exit(&spa_namespace_lock);
4174 
4175         return (0);
4176 }
4177 
4178 /*
4179  * Destroy a storage pool.
4180  */
4181 int
4182 spa_destroy(char *pool)
4183 {
4184         return (spa_export_common(pool, POOL_STATE_DESTROYED, NULL,
4185             B_FALSE, B_FALSE));
4186 }
4187 
4188 /*
4189  * Export a storage pool.
4190  */
4191 int
4192 spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
4193     boolean_t hardforce)
4194 {
4195         return (spa_export_common(pool, POOL_STATE_EXPORTED, oldconfig,
4196             force, hardforce));
4197 }
4198 
4199 /*
4200  * Similar to spa_export(), this unloads the spa_t without actually removing it
4201  * from the namespace in any way.
4202  */
4203 int
4204 spa_reset(char *pool)
4205 {
4206         return (spa_export_common(pool, POOL_STATE_UNINITIALIZED, NULL,
4207             B_FALSE, B_FALSE));
4208 }
4209 
4210 /*
4211  * ==========================================================================
4212  * Device manipulation
4213  * ==========================================================================
4214  */
4215 
4216 /*
4217  * Add a device to a storage pool.
4218  */
4219 int
4220 spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
4221 {
4222         uint64_t txg, id;
4223         int error;
4224         vdev_t *rvd = spa->spa_root_vdev;
4225         vdev_t *vd, *tvd;
4226         nvlist_t **spares, **l2cache;
4227         uint_t nspares, nl2cache;
4228 
4229         ASSERT(spa_writeable(spa));
4230 
4231         txg = spa_vdev_enter(spa);
4232 
4233         if ((error = spa_config_parse(spa, &vd, nvroot, NULL, 0,
4234             VDEV_ALLOC_ADD)) != 0)
4235                 return (spa_vdev_exit(spa, NULL, txg, error));
4236 
4237         spa->spa_pending_vdev = vd;  /* spa_vdev_exit() will clear this */
4238 
4239         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES, &spares,
4240             &nspares) != 0)
4241                 nspares = 0;
4242 
4243         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE, &l2cache,
4244             &nl2cache) != 0)
4245                 nl2cache = 0;
4246 
4247         if (vd->vdev_children == 0 && nspares == 0 && nl2cache == 0)
4248                 return (spa_vdev_exit(spa, vd, txg, EINVAL));
4249 
4250         if (vd->vdev_children != 0 &&
4251             (error = vdev_create(vd, txg, B_FALSE)) != 0)
4252                 return (spa_vdev_exit(spa, vd, txg, error));
4253 
4254         /*
4255          * We must validate the spares and l2cache devices after checking the
4256          * children.  Otherwise, vdev_inuse() will blindly overwrite the spare.
4257          */
4258         if ((error = spa_validate_aux(spa, nvroot, txg, VDEV_ALLOC_ADD)) != 0)
4259                 return (spa_vdev_exit(spa, vd, txg, error));
4260 
4261         /*
4262          * Transfer each new top-level vdev from vd to rvd.
4263          */
4264         for (int c = 0; c < vd->vdev_children; c++) {
4265 
4266                 /*
4267                  * Set the vdev id to the first hole, if one exists.
4268                  */
4269                 for (id = 0; id < rvd->vdev_children; id++) {
4270                         if (rvd->vdev_child[id]->vdev_ishole) {
4271                                 vdev_free(rvd->vdev_child[id]);
4272                                 break;
4273                         }
4274                 }
4275                 tvd = vd->vdev_child[c];
4276                 vdev_remove_child(vd, tvd);
4277                 tvd->vdev_id = id;
4278                 vdev_add_child(rvd, tvd);
4279                 vdev_config_dirty(tvd);
4280         }
4281 
4282         if (nspares != 0) {
4283                 spa_set_aux_vdevs(&spa->spa_spares, spares, nspares,
4284                     ZPOOL_CONFIG_SPARES);
4285                 spa_load_spares(spa);
4286                 spa->spa_spares.sav_sync = B_TRUE;
4287         }
4288 
4289         if (nl2cache != 0) {
4290                 spa_set_aux_vdevs(&spa->spa_l2cache, l2cache, nl2cache,
4291                     ZPOOL_CONFIG_L2CACHE);
4292                 spa_load_l2cache(spa);
4293                 spa->spa_l2cache.sav_sync = B_TRUE;
4294         }
4295 
4296         /*
4297          * We have to be careful when adding new vdevs to an existing pool.
4298          * If other threads start allocating from these vdevs before we
4299          * sync the config cache, and we lose power, then upon reboot we may
4300          * fail to open the pool because there are DVAs that the config cache
4301          * can't translate.  Therefore, we first add the vdevs without
4302          * initializing metaslabs; sync the config cache (via spa_vdev_exit());
4303          * and then let spa_config_update() initialize the new metaslabs.
4304          *
4305          * spa_load() checks for added-but-not-initialized vdevs, so that
4306          * if we lose power at any point in this sequence, the remaining
4307          * steps will be completed the next time we load the pool.
4308          */
4309         (void) spa_vdev_exit(spa, vd, txg, 0);
4310 
4311         mutex_enter(&spa_namespace_lock);
4312         spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
4313         mutex_exit(&spa_namespace_lock);
4314 
4315         return (0);
4316 }
4317 
4318 /*
4319  * Attach a device to a mirror.  The arguments are the path to any device
4320  * in the mirror, and the nvroot for the new device.  If the path specifies
4321  * a device that is not mirrored, we automatically insert the mirror vdev.
4322  *
4323  * If 'replacing' is specified, the new device is intended to replace the
4324  * existing device; in this case the two devices are made into their own
4325  * mirror using the 'replacing' vdev, which is functionally identical to
4326  * the mirror vdev (it actually reuses all the same ops) but has a few
4327  * extra rules: you can't attach to it after it's been created, and upon
4328  * completion of resilvering, the first disk (the one being replaced)
4329  * is automatically detached.
4330  */
4331 int
4332 spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
4333 {
4334         uint64_t txg, dtl_max_txg;
4335         vdev_t *rvd = spa->spa_root_vdev;
4336         vdev_t *oldvd, *newvd, *newrootvd, *pvd, *tvd;
4337         vdev_ops_t *pvops;
4338         char *oldvdpath, *newvdpath;
4339         int newvd_isspare;
4340         int error;
4341 
4342         ASSERT(spa_writeable(spa));
4343 
4344         txg = spa_vdev_enter(spa);
4345 
4346         oldvd = spa_lookup_by_guid(spa, guid, B_FALSE);
4347 
4348         if (oldvd == NULL)
4349                 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4350 
4351         if (!oldvd->vdev_ops->vdev_op_leaf)
4352                 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4353 
4354         pvd = oldvd->vdev_parent;
4355 
4356         if ((error = spa_config_parse(spa, &newrootvd, nvroot, NULL, 0,
4357             VDEV_ALLOC_ATTACH)) != 0)
4358                 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4359 
4360         if (newrootvd->vdev_children != 1)
4361                 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4362 
4363         newvd = newrootvd->vdev_child[0];
4364 
4365         if (!newvd->vdev_ops->vdev_op_leaf)
4366                 return (spa_vdev_exit(spa, newrootvd, txg, EINVAL));
4367 
4368         if ((error = vdev_create(newrootvd, txg, replacing)) != 0)
4369                 return (spa_vdev_exit(spa, newrootvd, txg, error));
4370 
4371         /*
4372          * Spares can't replace logs
4373          */
4374         if (oldvd->vdev_top->vdev_islog && newvd->vdev_isspare)
4375                 return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4376 
4377         if (!replacing) {
4378                 /*
4379                  * For attach, the only allowable parent is a mirror or the root
4380                  * vdev.
4381                  */
4382                 if (pvd->vdev_ops != &vdev_mirror_ops &&
4383                     pvd->vdev_ops != &vdev_root_ops)
4384                         return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4385 
4386                 pvops = &vdev_mirror_ops;
4387         } else {
4388                 /*
4389                  * Active hot spares can only be replaced by inactive hot
4390                  * spares.
4391                  */
4392                 if (pvd->vdev_ops == &vdev_spare_ops &&
4393                     oldvd->vdev_isspare &&
4394                     !spa_has_spare(spa, newvd->vdev_guid))
4395                         return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4396 
4397                 /*
4398                  * If the source is a hot spare, and the parent isn't already a
4399                  * spare, then we want to create a new hot spare.  Otherwise, we
4400                  * want to create a replacing vdev.  The user is not allowed to
4401                  * attach to a spared vdev child unless the 'isspare' state is
4402                  * the same (spare replaces spare, non-spare replaces
4403                  * non-spare).
4404                  */
4405                 if (pvd->vdev_ops == &vdev_replacing_ops &&
4406                     spa_version(spa) < SPA_VERSION_MULTI_REPLACE) {
4407                         return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4408                 } else if (pvd->vdev_ops == &vdev_spare_ops &&
4409                     newvd->vdev_isspare != oldvd->vdev_isspare) {
4410                         return (spa_vdev_exit(spa, newrootvd, txg, ENOTSUP));
4411                 }
4412 
4413                 if (newvd->vdev_isspare)
4414                         pvops = &vdev_spare_ops;
4415                 else
4416                         pvops = &vdev_replacing_ops;
4417         }
4418 
4419         /*
4420          * Make sure the new device is big enough.
4421          */
4422         if (newvd->vdev_asize < vdev_get_min_asize(oldvd))
4423                 return (spa_vdev_exit(spa, newrootvd, txg, EOVERFLOW));
4424 
4425         /*
4426          * The new device cannot have a higher alignment requirement
4427          * than the top-level vdev.
4428          */
4429         if (newvd->vdev_ashift > oldvd->vdev_top->vdev_ashift)
4430                 return (spa_vdev_exit(spa, newrootvd, txg, EDOM));
4431 
4432         /*
4433          * If this is an in-place replacement, update oldvd's path and devid
4434          * to make it distinguishable from newvd, and unopenable from now on.
4435          */
4436         if (strcmp(oldvd->vdev_path, newvd->vdev_path) == 0) {
4437                 spa_strfree(oldvd->vdev_path);
4438                 oldvd->vdev_path = kmem_alloc(strlen(newvd->vdev_path) + 5,
4439                     KM_SLEEP);
4440                 (void) sprintf(oldvd->vdev_path, "%s/%s",
4441                     newvd->vdev_path, "old");
4442                 if (oldvd->vdev_devid != NULL) {
4443                         spa_strfree(oldvd->vdev_devid);
4444                         oldvd->vdev_devid = NULL;
4445                 }
4446         }
4447 
4448         /* mark the device being resilvered */
4449         newvd->vdev_resilver_txg = txg;
4450 
4451         /*
4452          * If the parent is not a mirror, or if we're replacing, insert the new
4453          * mirror/replacing/spare vdev above oldvd.
4454          */
4455         if (pvd->vdev_ops != pvops)
4456                 pvd = vdev_add_parent(oldvd, pvops);
4457 
4458         ASSERT(pvd->vdev_top->vdev_parent == rvd);
4459         ASSERT(pvd->vdev_ops == pvops);
4460         ASSERT(oldvd->vdev_parent == pvd);
4461 
4462         /*
4463          * Extract the new device from its root and add it to pvd.
