]> granicus.if.org Git - zfs/blobdiff - module/zfs/vdev.c
OpenZFS 9591 - ms_shift can be incorrectly changed
[zfs] / module / zfs / vdev.c
index 0dea0b4e456bf5742543a77d6384542df7df1475..37bb5a0c5c5b235d2024639b14dad6944ccf744d 100644 (file)
@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
  * Copyright 2017 Nexenta Systems, Inc.
  * Copyright (c) 2014 Integros [integros.com]
  * Copyright 2016 Toomas Soome <tsoome@me.com>
@@ -74,6 +74,81 @@ unsigned int zfs_checksums_per_second = 20;
  */
 int zfs_scan_ignore_errors = 0;
 
+int vdev_validate_skip = B_FALSE;
+
+/*PRINTFLIKE2*/
+void
+vdev_dbgmsg(vdev_t *vd, const char *fmt, ...)
+{
+       va_list adx;
+       char buf[256];
+
+       va_start(adx, fmt);
+       (void) vsnprintf(buf, sizeof (buf), fmt, adx);
+       va_end(adx);
+
+       if (vd->vdev_path != NULL) {
+               zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type,
+                   vd->vdev_path, buf);
+       } else {
+               zfs_dbgmsg("%s-%llu vdev (guid %llu): %s",
+                   vd->vdev_ops->vdev_op_type,
+                   (u_longlong_t)vd->vdev_id,
+                   (u_longlong_t)vd->vdev_guid, buf);
+       }
+}
+
+void
+vdev_dbgmsg_print_tree(vdev_t *vd, int indent)
+{
+       char state[20];
+
+       if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) {
+               zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id,
+                   vd->vdev_ops->vdev_op_type);
+               return;
+       }
+
+       switch (vd->vdev_state) {
+       case VDEV_STATE_UNKNOWN:
+               (void) snprintf(state, sizeof (state), "unknown");
+               break;
+       case VDEV_STATE_CLOSED:
+               (void) snprintf(state, sizeof (state), "closed");
+               break;
+       case VDEV_STATE_OFFLINE:
+               (void) snprintf(state, sizeof (state), "offline");
+               break;
+       case VDEV_STATE_REMOVED:
+               (void) snprintf(state, sizeof (state), "removed");
+               break;
+       case VDEV_STATE_CANT_OPEN:
+               (void) snprintf(state, sizeof (state), "can't open");
+               break;
+       case VDEV_STATE_FAULTED:
+               (void) snprintf(state, sizeof (state), "faulted");
+               break;
+       case VDEV_STATE_DEGRADED:
+               (void) snprintf(state, sizeof (state), "degraded");
+               break;
+       case VDEV_STATE_HEALTHY:
+               (void) snprintf(state, sizeof (state), "healthy");
+               break;
+       default:
+               (void) snprintf(state, sizeof (state), "<state %u>",
+                   (uint_t)vd->vdev_state);
+       }
+
+       zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent,
+           "", vd->vdev_id, vd->vdev_ops->vdev_op_type,
+           vd->vdev_islog ? " (log)" : "",
+           (u_longlong_t)vd->vdev_guid,
+           vd->vdev_path ? vd->vdev_path : "N/A", state);
+
+       for (uint64_t i = 0; i < vd->vdev_children; i++)
+               vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2);
+}
+
 /*
  * Virtual device management.
  */
@@ -679,6 +754,8 @@ vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
                                    ZPOOL_CONFIG_AUX_STATE, &aux) == 0 &&
                                    strcmp(aux, "external") == 0)
                                        vd->vdev_label_aux = VDEV_AUX_EXTERNAL;
+                               else
+                                       vd->vdev_faulted = 0ULL;
                        }
                }
        }
@@ -1056,14 +1133,20 @@ vdev_metaslab_init(vdev_t *vd, uint64_t txg)
                        error = dmu_read(mos, vd->vdev_ms_array,
                            m * sizeof (uint64_t), sizeof (uint64_t), &object,
                            DMU_READ_PREFETCH);
-                       if (error)
+                       if (error != 0) {
+                               vdev_dbgmsg(vd, "unable to read the metaslab "
+                                   "array [error=%d]", error);
                                return (error);
+                       }
                }
 
