Illumos #1748: desire support for reguid in zfs
[zfs.git] / module / zfs / vdev.c
index a61f29b..18d2d5c 100644 (file)
@@ -21,6 +21,8 @@
 
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
+ * Copyright (c) 2011 by Delphix. All rights reserved.
  */
 
 #include <sys/zfs_context.h>
@@ -85,8 +87,9 @@ vdev_default_asize(vdev_t *vd, uint64_t psize)
 {
        uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift);
        uint64_t csize;
+       int c;
 
-       for (int c = 0; c < vd->vdev_children; c++) {
+       for (c = 0; c < vd->vdev_children; c++) {
                csize = vdev_psize_to_asize(vd->vdev_child[c], psize);
                asize = MAX(asize, csize);
        }
@@ -132,9 +135,10 @@ vdev_get_min_asize(vdev_t *vd)
 void
 vdev_set_min_asize(vdev_t *vd)
 {
+       int c;
        vd->vdev_min_asize = vdev_get_min_asize(vd);
 
-       for (int c = 0; c < vd->vdev_children; c++)
+       for (c = 0; c < vd->vdev_children; c++)
                vdev_set_min_asize(vd->vdev_child[c]);
 }
 
@@ -157,11 +161,12 @@ vdev_t *
 vdev_lookup_by_guid(vdev_t *vd, uint64_t guid)
 {
        vdev_t *mvd;
+       int c;
 
        if (vd->vdev_guid == guid)
                return (vd);
 
-       for (int c = 0; c < vd->vdev_children; c++)
+       for (c = 0; c < vd->vdev_children; c++)
                if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) !=
                    NULL)
                        return (mvd);
@@ -207,9 +212,6 @@ vdev_add_child(vdev_t *pvd, vdev_t *cvd)
         */
        for (; pvd != NULL; pvd = pvd->vdev_parent)
                pvd->vdev_guid_sum += cvd->vdev_guid_sum;
-
-       if (cvd->vdev_ops->vdev_op_leaf)
-               cvd->vdev_spa->spa_scrub_maxinflight += zfs_scrub_limit;
 }
 
 void
@@ -244,9 +246,6 @@ vdev_remove_child(vdev_t *pvd, vdev_t *cvd)
         */
        for (; pvd != NULL; pvd = pvd->vdev_parent)
                pvd->vdev_guid_sum -= cvd->vdev_guid_sum;
-
-       if (cvd->vdev_ops->vdev_op_leaf)
-               cvd->vdev_spa->spa_scrub_maxinflight -= zfs_scrub_limit;
 }
 
 /*
@@ -258,16 +257,17 @@ vdev_compact_children(vdev_t *pvd)
        vdev_t **newchild, *cvd;
        int oldc = pvd->vdev_children;
        int newc;
+       int c;
 
        ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL);
 
-       for (int c = newc = 0; c < oldc; c++)
+       for (c = newc = 0; c < oldc; c++)
                if (pvd->vdev_child[c])
                        newc++;
 
        newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP);
 
-       for (int c = newc = 0; c < oldc; c++) {
+       for (c = newc = 0; c < oldc; c++) {
                if ((cvd = pvd->vdev_child[c]) != NULL) {
                        newchild[newc] = cvd;
                        cvd->vdev_id = newc++;
@@ -286,12 +286,14 @@ vdev_t *
 vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
 {
        vdev_t *vd;
+       int t;
 
        vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
 
        if (spa->spa_root_vdev == NULL) {
                ASSERT(ops == &vdev_root_ops);
                spa->spa_root_vdev = vd;
+               spa->spa_load_guid = spa_generate_guid(NULL);
        }
 
        if (guid == 0 && ops != &vdev_hole_ops) {
@@ -318,10 +320,12 @@ vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops)
        vd->vdev_state = VDEV_STATE_CLOSED;
        vd->vdev_ishole = (ops == &vdev_hole_ops);
 
+       list_link_init(&vd->vdev_config_dirty_node);
+       list_link_init(&vd->vdev_state_dirty_node);
        mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL);
        mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL);
        mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL);
-       for (int t = 0; t < DTL_TYPES; t++) {
+       for (t = 0; t < DTL_TYPES; t++) {
                space_map_create(&vd->vdev_dtl[t], 0, -1ULL, 0,
                    &vd->vdev_dtl_lock);
        }
@@ -491,7 +495,7 @@ vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
                    &vd->vdev_removing);
        }
 
