]> git.proxmox.com Git - mirror_zfs.git/blobdiff - module/zfs/spa.c
Report pool suspended due to MMP
[mirror_zfs.git] / module / zfs / spa.c
index f82847c82fd5aa8e22ce80106e1353cd7b7f618e..561f4d04bfc9c0a17354880e419157afb06126ed 100644 (file)
 
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
  * Copyright (c) 2015, Nexenta Systems, Inc.  All rights reserved.
  * Copyright (c) 2013, 2014, Nexenta Systems, Inc.  All rights reserved.
  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  * Copyright 2013 Saso Kiselkov. All rights reserved.
+ * Copyright (c) 2014 Integros [integros.com]
+ * Copyright 2016 Toomas Soome <tsoome@me.com>
  * Copyright (c) 2016 Actifio, Inc. All rights reserved.
+ * Copyright (c) 2017 Datto Inc.
+ * Copyright 2017 Joyent, Inc.
  */
 
 /*
@@ -51,6 +55,7 @@
 #include <sys/vdev_disk.h>
 #include <sys/metaslab.h>
 #include <sys/metaslab_impl.h>
+#include <sys/mmp.h>
 #include <sys/uberblock_impl.h>
 #include <sys/txg.h>
 #include <sys/avl.h>
@@ -74,6 +79,8 @@
 #include <sys/zvol.h>
 
 #ifdef _KERNEL
+#include <sys/fm/protocol.h>
+#include <sys/fm/util.h>
 #include <sys/bootprops.h>
 #include <sys/callb.h>
 #include <sys/cpupart.h>
@@ -143,6 +150,9 @@ const zio_taskq_info_t zio_taskqs[ZIO_TYPES][ZIO_TASKQ_TYPES] = {
        { ZTI_ONE,      ZTI_NULL,       ZTI_ONE,        ZTI_NULL }, /* IOCTL */
 };
 
+static sysevent_t *spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl,
+    const char *name);
+static void spa_event_post(sysevent_t *ev);
 static void spa_sync_version(void *arg, dmu_tx_t *tx);
 static void spa_sync_props(void *arg, dmu_tx_t *tx);
 static boolean_t spa_has_active_shared_spare(spa_t *spa);
@@ -482,6 +492,16 @@ spa_prop_validate(spa_t *spa, nvlist_t *props)
                                error = SET_ERROR(EINVAL);
                        break;
 
+               case ZPOOL_PROP_MULTIHOST:
+                       error = nvpair_value_uint64(elem, &intval);
+                       if (!error && intval > 1)
+                               error = SET_ERROR(EINVAL);
+
+                       if (!error && !spa_get_hostid())
+                               error = SET_ERROR(ENOTSUP);
+
+                       break;
+
                case ZPOOL_PROP_BOOTFS:
                        /*
                         * If the pool version is less than SPA_VERSION_BOOTFS,
@@ -534,12 +554,6 @@ spa_prop_validate(spa_t *spa, nvlist_t *props)
                                    &propval)) == 0 &&
                                    !BOOTFS_COMPRESS_VALID(propval)) {
                                        error = SET_ERROR(ENOTSUP);
-                               } else if ((error =
-                                   dsl_prop_get_int_ds(dmu_objset_ds(os),
-                                   zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
-                                   &propval)) == 0 &&
-                                   propval > SPA_OLD_MAXBLOCKSIZE) {
-                                       error = SET_ERROR(ENOTSUP);
                                } else if ((error =
                                    dsl_prop_get_int_ds(dmu_objset_ds(os),
                                    zfs_prop_to_name(ZFS_PROP_DNODESIZE),
@@ -804,7 +818,7 @@ spa_change_guid(spa_t *spa)
 
        if (error == 0) {
                spa_config_sync(spa, B_FALSE, B_TRUE);
-               spa_event_notify(spa, NULL, ESC_ZFS_POOL_REGUID);
+               spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_REGUID);
        }
 
        mutex_exit(&spa_namespace_lock);
@@ -1140,7 +1154,7 @@ spa_activate(spa_t *spa, int mode)
        list_create(&spa->spa_state_dirty_list, sizeof (vdev_t),
            offsetof(vdev_t, vdev_state_dirty_node));
 