                error = metaslab_init(vd->vdev_mg, m, object, txg,
                    &(vd->vdev_ms[m]));
-               if (error)
+               if (error != 0) {
+                       vdev_dbgmsg(vd, "metaslab_init failed [error=%d]",
+                           error);
                        return (error);
+               }
        }
 
        if (txg == 0)
@@ -1147,8 +1230,7 @@ vdev_probe_done(zio_t *zio)
                        zio->io_error = 0;
                } else {
                        ASSERT(zio->io_error != 0);
-                       zfs_dbgmsg("failed probe on vdev %llu",
-                           (longlong_t)vd->vdev_id);
+                       vdev_dbgmsg(vd, "failed probe");
                        zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE,
                            spa, vd, NULL, NULL, 0, 0);
                        zio->io_error = SET_ERROR(ENXIO);
@@ -1397,8 +1479,13 @@ vdev_open(vdev_t *vd)
                    vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED)
                        vd->vdev_removed = B_FALSE;
 
-               vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
-                   vd->vdev_stat.vs_aux);
+               if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) {
+                       vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE,
+                           vd->vdev_stat.vs_aux);
+               } else {
+                       vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
+                           vd->vdev_stat.vs_aux);
+               }
                return (error);
        }
 
@@ -1544,14 +1631,6 @@ vdev_open(vdev_t *vd)
                return (error);
        }
 
-       if (vd->vdev_top == vd && vd->vdev_ashift != 0 &&
-           !vd->vdev_isl2cache && !vd->vdev_islog) {
-               if (vd->vdev_ashift > spa->spa_max_ashift)
-                       spa->spa_max_ashift = vd->vdev_ashift;
-               if (vd->vdev_ashift < spa->spa_min_ashift)
-                       spa->spa_min_ashift = vd->vdev_ashift;
-       }
-
        /*
         * Track the min and max ashift values for normal data devices.
         */
@@ -1577,29 +1656,29 @@ vdev_open(vdev_t *vd)
 
 /*
  * Called once the vdevs are all opened, this routine validates the label
- * contents.  This needs to be done before vdev_load() so that we don't
+ * contents. This needs to be done before vdev_load() so that we don't
  * inadvertently do repair I/Os to the wrong device.
  *
- * If 'strict' is false ignore the spa guid check. This is necessary because
- * if the machine crashed during a re-guid the new guid might have been written
- * to all of the vdev labels, but not the cached config. The strict check
- * will be performed when the pool is opened again using the mos config.
- *
  * This function will only return failure if one of the vdevs indicates that it
  * has since been destroyed or exported.  This is only possible if
  * /etc/zfs/zpool.cache was readonly at the time.  Otherwise, the vdev state
  * will be updated but the function will return 0.
  */
 int
-vdev_validate(vdev_t *vd, boolean_t strict)
+vdev_validate(vdev_t *vd)
 {
        spa_t *spa = vd->vdev_spa;
        nvlist_t *label;
-       uint64_t guid = 0, top_guid;
+       uint64_t guid = 0, aux_guid = 0, top_guid;
        uint64_t state;
+       nvlist_t *nvl;
+       uint64_t txg;
 
-       for (int c = 0; c < vd->vdev_children; c++)
-               if (vdev_validate(vd->vdev_child[c], strict) != 0)
+       if (vdev_validate_skip)
+               return (0);
+
+       for (uint64_t c = 0; c < vd->vdev_children; c++)
+               if (vdev_validate(vd->vdev_child[c]) != 0)
                        return (SET_ERROR(EBADF));
 
        /*
@@ -1607,99 +1686,280 @@ vdev_validate(vdev_t *vd, boolean_t strict)
         * any further validation.  Otherwise, label I/O will fail and we will
         * overwrite the previous state.
         */
-       if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
-               uint64_t aux_guid = 0;
-               nvlist_t *nvl;
-               uint64_t txg = spa_last_synced_txg(spa) != 0 ?
-                   spa_last_synced_txg(spa) : -1ULL;
+       if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd))
+               return (0);
 
-               if ((label = vdev_label_read_config(vd, txg)) == NULL) {
-                       vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
-                           VDEV_AUX_BAD_LABEL);
-                       return (0);
-               }
+       /*
+        * If we are performing an extreme rewind, we allow for a label that
+        * was modified at a point after the current txg.
+        * If config lock is not held do not check for the txg. spa_sync could
+        * be updating the vdev's label before updating spa_last_synced_txg.
+        */
+       if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0 ||
+           spa_config_held(spa, SCL_CONFIG, RW_WRITER) != SCL_CONFIG)
+               txg = UINT64_MAX;
+       else
+               txg = spa_last_synced_txg(spa);
 