-       if (parent && !parent->vdev_parent) {
+       if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) {
                ASSERT(alloctype == VDEV_ALLOC_LOAD ||
                    alloctype == VDEV_ALLOC_ADD ||
                    alloctype == VDEV_ALLOC_SPLIT ||
@@ -524,6 +528,9 @@ vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
                (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE,
                    &vd->vdev_offline);
 
+               (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVERING,
+                   &vd->vdev_resilvering);
+
                /*
                 * When importing a pool, we want to ignore the persistent fault
                 * state, as the diagnosis made on another system may not be
@@ -564,6 +571,7 @@ vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id,
 void
 vdev_free(vdev_t *vd)
 {
+       int c, t;
        spa_t *spa = vd->vdev_spa;
 
        /*
@@ -578,7 +586,7 @@ vdev_free(vdev_t *vd)
        /*
         * Free all children.
         */
-       for (int c = 0; c < vd->vdev_children; c++)
+       for (c = 0; c < vd->vdev_children; c++)
                vdev_free(vd->vdev_child[c]);
 
        ASSERT(vd->vdev_child == NULL);
@@ -627,7 +635,7 @@ vdev_free(vdev_t *vd)
        txg_list_destroy(&vd->vdev_dtl_list);
 
        mutex_enter(&vd->vdev_dtl_lock);
-       for (int t = 0; t < DTL_TYPES; t++) {
+       for (t = 0; t < DTL_TYPES; t++) {
                space_map_unload(&vd->vdev_dtl[t]);
                space_map_destroy(&vd->vdev_dtl[t]);
        }
@@ -664,6 +672,8 @@ vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
        svd->vdev_ms_shift = 0;
        svd->vdev_ms_count = 0;
 
+       if (tvd->vdev_mg)
+               ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg);
        tvd->vdev_mg = svd->vdev_mg;
        tvd->vdev_ms = svd->vdev_ms;
 
@@ -710,12 +720,14 @@ vdev_top_transfer(vdev_t *svd, vdev_t *tvd)
 static void
 vdev_top_update(vdev_t *tvd, vdev_t *vd)
 {
+       int c;
+
        if (vd == NULL)
                return;
 
        vd->vdev_top = tvd;
 
-       for (int c = 0; c < vd->vdev_children; c++)
+       for (c = 0; c < vd->vdev_children; c++)
                vdev_top_update(tvd, vd->vdev_child[c]);
 }
 
@@ -826,7 +838,7 @@ vdev_metaslab_init(vdev_t *vd, uint64_t txg)
 
        ASSERT(oldc <= newc);
 
-       mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP);
+       mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP | KM_NODEBUG);
 
        if (oldc != 0) {
                bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp));
@@ -963,6 +975,7 @@ vdev_probe(vdev_t *vd, zio_t *zio)
        spa_t *spa = vd->vdev_spa;
        vdev_probe_stats_t *vps = NULL;
        zio_t *pio;
+       int l;
 
        ASSERT(vd->vdev_ops->vdev_op_leaf);
 
@@ -1032,7 +1045,7 @@ vdev_probe(vdev_t *vd, zio_t *zio)
                return (NULL);
        }
 
-       for (int l = 1; l < VDEV_LABELS; l++) {
+       for (l = 1; l < VDEV_LABELS; l++) {
                zio_nowait(zio_read_phys(pio, vd,
                    vdev_label_offset(vd->vdev_psize, l,
                    offsetof(vdev_label_t, vl_pad2)),
@@ -1061,12 +1074,24 @@ vdev_open_child(void *arg)
 boolean_t
 vdev_uses_zvols(vdev_t *vd)
 {
+/*
+ * Stacking zpools on top of zvols is unsupported until we implement a method
+ * for determining if an arbitrary block device is a zvol without using the
+ * path.  Solaris would check the 'zvol' path component but this does not
+ * exist in the Linux port, so we really should do something like stat the
+ * file and check the major number.  This is complicated by the fact that
+ * we need to do this portably in user or kernel space.
+ */
+#if 0
+       int c;
+
        if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR,
            strlen(ZVOL_DIR)) == 0)
                return (B_TRUE);
-       for (int c = 0; c < vd->vdev_children; c++)
+       for (c = 0; c < vd->vdev_children; c++)
                if (vdev_uses_zvols(vd->vdev_child[c]))
                        return (B_TRUE);
+#endif
        return (B_FALSE);
 }
 