-       txg_list_create(&spa->spa_vdev_txg_list,
+       txg_list_create(&spa->spa_vdev_txg_list, spa,
            offsetof(struct vdev, vdev_txg_node));
 
        avl_create(&spa->spa_errlist_scrub,
@@ -1343,6 +1357,9 @@ spa_unload(spa_t *spa)
                spa_config_exit(spa, SCL_ALL, FTAG);
        }
 
+       if (spa->spa_mmp.mmp_thread)
+               mmp_thread_stop(spa);
+
        /*
         * Wait for any outstanding async I/O to complete.
         */
@@ -1704,7 +1721,7 @@ spa_check_removed(vdev_t *vd)
        if (vd->vdev_ops->vdev_op_leaf && vdev_is_dead(vd) &&
            !vd->vdev_ishole) {
                zfs_post_autoreplace(vd->vdev_spa, vd);
-               spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_CHECK);
+               spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_CHECK);
        }
 }
 
@@ -2320,6 +2337,201 @@ vdev_count_verify_zaps(vdev_t *vd)
 }
 #endif
 
+/*
+ * Determine whether the activity check is required.
+ */
+static boolean_t
+spa_activity_check_required(spa_t *spa, uberblock_t *ub, nvlist_t *label,
+    nvlist_t *config)
+{
+       uint64_t state = 0;
+       uint64_t hostid = 0;
+       uint64_t tryconfig_txg = 0;
+       uint64_t tryconfig_timestamp = 0;
+       nvlist_t *nvinfo;
+
+       if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) {
+               nvinfo = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO);
+               (void) nvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG,
+                   &tryconfig_txg);
+               (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_TIMESTAMP,
+                   &tryconfig_timestamp);
+       }
+
+       (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE, &state);
+
+       /*
+        * Disable the MMP activity check - This is used by zdb which
+        * is intended to be used on potentially active pools.
+        */
+       if (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP)
+               return (B_FALSE);
+
+       /*
+        * Skip the activity check when the MMP feature is disabled.
+        */
+       if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay == 0)
+               return (B_FALSE);
+       /*
+        * If the tryconfig_* values are nonzero, they are the results of an
+        * earlier tryimport.  If they match the uberblock we just found, then
+        * the pool has not changed and we return false so we do not test a
+        * second time.
+        */
+       if (tryconfig_txg && tryconfig_txg == ub->ub_txg &&
+           tryconfig_timestamp && tryconfig_timestamp == ub->ub_timestamp)
+               return (B_FALSE);
+
+       /*
+        * Allow the activity check to be skipped when importing the pool
+        * on the same host which last imported it.  Since the hostid from
+        * configuration may be stale use the one read from the label.
+        */
+       if (nvlist_exists(label, ZPOOL_CONFIG_HOSTID))
+               hostid = fnvlist_lookup_uint64(label, ZPOOL_CONFIG_HOSTID);
+
+       if (hostid == spa_get_hostid())
+               return (B_FALSE);
+
+       /*
+        * Skip the activity test when the pool was cleanly exported.
+        */
+       if (state != POOL_STATE_ACTIVE)
+               return (B_FALSE);
+
+       return (B_TRUE);
+}
+
+/*
+ * Perform the import activity check.  If the user canceled the import or
+ * we detected activity then fail.
+ */
+static int
+spa_activity_check(spa_t *spa, uberblock_t *ub, nvlist_t *config)
+{
+       uint64_t import_intervals = MAX(zfs_multihost_import_intervals, 1);
+       uint64_t txg = ub->ub_txg;
+       uint64_t timestamp = ub->ub_timestamp;
+       uint64_t import_delay = NANOSEC;
+       hrtime_t import_expire;
+       nvlist_t *mmp_label = NULL;
+       vdev_t *rvd = spa->spa_root_vdev;
+       kcondvar_t cv;
+       kmutex_t mtx;
+       int error = 0;
+
+       cv_init(&cv, NULL, CV_DEFAULT, NULL);
+       mutex_init(&mtx, NULL, MUTEX_DEFAULT, NULL);
+       mutex_enter(&mtx);
+
+       /*
+        * If ZPOOL_CONFIG_MMP_TXG is present an activity check was performed
+        * during the earlier tryimport.  If the txg recorded there is 0 then
+        * the pool is known to be active on another host.
+        *
+        * Otherwise, the pool might be in use on another node.  Check for
+        * changes in the uberblocks on disk if necessary.
+        */