-               /*
-                * Determine if this vdev has been split off into another
-                * pool.  If so, then refuse to open it.
-                */
-               if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
-                   &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
-                       vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
-                           VDEV_AUX_SPLIT_POOL);
-                       nvlist_free(label);
-                       return (0);
-               }
+       if ((label = vdev_label_read_config(vd, txg)) == NULL) {
+               vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN,
+                   VDEV_AUX_BAD_LABEL);
+               vdev_dbgmsg(vd, "vdev_validate: failed reading config for "
+                   "txg %llu", (u_longlong_t)txg);
+               return (0);
+       }
 
-               if (strict && (nvlist_lookup_uint64(label,
-                   ZPOOL_CONFIG_POOL_GUID, &guid) != 0 ||
-                   guid != spa_guid(spa))) {
-                       vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
-                           VDEV_AUX_CORRUPT_DATA);
-                       nvlist_free(label);
-                       return (0);
-               }
+       /*
+        * Determine if this vdev has been split off into another
+        * pool.  If so, then refuse to open it.
+        */
+       if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID,
+           &aux_guid) == 0 && aux_guid == spa_guid(spa)) {
+               vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
+                   VDEV_AUX_SPLIT_POOL);
+               nvlist_free(label);
+               vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool");
+               return (0);
+       }
 
-               if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
-                   != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
-                   &aux_guid) != 0)
-                       aux_guid = 0;
+       if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) {
+               vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
+                   VDEV_AUX_CORRUPT_DATA);
+               nvlist_free(label);
+               vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
+                   ZPOOL_CONFIG_POOL_GUID);
+               return (0);
+       }
 
-               /*
-                * If this vdev just became a top-level vdev because its
-                * sibling was detached, it will have adopted the parent's
-                * vdev guid -- but the label may or may not be on disk yet.
-                * Fortunately, either version of the label will have the
-                * same top guid, so if we're a top-level vdev, we can
-                * safely compare to that instead.
-                *
-                * If we split this vdev off instead, then we also check the
-                * original pool's guid.  We don't want to consider the vdev
-                * corrupt if it is partway through a split operation.
-                */
-               if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
-                   &guid) != 0 ||
-                   nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID,
-                   &top_guid) != 0 ||
-                   ((vd->vdev_guid != guid && vd->vdev_guid != aux_guid) &&
-                   (vd->vdev_guid != top_guid || vd != vd->vdev_top))) {
-                       vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
-                           VDEV_AUX_CORRUPT_DATA);
-                       nvlist_free(label);
-                       return (0);
+       /*
+        * If config is not trusted then ignore the spa guid check. This is
+        * necessary because if the machine crashed during a re-guid the new
+        * guid might have been written to all of the vdev labels, but not the
+        * cached config. The check will be performed again once we have the
+        * trusted config from the MOS.
+        */
+       if (spa->spa_trust_config && guid != spa_guid(spa)) {
+               vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
+                   VDEV_AUX_CORRUPT_DATA);
+               nvlist_free(label);
+               vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't "
+                   "match config (%llu != %llu)", (u_longlong_t)guid,
+                   (u_longlong_t)spa_guid(spa));
+               return (0);
+       }
+
+       if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl)
+           != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID,
+           &aux_guid) != 0)
+               aux_guid = 0;
+
+       if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) {
+               vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
+                   VDEV_AUX_CORRUPT_DATA);
+               nvlist_free(label);
+               vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
+                   ZPOOL_CONFIG_GUID);
+               return (0);
+       }
+
+       if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid)
+           != 0) {
+               vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
+                   VDEV_AUX_CORRUPT_DATA);
+               nvlist_free(label);
+               vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
+                   ZPOOL_CONFIG_TOP_GUID);
+               return (0);
+       }
+
+       /*
+        * If this vdev just became a top-level vdev because its sibling was
+        * detached, it will have adopted the parent's vdev guid -- but the
+        * label may or may not be on disk yet. Fortunately, either version
+        * of the label will have the same top guid, so if we're a top-level
+        * vdev, we can safely compare to that instead.
+        * However, if the config comes from a cachefile that failed to update
+        * after the detach, a top-level vdev will appear as a non top-level
+        * vdev in the config. Also relax the constraints if we perform an
+        * extreme rewind.
+        *
+        * If we split this vdev off instead, then we also check the
+        * original pool's guid. We don't want to consider the vdev
+        * corrupt if it is partway through a split operation.
+        */
+       if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) {
+               boolean_t mismatch = B_FALSE;
+               if (spa->spa_trust_config && !spa->spa_extreme_rewind) {
+                       if (vd != vd->vdev_top || vd->vdev_guid != top_guid)
+                               mismatch = B_TRUE;
+               } else {
+                       if (vd->vdev_guid != top_guid &&
+                           vd->vdev_top->vdev_guid != guid)
+                               mismatch = B_TRUE;
                }
 