4464          */
4465         vdev_remove_child(newrootvd, newvd);
4466         newvd->vdev_id = pvd->vdev_children;
4467         newvd->vdev_crtxg = oldvd->vdev_crtxg;
4468         vdev_add_child(pvd, newvd);
4469 
4470         tvd = newvd->vdev_top;
4471         ASSERT(pvd->vdev_top == tvd);
4472         ASSERT(tvd->vdev_parent == rvd);
4473 
4474         vdev_config_dirty(tvd);
4475 
4476         /*
4477          * Set newvd's DTL to [TXG_INITIAL, dtl_max_txg) so that we account
4478          * for any dmu_sync-ed blocks.  It will propagate upward when
4479          * spa_vdev_exit() calls vdev_dtl_reassess().
4480          */
4481         dtl_max_txg = txg + TXG_CONCURRENT_STATES;
4482 
4483         vdev_dtl_dirty(newvd, DTL_MISSING, TXG_INITIAL,
4484             dtl_max_txg - TXG_INITIAL);
4485 
4486         if (newvd->vdev_isspare) {
4487                 spa_spare_activate(newvd);
4488                 spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
4489         }
4490 
4491         oldvdpath = spa_strdup(oldvd->vdev_path);
4492         newvdpath = spa_strdup(newvd->vdev_path);
4493         newvd_isspare = newvd->vdev_isspare;
4494 
4495         /*
4496          * Mark newvd's DTL dirty in this txg.
4497          */
4498         vdev_dirty(tvd, VDD_DTL, newvd, txg);
4499 
4500         /*
4501          * Restart the resilver
4502          */
4503         dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
4504 
4505         /*
4506          * Commit the config
4507          */
4508         (void) spa_vdev_exit(spa, newrootvd, dtl_max_txg, 0);
4509 
4510         spa_history_log_internal(spa, "vdev attach", NULL,
4511             "%s vdev=%s %s vdev=%s",
4512             replacing && newvd_isspare ? "spare in" :
4513             replacing ? "replace" : "attach", newvdpath,
4514             replacing ? "for" : "to", oldvdpath);
4515 
4516         spa_strfree(oldvdpath);
4517         spa_strfree(newvdpath);
4518 
4519         if (spa->spa_bootfs)
4520                 spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
4521 
4522         return (0);
4523 }
4524 
4525 /*
4526  * Detach a device from a mirror or replacing vdev.
4527  *
4528  * If 'replace_done' is specified, only detach if the parent
4529  * is a replacing vdev.
4530  */
4531 int
4532 spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
4533 {
4534         uint64_t txg;
4535         int error;
4536         vdev_t *rvd = spa->spa_root_vdev;
4537         vdev_t *vd, *pvd, *cvd, *tvd;
4538         boolean_t unspare = B_FALSE;
4539         uint64_t unspare_guid = 0;
4540         char *vdpath;
4541 
4542         ASSERT(spa_writeable(spa));
4543 
4544         txg = spa_vdev_enter(spa);
4545 
4546         vd = spa_lookup_by_guid(spa, guid, B_FALSE);
4547 
4548         if (vd == NULL)
4549                 return (spa_vdev_exit(spa, NULL, txg, ENODEV));
4550 
4551         if (!vd->vdev_ops->vdev_op_leaf)
4552                 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4553 
4554         pvd = vd->vdev_parent;
4555 
4556         /*
4557          * If the parent/child relationship is not as expected, don't do it.
4558          * Consider M(A,R(B,C)) -- that is, a mirror of A with a replacing
4559          * vdev that's replacing B with C.  The user's intent in replacing
4560          * is to go from M(A,B) to M(A,C).  If the user decides to cancel
4561          * the replace by detaching C, the expected behavior is to end up
4562          * M(A,B).  But suppose that right after deciding to detach C,
4563          * the replacement of B completes.  We would have M(A,C), and then
4564          * ask to detach C, which would leave us with just A -- not what
4565          * the user wanted.  To prevent this, we make sure that the
4566          * parent/child relationship hasn't changed -- in this example,
4567          * that C's parent is still the replacing vdev R.
4568          */
4569         if (pvd->vdev_guid != pguid && pguid != 0)
4570                 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4571 
4572         /*
4573          * Only 'replacing' or 'spare' vdevs can be replaced.
4574          */
4575         if (replace_done && pvd->vdev_ops != &vdev_replacing_ops &&
4576             pvd->vdev_ops != &vdev_spare_ops)
4577                 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4578 
4579         ASSERT(pvd->vdev_ops != &vdev_spare_ops ||
4580             spa_version(spa) >= SPA_VERSION_SPARES);
4581 
4582         /*
4583          * Only mirror, replacing, and spare vdevs support detach.
4584          */
4585         if (pvd->vdev_ops != &vdev_replacing_ops &&
4586             pvd->vdev_ops != &vdev_mirror_ops &&
4587             pvd->vdev_ops != &vdev_spare_ops)
4588                 return (spa_vdev_exit(spa, NULL, txg, ENOTSUP));
4589 
4590         /*
4591          * If this device has the only valid copy of some data,
4592          * we cannot safely detach it.
4593          */
4594         if (vdev_dtl_required(vd))
4595                 return (spa_vdev_exit(spa, NULL, txg, EBUSY));
4596 
4597         ASSERT(pvd->vdev_children >= 2);
4598 
4599         /*
4600          * If we are detaching the second disk from a replacing vdev, then
4601          * check to see if we changed the original vdev's path to have "/old"
4602          * at the end in spa_vdev_attach().  If so, undo that change now.
4603          */
4604         if (pvd->vdev_ops == &vdev_replacing_ops && vd->vdev_id > 0 &&
4605             vd->vdev_path != NULL) {
4606                 size_t len = strlen(vd->vdev_path);
4607 
4608                 for (int c = 0; c < pvd->vdev_children; c++) {
4609                         cvd = pvd->vdev_child[c];
4610 
4611                         if (cvd == vd || cvd->vdev_path == NULL)
4612                                 continue;
4613 
4614                         if (strncmp(cvd->vdev_path, vd->vdev_path, len) == 0 &&
4615                             strcmp(cvd->vdev_path + len, "/old") == 0) {
4616                                 spa_strfree(cvd->vdev_path);
4617                                 cvd->vdev_path = spa_strdup(vd->vdev_path);
4618                                 break;
4619                         }
4620                 }
4621         }
4622 
4623         /*
4624          * If we are detaching the original disk from a spare, then it implies
4625          * that the spare should become a real disk, and be removed from the
4626          * active spare list for the pool.
4627          */
4628         if (pvd->vdev_ops == &vdev_spare_ops &&
4629             vd->vdev_id == 0 &&
4630             pvd->vdev_child[pvd->vdev_children - 1]->vdev_isspare)
4631                 unspare = B_TRUE;
4632 
4633         /*
4634          * Erase the disk labels so the disk can be used for other things.
4635          * This must be done after all other error cases are handled,
4636          * but before we disembowel vd (so we can still do I/O to it).
4637          * But if we can't do it, don't treat the error as fatal --
4638          * it may be that the unwritability of the disk is the reason
4639          * it's being detached!
4640          */
4641         error = vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
4642 
4643         /*
4644          * Remove vd from its parent and compact the parent's children.
4645          */
4646         vdev_remove_child(pvd, vd);
4647         vdev_compact_children(pvd);
4648 
4649         /*
4650          * Remember one of the remaining children so we can get tvd below.
4651          */
4652         cvd = pvd->vdev_child[pvd->vdev_children - 1];
4653 
4654         /*
4655          * If we need to remove the remaining child from the list of hot spares,
4656          * do it now, marking the vdev as no longer a spare in the process.
4657          * We must do this before vdev_remove_parent(), because that can
4658          * change the GUID if it creates a new toplevel GUID.  For a similar
4659          * reason, we must remove the spare now, in the same txg as the detach;
4660          * otherwise someone could attach a new sibling, change the GUID, and
4661          * the subsequent attempt to spa_vdev_remove(unspare_guid) would fail.
4662          */
4663         if (unspare) {
4664                 ASSERT(cvd->vdev_isspare);
4665                 spa_spare_remove(cvd);
4666                 unspare_guid = cvd->vdev_guid;
4667                 (void) spa_vdev_remove(spa, unspare_guid, B_TRUE);
4668                 cvd->vdev_unspare = B_TRUE;
4669         }
4670 
4671         /*
4672          * If the parent mirror/replacing vdev only has one child,
4673          * the parent is no longer needed.  Remove it from the tree.
4674          */
4675         if (pvd->vdev_children == 1) {
4676                 if (pvd->vdev_ops == &vdev_spare_ops)
4677                         cvd->vdev_unspare = B_FALSE;
4678                 vdev_remove_parent(cvd);
4679         }
4680 
4681 
4682         /*
4683          * We don't set tvd until now because the parent we just removed
4684          * may have been the previous top-level vdev.
4685          */
4686         tvd = cvd->vdev_top;
4687         ASSERT(tvd->vdev_parent == rvd);
4688 
4689         /*
4690          * Reevaluate the parent vdev state.
4691          */
4692         vdev_propagate_state(cvd);
4693 
4694         /*
4695          * If the 'autoexpand' property is set on the pool then automatically
4696          * try to expand the size of the pool. For example if the device we
4697          * just detached was smaller than the others, it may be possible to
4698          * add metaslabs (i.e. grow the pool). We need to reopen the vdev
4699          * first so that we can obtain the updated sizes of the leaf vdevs.
4700          */
4701         if (spa->spa_autoexpand) {
4702                 vdev_reopen(tvd);
4703                 vdev_expand(tvd, txg);
4704         }
4705 
4706         vdev_config_dirty(tvd);
4707 
4708         /*
4709          * Mark vd's DTL as dirty in this txg.  vdev_dtl_sync() will see that
4710          * vd->vdev_detached is set and free vd's DTL object in syncing context.
4711          * But first make sure we're not on any *other* txg's DTL list, to
4712          * prevent vd from being accessed after it's freed.
4713          */
4714         vdpath = spa_strdup(vd->vdev_path);
4715         for (int t = 0; t < TXG_SIZE; t++)
4716                 (void) txg_list_remove_this(&tvd->vdev_dtl_list, vd, t);
4717         vd->vdev_detached = B_TRUE;
4718         vdev_dirty(tvd, VDD_DTL, vd, txg);
4719 
4720         spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
4721 
4722         /* hang on to the spa before we release the lock */
4723         spa_open_ref(spa, FTAG);
4724 
4725         error = spa_vdev_exit(spa, vd, txg, 0);
4726 
4727         spa_history_log_internal(spa, "detach", NULL,
4728             "vdev=%s", vdpath);
4729         spa_strfree(vdpath);
4730 
4731         /*
4732          * If this was the removal of the original device in a hot spare vdev,
4733          * then we want to go through and remove the device from the hot spare
4734          * list of every other pool.
4735          */
4736         if (unspare) {
4737                 spa_t *altspa = NULL;
4738 
4739                 mutex_enter(&spa_namespace_lock);
4740                 while ((altspa = spa_next(altspa)) != NULL) {
4741                         if (altspa->spa_state != POOL_STATE_ACTIVE ||
4742                             altspa == spa)
4743                                 continue;
4744 
4745                         spa_open_ref(altspa, FTAG);
4746                         mutex_exit(&spa_namespace_lock);
4747                         (void) spa_vdev_remove(altspa, unspare_guid, B_TRUE);
4748                         mutex_enter(&spa_namespace_lock);
4749                         spa_close(altspa, FTAG);
4750                 }
4751                 mutex_exit(&spa_namespace_lock);
4752 
4753                 /* search the rest of the vdevs for spares to remove */
4754                 spa_vdev_resilver_done(spa);
4755         }
4756 
4757         /* all done with the spa; OK to release */
4758         mutex_enter(&spa_namespace_lock);
4759         spa_close(spa, FTAG);
4760         mutex_exit(&spa_namespace_lock);
4761 
4762         return (error);
4763 }
4764 
4765 /*
4766  * Split a set of devices from their mirrors, and create a new pool from them.