+       if (nvlist_exists(config, ZPOOL_CONFIG_LOAD_INFO)) {
+               nvlist_t *nvinfo = fnvlist_lookup_nvlist(config,
+                   ZPOOL_CONFIG_LOAD_INFO);
+
+               if (nvlist_exists(nvinfo, ZPOOL_CONFIG_MMP_TXG) &&
+                   fnvlist_lookup_uint64(nvinfo, ZPOOL_CONFIG_MMP_TXG) == 0) {
+                       vdev_uberblock_load(rvd, ub, &mmp_label);
+                       error = SET_ERROR(EREMOTEIO);
+                       goto out;
+               }
+       }
+
+       /*
+        * Preferentially use the zfs_multihost_interval from the node which
+        * last imported the pool.  This value is stored in an MMP uberblock as.
+        *
+        * ub_mmp_delay * vdev_count_leaves() == zfs_multihost_interval
+        */
+       if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay)
+               import_delay = MAX(import_delay, import_intervals *
+                   ub->ub_mmp_delay * MAX(vdev_count_leaves(spa), 1));
+
+       /* Apply a floor using the local default values. */
+       import_delay = MAX(import_delay, import_intervals *
+           MSEC2NSEC(MAX(zfs_multihost_interval, MMP_MIN_INTERVAL)));
+
+       /* Add a small random factor in case of simultaneous imports (0-25%) */
+       import_expire = gethrtime() + import_delay +
+           (import_delay * spa_get_random(250) / 1000);
+
+       while (gethrtime() < import_expire) {
+               vdev_uberblock_load(rvd, ub, &mmp_label);
+
+               if (txg != ub->ub_txg || timestamp != ub->ub_timestamp) {
+                       error = SET_ERROR(EREMOTEIO);
+                       break;
+               }
+
+               if (mmp_label) {
+                       nvlist_free(mmp_label);
+                       mmp_label = NULL;
+               }
+
+               error = cv_timedwait_sig(&cv, &mtx, ddi_get_lbolt() + hz);
+               if (error != -1) {
+                       error = SET_ERROR(EINTR);
+                       break;
+               }
+               error = 0;
+       }
+
+out:
+       mutex_exit(&mtx);
+       mutex_destroy(&mtx);
+       cv_destroy(&cv);
+
+       /*
+        * If the pool is determined to be active store the status in the
+        * spa->spa_load_info nvlist.  If the remote hostname or hostid are
+        * available from configuration read from disk store them as well.
+        * This allows 'zpool import' to generate a more useful message.
+        *
+        * ZPOOL_CONFIG_MMP_STATE    - observed pool status (mandatory)
+        * ZPOOL_CONFIG_MMP_HOSTNAME - hostname from the active pool
+        * ZPOOL_CONFIG_MMP_HOSTID   - hostid from the active pool
+        */
+       if (error == EREMOTEIO) {
+               char *hostname = "<unknown>";
+               uint64_t hostid = 0;
+
+               if (mmp_label) {
+                       if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTNAME)) {
+                               hostname = fnvlist_lookup_string(mmp_label,
+                                   ZPOOL_CONFIG_HOSTNAME);
+                               fnvlist_add_string(spa->spa_load_info,
+                                   ZPOOL_CONFIG_MMP_HOSTNAME, hostname);
+                       }
+
+                       if (nvlist_exists(mmp_label, ZPOOL_CONFIG_HOSTID)) {
+                               hostid = fnvlist_lookup_uint64(mmp_label,
+                                   ZPOOL_CONFIG_HOSTID);
+                               fnvlist_add_uint64(spa->spa_load_info,
+                                   ZPOOL_CONFIG_MMP_HOSTID, hostid);
+                       }
+               }
+
+               fnvlist_add_uint64(spa->spa_load_info,
+                   ZPOOL_CONFIG_MMP_STATE, MMP_STATE_ACTIVE);
+               fnvlist_add_uint64(spa->spa_load_info,
+                   ZPOOL_CONFIG_MMP_TXG, 0);
+
+               error = spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO);
+       }
+
+       if (mmp_label)
+               nvlist_free(mmp_label);
+
+       return (error);
+}
+
 /*
  * Load an existing storage pool, using the pool's builtin spa_config as a
  * source of configuration information.
@@ -2340,6 +2552,7 @@ spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
        int parse, i;
        uint64_t obj;
        boolean_t missing_feat_write = B_FALSE;
+       boolean_t activity_check = B_FALSE;
        nvlist_t *mos_config;
 