@@ -1075,6 +1100,7 @@ vdev_open_children(vdev_t *vd)
 {
        taskq_t *tq;
        int children = vd->vdev_children;
+       int c;
 
        /*
         * in order to handle pools on top of zvols, do the opens
@@ -1082,7 +1108,7 @@ vdev_open_children(vdev_t *vd)
         * spa_namespace_lock
         */
        if (vdev_uses_zvols(vd)) {
-               for (int c = 0; c < children; c++)
+               for (c = 0; c < children; c++)
                        vd->vdev_child[c]->vdev_open_error =
                            vdev_open(vd->vdev_child[c]);
                return;
@@ -1090,9 +1116,9 @@ vdev_open_children(vdev_t *vd)
        tq = taskq_create("vdev_open", children, minclsyspri,
            children, children, TASKQ_PREPOPULATE);
 
-       for (int c = 0; c < children; c++)
+       for (c = 0; c < children; c++)
                VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c],
-                   TQ_SLEEP) != NULL);
+                   TQ_SLEEP) != 0);
 
        taskq_destroy(tq);
 }
@@ -1108,6 +1134,7 @@ vdev_open(vdev_t *vd)
        uint64_t osize = 0;
        uint64_t asize, psize;
        uint64_t ashift = 0;
+       int c;
 
        ASSERT(vd->vdev_open_thread == curthread ||
            spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
@@ -1186,7 +1213,7 @@ vdev_open(vdev_t *vd)
        if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops)
                return (0);
 
-       for (int c = 0; c < vd->vdev_children; c++) {
+       for (c = 0; c < vd->vdev_children; c++) {
                if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) {
                        vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED,
                            VDEV_AUX_NONE);
@@ -1295,8 +1322,9 @@ vdev_validate(vdev_t *vd)
        nvlist_t *label;
        uint64_t guid = 0, top_guid;
        uint64_t state;
+       int c;
 
-       for (int c = 0; c < vd->vdev_children; c++)
+       for (c = 0; c < vd->vdev_children; c++)
                if (vdev_validate(vd->vdev_child[c]) != 0)
                        return (EBADF);
 
@@ -1375,10 +1403,10 @@ vdev_validate(vdev_t *vd)
                nvlist_free(label);
 
                /*
-                * If spa->spa_load_verbatim is true, no need to check the
+                * If this is a verbatim import, no need to check the
                 * state of the pool.
                 */
-               if (!spa->spa_load_verbatim &&
+               if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) &&
                    spa_load_state(spa) == SPA_LOAD_OPEN &&
                    state != POOL_STATE_ACTIVE)
                        return (EBADF);
@@ -1401,8 +1429,8 @@ vdev_validate(vdev_t *vd)
 void
 vdev_close(vdev_t *vd)
 {
-       spa_t *spa = vd->vdev_spa;
        vdev_t *pvd = vd->vdev_parent;
+       ASSERTV(spa_t *spa = vd->vdev_spa);
 
        ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
 
@@ -1435,12 +1463,13 @@ void
 vdev_hold(vdev_t *vd)
 {
        spa_t *spa = vd->vdev_spa;
+       int c;
 
        ASSERT(spa_is_root(spa));
        if (spa->spa_state == POOL_STATE_UNINITIALIZED)
                return;
 
-       for (int c = 0; c < vd->vdev_children; c++)
+       for (c = 0; c < vd->vdev_children; c++)
                vdev_hold(vd->vdev_child[c]);
 
        if (vd->vdev_ops->vdev_op_leaf)
@@ -1450,10 +1479,10 @@ vdev_hold(vdev_t *vd)
 void
 vdev_rele(vdev_t *vd)
 {
-       spa_t *spa = vd->vdev_spa;
+       int c;
 