4767  */
4768 int
4769 spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
4770     nvlist_t *props, boolean_t exp)
4771 {
4772         int error = 0;
4773         uint64_t txg, *glist;
4774         spa_t *newspa;
4775         uint_t c, children, lastlog;
4776         nvlist_t **child, *nvl, *tmp;
4777         dmu_tx_t *tx;
4778         char *altroot = NULL;
4779         vdev_t *rvd, **vml = NULL;                      /* vdev modify list */
4780         boolean_t activate_slog;
4781 
4782         ASSERT(spa_writeable(spa));
4783 
4784         txg = spa_vdev_enter(spa);
4785 
4786         /* clear the log and flush everything up to now */
4787         activate_slog = spa_passivate_log(spa);
4788         (void) spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4789         error = spa_offline_log(spa);
4790         txg = spa_vdev_config_enter(spa);
4791 
4792         if (activate_slog)
4793                 spa_activate_log(spa);
4794 
4795         if (error != 0)
4796                 return (spa_vdev_exit(spa, NULL, txg, error));
4797 
4798         /* check new spa name before going any further */
4799         if (spa_lookup(newname) != NULL)
4800                 return (spa_vdev_exit(spa, NULL, txg, EEXIST));
4801 
4802         /*
4803          * scan through all the children to ensure they're all mirrors
4804          */
4805         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &nvl) != 0 ||
4806             nvlist_lookup_nvlist_array(nvl, ZPOOL_CONFIG_CHILDREN, &child,
4807             &children) != 0)
4808                 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4809 
4810         /* first, check to ensure we've got the right child count */
4811         rvd = spa->spa_root_vdev;
4812         lastlog = 0;
4813         for (c = 0; c < rvd->vdev_children; c++) {
4814                 vdev_t *vd = rvd->vdev_child[c];
4815 
4816                 /* don't count the holes & logs as children */
4817                 if (vd->vdev_islog || vd->vdev_ishole) {
4818                         if (lastlog == 0)
4819                                 lastlog = c;
4820                         continue;
4821                 }
4822 
4823                 lastlog = 0;
4824         }
4825         if (children != (lastlog != 0 ? lastlog : rvd->vdev_children))
4826                 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4827 
4828         /* next, ensure no spare or cache devices are part of the split */
4829         if (nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_SPARES, &tmp) == 0 ||
4830             nvlist_lookup_nvlist(nvl, ZPOOL_CONFIG_L2CACHE, &tmp) == 0)
4831                 return (spa_vdev_exit(spa, NULL, txg, EINVAL));
4832 
4833         vml = kmem_zalloc(children * sizeof (vdev_t *), KM_SLEEP);
4834         glist = kmem_zalloc(children * sizeof (uint64_t), KM_SLEEP);
4835 
4836         /* then, loop over each vdev and validate it */
4837         for (c = 0; c < children; c++) {
4838                 uint64_t is_hole = 0;
4839 
4840                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
4841                     &is_hole);
4842 
4843                 if (is_hole != 0) {
4844                         if (spa->spa_root_vdev->vdev_child[c]->vdev_ishole ||
4845                             spa->spa_root_vdev->vdev_child[c]->vdev_islog) {
4846                                 continue;
4847                         } else {
4848                                 error = SET_ERROR(EINVAL);
4849                                 break;
4850                         }
4851                 }
4852 
4853                 /* which disk is going to be split? */
4854                 if (nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_GUID,
4855                     &glist[c]) != 0) {
4856                         error = SET_ERROR(EINVAL);
4857                         break;
4858                 }
4859 
4860                 /* look it up in the spa */
4861                 vml[c] = spa_lookup_by_guid(spa, glist[c], B_FALSE);
4862                 if (vml[c] == NULL) {
4863                         error = SET_ERROR(ENODEV);
4864                         break;
4865                 }
4866 
4867                 /* make sure there's nothing stopping the split */
4868                 if (vml[c]->vdev_parent->vdev_ops != &vdev_mirror_ops ||
4869                     vml[c]->vdev_islog ||
4870                     vml[c]->vdev_ishole ||
4871                     vml[c]->vdev_isspare ||
4872                     vml[c]->vdev_isl2cache ||
4873                     !vdev_writeable(vml[c]) ||
4874                     vml[c]->vdev_children != 0 ||
4875                     vml[c]->vdev_state != VDEV_STATE_HEALTHY ||
4876                     c != spa->spa_root_vdev->vdev_child[c]->vdev_id) {
4877                         error = SET_ERROR(EINVAL);
4878                         break;
4879                 }
4880 
4881                 if (vdev_dtl_required(vml[c])) {
4882                         error = SET_ERROR(EBUSY);
4883                         break;
4884                 }
4885 
4886                 /* we need certain info from the top level */
4887                 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_ARRAY,
4888                     vml[c]->vdev_top->vdev_ms_array) == 0);
4889                 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_METASLAB_SHIFT,
4890                     vml[c]->vdev_top->vdev_ms_shift) == 0);
4891                 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASIZE,
4892                     vml[c]->vdev_top->vdev_asize) == 0);
4893                 VERIFY(nvlist_add_uint64(child[c], ZPOOL_CONFIG_ASHIFT,
4894                     vml[c]->vdev_top->vdev_ashift) == 0);
4895         }
4896 
4897         if (error != 0) {
4898                 kmem_free(vml, children * sizeof (vdev_t *));
4899                 kmem_free(glist, children * sizeof (uint64_t));
4900                 return (spa_vdev_exit(spa, NULL, txg, error));
4901         }
4902 
4903         /* stop writers from using the disks */
4904         for (c = 0; c < children; c++) {
4905                 if (vml[c] != NULL)
4906                         vml[c]->vdev_offline = B_TRUE;
4907         }
4908         vdev_reopen(spa->spa_root_vdev);
4909 
4910         /*
4911          * Temporarily record the splitting vdevs in the spa config.  This
4912          * will disappear once the config is regenerated.
4913          */
4914         VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
4915         VERIFY(nvlist_add_uint64_array(nvl, ZPOOL_CONFIG_SPLIT_LIST,
4916             glist, children) == 0);
4917         kmem_free(glist, children * sizeof (uint64_t));
4918 
4919         mutex_enter(&spa->spa_props_lock);
4920         VERIFY(nvlist_add_nvlist(spa->spa_config, ZPOOL_CONFIG_SPLIT,
4921             nvl) == 0);
4922         mutex_exit(&spa->spa_props_lock);
4923         spa->spa_config_splitting = nvl;
4924         vdev_config_dirty(spa->spa_root_vdev);
4925 
4926         /* configure and create the new pool */
4927         VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, newname) == 0);
4928         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
4929             exp ? POOL_STATE_EXPORTED : POOL_STATE_ACTIVE) == 0);
4930         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
4931             spa_version(spa)) == 0);
4932         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
4933             spa->spa_config_txg) == 0);
4934         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
4935             spa_generate_guid(NULL)) == 0);
4936         (void) nvlist_lookup_string(props,
4937             zpool_prop_to_name(ZPOOL_PROP_ALTROOT), &altroot);
4938 
4939         /* add the new pool to the namespace */
4940         newspa = spa_add(newname, config, altroot);
4941         newspa->spa_config_txg = spa->spa_config_txg;
4942         spa_set_log_state(newspa, SPA_LOG_CLEAR);
4943 
4944         /* release the spa config lock, retaining the namespace lock */
4945         spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
4946 
4947         if (zio_injection_enabled)
4948                 zio_handle_panic_injection(spa, FTAG, 1);
4949 
4950         spa_activate(newspa, spa_mode_global);
4951         spa_async_suspend(newspa);
4952 
4953         /* create the new pool from the disks of the original pool */
4954         error = spa_load(newspa, SPA_LOAD_IMPORT, SPA_IMPORT_ASSEMBLE, B_TRUE);
4955         if (error)
4956                 goto out;
4957 
4958         /* if that worked, generate a real config for the new pool */
4959         if (newspa->spa_root_vdev != NULL) {
4960                 VERIFY(nvlist_alloc(&newspa->spa_config_splitting,
4961                     NV_UNIQUE_NAME, KM_SLEEP) == 0);
4962                 VERIFY(nvlist_add_uint64(newspa->spa_config_splitting,
4963                     ZPOOL_CONFIG_SPLIT_GUID, spa_guid(spa)) == 0);
4964                 spa_config_set(newspa, spa_config_generate(newspa, NULL, -1ULL,
4965                     B_TRUE));
4966         }
4967 
4968         /* set the props */
4969         if (props != NULL) {
4970                 spa_configfile_set(newspa, props, B_FALSE);
4971                 error = spa_prop_set(newspa, props);
4972                 if (error)
4973                         goto out;
4974         }
4975 
4976         /* flush everything */
4977         txg = spa_vdev_config_enter(newspa);
4978         vdev_config_dirty(newspa->spa_root_vdev);
4979         (void) spa_vdev_config_exit(newspa, NULL, txg, 0, FTAG);
4980 
4981         if (zio_injection_enabled)
4982                 zio_handle_panic_injection(spa, FTAG, 2);
4983 
4984         spa_async_resume(newspa);
4985 
4986         /* finally, update the original pool's config */
4987         txg = spa_vdev_config_enter(spa);
4988         tx = dmu_tx_create_dd(spa_get_dsl(spa)->dp_mos_dir);
4989         error = dmu_tx_assign(tx, TXG_WAIT);
4990         if (error != 0)
4991                 dmu_tx_abort(tx);
4992         for (c = 0; c < children; c++) {
4993                 if (vml[c] != NULL) {
4994                         vdev_split(vml[c]);
4995                         if (error == 0)
4996                                 spa_history_log_internal(spa, "detach", tx,
4997                                     "vdev=%s", vml[c]->vdev_path);
4998                         vdev_free(vml[c]);
4999                 }
5000         }
5001         vdev_config_dirty(spa->spa_root_vdev);
5002         spa->spa_config_splitting = NULL;
5003         nvlist_free(nvl);
5004         if (error == 0)
5005                 dmu_tx_commit(tx);
5006         (void) spa_vdev_exit(spa, NULL, txg, 0);
5007 
5008         if (zio_injection_enabled)
5009                 zio_handle_panic_injection(spa, FTAG, 3);
5010 
5011         /* split is complete; log a history record */
5012         spa_history_log_internal(newspa, "split", NULL,
5013             "from pool %s", spa_name(spa));
5014 
5015         kmem_free(vml, children * sizeof (vdev_t *));
5016 
5017         /* if we're not going to mount the filesystems in userland, export */
5018         if (exp)
5019                 error = spa_export_common(newname, POOL_STATE_EXPORTED, NULL,
5020                     B_FALSE, B_FALSE);
5021 
5022         return (error);
5023 
5024 out:
5025         spa_unload(newspa);
5026         spa_deactivate(newspa);
5027         spa_remove(newspa);
5028 
5029         txg = spa_vdev_config_enter(spa);
5030 
5031         /* re-online all offlined disks */
5032         for (c = 0; c < children; c++) {
5033                 if (vml[c] != NULL)
5034                         vml[c]->vdev_offline = B_FALSE;
5035         }
5036         vdev_reopen(spa->spa_root_vdev);
5037 
5038         nvlist_free(spa->spa_config_splitting);
5039         spa->spa_config_splitting = NULL;
5040         (void) spa_vdev_exit(spa, NULL, txg, error);
5041 
5042         kmem_free(vml, children * sizeof (vdev_t *));
5043         return (error);
5044 }
5045 
5046 static nvlist_t *
5047 spa_nvlist_lookup_by_guid(nvlist_t **nvpp, int count, uint64_t target_guid)
5048 {
5049         for (int i = 0; i < count; i++) {
5050                 uint64_t guid;
5051 
5052                 VERIFY(nvlist_lookup_uint64(nvpp[i], ZPOOL_CONFIG_GUID,
5053                     &guid) == 0);
5054 
5055                 if (guid == target_guid)
5056                         return (nvpp[i]);
5057         }
5058 
5059         return (NULL);
5060 }
5061 
5062 static void
5063 spa_vdev_remove_aux(nvlist_t *config, char *name, nvlist_t **dev, int count,
5064         nvlist_t *dev_to_remove)
5065 {
5066         nvlist_t **newdev = NULL;
5067 
5068         if (count > 1)
5069                 newdev = kmem_alloc((count - 1) * sizeof (void *), KM_SLEEP);
5070 
5071         for (int i = 0, j = 0; i < count; i++) {
5072                 if (dev[i] == dev_to_remove)
5073                         continue;
5074                 VERIFY(nvlist_dup(dev[i], &newdev[j++], KM_SLEEP) == 0);
5075         }
5076 
5077         VERIFY(nvlist_remove(config, name, DATA_TYPE_NVLIST_ARRAY) == 0);
5078         VERIFY(nvlist_add_nvlist_array(config, name, newdev, count - 1) == 0);
5079 
5080         for (int i = 0; i < count - 1; i++)
5081                 nvlist_free(newdev[i]);
5082 
5083         if (count > 1)
5084                 kmem_free(newdev, (count - 1) * sizeof (void *));
5085 }
5086 
5087 /*
5088  * Evacuate the device.