        /*
@@ -2437,6 +2650,33 @@ spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
                return (spa_vdev_err(rvd, VDEV_AUX_CORRUPT_DATA, ENXIO));
        }
 
+       /*
+        * For pools which have the multihost property on determine if the
+        * pool is truly inactive and can be safely imported.  Prevent
+        * hosts which don't have a hostid set from importing the pool.
+        */
+       activity_check = spa_activity_check_required(spa, ub, label, config);
+       if (activity_check) {
+               if (ub->ub_mmp_magic == MMP_MAGIC && ub->ub_mmp_delay &&
+                   spa_get_hostid() == 0) {
+                       nvlist_free(label);
+                       fnvlist_add_uint64(spa->spa_load_info,
+                           ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID);
+                       return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO));
+               }
+
+               error = spa_activity_check(spa, ub, config);
+               if (error) {
+                       nvlist_free(label);
+                       return (error);
+               }
+
+               fnvlist_add_uint64(spa->spa_load_info,
+                   ZPOOL_CONFIG_MMP_STATE, MMP_STATE_INACTIVE);
+               fnvlist_add_uint64(spa->spa_load_info,
+                   ZPOOL_CONFIG_MMP_TXG, ub->ub_txg);
+       }
+
        /*
         * If the pool has an unsupported version we can't open it.
         */
@@ -2664,24 +2904,9 @@ spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
                        VERIFY(nvlist_lookup_string(nvconfig,
                            ZPOOL_CONFIG_HOSTNAME, &hostname) == 0);
 
-#ifdef _KERNEL
-                       myhostid = zone_get_hostid(NULL);
-#else  /* _KERNEL */
-                       /*
-                        * We're emulating the system's hostid in userland, so
-                        * we can't use zone_get_hostid().
-                        */
-                       (void) ddi_strtoul(hw_serial, NULL, 10, &myhostid);
-#endif /* _KERNEL */
-                       if (hostid != 0 && myhostid != 0 &&
-                           hostid != myhostid) {
+                       myhostid = spa_get_hostid();
+                       if (hostid && myhostid && hostid != myhostid) {
                                nvlist_free(nvconfig);
-                               cmn_err(CE_WARN, "pool '%s' could not be "
-                                   "loaded as it was last accessed by another "
-                                   "system (host: %s hostid: 0x%lx). See: "
-                                   "http://zfsonlinux.org/msg/ZFS-8000-EY",
-                                   spa_name(spa), hostname,
-                                   (unsigned long)hostid);
                                return (SET_ERROR(EBADF));
                        }
                }
@@ -2847,12 +3072,25 @@ spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
                spa_prop_find(spa, ZPOOL_PROP_DELEGATION, &spa->spa_delegation);
                spa_prop_find(spa, ZPOOL_PROP_FAILUREMODE, &spa->spa_failmode);
                spa_prop_find(spa, ZPOOL_PROP_AUTOEXPAND, &spa->spa_autoexpand);
+               spa_prop_find(spa, ZPOOL_PROP_MULTIHOST, &spa->spa_multihost);
                spa_prop_find(spa, ZPOOL_PROP_DEDUPDITTO,
                    &spa->spa_dedup_ditto);
 