-               if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
-                   &state) != 0) {
+               if (mismatch) {
                        vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
                            VDEV_AUX_CORRUPT_DATA);
                        nvlist_free(label);
+                       vdev_dbgmsg(vd, "vdev_validate: config guid "
+                           "doesn't match label guid");
+                       vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu",
+                           (u_longlong_t)vd->vdev_guid,
+                           (u_longlong_t)vd->vdev_top->vdev_guid);
+                       vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, "
+                           "aux_guid %llu", (u_longlong_t)guid,
+                           (u_longlong_t)top_guid, (u_longlong_t)aux_guid);
                        return (0);
                }
+       }
 
+       if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
+           &state) != 0) {
+               vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
+                   VDEV_AUX_CORRUPT_DATA);
                nvlist_free(label);
+               vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label",
+                   ZPOOL_CONFIG_POOL_STATE);
+               return (0);
+       }
 
-               /*
-                * If this is a verbatim import, no need to check the
-                * state of the pool.
-                */
-               if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
-                   spa_load_state(spa) == SPA_LOAD_OPEN &&
-                   state != POOL_STATE_ACTIVE)
-                       return (SET_ERROR(EBADF));
+       nvlist_free(label);
 
-               /*
-                * If we were able to open and validate a vdev that was
-                * previously marked permanently unavailable, clear that state
-                * now.
-                */
-               if (vd->vdev_not_present)
-                       vd->vdev_not_present = 0;
+       /*
+        * If this is a verbatim import, no need to check the
+        * state of the pool.
+        */
+       if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
+           spa_load_state(spa) == SPA_LOAD_OPEN &&
+           state != POOL_STATE_ACTIVE) {
+               vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) "
+                   "for spa %s", (u_longlong_t)state, spa->spa_name);
+               return (SET_ERROR(EBADF));
+       }
+
+       /*
+        * If we were able to open and validate a vdev that was
+        * previously marked permanently unavailable, clear that state
+        * now.
+        */
+       if (vd->vdev_not_present)
+               vd->vdev_not_present = 0;
+
+       return (0);
+}
+
+static void
+vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd)
+{
+       if (svd->vdev_path != NULL && dvd->vdev_path != NULL) {
+               if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) {
+                       zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed "
+                           "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid,
+                           dvd->vdev_path, svd->vdev_path);
+                       spa_strfree(dvd->vdev_path);
+                       dvd->vdev_path = spa_strdup(svd->vdev_path);
+               }
+       } else if (svd->vdev_path != NULL) {
+               dvd->vdev_path = spa_strdup(svd->vdev_path);
+               zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'",
+                   (u_longlong_t)dvd->vdev_guid, dvd->vdev_path);
+       }
+}
+
+/*
+ * Recursively copy vdev paths from one vdev to another. Source and destination
+ * vdev trees must have same geometry otherwise return error. Intended to copy
+ * paths from userland config into MOS config.
+ */
+int
+vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd)
+{
+       if ((svd->vdev_ops == &vdev_missing_ops) ||
+           (svd->vdev_ishole && dvd->vdev_ishole) ||
+           (dvd->vdev_ops == &vdev_indirect_ops))
+               return (0);
+
+       if (svd->vdev_ops != dvd->vdev_ops) {
+               vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s",
+                   svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type);
+               return (SET_ERROR(EINVAL));
+       }
+
+       if (svd->vdev_guid != dvd->vdev_guid) {
+               vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != "
+                   "%llu)", (u_longlong_t)svd->vdev_guid,
+                   (u_longlong_t)dvd->vdev_guid);
+               return (SET_ERROR(EINVAL));
+       }
+
+       if (svd->vdev_children != dvd->vdev_children) {
+               vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: "
+                   "%llu != %llu", (u_longlong_t)svd->vdev_children,
+                   (u_longlong_t)dvd->vdev_children);
+               return (SET_ERROR(EINVAL));
+       }
+
+       for (uint64_t i = 0; i < svd->vdev_children; i++) {
+               int error = vdev_copy_path_strict(svd->vdev_child[i],
+                   dvd->vdev_child[i]);
+               if (error != 0)
+                       return (error);
        }
 