-       ASSERT(spa_is_root(spa));
-       for (int c = 0; c < vd->vdev_children; c++)
+       ASSERT(spa_is_root(vd->vdev_spa));
+       for (c = 0; c < vd->vdev_children; c++)
                vdev_rele(vd->vdev_child[c]);
 
        if (vd->vdev_ops->vdev_op_leaf)
@@ -1544,6 +1573,7 @@ vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg)
        ASSERT(vd == vd->vdev_top);
        ASSERT(!vd->vdev_ishole);
        ASSERT(ISP2(flags));
+       ASSERT(spa_writeable(vd->vdev_spa));
 
        if (flags & VDD_METASLAB)
                (void) txg_list_add(&vd->vdev_ms_list, arg, txg);
@@ -1599,6 +1629,7 @@ vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size)
 
        ASSERT(t < DTL_TYPES);
        ASSERT(vd != vd->vdev_spa->spa_root_vdev);
+       ASSERT(spa_writeable(vd->vdev_spa));
 
        mutex_enter(sm->sm_lock);
        if (!space_map_contains(sm, txg, size))
@@ -1644,11 +1675,11 @@ vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
 {
        spa_t *spa = vd->vdev_spa;
        avl_tree_t reftree;
-       int minref;
+       int c, t, minref;
 
        ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0);
 
-       for (int c = 0; c < vd->vdev_children; c++)
+       for (c = 0; c < vd->vdev_children; c++)
                vdev_dtl_reassess(vd->vdev_child[c], txg,
                    scrub_txg, scrub_done);
 
@@ -1708,7 +1739,7 @@ vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
        }
 
        mutex_enter(&vd->vdev_dtl_lock);
-       for (int t = 0; t < DTL_TYPES; t++) {
+       for (t = 0; t < DTL_TYPES; t++) {
                /* account for child's outage in parent's missing map */
                int s = (t == DTL_MISSING) ? DTL_OUTAGE: t;
                if (t == DTL_SCRUB)
@@ -1720,7 +1751,7 @@ vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done)
                else
                        minref = vd->vdev_children;     /* any kind of mirror */
                space_map_ref_create(&reftree);
-               for (int c = 0; c < vd->vdev_children; c++) {
+               for (c = 0; c < vd->vdev_children; c++) {
                        vdev_t *cvd = vd->vdev_child[c];
                        mutex_enter(&cvd->vdev_dtl_lock);
                        space_map_ref_add_map(&reftree, &cvd->vdev_dtl[s], 1);
@@ -1781,8 +1812,7 @@ vdev_dtl_sync(vdev_t *vd, uint64_t txg)
 
        if (vd->vdev_detached) {
                if (smo->smo_object != 0) {
-                       int err = dmu_object_free(mos, smo->smo_object, tx);
-                       ASSERT3U(err, ==, 0);
+                       VERIFY(0 == dmu_object_free(mos, smo->smo_object, tx));
                        smo->smo_object = 0;
                }
                dmu_tx_commit(tx);
@@ -1855,6 +1885,9 @@ vdev_dtl_required(vdev_t *vd)
        vd->vdev_cant_read = cant_read;
        vdev_dtl_reassess(tvd, 0, 0, B_FALSE);
 
+       if (!required && zio_injection_enabled)
+               required = !!zio_handle_device_injection(vd, NULL, ECHILD);
+
        return (required);
 }
 
@@ -1867,6 +1900,7 @@ vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
        boolean_t needed = B_FALSE;
        uint64_t thismin = UINT64_MAX;
        uint64_t thismax = 0;
+       int c;
 
        if (vd->vdev_children == 0) {
                mutex_enter(&vd->vdev_dtl_lock);
@@ -1882,7 +1916,7 @@ vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
                }
                mutex_exit(&vd->vdev_dtl_lock);
        } else {
-               for (int c = 0; c < vd->vdev_children; c++) {
+               for (c = 0; c < vd->vdev_children; c++) {
                        vdev_t *cvd = vd->vdev_child[c];
                        uint64_t cmin, cmax;
 
@@ -1904,10 +1938,12 @@ vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp)
 void
 vdev_load(vdev_t *vd)
 {
+       int c;
+
        /*
         * Recursively load all children.
         */
-       for (int c = 0; c < vd->vdev_children; c++)
+       for (c = 0; c < vd->vdev_children; c++)
                vdev_load(vd->vdev_child[c]);
 