5089  */
5090 static int
5091 spa_vdev_remove_evacuate(spa_t *spa, vdev_t *vd)
5092 {
5093         uint64_t txg;
5094         int error = 0;
5095 
5096         ASSERT(MUTEX_HELD(&spa_namespace_lock));
5097         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5098         ASSERT(vd == vd->vdev_top);
5099 
5100         /*
5101          * Evacuate the device.  We don't hold the config lock as writer
5102          * since we need to do I/O but we do keep the
5103          * spa_namespace_lock held.  Once this completes the device
5104          * should no longer have any blocks allocated on it.
5105          */
5106         if (vd->vdev_islog) {
5107                 if (vd->vdev_stat.vs_alloc != 0)
5108                         error = spa_offline_log(spa);
5109         } else {
5110                 error = SET_ERROR(ENOTSUP);
5111         }
5112 
5113         if (error)
5114                 return (error);
5115 
5116         /*
5117          * The evacuation succeeded.  Remove any remaining MOS metadata
5118          * associated with this vdev, and wait for these changes to sync.
5119          */
5120         ASSERT0(vd->vdev_stat.vs_alloc);
5121         txg = spa_vdev_config_enter(spa);
5122         vd->vdev_removing = B_TRUE;
5123         vdev_dirty(vd, 0, NULL, txg);
5124         vdev_config_dirty(vd);
5125         spa_vdev_config_exit(spa, NULL, txg, 0, FTAG);
5126 
5127         return (0);
5128 }
5129 
5130 /*
5131  * Complete the removal by cleaning up the namespace.
5132  */
5133 static void
5134 spa_vdev_remove_from_namespace(spa_t *spa, vdev_t *vd)
5135 {
5136         vdev_t *rvd = spa->spa_root_vdev;
5137         uint64_t id = vd->vdev_id;
5138         boolean_t last_vdev = (id == (rvd->vdev_children - 1));
5139 
5140         ASSERT(MUTEX_HELD(&spa_namespace_lock));
5141         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5142         ASSERT(vd == vd->vdev_top);
5143 
5144         /*
5145          * Only remove any devices which are empty.
5146          */
5147         if (vd->vdev_stat.vs_alloc != 0)
5148                 return;
5149 
5150         (void) vdev_label_init(vd, 0, VDEV_LABEL_REMOVE);
5151 
5152         if (list_link_active(&vd->vdev_state_dirty_node))
5153                 vdev_state_clean(vd);
5154         if (list_link_active(&vd->vdev_config_dirty_node))
5155                 vdev_config_clean(vd);
5156 
5157         vdev_free(vd);
5158 
5159         if (last_vdev) {
5160                 vdev_compact_children(rvd);
5161         } else {
5162                 vd = vdev_alloc_common(spa, id, 0, &vdev_hole_ops);
5163                 vdev_add_child(rvd, vd);
5164         }
5165         vdev_config_dirty(rvd);
5166 
5167         /*
5168          * Reassess the health of our root vdev.
5169          */
5170         vdev_reopen(rvd);
5171 }
5172 
5173 /*
5174  * Remove a device from the pool -
5175  *
5176  * Removing a device from the vdev namespace requires several steps
5177  * and can take a significant amount of time.  As a result we use
5178  * the spa_vdev_config_[enter/exit] functions which allow us to
5179  * grab and release the spa_config_lock while still holding the namespace
5180  * lock.  During each step the configuration is synced out.
5181  *
5182  * Currently, this supports removing only hot spares, slogs, and level 2 ARC
5183  * devices.
5184  */
5185 int
5186 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
5187 {
5188         vdev_t *vd;
5189         metaslab_group_t *mg;
5190         nvlist_t **spares, **l2cache, *nv;
5191         uint64_t txg = 0;
5192         uint_t nspares, nl2cache;
5193         int error = 0;
5194         boolean_t locked = MUTEX_HELD(&spa_namespace_lock);
5195 
5196         ASSERT(spa_writeable(spa));
5197 
5198         if (!locked)
5199                 txg = spa_vdev_enter(spa);
5200 
5201         vd = spa_lookup_by_guid(spa, guid, B_FALSE);
5202 
5203         if (spa->spa_spares.sav_vdevs != NULL &&
5204             nvlist_lookup_nvlist_array(spa->spa_spares.sav_config,
5205             ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0 &&
5206             (nv = spa_nvlist_lookup_by_guid(spares, nspares, guid)) != NULL) {
5207                 /*
5208                  * Only remove the hot spare if it's not currently in use
5209                  * in this pool.
5210                  */
5211                 if (vd == NULL || unspare) {
5212                         spa_vdev_remove_aux(spa->spa_spares.sav_config,
5213                             ZPOOL_CONFIG_SPARES, spares, nspares, nv);
5214                         spa_load_spares(spa);
5215                         spa->spa_spares.sav_sync = B_TRUE;
5216                 } else {
5217                         error = SET_ERROR(EBUSY);
5218                 }
5219         } else if (spa->spa_l2cache.sav_vdevs != NULL &&
5220             nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
5221             ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
5222             (nv = spa_nvlist_lookup_by_guid(l2cache, nl2cache, guid)) != NULL) {
5223                 /*
5224                  * Cache devices can always be removed.
5225                  */
5226                 spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
5227                     ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
5228                 spa_load_l2cache(spa);
5229                 spa->spa_l2cache.sav_sync = B_TRUE;
5230         } else if (vd != NULL && vd->vdev_islog) {
5231                 ASSERT(!locked);
5232                 ASSERT(vd == vd->vdev_top);
5233 
5234                 /*
5235                  * XXX - Once we have bp-rewrite this should
5236                  * become the common case.
5237                  */
5238 
5239                 mg = vd->vdev_mg;
5240 
5241                 /*
5242                  * Stop allocating from this vdev.
5243                  */
5244                 metaslab_group_passivate(mg);
5245 
5246                 /*
5247                  * Wait for the youngest allocations and frees to sync,
5248                  * and then wait for the deferral of those frees to finish.
5249                  */
5250                 spa_vdev_config_exit(spa, NULL,
5251                     txg + TXG_CONCURRENT_STATES + TXG_DEFER_SIZE, 0, FTAG);
5252 
5253                 /*
5254                  * Attempt to evacuate the vdev.
5255                  */
5256                 error = spa_vdev_remove_evacuate(spa, vd);
5257 
5258                 txg = spa_vdev_config_enter(spa);
5259 
5260                 /*
5261                  * If we couldn't evacuate the vdev, unwind.
5262                  */
5263                 if (error) {
5264                         metaslab_group_activate(mg);
5265                         return (spa_vdev_exit(spa, NULL, txg, error));
5266                 }
5267 
5268                 /*
5269                  * Clean up the vdev namespace.
5270                  */
5271                 spa_vdev_remove_from_namespace(spa, vd);
5272 
5273         } else if (vd != NULL) {
5274                 /*
5275                  * Normal vdevs cannot be removed (yet).
5276                  */
5277                 error = SET_ERROR(ENOTSUP);
5278         } else {
5279                 /*
5280                  * There is no vdev of any kind with the specified guid.
5281                  */
5282                 error = SET_ERROR(ENOENT);
5283         }
5284 
5285         if (!locked)
5286                 return (spa_vdev_exit(spa, NULL, txg, error));
5287 
5288         return (error);
5289 }
5290 
5291 /*
5292  * Find any device that's done replacing, or a vdev marked 'unspare' that's
5293  * currently spared, so we can detach it.
5294  */
5295 static vdev_t *
5296 spa_vdev_resilver_done_hunt(vdev_t *vd)
5297 {
5298         vdev_t *newvd, *oldvd;
5299 
5300         for (int c = 0; c < vd->vdev_children; c++) {
5301                 oldvd = spa_vdev_resilver_done_hunt(vd->vdev_child[c]);
5302                 if (oldvd != NULL)
5303                         return (oldvd);
5304         }
5305 
5306         /*
5307          * Check for a completed replacement.  We always consider the first
5308          * vdev in the list to be the oldest vdev, and the last one to be
5309          * the newest (see spa_vdev_attach() for how that works).  In
5310          * the case where the newest vdev is faulted, we will not automatically
5311          * remove it after a resilver completes.  This is OK as it will require
5312          * user intervention to determine which disk the admin wishes to keep.
5313          */
5314         if (vd->vdev_ops == &vdev_replacing_ops) {
5315                 ASSERT(vd->vdev_children > 1);
5316 
5317                 newvd = vd->vdev_child[vd->vdev_children - 1];
5318                 oldvd = vd->vdev_child[0];
5319 
5320                 if (vdev_dtl_empty(newvd, DTL_MISSING) &&
5321                     vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5322                     !vdev_dtl_required(oldvd))
5323                         return (oldvd);
5324         }
5325 
5326         /*
5327          * Check for a completed resilver with the 'unspare' flag set.