                spa->spa_autoreplace = (autoreplace != 0);
        }
 
+       /*
+        * If the 'multihost' property is set, then never allow a pool to
+        * be imported when the system hostid is zero.  The exception to
+        * this rule is zdb which is always allowed to access pools.
+        */
+       if (spa_multihost(spa) && spa_get_hostid() == 0 &&
+           (spa->spa_import_flags & ZFS_IMPORT_SKIP_MMP) == 0) {
+               fnvlist_add_uint64(spa->spa_load_info,
+                   ZPOOL_CONFIG_MMP_STATE, MMP_STATE_NO_HOSTID);
+               return (spa_vdev_err(rvd, VDEV_AUX_ACTIVE, EREMOTEIO));
+       }
+
        /*
         * If the 'autoreplace' property is set, then post a resource notifying
         * the ZFS DE that it should not issue any faults for unopenable
@@ -2977,6 +3215,7 @@ spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
                spa_set_log_state(spa, SPA_LOG_GOOD);
                spa->spa_sync_on = B_TRUE;
                txg_sync_start(spa->spa_dsl_pool);
+               mmp_thread_start(spa);
 
                /*
                 * Wait for all claims to sync.  We sync up to the highest
@@ -3022,7 +3261,7 @@ spa_load_impl(spa_t *spa, uint64_t pool_guid, nvlist_t *config,
                 * Log the fact that we booted up (so that we can detect if
                 * we rebooted in the middle of an operation).
                 */
-               spa_history_log_version(spa, "open");
+               spa_history_log_version(spa, "open", NULL);
 
                /*
                 * Delete any inconsistent datasets.
@@ -3539,10 +3778,14 @@ spa_get_stats(const char *name, nvlist_t **config,
                            ZPOOL_CONFIG_ERRCOUNT,
                            spa_get_errlog_size(spa)) == 0);
 
-                       if (spa_suspended(spa))
+                       if (spa_suspended(spa)) {
                                VERIFY(nvlist_add_uint64(*config,
                                    ZPOOL_CONFIG_SUSPENDED,
                                    spa->spa_failmode) == 0);
+                               VERIFY(nvlist_add_uint64(*config,
+                                   ZPOOL_CONFIG_SUSPENDED_REASON,
+                                   spa->spa_suspended) == 0);
+                       }
 
                        spa_add_spares(spa, *config);
                        spa_add_l2cache(spa, *config);
@@ -3629,18 +3872,6 @@ spa_validate_aux_devs(spa_t *spa, nvlist_t *nvroot, uint64_t crtxg, int mode,
                        goto out;
                }
 
-               /*
-                * The L2ARC currently only supports disk devices in
-                * kernel context.  For user-level testing, we allow it.
-                */
-#ifdef _KERNEL
-               if ((strcmp(config, ZPOOL_CONFIG_L2CACHE) == 0) &&
-                   strcmp(vd->vdev_ops->vdev_op_type, VDEV_TYPE_DISK) != 0) {
-                       error = SET_ERROR(ENOTBLK);
-                       vdev_free(vd);
-                       goto out;
-               }
-#endif
                vd->vdev_top = vd;
 
                if ((error = vdev_open(vd)) == 0 &&
@@ -3916,6 +4147,15 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
 
        tx = dmu_tx_create_assigned(dp, txg);
 
+       /*
+        * Create the pool's history object.
+        */
+       if (version >= SPA_VERSION_ZPOOL_HISTORY && !spa->spa_history)
+               spa_history_create_obj(spa, tx);
+
+       spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_CREATE);
+       spa_history_log_version(spa, "create", tx);
+
        /*
         * Create the pool config object.
         */
@@ -3964,12 +4204,6 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
        VERIFY3U(0, ==, bpobj_open(&spa->spa_deferred_bpobj,
            spa->spa_meta_objset, obj));
 
-       /*
-        * Create the pool's history object.
-        */
-       if (version >= SPA_VERSION_ZPOOL_HISTORY)
-               spa_history_create_obj(spa, tx);
-
        /*
         * Generate some random noise for salted checksums to operate on.
         */
@@ -3983,6 +4217,7 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
        spa->spa_delegation = zpool_prop_default_numeric(ZPOOL_PROP_DELEGATION);
        spa->spa_failmode = zpool_prop_default_numeric(ZPOOL_PROP_FAILUREMODE);
        spa->spa_autoexpand = zpool_prop_default_numeric(ZPOOL_PROP_AUTOEXPAND);
+       spa->spa_multihost = zpool_prop_default_numeric(ZPOOL_PROP_MULTIHOST);
 
        if (props != NULL) {
                spa_configfile_set(spa, props, B_FALSE);
@@ -3993,6 +4228,7 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
 
        spa->spa_sync_on = B_TRUE;
        txg_sync_start(spa->spa_dsl_pool);
+       mmp_thread_start(spa);
 