        /*
@@ -1975,6 +2011,7 @@ vdev_remove(vdev_t *vd, uint64_t txg)
        spa_t *spa = vd->vdev_spa;
        objset_t *mos = spa->spa_meta_objset;
        dmu_tx_t *tx;
+       int m;
 
        tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg);
 
@@ -1985,7 +2022,7 @@ vdev_remove(vdev_t *vd, uint64_t txg)
        }
 
        if (vd->vdev_ms != NULL) {
-               for (int m = 0; m < vd->vdev_ms_count; m++) {
+               for (m = 0; m < vd->vdev_ms_count; m++) {
                        metaslab_t *msp = vd->vdev_ms[m];
 
                        if (msp == NULL || msp->ms_smo.smo_object == 0)
@@ -2013,7 +2050,7 @@ vdev_sync_done(vdev_t *vd, uint64_t txg)
 
        ASSERT(!vd->vdev_ishole);
 
-       while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg)))
+       while ((msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))))
                metaslab_sync_done(msp, txg);
 
        if (reassess)
@@ -2070,7 +2107,7 @@ vdev_psize_to_asize(vdev_t *vd, uint64_t psize)
 int
 vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
 {
-       vdev_t *vd;
+       vdev_t *vd, *tvd;
 
        spa_vdev_state_enter(spa, SCL_NONE);
 
@@ -2080,6 +2117,8 @@ vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
        if (!vd->vdev_ops->vdev_op_leaf)
                return (spa_vdev_state_exit(spa, NULL, ENOTSUP));
 
+       tvd = vd->vdev_top;
+
        /*
         * We don't directly use the aux state here, but if we do a
         * vdev_reopen(), we need this value to be present to remember why we
@@ -2099,7 +2138,7 @@ vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
         * If this device has the only valid copy of the data, then
         * back off and simply mark the vdev as degraded instead.
         */
-       if (!vd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
+       if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) {
                vd->vdev_degraded = 1ULL;
                vd->vdev_faulted = 0ULL;
 
@@ -2107,7 +2146,7 @@ vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux)
                 * If we reopen the device and it's not dead, only then do we
                 * mark it degraded.
                 */
-               vdev_reopen(vd);
+               vdev_reopen(tvd);
 
                if (vdev_readable(vd))
                        vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux);
@@ -2320,6 +2359,7 @@ void
 vdev_clear(spa_t *spa, vdev_t *vd)
 {
        vdev_t *rvd = spa->spa_root_vdev;
+       int c;
 
        ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
 
@@ -2330,7 +2370,7 @@ vdev_clear(spa_t *spa, vdev_t *vd)
        vd->vdev_stat.vs_write_errors = 0;
        vd->vdev_stat.vs_checksum_errors = 0;
 
-       for (int c = 0; c < vd->vdev_children; c++)
+       for (c = 0; c < vd->vdev_children; c++)
                vdev_clear(spa, vd->vdev_child[c]);
 
        /*
@@ -2349,21 +2389,21 @@ vdev_clear(spa_t *spa, vdev_t *vd)
                 */
                vd->vdev_forcefault = B_TRUE;
 
-               vd->vdev_faulted = vd->vdev_degraded = 0;
+               vd->vdev_faulted = vd->vdev_degraded = 0ULL;
                vd->vdev_cant_read = B_FALSE;
                vd->vdev_cant_write = B_FALSE;
 
-               vdev_reopen(vd);
+               vdev_reopen(vd == rvd ? rvd : vd->vdev_top);
 
                vd->vdev_forcefault = B_FALSE;
 
-               if (vd != rvd)
+               if (vd != rvd && vdev_writeable(vd->vdev_top))
                        vdev_state_dirty(vd->vdev_top);
 
                if (vd->vdev_aux == NULL && !vdev_is_dead(vd))
                        spa_async_request(spa, SPA_ASYNC_RESILVER);
 
-               spa_event_notify(spa, vd, ESC_ZFS_VDEV_CLEAR);
+               spa_event_notify(spa, vd, FM_EREPORT_ZFS_DEVICE_CLEAR);
        }
 