5328          */
5329         if (vd->vdev_ops == &vdev_spare_ops) {
5330                 vdev_t *first = vd->vdev_child[0];
5331                 vdev_t *last = vd->vdev_child[vd->vdev_children - 1];
5332 
5333                 if (last->vdev_unspare) {
5334                         oldvd = first;
5335                         newvd = last;
5336                 } else if (first->vdev_unspare) {
5337                         oldvd = last;
5338                         newvd = first;
5339                 } else {
5340                         oldvd = NULL;
5341                 }
5342 
5343                 if (oldvd != NULL &&
5344                     vdev_dtl_empty(newvd, DTL_MISSING) &&
5345                     vdev_dtl_empty(newvd, DTL_OUTAGE) &&
5346                     !vdev_dtl_required(oldvd))
5347                         return (oldvd);
5348 
5349                 /*
5350                  * If there are more than two spares attached to a disk,
5351                  * and those spares are not required, then we want to
5352                  * attempt to free them up now so that they can be used
5353                  * by other pools.  Once we're back down to a single
5354                  * disk+spare, we stop removing them.
5355                  */
5356                 if (vd->vdev_children > 2) {
5357                         newvd = vd->vdev_child[1];
5358 
5359                         if (newvd->vdev_isspare && last->vdev_isspare &&
5360                             vdev_dtl_empty(last, DTL_MISSING) &&
5361                             vdev_dtl_empty(last, DTL_OUTAGE) &&
5362                             !vdev_dtl_required(newvd))
5363                                 return (newvd);
5364                 }
5365         }
5366 
5367         return (NULL);
5368 }
5369 
5370 static void
5371 spa_vdev_resilver_done(spa_t *spa)
5372 {
5373         vdev_t *vd, *pvd, *ppvd;
5374         uint64_t guid, sguid, pguid, ppguid;
5375 
5376         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5377 
5378         while ((vd = spa_vdev_resilver_done_hunt(spa->spa_root_vdev)) != NULL) {
5379                 pvd = vd->vdev_parent;
5380                 ppvd = pvd->vdev_parent;
5381                 guid = vd->vdev_guid;
5382                 pguid = pvd->vdev_guid;
5383                 ppguid = ppvd->vdev_guid;
5384                 sguid = 0;
5385                 /*
5386                  * If we have just finished replacing a hot spared device, then
5387                  * we need to detach the parent's first child (the original hot
5388                  * spare) as well.
5389                  */
5390                 if (ppvd->vdev_ops == &vdev_spare_ops && pvd->vdev_id == 0 &&
5391                     ppvd->vdev_children == 2) {
5392                         ASSERT(pvd->vdev_ops == &vdev_replacing_ops);
5393                         sguid = ppvd->vdev_child[1]->vdev_guid;
5394                 }
5395                 ASSERT(vd->vdev_resilver_txg == 0 || !vdev_dtl_required(vd));
5396 
5397                 spa_config_exit(spa, SCL_ALL, FTAG);
5398                 if (spa_vdev_detach(spa, guid, pguid, B_TRUE) != 0)
5399                         return;
5400                 if (sguid && spa_vdev_detach(spa, sguid, ppguid, B_TRUE) != 0)
5401                         return;
5402                 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
5403         }
5404 
5405         spa_config_exit(spa, SCL_ALL, FTAG);
5406 }
5407 
5408 /*
5409  * Update the stored path or FRU for this vdev.
5410  */
5411 int
5412 spa_vdev_set_common(spa_t *spa, uint64_t guid, const char *value,
5413     boolean_t ispath)
5414 {
5415         vdev_t *vd;
5416         boolean_t sync = B_FALSE;
5417 
5418         ASSERT(spa_writeable(spa));
5419 
5420         spa_vdev_state_enter(spa, SCL_ALL);
5421 
5422         if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL)
5423                 return (spa_vdev_state_exit(spa, NULL, ENOENT));
5424 
5425         if (!vd->vdev_ops->vdev_op_leaf)
5426                 return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
5427 
5428         if (ispath) {
5429                 if (strcmp(value, vd->vdev_path) != 0) {
5430                         spa_strfree(vd->vdev_path);
5431                         vd->vdev_path = spa_strdup(value);
5432                         sync = B_TRUE;
5433                 }
5434         } else {
5435                 if (vd->vdev_fru == NULL) {
5436                         vd->vdev_fru = spa_strdup(value);
5437                         sync = B_TRUE;
5438                 } else if (strcmp(value, vd->vdev_fru) != 0) {
5439                         spa_strfree(vd->vdev_fru);
5440                         vd->vdev_fru = spa_strdup(value);
5441                         sync = B_TRUE;
5442                 }
5443         }
5444 
5445         return (spa_vdev_state_exit(spa, sync ? vd : NULL, 0));
5446 }
5447 
5448 int
5449 spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath)
5450 {
5451         return (spa_vdev_set_common(spa, guid, newpath, B_TRUE));
5452 }
5453 
5454 int
5455 spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
5456 {
5457         return (spa_vdev_set_common(spa, guid, newfru, B_FALSE));
5458 }
5459 
5460 /*
5461  * ==========================================================================
5462  * SPA Scanning
5463  * ==========================================================================
5464  */
5465 
5466 int
5467 spa_scan_stop(spa_t *spa)
5468 {
5469         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5470         if (dsl_scan_resilvering(spa->spa_dsl_pool))
5471                 return (SET_ERROR(EBUSY));
5472         return (dsl_scan_cancel(spa->spa_dsl_pool));
5473 }
5474 
5475 int
5476 spa_scan(spa_t *spa, pool_scan_func_t func)
5477 {
5478         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
5479 
5480         if (func >= POOL_SCAN_FUNCS || func == POOL_SCAN_NONE)
5481                 return (SET_ERROR(ENOTSUP));
5482 
5483         /*
5484          * If a resilver was requested, but there is no DTL on a
5485          * writeable leaf device, we have nothing to do.
5486          */
5487         if (func == POOL_SCAN_RESILVER &&
5488             !vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL)) {
5489                 spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
5490                 return (0);
5491         }
5492 
5493         return (dsl_scan(spa->spa_dsl_pool, func));
5494 }
5495 
5496 /*
5497  * ==========================================================================
5498  * SPA async task processing
5499  * ==========================================================================
5500  */
5501 
5502 static void
5503 spa_async_remove(spa_t *spa, vdev_t *vd)
5504 {
5505         if (vd->vdev_remove_wanted) {
5506                 vd->vdev_remove_wanted = B_FALSE;
5507                 vd->vdev_delayed_close = B_FALSE;
5508                 vdev_set_state(vd, B_FALSE, VDEV_STATE_REMOVED, VDEV_AUX_NONE);
5509 
5510                 /*
5511                  * We want to clear the stats, but we don't want to do a full
5512                  * vdev_clear() as that will cause us to throw away
5513                  * degraded/faulted state as well as attempt to reopen the
5514                  * device, all of which is a waste.
5515                  */
5516                 vd->vdev_stat.vs_read_errors = 0;
5517                 vd->vdev_stat.vs_write_errors = 0;
5518                 vd->vdev_stat.vs_checksum_errors = 0;
5519 
5520                 vdev_state_dirty(vd->vdev_top);
5521         }
5522 
5523         for (int c = 0; c < vd->vdev_children; c++)
5524                 spa_async_remove(spa, vd->vdev_child[c]);
5525 }
5526 
5527 static void
5528 spa_async_probe(spa_t *spa, vdev_t *vd)
5529 {
5530         if (vd->vdev_probe_wanted) {
5531                 vd->vdev_probe_wanted = B_FALSE;
5532                 vdev_reopen(vd);        /* vdev_open() does the actual probe */
5533         }
5534 
5535         for (int c = 0; c < vd->vdev_children; c++)
5536                 spa_async_probe(spa, vd->vdev_child[c]);
5537 }
5538 
5539 static void
5540 spa_async_autoexpand(spa_t *spa, vdev_t *vd)
5541 {
5542         sysevent_id_t eid;
5543         nvlist_t *attr;
5544         char *physpath;
5545 
5546         if (!spa->spa_autoexpand)
5547                 return;
5548 
5549         for (int c = 0; c < vd->vdev_children; c++) {
5550                 vdev_t *cvd = vd->vdev_child[c];
5551                 spa_async_autoexpand(spa, cvd);
5552         }
5553 
5554         if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
5555                 return;
5556 
5557         physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
5558         (void) snprintf(physpath, MAXPATHLEN, "/devices%s", vd->vdev_physpath);
5559 
5560         VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5561         VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
5562 
5563         (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
5564             ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
5565 
5566         nvlist_free(attr);
5567         kmem_free(physpath, MAXPATHLEN);
5568 }
5569 
5570 static void
5571 spa_async_thread(spa_t *spa)
5572 {
5573         int tasks;
5574 
5575         ASSERT(spa->spa_sync_on);
5576 
5577         mutex_enter(&spa->spa_async_lock);
5578         tasks = spa->spa_async_tasks;
5579         spa->spa_async_tasks = 0;
5580         mutex_exit(&spa->spa_async_lock);
5581 
5582         /*
5583          * See if the config needs to be updated.
5584          */
5585         if (tasks & SPA_ASYNC_CONFIG_UPDATE) {
5586                 uint64_t old_space, new_space;
5587 
5588                 mutex_enter(&spa_namespace_lock);
5589                 old_space = metaslab_class_get_space(spa_normal_class(spa));
5590                 spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
5591                 new_space = metaslab_class_get_space(spa_normal_class(spa));
5592                 mutex_exit(&spa_namespace_lock);
5593 
5594                 /*
5595                  * If the pool grew as a result of the config update,
5596                  * then log an internal history event.
5597                  */
5598                 if (new_space != old_space) {
5599                         spa_history_log_internal(spa, "vdev online", NULL,
5600                             "pool '%s' size: %llu(+%llu)",
5601                             spa_name(spa), new_space, new_space - old_space);
5602                 }
5603         }
5604 
5605         /*
5606          * See if any devices need to be marked REMOVED.
5607          */
5608         if (tasks & SPA_ASYNC_REMOVE) {
5609                 spa_vdev_state_enter(spa, SCL_NONE);
5610                 spa_async_remove(spa, spa->spa_root_vdev);
5611                 for (int i = 0; i < spa->spa_l2cache.sav_count; i++)
5612                         spa_async_remove(spa, spa->spa_l2cache.sav_vdevs[i]);
5613                 for (int i = 0; i < spa->spa_spares.sav_count; i++)
5614                         spa_async_remove(spa, spa->spa_spares.sav_vdevs[i]);
5615                 (void) spa_vdev_state_exit(spa, NULL, 0);
5616         }
5617 
5618         if ((tasks & SPA_ASYNC_AUTOEXPAND) && !spa_suspended(spa)) {
5619                 spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
5620                 spa_async_autoexpand(spa, spa->spa_root_vdev);
5621                 spa_config_exit(spa, SCL_CONFIG, FTAG);
5622         }
5623 
5624         /*
5625          * See if any devices need to be probed.
5626          */
5627         if (tasks & SPA_ASYNC_PROBE) {
5628                 spa_vdev_state_enter(spa, SCL_NONE);
5629                 spa_async_probe(spa, spa->spa_root_vdev);
5630                 (void) spa_vdev_state_exit(spa, NULL, 0);
5631         }
5632 
5633         /*
5634          * If any devices are done replacing, detach them.
5635          */
5636         if (tasks & SPA_ASYNC_RESILVER_DONE)
5637                 spa_vdev_resilver_done(spa);
5638 
5639         /*
5640          * Kick off a resilver.
5641          */
5642         if (tasks & SPA_ASYNC_RESILVER)
5643                 dsl_resilver_restart(spa->spa_dsl_pool, 0);
5644 
5645         /*
5646          * Let the world know that we're done.