        /*
         * We explicitly wait for the first transaction to complete so that our
@@ -4001,9 +4237,6 @@ spa_create(const char *pool, nvlist_t *nvroot, nvlist_t *props,
        txg_wait_synced(spa->spa_dsl_pool, txg);
 
        spa_config_sync(spa, B_FALSE, B_TRUE);
-       spa_event_notify(spa, NULL, ESC_ZFS_POOL_CREATE);
-
-       spa_history_log_version(spa, "create");
 
        /*
         * Don't count references from objsets that are already closed
@@ -4065,7 +4298,7 @@ spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
                        spa_configfile_set(spa, props, B_FALSE);
 
                spa_config_sync(spa, B_FALSE, B_TRUE);
-               spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
+               spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
 
                mutex_exit(&spa_namespace_lock);
                return (0);
@@ -4118,12 +4351,6 @@ spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
 
        VERIFY(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
            &nvroot) == 0);
-       if (error == 0)
-               error = spa_validate_aux(spa, nvroot, -1ULL,
-                   VDEV_ALLOC_SPARE);
-       if (error == 0)
-               error = spa_validate_aux(spa, nvroot, -1ULL,
-                   VDEV_ALLOC_L2CACHE);
        spa_config_exit(spa, SCL_ALL, FTAG);
 
        if (props != NULL)
@@ -4196,9 +4423,9 @@ spa_import(char *pool, nvlist_t *config, nvlist_t *props, uint64_t flags)
         */
        spa_async_request(spa, SPA_ASYNC_AUTOEXPAND);
 
-       spa_history_log_version(spa, "import");
+       spa_history_log_version(spa, "import", NULL);
 
-       spa_event_notify(spa, NULL, ESC_ZFS_POOL_IMPORT);
+       spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_IMPORT);
 
        zvol_create_minors(spa, pool, B_TRUE);
 
@@ -4399,7 +4626,10 @@ spa_export_common(char *pool, int new_state, nvlist_t **oldconfig,
        }
 
 export_spa:
-       spa_event_notify(spa, NULL, ESC_ZFS_POOL_DESTROY);
+       if (new_state == POOL_STATE_DESTROYED)
+               spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_DESTROY);
+       else if (new_state == POOL_STATE_EXPORTED)
+               spa_event_notify(spa, NULL, NULL, ESC_ZFS_POOL_EXPORT);
 
        if (spa->spa_state != POOL_STATE_UNINITIALIZED) {
                spa_unload(spa);
@@ -4555,7 +4785,7 @@ spa_vdev_add(spa_t *spa, nvlist_t *nvroot)
 
        mutex_enter(&spa_namespace_lock);
        spa_config_update(spa, SPA_CONFIG_UPDATE_POOL);
-       spa_event_notify(spa, NULL, ESC_ZFS_VDEV_ADD);
+       spa_event_notify(spa, NULL, NULL, ESC_ZFS_VDEV_ADD);
        mutex_exit(&spa_namespace_lock);
 
        return (0);
@@ -4713,6 +4943,11 @@ spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
        newvd->vdev_crtxg = oldvd->vdev_crtxg;
        vdev_add_child(pvd, newvd);
 