        /*
@@ -2444,6 +2484,7 @@ void
 vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
 {
        vdev_t *rvd = vd->vdev_spa->spa_root_vdev;
+       int c, t;
 
        mutex_enter(&vd->vdev_stat_lock);
        bcopy(&vd->vdev_stat, vs, sizeof (*vs));
@@ -2459,12 +2500,12 @@ vdev_get_stats(vdev_t *vd, vdev_stat_t *vs)
         * over all top-level vdevs (i.e. the direct children of the root).
         */
        if (vd == rvd) {
-               for (int c = 0; c < rvd->vdev_children; c++) {
+               for (c = 0; c < rvd->vdev_children; c++) {
                        vdev_t *cvd = rvd->vdev_child[c];
                        vdev_stat_t *cvs = &cvd->vdev_stat;
 
                        mutex_enter(&vd->vdev_stat_lock);
-                       for (int t = 0; t < ZIO_TYPES; t++) {
+                       for (t = 0; t < ZIO_TYPES; t++) {
                                vs->vs_ops[t] += cvs->vs_ops[t];
                                vs->vs_bytes[t] += cvs->vs_bytes[t];
                        }
@@ -2488,8 +2529,9 @@ void
 vdev_scan_stat_init(vdev_t *vd)
 {
        vdev_stat_t *vs = &vd->vdev_stat;
+       int c;
 
-       for (int c = 0; c < vd->vdev_children; c++)
+       for (c = 0; c < vd->vdev_children; c++)
                vdev_scan_stat_init(vd->vdev_child[c]);
 
        mutex_enter(&vd->vdev_stat_lock);
@@ -2541,7 +2583,7 @@ vdev_stat_update(zio_t *zio, uint64_t psize)
                mutex_enter(&vd->vdev_stat_lock);
 
                if (flags & ZIO_FLAG_IO_REPAIR) {
-                       if (flags & ZIO_FLAG_SCRUB_THREAD) {
+                       if (flags & ZIO_FLAG_SCAN_THREAD) {
                                dsl_scan_phys_t *scn_phys =
                                    &spa->spa_dsl_pool->dp_scan->scn_phys;
                                uint64_t *processed = &scn_phys->scn_processed;
@@ -2597,7 +2639,7 @@ vdev_stat_update(zio_t *zio, uint64_t psize)
 
        if (type == ZIO_TYPE_WRITE && txg != 0 &&
            (!(flags & ZIO_FLAG_IO_REPAIR) ||
-           (flags & ZIO_FLAG_SCRUB_THREAD) ||
+           (flags & ZIO_FLAG_SCAN_THREAD) ||
            spa->spa_claiming)) {
                /*
                 * This is either a normal write (not a repair), or it's
@@ -2616,7 +2658,7 @@ vdev_stat_update(zio_t *zio, uint64_t psize)
                 */
                if (vd->vdev_ops->vdev_op_leaf) {
                        uint64_t commit_txg = txg;
-                       if (flags & ZIO_FLAG_SCRUB_THREAD) {
+                       if (flags & ZIO_FLAG_SCAN_THREAD) {
                                ASSERT(flags & ZIO_FLAG_IO_REPAIR);
                                ASSERT(spa_sync_pass(spa) == 1);
                                vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1);
@@ -2699,6 +2741,8 @@ vdev_config_dirty(vdev_t *vd)
        vdev_t *rvd = spa->spa_root_vdev;
        int c;
 
+       ASSERT(spa_writeable(spa));
+
        /*
         * If this is an aux vdev (as with l2cache and spare devices), then we
         * update the vdev config manually and set the sync flag.
@@ -2787,6 +2831,7 @@ vdev_state_dirty(vdev_t *vd)
 {
        spa_t *spa = vd->vdev_spa;
 
+       ASSERT(spa_writeable(spa));
        ASSERT(vd == vd->vdev_top);
 
        /*
@@ -2827,9 +2872,10 @@ vdev_propagate_state(vdev_t *vd)
        int degraded = 0, faulted = 0;
        int corrupted = 0;
        vdev_t *child;
+       int c;
 
        if (vd->vdev_children > 0) {
-               for (int c = 0; c < vd->vdev_children; c++) {
+               for (c = 0; c < vd->vdev_children; c++) {
                        child = vd->vdev_child[c];
 