5647          */
5648         mutex_enter(&spa->spa_async_lock);
5649         spa->spa_async_thread = NULL;
5650         cv_broadcast(&spa->spa_async_cv);
5651         mutex_exit(&spa->spa_async_lock);
5652         thread_exit();
5653 }
5654 
5655 void
5656 spa_async_suspend(spa_t *spa)
5657 {
5658         mutex_enter(&spa->spa_async_lock);
5659         spa->spa_async_suspended++;
5660         while (spa->spa_async_thread != NULL)
5661                 cv_wait(&spa->spa_async_cv, &spa->spa_async_lock);
5662         mutex_exit(&spa->spa_async_lock);
5663 }
5664 
5665 void
5666 spa_async_resume(spa_t *spa)
5667 {
5668         mutex_enter(&spa->spa_async_lock);
5669         ASSERT(spa->spa_async_suspended != 0);
5670         spa->spa_async_suspended--;
5671         mutex_exit(&spa->spa_async_lock);
5672 }
5673 
5674 static boolean_t
5675 spa_async_tasks_pending(spa_t *spa)
5676 {
5677         uint_t non_config_tasks;
5678         uint_t config_task;
5679         boolean_t config_task_suspended;
5680 
5681         non_config_tasks = spa->spa_async_tasks & ~SPA_ASYNC_CONFIG_UPDATE;
5682         config_task = spa->spa_async_tasks & SPA_ASYNC_CONFIG_UPDATE;
5683         if (spa->spa_ccw_fail_time == 0) {
5684                 config_task_suspended = B_FALSE;
5685         } else {
5686                 config_task_suspended =
5687                     (gethrtime() - spa->spa_ccw_fail_time) <
5688                     (zfs_ccw_retry_interval * NANOSEC);
5689         }
5690 
5691         return (non_config_tasks || (config_task && !config_task_suspended));
5692 }
5693 
5694 static void
5695 spa_async_dispatch(spa_t *spa)
5696 {
5697         mutex_enter(&spa->spa_async_lock);
5698         if (spa_async_tasks_pending(spa) &&
5699             !spa->spa_async_suspended &&
5700             spa->spa_async_thread == NULL &&
5701             rootdir != NULL)
5702                 spa->spa_async_thread = thread_create(NULL, 0,
5703                     spa_async_thread, spa, 0, &p0, TS_RUN, maxclsyspri);
5704         mutex_exit(&spa->spa_async_lock);
5705 }
5706 
5707 void
5708 spa_async_request(spa_t *spa, int task)
5709 {
5710         zfs_dbgmsg("spa=%s async request task=%u", spa->spa_name, task);
5711         mutex_enter(&spa->spa_async_lock);
5712         spa->spa_async_tasks |= task;
5713         mutex_exit(&spa->spa_async_lock);
5714 }
5715 
5716 /*
5717  * ==========================================================================
5718  * SPA syncing routines
5719  * ==========================================================================
5720  */
5721 
5722 static int
5723 bpobj_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5724 {
5725         bpobj_t *bpo = arg;
5726         bpobj_enqueue(bpo, bp, tx);
5727         return (0);
5728 }
5729 
5730 static int
5731 spa_free_sync_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
5732 {
5733         zio_t *zio = arg;
5734 
5735         zio_nowait(zio_free_sync(zio, zio->io_spa, dmu_tx_get_txg(tx), bp,
5736             zio->io_flags));
5737         return (0);
5738 }
5739 
5740 static void
5741 spa_sync_nvlist(spa_t *spa, uint64_t obj, nvlist_t *nv, dmu_tx_t *tx)
5742 {
5743         char *packed = NULL;
5744         size_t bufsize;
5745         size_t nvsize = 0;
5746         dmu_buf_t *db;
5747 
5748         VERIFY(nvlist_size(nv, &nvsize, NV_ENCODE_XDR) == 0);
5749 
5750         /*
5751          * Write full (SPA_CONFIG_BLOCKSIZE) blocks of configuration
5752          * information.  This avoids the dbuf_will_dirty() path and
5753          * saves us a pre-read to get data we don't actually care about.
5754          */
5755         bufsize = P2ROUNDUP((uint64_t)nvsize, SPA_CONFIG_BLOCKSIZE);
5756         packed = kmem_alloc(bufsize, KM_SLEEP);
5757 
5758         VERIFY(nvlist_pack(nv, &packed, &nvsize, NV_ENCODE_XDR,
5759             KM_SLEEP) == 0);
5760         bzero(packed + nvsize, bufsize - nvsize);
5761 
5762         dmu_write(spa->spa_meta_objset, obj, 0, bufsize, packed, tx);
5763 
5764         kmem_free(packed, bufsize);
5765 
5766         VERIFY(0 == dmu_bonus_hold(spa->spa_meta_objset, obj, FTAG, &db));
5767         dmu_buf_will_dirty(db, tx);
5768         *(uint64_t *)db->db_data = nvsize;
5769         dmu_buf_rele(db, FTAG);
5770 }
5771 
5772 static void
5773 spa_sync_aux_dev(spa_t *spa, spa_aux_vdev_t *sav, dmu_tx_t *tx,
5774     const char *config, const char *entry)
5775 {
5776         nvlist_t *nvroot;
5777         nvlist_t **list;
5778         int i;
5779 
5780         if (!sav->sav_sync)
5781                 return;
5782 
5783         /*
5784          * Update the MOS nvlist describing the list of available devices.
5785          * spa_validate_aux() will have already made sure this nvlist is
5786          * valid and the vdevs are labeled appropriately.
5787          */
5788         if (sav->sav_object == 0) {
5789                 sav->sav_object = dmu_object_alloc(spa->spa_meta_objset,
5790                     DMU_OT_PACKED_NVLIST, 1 << 14, DMU_OT_PACKED_NVLIST_SIZE,
5791                     sizeof (uint64_t), tx);
5792                 VERIFY(zap_update(spa->spa_meta_objset,
5793                     DMU_POOL_DIRECTORY_OBJECT, entry, sizeof (uint64_t), 1,
5794                     &sav->sav_object, tx) == 0);
5795         }
5796 
5797         VERIFY(nvlist_alloc(&nvroot, NV_UNIQUE_NAME, KM_SLEEP) == 0);
5798         if (sav->sav_count == 0) {
5799                 VERIFY(nvlist_add_nvlist_array(nvroot, config, NULL, 0) == 0);
5800         } else {
5801                 list = kmem_alloc(sav->sav_count * sizeof (void *), KM_SLEEP);
5802                 for (i = 0; i < sav->sav_count; i++)
5803                         list[i] = vdev_config_generate(spa, sav->sav_vdevs[i],
5804                             B_FALSE, VDEV_CONFIG_L2CACHE);
5805                 VERIFY(nvlist_add_nvlist_array(nvroot, config, list,
5806                     sav->sav_count) == 0);
5807                 for (i = 0; i < sav->sav_count; i++)
5808                         nvlist_free(list[i]);
5809                 kmem_free(list, sav->sav_count * sizeof (void *));
5810         }
5811 
5812         spa_sync_nvlist(spa, sav->sav_object, nvroot, tx);
5813         nvlist_free(nvroot);
5814 
5815         sav->sav_sync = B_FALSE;
5816 }
5817 
5818 static void
5819 spa_sync_config_object(spa_t *spa, dmu_tx_t *tx)
5820 {
5821         nvlist_t *config;
5822 
5823         if (list_is_empty(&spa->spa_config_dirty_list))
5824                 return;
5825 
5826         spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
5827 
5828         config = spa_config_generate(spa, spa->spa_root_vdev,
5829             dmu_tx_get_txg(tx), B_FALSE);
5830 
5831         /*
5832          * If we're upgrading the spa version then make sure that
5833          * the config object gets updated with the correct version.
5834          */
5835         if (spa->spa_ubsync.ub_version < spa->spa_uberblock.ub_version)
5836                 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
5837                     spa->spa_uberblock.ub_version);
5838 
5839         spa_config_exit(spa, SCL_STATE, FTAG);
5840 
5841         if (spa->spa_config_syncing)
5842                 nvlist_free(spa->spa_config_syncing);
5843         spa->spa_config_syncing = config;
5844 
5845         spa_sync_nvlist(spa, spa->spa_config_object, config, tx);
5846 }
5847 
5848 static void
5849 spa_sync_version(void *arg, dmu_tx_t *tx)
5850 {
5851         uint64_t *versionp = arg;
5852         uint64_t version = *versionp;
5853         spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5854 
5855         /*
5856          * Setting the version is special cased when first creating the pool.
5857          */
5858         ASSERT(tx->tx_txg != TXG_INITIAL);
5859 
5860         ASSERT(SPA_VERSION_IS_SUPPORTED(version));
5861         ASSERT(version >= spa_version(spa));
5862 
5863         spa->spa_uberblock.ub_version = version;
5864         vdev_config_dirty(spa->spa_root_vdev);
5865         spa_history_log_internal(spa, "set", tx, "version=%lld", version);
5866 }
5867 
5868 /*
5869  * Set zpool properties.
5870  */
5871 static void
5872 spa_sync_props(void *arg, dmu_tx_t *tx)
5873 {
5874         nvlist_t *nvp = arg;
5875         spa_t *spa = dmu_tx_pool(tx)->dp_spa;
5876         objset_t *mos = spa->spa_meta_objset;
5877         nvpair_t *elem = NULL;
5878 
5879         mutex_enter(&spa->spa_props_lock);
5880 
5881         while ((elem = nvlist_next_nvpair(nvp, elem))) {
5882                 uint64_t intval;
5883                 char *strval, *fname;
5884                 zpool_prop_t prop;
5885                 const char *propname;
5886                 zprop_type_t proptype;
5887                 zfeature_info_t *feature;
5888 
5889                 switch (prop = zpool_name_to_prop(nvpair_name(elem))) {
5890                 case ZPROP_INVAL:
5891                         /*
5892                          * We checked this earlier in spa_prop_validate().
5893                          */
5894                         ASSERT(zpool_prop_feature(nvpair_name(elem)));
5895 
5896                         fname = strchr(nvpair_name(elem), '@') + 1;
5897                         VERIFY3U(0, ==, zfeature_lookup_name(fname, &feature));
5898 
5899                         spa_feature_enable(spa, feature, tx);
5900                         spa_history_log_internal(spa, "set", tx,
5901                             "%s=enabled", nvpair_name(elem));
5902                         break;
5903 
5904                 case ZPOOL_PROP_VERSION:
5905                         VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5906                         /*
5907                          * The version is synced seperatly before other
5908                          * properties and should be correct by now.
5909                          */
5910                         ASSERT3U(spa_version(spa), >=, intval);
5911                         break;
5912 
5913                 case ZPOOL_PROP_ALTROOT:
5914                         /*
5915                          * 'altroot' is a non-persistent property. It should
5916                          * have been set temporarily at creation or import time.
5917                          */
5918                         ASSERT(spa->spa_root != NULL);
5919                         break;
5920 
5921                 case ZPOOL_PROP_READONLY:
5922                 case ZPOOL_PROP_CACHEFILE:
5923                         /*
5924                          * 'readonly' and 'cachefile' are also non-persisitent
5925                          * properties.