+       /*
+        * Reevaluate the parent vdev state.
+        */
+       vdev_propagate_state(pvd);
+
        tvd = newvd->vdev_top;
        ASSERT(pvd->vdev_top == tvd);
        ASSERT(tvd->vdev_parent == rvd);
@@ -4731,7 +4966,7 @@ spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
 
        if (newvd->vdev_isspare) {
                spa_spare_activate(newvd);
-               spa_event_notify(spa, newvd, ESC_ZFS_VDEV_SPARE);
+               spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_SPARE);
        }
 
        oldvdpath = spa_strdup(oldvd->vdev_path);
@@ -4751,9 +4986,9 @@ spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot, int replacing)
        dsl_resilver_restart(spa->spa_dsl_pool, dtl_max_txg);
 
        if (spa->spa_bootfs)
-               spa_event_notify(spa, newvd, ESC_ZFS_BOOTFS_VDEV_ATTACH);
+               spa_event_notify(spa, newvd, NULL, ESC_ZFS_BOOTFS_VDEV_ATTACH);
 
-       spa_event_notify(spa, newvd, ESC_ZFS_VDEV_ATTACH);
+       spa_event_notify(spa, newvd, NULL, ESC_ZFS_VDEV_ATTACH);
 
        /*
         * Commit the config
@@ -4967,7 +5202,7 @@ spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid, int replace_done)
        vd->vdev_detached = B_TRUE;
        vdev_dirty(tvd, VDD_DTL, vd, txg);
 
-       spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE);
+       spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE);
 
        /* hang on to the spa before we release the lock */
        spa_open_ref(spa, FTAG);
@@ -5453,6 +5688,7 @@ int
 spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
 {
        vdev_t *vd;
+       sysevent_t *ev = NULL;
        metaslab_group_t *mg;
        nvlist_t **spares, **l2cache, *nv;
        uint64_t txg = 0;
@@ -5476,6 +5712,10 @@ spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
                 * in this pool.
                 */
                if (vd == NULL || unspare) {
+                       if (vd == NULL)
+                               vd = spa_lookup_by_guid(spa, guid, B_TRUE);
+                       ev = spa_event_create(spa, vd, NULL,
+                           ESC_ZFS_VDEV_REMOVE_AUX);
                        spa_vdev_remove_aux(spa->spa_spares.sav_config,
                            ZPOOL_CONFIG_SPARES, spares, nspares, nv);
                        spa_load_spares(spa);
@@ -5483,7 +5723,6 @@ spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
                } else {
                        error = SET_ERROR(EBUSY);
                }
-               spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX);
        } else if (spa->spa_l2cache.sav_vdevs != NULL &&
            nvlist_lookup_nvlist_array(spa->spa_l2cache.sav_config,
            ZPOOL_CONFIG_L2CACHE, &l2cache, &nl2cache) == 0 &&
@@ -5491,11 +5730,12 @@ spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
                /*
                 * Cache devices can always be removed.
                 */
+               vd = spa_lookup_by_guid(spa, guid, B_TRUE);
+               ev = spa_event_create(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE_AUX);
                spa_vdev_remove_aux(spa->spa_l2cache.sav_config,
                    ZPOOL_CONFIG_L2CACHE, l2cache, nl2cache, nv);
                spa_load_l2cache(spa);
                spa->spa_l2cache.sav_sync = B_TRUE;
-               spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE_AUX);
        } else if (vd != NULL && vd->vdev_islog) {
                ASSERT(!locked);
                ASSERT(vd == vd->vdev_top);
@@ -5532,9 +5772,9 @@ spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
                /*
                 * Clean up the vdev namespace.
                 */
+               ev = spa_event_create(spa, vd, NULL, ESC_ZFS_VDEV_REMOVE_DEV);
                spa_vdev_remove_from_namespace(spa, vd);
 
-               spa_event_notify(spa, vd, ESC_ZFS_VDEV_REMOVE_DEV);
        } else if (vd != NULL) {
                /*
                 * Normal vdevs cannot be removed (yet).
@@ -5550,6 +5790,9 @@ spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare)
        if (!locked)
                error = spa_vdev_exit(spa, NULL, txg, error);
 