                        /*
@@ -2944,12 +2990,13 @@ vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
                vd->vdev_removed = B_TRUE;
        } else if (state == VDEV_STATE_CANT_OPEN) {
                /*
-                * If we fail to open a vdev during an import, we mark it as
-                * "not available", which signifies that it was never there to
-                * begin with.  Failure to open such a device is not considered
-                * an error.
+                * If we fail to open a vdev during an import or recovery, we
+                * mark it as "not available", which signifies that it was
+                * never there to begin with.  Failure to open such a device
+                * is not considered an error.
                 */
-               if (spa_load_state(spa) == SPA_LOAD_IMPORT &&
+               if ((spa_load_state(spa) == SPA_LOAD_IMPORT ||
+                   spa_load_state(spa) == SPA_LOAD_RECOVER) &&
                    vd->vdev_ops->vdev_op_leaf)
                        vd->vdev_not_present = 1;
 
@@ -3011,13 +3058,19 @@ vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux)
 
 /*
  * Check the vdev configuration to ensure that it's capable of supporting
- * a root pool. Currently, we do not support RAID-Z or partial configuration.
- * In addition, only a single top-level vdev is allowed and none of the leaves
- * can be wholedisks.
+ * a root pool.
  */
 boolean_t
 vdev_is_bootable(vdev_t *vd)
 {
+#if defined(__sun__) || defined(__sun)
+       /*
+        * Currently, we do not support RAID-Z or partial configuration.
+        * In addition, only a single top-level vdev is allowed and none of the
+        * leaves can be wholedisks.
+        */
+       int c;
+
        if (!vd->vdev_ops->vdev_op_leaf) {
                char *vdev_type = vd->vdev_ops->vdev_op_type;
 
@@ -3032,42 +3085,66 @@ vdev_is_bootable(vdev_t *vd)
                return (B_FALSE);
        }
 
-       for (int c = 0; c < vd->vdev_children; c++) {
+       for (c = 0; c < vd->vdev_children; c++) {
                if (!vdev_is_bootable(vd->vdev_child[c]))
                        return (B_FALSE);
        }
+#endif /* __sun__ || __sun */
        return (B_TRUE);
 }
 
 /*
  * Load the state from the original vdev tree (ovd) which
  * we've retrieved from the MOS config object. If the original
- * vdev was offline then we transfer that state to the device
- * in the current vdev tree (nvd).
+ * 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)
 {
-       spa_t *spa = nvd->vdev_spa;
+       int c;
 
-       ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
+       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++)
+       for (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 && ovd->vdev_offline) {
+       if (nvd->vdev_ops->vdev_op_leaf) {
                /*
-                * It would be nice to call vdev_offline()
-                * directly but the pool isn't fully loaded and
-                * the txg threads have not been started yet.
+                * Restore the persistent vdev state
                 */
                nvd->vdev_offline = ovd->vdev_offline;
-               vdev_reopen(nvd->vdev_top);
+               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
+ * the content on the log device has already been written to the pool.
+ */
+boolean_t
+vdev_log_state_valid(vdev_t *vd)
+{
+       int c;
+
+       if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted &&
+           !vd->vdev_removed)
+               return (B_TRUE);
+
+       for (c = 0; c < vd->vdev_children; c++)
+               if (vdev_log_state_valid(vd->vdev_child[c]))
+                       return (B_TRUE);
+
+       return (B_FALSE);
+}
+
+/*
  * Expand a vdev if possible.
  */
 void
@@ -3100,3 +3177,14 @@ vdev_split(vdev_t *vd)
        }
        vdev_propagate_state(cvd);
 }
+
+#if defined(_KERNEL) && defined(HAVE_SPL)
+EXPORT_SYMBOL(vdev_fault);
+EXPORT_SYMBOL(vdev_degrade);
+EXPORT_SYMBOL(vdev_online);
+EXPORT_SYMBOL(vdev_offline);
+EXPORT_SYMBOL(vdev_clear);
+
+module_param(zfs_scrub_limit, int, 0644);
+MODULE_PARM_DESC(zfs_scrub_limit, "Max scrub/resilver I/O per leaf vdev");
+#endif