5926                          */
5927                         break;
5928                 case ZPOOL_PROP_COMMENT:
5929                         VERIFY(nvpair_value_string(elem, &strval) == 0);
5930                         if (spa->spa_comment != NULL)
5931                                 spa_strfree(spa->spa_comment);
5932                         spa->spa_comment = spa_strdup(strval);
5933                         /*
5934                          * We need to dirty the configuration on all the vdevs
5935                          * so that their labels get updated.  It's unnecessary
5936                          * to do this for pool creation since the vdev's
5937                          * configuratoin has already been dirtied.
5938                          */
5939                         if (tx->tx_txg != TXG_INITIAL)
5940                                 vdev_config_dirty(spa->spa_root_vdev);
5941                         spa_history_log_internal(spa, "set", tx,
5942                             "%s=%s", nvpair_name(elem), strval);
5943                         break;
5944                 default:
5945                         /*
5946                          * Set pool property values in the poolprops mos object.
5947                          */
5948                         if (spa->spa_pool_props_object == 0) {
5949                                 spa->spa_pool_props_object =
5950                                     zap_create_link(mos, DMU_OT_POOL_PROPS,
5951                                     DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_PROPS,
5952                                     tx);
5953                         }
5954 
5955                         /* normalize the property name */
5956                         propname = zpool_prop_to_name(prop);
5957                         proptype = zpool_prop_get_type(prop);
5958 
5959                         if (nvpair_type(elem) == DATA_TYPE_STRING) {
5960                                 ASSERT(proptype == PROP_TYPE_STRING);
5961                                 VERIFY(nvpair_value_string(elem, &strval) == 0);
5962                                 VERIFY(zap_update(mos,
5963                                     spa->spa_pool_props_object, propname,
5964                                     1, strlen(strval) + 1, strval, tx) == 0);
5965                                 spa_history_log_internal(spa, "set", tx,
5966                                     "%s=%s", nvpair_name(elem), strval);
5967                         } else if (nvpair_type(elem) == DATA_TYPE_UINT64) {
5968                                 VERIFY(nvpair_value_uint64(elem, &intval) == 0);
5969 
5970                                 if (proptype == PROP_TYPE_INDEX) {
5971                                         const char *unused;
5972                                         VERIFY(zpool_prop_index_to_string(
5973                                             prop, intval, &unused) == 0);
5974                                 }
5975                                 VERIFY(zap_update(mos,
5976                                     spa->spa_pool_props_object, propname,
5977                                     8, 1, &intval, tx) == 0);
5978                                 spa_history_log_internal(spa, "set", tx,
5979                                     "%s=%lld", nvpair_name(elem), intval);
5980                         } else {
5981                                 ASSERT(0); /* not allowed */
5982                         }
5983 
5984                         switch (prop) {
5985                         case ZPOOL_PROP_DELEGATION:
5986                                 spa->spa_delegation = intval;
5987                                 break;
5988                         case ZPOOL_PROP_BOOTFS:
5989                                 spa->spa_bootfs = intval;
5990                                 break;
5991                         case ZPOOL_PROP_FAILUREMODE:
5992                                 spa->spa_failmode = intval;
5993                                 break;
5994                         case ZPOOL_PROP_AUTOEXPAND:
5995                                 spa->spa_autoexpand = intval;
5996                                 if (tx->tx_txg != TXG_INITIAL)
5997                                         spa_async_request(spa,
5998                                             SPA_ASYNC_AUTOEXPAND);
5999                                 break;
6000                         case ZPOOL_PROP_DEDUPDITTO:
6001                                 spa->spa_dedup_ditto = intval;
6002                                 break;
6003                         default:
6004                                 break;
6005                         }
6006                 }
6007 
6008         }
6009 
6010         mutex_exit(&spa->spa_props_lock);
6011 }
6012 
6013 /*
6014  * Perform one-time upgrade on-disk changes.  spa_version() does not
6015  * reflect the new version this txg, so there must be no changes this
6016  * txg to anything that the upgrade code depends on after it executes.
6017  * Therefore this must be called after dsl_pool_sync() does the sync
6018  * tasks.
6019  */
6020 static void
6021 spa_sync_upgrades(spa_t *spa, dmu_tx_t *tx)
6022 {
6023         dsl_pool_t *dp = spa->spa_dsl_pool;
6024 
6025         ASSERT(spa->spa_sync_pass == 1);
6026 
6027         rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
6028 
6029         if (spa->spa_ubsync.ub_version < SPA_VERSION_ORIGIN &&
6030             spa->spa_uberblock.ub_version >= SPA_VERSION_ORIGIN) {
6031                 dsl_pool_create_origin(dp, tx);
6032 
6033                 /* Keeping the origin open increases spa_minref */
6034                 spa->spa_minref += 3;
6035         }
6036 
6037         if (spa->spa_ubsync.ub_version < SPA_VERSION_NEXT_CLONES &&
6038             spa->spa_uberblock.ub_version >= SPA_VERSION_NEXT_CLONES) {
6039                 dsl_pool_upgrade_clones(dp, tx);
6040         }
6041 
6042         if (spa->spa_ubsync.ub_version < SPA_VERSION_DIR_CLONES &&
6043             spa->spa_uberblock.ub_version >= SPA_VERSION_DIR_CLONES) {
6044                 dsl_pool_upgrade_dir_clones(dp, tx);
6045 
6046                 /* Keeping the freedir open increases spa_minref */
6047                 spa->spa_minref += 3;
6048         }
6049 
6050         if (spa->spa_ubsync.ub_version < SPA_VERSION_FEATURES &&
6051             spa->spa_uberblock.ub_version >= SPA_VERSION_FEATURES) {
6052                 spa_feature_create_zap_objects(spa, tx);
6053         }
6054         rrw_exit(&dp->dp_config_rwlock, FTAG);
6055 }
6056 
6057 /*
6058  * Sync the specified transaction group.  New blocks may be dirtied as
6059  * part of the process, so we iterate until it converges.
6060  */
6061 void
6062 spa_sync(spa_t *spa, uint64_t txg)
6063 {
6064         dsl_pool_t *dp = spa->spa_dsl_pool;
6065         objset_t *mos = spa->spa_meta_objset;
6066         bpobj_t *defer_bpo = &spa->spa_deferred_bpobj;
6067         bplist_t *free_bpl = &spa->spa_free_bplist[txg & TXG_MASK];
6068         vdev_t *rvd = spa->spa_root_vdev;
6069         vdev_t *vd;
6070         dmu_tx_t *tx;
6071         int error;
6072 
6073         VERIFY(spa_writeable(spa));
6074 
6075         /*
6076          * Lock out configuration changes.
6077          */
6078         spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
6079 
6080         spa->spa_syncing_txg = txg;
6081         spa->spa_sync_pass = 0;
6082 
6083         /*
6084          * If there are any pending vdev state changes, convert them
6085          * into config changes that go out with this transaction group.
6086          */
6087         spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6088         while (list_head(&spa->spa_state_dirty_list) != NULL) {
6089                 /*
6090                  * We need the write lock here because, for aux vdevs,
6091                  * calling vdev_config_dirty() modifies sav_config.
6092                  * This is ugly and will become unnecessary when we
6093                  * eliminate the aux vdev wart by integrating all vdevs
6094                  * into the root vdev tree.
6095                  */
6096                 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6097                 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_WRITER);
6098                 while ((vd = list_head(&spa->spa_state_dirty_list)) != NULL) {
6099                         vdev_state_clean(vd);
6100                         vdev_config_dirty(vd);
6101                 }
6102                 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
6103                 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
6104         }
6105         spa_config_exit(spa, SCL_STATE, FTAG);
6106 
6107         tx = dmu_tx_create_assigned(dp, txg);
6108 
6109         spa->spa_sync_starttime = gethrtime();
6110         VERIFY(cyclic_reprogram(spa->spa_deadman_cycid,
6111             spa->spa_sync_starttime + spa->spa_deadman_synctime));
6112 
6113         /*
6114          * If we are upgrading to SPA_VERSION_RAIDZ_DEFLATE this txg,
6115          * set spa_deflate if we have no raid-z vdevs.
6116          */
6117         if (spa->spa_ubsync.ub_version < SPA_VERSION_RAIDZ_DEFLATE &&
6118             spa->spa_uberblock.ub_version >= SPA_VERSION_RAIDZ_DEFLATE) {
6119                 int i;
6120 
6121                 for (i = 0; i < rvd->vdev_children; i++) {
6122                         vd = rvd->vdev_child[i];
6123                         if (vd->vdev_deflate_ratio != SPA_MINBLOCKSIZE)
6124                                 break;
6125                 }
6126                 if (i == rvd->vdev_children) {
6127                         spa->spa_deflate = TRUE;
6128                         VERIFY(0 == zap_add(spa->spa_meta_objset,
6129                             DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_DEFLATE,
6130                             sizeof (uint64_t), 1, &spa->spa_deflate, tx));
6131                 }
6132         }
6133 
6134         /*
6135          * If anything has changed in this txg, or if someone is waiting
6136          * for this txg to sync (eg, spa_vdev_remove()), push the
6137          * deferred frees from the previous txg.  If not, leave them
6138          * alone so that we don't generate work on an otherwise idle
6139          * system.
6140          */
6141         if (!txg_list_empty(&dp->dp_dirty_datasets, txg) ||
6142             !txg_list_empty(&dp->dp_dirty_dirs, txg) ||
6143             !txg_list_empty(&dp->dp_sync_tasks, txg) ||
6144             ((dsl_scan_active(dp->dp_scan) ||
6145             txg_sync_waiting(dp)) && !spa_shutting_down(spa))) {
6146                 zio_t *zio = zio_root(spa, NULL, NULL, 0);
6147                 VERIFY3U(bpobj_iterate(defer_bpo,
6148                     spa_free_sync_cb, zio, tx), ==, 0);
6149                 VERIFY0(zio_wait(zio));
6150         }
6151 
6152         /*
6153          * Iterate to convergence.
6154          */
6155         do {
6156                 int pass = ++spa->spa_sync_pass;
6157 
6158                 spa_sync_config_object(spa, tx);
6159                 spa_sync_aux_dev(spa, &spa->spa_spares, tx,
6160                     ZPOOL_CONFIG_SPARES, DMU_POOL_SPARES);
6161                 spa_sync_aux_dev(spa, &spa->spa_l2cache, tx,
6162                     ZPOOL_CONFIG_L2CACHE, DMU_POOL_L2CACHE);
6163                 spa_errlog_sync(spa, txg);
6164                 dsl_pool_sync(dp, txg);
6165 
6166                 if (pass < zfs_sync_pass_deferred_free) {
6167                         zio_t *zio = zio_root(spa, NULL, NULL, 0);
6168                         bplist_iterate(free_bpl, spa_free_sync_cb,
6169                             zio, tx);
6170                         VERIFY(zio_wait(zio) == 0);
6171                 } else {
6172                         bplist_iterate(free_bpl, bpobj_enqueue_cb,
6173                             defer_bpo, tx);
6174                 }
6175 
6176                 ddt_sync(spa, txg);
6177                 dsl_scan_sync(dp, tx);
6178 
6179                 while (vd = txg_list_remove(&spa->spa_vdev_txg_list, txg))
6180                         vdev_sync(vd, txg);
6181 
6182                 if (pass == 1)
6183                         spa_sync_upgrades(spa, tx);
6184 
6185         } while (dmu_objset_is_dirty(mos, txg));
6186 
6187         /*
6188          * Rewrite the vdev configuration (which includes the uberblock)
6189          * to commit the transaction group.