+       if (svd->vdev_ops->vdev_op_leaf)
+               vdev_copy_path_impl(svd, dvd);
+
        return (0);
 }
 
+static void
+vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd)
+{
+       ASSERT(stvd->vdev_top == stvd);
+       ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id);
+
+       for (uint64_t i = 0; i < dvd->vdev_children; i++) {
+               vdev_copy_path_search(stvd, dvd->vdev_child[i]);
+       }
+
+       if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd))
+               return;
+
+       /*
+        * The idea here is that while a vdev can shift positions within
+        * a top vdev (when replacing, attaching mirror, etc.) it cannot
+        * step outside of it.
+        */
+       vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid);
+
+       if (vd == NULL || vd->vdev_ops != dvd->vdev_ops)
+               return;
+
+       ASSERT(vd->vdev_ops->vdev_op_leaf);
+
+       vdev_copy_path_impl(vd, dvd);
+}
+
+/*
+ * Recursively copy vdev paths from one root vdev to another. Source and
+ * destination vdev trees may differ in geometry. For each destination leaf
+ * vdev, search a vdev with the same guid and top vdev id in the source.
+ * Intended to copy paths from userland config into MOS config.
+ */
+void
+vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd)
+{
+       uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children);
+       ASSERT(srvd->vdev_ops == &vdev_root_ops);
+       ASSERT(drvd->vdev_ops == &vdev_root_ops);
+
+       for (uint64_t i = 0; i < children; i++) {
+               vdev_copy_path_search(srvd->vdev_child[i],
+                   drvd->vdev_child[i]);
+       }
+}
+
 /*
  * Close a virtual device.
  */
@@ -1793,7 +2053,7 @@ vdev_reopen(vdev_t *vd)
                    !l2arc_vdev_present(vd))
                        l2arc_add_vdev(spa, vd);
        } else {
-               (void) vdev_validate(vd, B_TRUE);
+               (void) vdev_validate(vd);
        }
 
        /*
@@ -2322,9 +2582,10 @@ vdev_dtl_sync(vdev_t *vd, uint64_t txg)
         * the top level so that we update the config.
         */
        if (object != space_map_object(vd->vdev_dtl_sm)) {
-               zfs_dbgmsg("txg %llu, spa %s, DTL old object %llu, "
-                   "new object %llu", txg, spa_name(spa), object,
-                   space_map_object(vd->vdev_dtl_sm));
+               vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, "
+                   "new object %llu", (u_longlong_t)txg, spa_name(spa),
+                   (u_longlong_t)object,
+                   (u_longlong_t)space_map_object(vd->vdev_dtl_sm));
                vdev_config_dirty(vd->vdev_top);
        }
 
@@ -2433,8 +2694,13 @@ vdev_load(vdev_t *vd)
                if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) {
                        vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
                            VDEV_AUX_CORRUPT_DATA);
+                       vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, "
+                           "asize=%llu", (u_longlong_t)vd->vdev_ashift,
+                           (u_longlong_t)vd->vdev_asize);
                        return (SET_ERROR(ENXIO));
                } else if ((error = vdev_metaslab_init(vd, 0)) != 0) {
+                       vdev_dbgmsg(vd, "vdev_load: metaslab_init failed "
+                           "[error=%d]", error);
                        vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
                            VDEV_AUX_CORRUPT_DATA);
                        return (error);
@@ -2447,6 +2713,8 @@ vdev_load(vdev_t *vd)
        if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) {
                vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
                    VDEV_AUX_CORRUPT_DATA);
+               vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed "
+                   "[error=%d]", error);
                return (error);
        }
 