+       if (ev)
+               spa_event_post(ev);
+
        return (error);
 }
 
@@ -5728,6 +5971,16 @@ spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru)
  * SPA Scanning
  * ==========================================================================
  */
+int
+spa_scrub_pause_resume(spa_t *spa, pool_scrub_cmd_t cmd)
+{
+       ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == 0);
+
+       if (dsl_scan_resilvering(spa->spa_dsl_pool))
+               return (SET_ERROR(EBUSY));
+
+       return (dsl_scrub_set_pause_resume(spa->spa_dsl_pool, cmd));
+}
 
 int
 spa_scan_stop(spa_t *spa)
@@ -5822,7 +6075,7 @@ spa_async_autoexpand(spa_t *spa, vdev_t *vd)
        if (!vd->vdev_ops->vdev_op_leaf || vd->vdev_physpath == NULL)
                return;
 
-       spa_event_notify(vd->vdev_spa, vd, ESC_ZFS_VDEV_AUTOEXPAND);
+       spa_event_notify(vd->vdev_spa, vd, NULL, ESC_ZFS_VDEV_AUTOEXPAND);
 }
 
 static void
@@ -6385,6 +6638,9 @@ spa_sync_props(void *arg, dmu_tx_t *tx)
                                        spa_async_request(spa,
                                            SPA_ASYNC_AUTOEXPAND);
                                break;
+                       case ZPOOL_PROP_MULTIHOST:
+                               spa->spa_multihost = intval;
+                               break;
                        case ZPOOL_PROP_DEDUPDITTO:
                                spa->spa_dedup_ditto = intval;
                                break;
@@ -6717,7 +6973,7 @@ spa_sync(spa_t *spa, uint64_t txg)
 
                if (error == 0)
                        break;
-               zio_suspend(spa, NULL);
+               zio_suspend(spa, NULL, ZIO_SUSPEND_IOERR);
                zio_resume_wait(spa);
        }
        dmu_tx_commit(tx);
@@ -6741,8 +6997,6 @@ spa_sync(spa_t *spa, uint64_t txg)
                spa->spa_config_syncing = NULL;
        }
 
-       spa->spa_ubsync = spa->spa_uberblock;
-
        dsl_pool_sync_done(dp, txg);
 
        mutex_enter(&spa->spa_alloc_lock);
@@ -6767,6 +7021,13 @@ spa_sync(spa_t *spa, uint64_t txg)
 
        spa->spa_sync_pass = 0;
 
+       /*
+        * Update the last synced uberblock here. We want to do this at
+        * the end of spa_sync() so that consumers of spa_last_synced_txg()
+        * will be guaranteed that all the processing associated with
+        * that txg has been completed.
+        */
+       spa->spa_ubsync = spa->spa_uberblock;
        spa_config_exit(spa, SCL_CONFIG, FTAG);
 
        spa_handle_ignored_writes(spa);
@@ -6930,6 +7191,33 @@ spa_has_active_shared_spare(spa_t *spa)
        return (B_FALSE);
 }
 
+static sysevent_t *
+spa_event_create(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
+{
+       sysevent_t *ev = NULL;
+#ifdef _KERNEL
+       nvlist_t *resource;
+
+       resource = zfs_event_create(spa, vd, FM_SYSEVENT_CLASS, name, hist_nvl);
+       if (resource) {
+               ev = kmem_alloc(sizeof (sysevent_t), KM_SLEEP);
+               ev->resource = resource;
+       }
+#endif
+       return (ev);
+}
+
+static void
+spa_event_post(sysevent_t *ev)
+{
+#ifdef _KERNEL
+       if (ev) {
+               zfs_zevent_post(ev->resource, NULL, zfs_zevent_post_cb);
+               kmem_free(ev, sizeof (*ev));
+       }
+#endif
+}
+
 /*
  * Post a zevent corresponding to the given sysevent.   The 'name' must be one
  * of the event definitions in sys/sysevent/eventdefs.h.  The payload will be
@@ -6938,9 +7226,9 @@ spa_has_active_shared_spare(spa_t *spa)
  * or zdb as real changes.
  */
 void
-spa_event_notify(spa_t *spa, vdev_t *vd, const char *name)
+spa_event_notify(spa_t *spa, vdev_t *vd, nvlist_t *hist_nvl, const char *name)
 {
-       zfs_post_sysevent(spa, vd, name);
+       spa_event_post(spa_event_create(spa, vd, hist_nvl, name));
 }
 
 #if defined(_KERNEL) && defined(HAVE_SPL)