6190          *
6191          * If there are no dirty vdevs, we sync the uberblock to a few
6192          * random top-level vdevs that are known to be visible in the
6193          * config cache (see spa_vdev_add() for a complete description).
6194          * If there *are* dirty vdevs, sync the uberblock to all vdevs.
6195          */
6196         for (;;) {
6197                 /*
6198                  * We hold SCL_STATE to prevent vdev open/close/etc.
6199                  * while we're attempting to write the vdev labels.
6200                  */
6201                 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
6202 
6203                 if (list_is_empty(&spa->spa_config_dirty_list)) {
6204                         vdev_t *svd[SPA_DVAS_PER_BP];
6205                         int svdcount = 0;
6206                         int children = rvd->vdev_children;
6207                         int c0 = spa_get_random(children);
6208 
6209                         for (int c = 0; c < children; c++) {
6210                                 vd = rvd->vdev_child[(c0 + c) % children];
6211                                 if (vd->vdev_ms_array == 0 || vd->vdev_islog)
6212                                         continue;
6213                                 svd[svdcount++] = vd;
6214                                 if (svdcount == SPA_DVAS_PER_BP)
6215                                         break;
6216                         }
6217                         error = vdev_config_sync(svd, svdcount, txg, B_FALSE);
6218                         if (error != 0)
6219                                 error = vdev_config_sync(svd, svdcount, txg,
6220                                     B_TRUE);
6221                 } else {
6222                         error = vdev_config_sync(rvd->vdev_child,
6223                             rvd->vdev_children, txg, B_FALSE);
6224                         if (error != 0)
6225                                 error = vdev_config_sync(rvd->vdev_child,
6226                                     rvd->vdev_children, txg, B_TRUE);
6227                 }
6228 
6229                 if (error == 0)
6230                         spa->spa_last_synced_guid = rvd->vdev_guid;
6231 
6232                 spa_config_exit(spa, SCL_STATE, FTAG);
6233 
6234                 if (error == 0)
6235                         break;
6236                 zio_suspend(spa, NULL);
6237                 zio_resume_wait(spa);
6238         }
6239         dmu_tx_commit(tx);
6240 
6241         VERIFY(cyclic_reprogram(spa->spa_deadman_cycid, CY_INFINITY));
6242 
6243         /*
6244          * Clear the dirty config list.
6245          */
6246         while ((vd = list_head(&spa->spa_config_dirty_list)) != NULL)
6247                 vdev_config_clean(vd);
6248 
6249         /*
6250          * Now that the new config has synced transactionally,
6251          * let it become visible to the config cache.
6252          */
6253         if (spa->spa_config_syncing != NULL) {
6254                 spa_config_set(spa, spa->spa_config_syncing);
6255                 spa->spa_config_txg = txg;
6256                 spa->spa_config_syncing = NULL;
6257         }
6258 
6259         spa->spa_ubsync = spa->spa_uberblock;
6260 
6261         dsl_pool_sync_done(dp, txg);
6262 
6263         /*
6264          * Update usable space statistics.
6265          */
6266         while (vd = txg_list_remove(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)))
6267                 vdev_sync_done(vd, txg);
6268 
6269         spa_update_dspace(spa);
6270 
6271         /*
6272          * It had better be the case that we didn't dirty anything
6273          * since vdev_config_sync().
6274          */
6275         ASSERT(txg_list_empty(&dp->dp_dirty_datasets, txg));
6276         ASSERT(txg_list_empty(&dp->dp_dirty_dirs, txg));
6277         ASSERT(txg_list_empty(&spa->spa_vdev_txg_list, txg));
6278 
6279         spa->spa_sync_pass = 0;
6280 
6281         spa_config_exit(spa, SCL_CONFIG, FTAG);
6282 
6283         spa_handle_ignored_writes(spa);
6284 
6285         /*
6286          * If any async tasks have been requested, kick them off.
6287          */
6288         spa_async_dispatch(spa);
6289 }
6290 
6291 /*
6292  * Sync all pools.  We don't want to hold the namespace lock across these
6293  * operations, so we take a reference on the spa_t and drop the lock during the
6294  * sync.
6295  */
6296 void
6297 spa_sync_allpools(void)
6298 {
6299         spa_t *spa = NULL;
6300         mutex_enter(&spa_namespace_lock);
6301         while ((spa = spa_next(spa)) != NULL) {
6302                 if (spa_state(spa) != POOL_STATE_ACTIVE ||
6303                     !spa_writeable(spa) || spa_suspended(spa))
6304                         continue;
6305                 spa_open_ref(spa, FTAG);
6306                 mutex_exit(&spa_namespace_lock);
6307                 txg_wait_synced(spa_get_dsl(spa), 0);
6308                 mutex_enter(&spa_namespace_lock);
6309                 spa_close(spa, FTAG);
6310         }
6311         mutex_exit(&spa_namespace_lock);
6312 }
6313 
6314 /*
6315  * ==========================================================================
6316  * Miscellaneous routines
6317  * ==========================================================================
6318  */
6319 
6320 /*
6321  * Remove all pools in the system.
6322  */
6323 void
6324 spa_evict_all(void)
6325 {
6326         spa_t *spa;
6327 
6328         /*
6329          * Remove all cached state.  All pools should be closed now,
6330          * so every spa in the AVL tree should be unreferenced.
6331          */
6332         mutex_enter(&spa_namespace_lock);
6333         while ((spa = spa_next(NULL)) != NULL) {
6334                 /*
6335                  * Stop async tasks.  The async thread may need to detach
6336                  * a device that's been replaced, which requires grabbing
6337                  * spa_namespace_lock, so we must drop it here.
6338                  */
6339                 spa_open_ref(spa, FTAG);
6340                 mutex_exit(&spa_namespace_lock);
6341                 spa_async_suspend(spa);
6342                 mutex_enter(&spa_namespace_lock);
6343                 spa_close(spa, FTAG);
6344 
6345                 if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
6346                         spa_unload(spa);
6347                         spa_deactivate(spa);
6348                 }
6349                 spa_remove(spa);
6350         }
6351         mutex_exit(&spa_namespace_lock);
6352 }
6353 
6354 vdev_t *
6355 spa_lookup_by_guid(spa_t *spa, uint64_t guid, boolean_t aux)
6356 {
6357         vdev_t *vd;
6358         int i;
6359 
6360         if ((vd = vdev_lookup_by_guid(spa->spa_root_vdev, guid)) != NULL)
6361                 return (vd);
6362 
6363         if (aux) {
6364                 for (i = 0; i < spa->spa_l2cache.sav_count; i++) {
6365                         vd = spa->spa_l2cache.sav_vdevs[i];
6366                         if (vd->vdev_guid == guid)
6367                                 return (vd);
6368                 }
6369 
6370                 for (i = 0; i < spa->spa_spares.sav_count; i++) {
6371                         vd = spa->spa_spares.sav_vdevs[i];
6372                         if (vd->vdev_guid == guid)
6373                                 return (vd);
6374                 }
6375         }
6376 
6377         return (NULL);
6378 }
6379 
6380 void
6381 spa_upgrade(spa_t *spa, uint64_t version)
6382 {
6383         ASSERT(spa_writeable(spa));
6384 
6385         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
6386 
6387         /*
6388          * This should only be called for a non-faulted pool, and since a
6389          * future version would result in an unopenable pool, this shouldn't be
6390          * possible.
6391          */
6392         ASSERT(SPA_VERSION_IS_SUPPORTED(spa->spa_uberblock.ub_version));
6393         ASSERT(version >= spa->spa_uberblock.ub_version);
6394 
6395         spa->spa_uberblock.ub_version = version;
6396         vdev_config_dirty(spa->spa_root_vdev);
6397 
6398         spa_config_exit(spa, SCL_ALL, FTAG);
6399 
6400         txg_wait_synced(spa_get_dsl(spa), 0);
6401 }
6402 
6403 boolean_t
6404 spa_has_spare(spa_t *spa, uint64_t guid)
6405 {
6406         int i;
6407         uint64_t spareguid;
6408         spa_aux_vdev_t *sav = &spa->spa_spares;
6409 
6410         for (i = 0; i < sav->sav_count; i++)
6411                 if (sav->sav_vdevs[i]->vdev_guid == guid)
6412                         return (B_TRUE);
6413 
6414         for (i = 0; i < sav->sav_npending; i++) {
6415                 if (nvlist_lookup_uint64(sav->sav_pending[i], ZPOOL_CONFIG_GUID,
6416                     &spareguid) == 0 && spareguid == guid)
6417                         return (B_TRUE);
6418         }
6419 
6420         return (B_FALSE);
6421 }
6422 
6423 /*
6424  * Check if a pool has an active shared spare device.
6425  * Note: reference count of an active spare is 2, as a spare and as a replace
6426  */
6427 static boolean_t
6428 spa_has_active_shared_spare(spa_t *spa)
6429 {
6430         int i, refcnt;
6431         uint64_t pool;
6432         spa_aux_vdev_t *sav = &spa->spa_spares;
6433 
6434         for (i = 0; i < sav->sav_count; i++) {
6435                 if (spa_spare_exists(sav->sav_vdevs[i]->vdev_guid, &pool,
6436                     &refcnt) && pool != 0ULL && pool == spa_guid(spa) &&
6437                     refcnt > 2)
6438                         return (B_TRUE);
6439         }
6440 
6441         return (B_FALSE);
6442 }
6443 
6444 /*
6445  * Post a sysevent corresponding to the given event.  The 'name' must be one of
6446  * the event definitions in sys/sysevent/eventdefs.h.  The payload will be
6447  * filled in from the spa and (optionally) the vdev.  This doesn't do anything
6448  * in the userland libzpool, as we don't want consumers to misinterpret ztest
6449  * or zdb as real changes.
6450  */
6451 void
6452 spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
6453 {
6454 #ifdef _KERNEL
6455         sysevent_t              *ev;
6456         sysevent_attr_list_t    *attr = NULL;
6457         sysevent_value_t        value;
6458         sysevent_id_t           eid;
6459 
6460         ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
6461             SE_SLEEP);
6462 
6463         value.value_type = SE_DATA_TYPE_STRING;
6464         value.value.sv_string = spa_name(spa);
6465         if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
6466                 goto done;
6467 
6468         value.value_type = SE_DATA_TYPE_UINT64;
6469         value.value.sv_uint64 = spa_guid(spa);
6470         if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
6471                 goto done;
6472 
6473         if (vd) {
6474                 value.value_type = SE_DATA_TYPE_UINT64;
6475                 value.value.sv_uint64 = vd->vdev_guid;
6476                 if (sysevent_add_attr(&attr, ZFS_EV_VDEV_GUID, &value,
6477                     SE_SLEEP) != 0)
6478                         goto done;
6479 
6480                 if (vd->vdev_path) {
6481                         value.value_type = SE_DATA_TYPE_STRING;
6482                         value.value.sv_string = vd->vdev_path;
6483                         if (sysevent_add_attr(&attr, ZFS_EV_VDEV_PATH,
6484                             &value, SE_SLEEP) != 0)
6485                                 goto done;
6486                 }
6487         }
6488 
6489         if (sysevent_attach_attributes(ev, attr) != 0)
6490                 goto done;
6491         attr = NULL;
6492 
6493         (void) log_sysevent(ev, SE_SLEEP, &eid);
6494 
6495 done:
6496         if (attr)
6497                 sysevent_free_attr(attr);
6498         sysevent_free(ev);
6499 #endif
6500 }