@@ -2460,6 +2728,9 @@ vdev_load(vdev_t *vd)
                    obsolete_sm_object, 0, vd->vdev_asize, 0))) {
                        vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN,
                            VDEV_AUX_CORRUPT_DATA);
+                       vdev_dbgmsg(vd, "vdev_load: space_map_open failed for "
+                           "obsolete spacemap (obj %llu) [error=%d]",
+                           (u_longlong_t)obsolete_sm_object, error);
                        return (error);
                }
                space_map_update(vd->vdev_obsolete_sm);
@@ -3222,7 +3493,6 @@ vdev_get_stats_ex(vdev_t *vd, vdev_stat_t *vs, vdev_stat_ex_t *vsx)
                            vd->vdev_max_asize - vd->vdev_asize,
                            1ULL << tvd->vdev_ms_shift);
                }
-               vs->vs_esize = vd->vdev_max_asize - vd->vdev_asize;
                if (vd->vdev_aux == NULL && vd == vd->vdev_top &&
                    vdev_is_concrete(vd)) {
                        vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation;
@@ -3827,6 +4097,19 @@ vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
                vdev_propagate_state(vd->vdev_parent);
 }
 
+boolean_t
+vdev_children_are_offline(vdev_t *vd)
+{
+       ASSERT(!vd->vdev_ops->vdev_op_leaf);
+
+       for (uint64_t i = 0; i < vd->vdev_children; i++) {
+               if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE)
+                       return (B_FALSE);
+       }
+
+       return (B_TRUE);
+}
+
 /*
  * Check the vdev configuration to ensure that it's capable of supporting
  * a root pool. We do not support partial configuration.
@@ -3862,34 +4145,6 @@ vdev_is_concrete(vdev_t *vd)
        }
 }
 
-/*
- * Load the state from the original vdev tree (ovd) which
- * we've retrieved from the MOS config object. If the original
- * vdev was offline or faulted then we transfer that state to the
- * device in the current vdev tree (nvd).
- */
-void
-vdev_load_log_state(vdev_t *nvd, vdev_t *ovd)
-{
-       ASSERT(nvd->vdev_top->vdev_islog);
-       ASSERT(spa_config_held(nvd->vdev_spa,
-           SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
-       ASSERT3U(nvd->vdev_guid, ==, ovd->vdev_guid);
-
-       for (int c = 0; c < nvd->vdev_children; c++)
-               vdev_load_log_state(nvd->vdev_child[c], ovd->vdev_child[c]);
-
-       if (nvd->vdev_ops->vdev_op_leaf) {
-               /*
-                * Restore the persistent vdev state
-                */
-               nvd->vdev_offline = ovd->vdev_offline;
-               nvd->vdev_faulted = ovd->vdev_faulted;
-               nvd->vdev_degraded = ovd->vdev_degraded;
-               nvd->vdev_removed = ovd->vdev_removed;
-       }
-}
-
 /*
  * Determine if a log device has valid content.  If the vdev was
  * removed or faulted in the MOS config then we know that
@@ -3917,11 +4172,11 @@ vdev_expand(vdev_t *vd, uint64_t txg)
 {
        ASSERT(vd->vdev_top == vd);
        ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
+       ASSERT(vdev_is_concrete(vd));
 
        vdev_set_deflate_ratio(vd);
 
-       if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count &&
-           vdev_is_concrete(vd)) {
+       if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count) {
                VERIFY(vdev_metaslab_init(vd, txg) == 0);
                vdev_config_dirty(vd);
        }
@@ -3981,7 +4236,7 @@ vdev_deadman(vdev_t *vd, char *tag)
        }
 }
 
-#if defined(_KERNEL) && defined(HAVE_SPL)
+#if defined(_KERNEL)
 EXPORT_SYMBOL(vdev_fault);
 EXPORT_SYMBOL(vdev_degrade);
 EXPORT_SYMBOL(vdev_online);
@@ -4005,5 +4260,9 @@ module_param(zfs_checksums_per_second, uint, 0644);
 module_param(zfs_scan_ignore_errors, int, 0644);
 MODULE_PARM_DESC(zfs_scan_ignore_errors,
        "Ignore errors during resilver/scrub");
+
+module_param(vdev_validate_skip, int, 0644);
+MODULE_PARM_DESC(vdev_validate_skip,
+       "Bypass vdev_validate()");
 /* END CSTYLED */
 #endif