zfs_retire.c revision 88ecc943b4eb72f7c4fbbd8435997b85ef171fc3
99653d4ee642c6528e88224f12409a5f23060994eschrock/*
99653d4ee642c6528e88224f12409a5f23060994eschrock * CDDL HEADER START
99653d4ee642c6528e88224f12409a5f23060994eschrock *
99653d4ee642c6528e88224f12409a5f23060994eschrock * The contents of this file are subject to the terms of the
99653d4ee642c6528e88224f12409a5f23060994eschrock * Common Development and Distribution License (the "License").
99653d4ee642c6528e88224f12409a5f23060994eschrock * You may not use this file except in compliance with the License.
99653d4ee642c6528e88224f12409a5f23060994eschrock *
99653d4ee642c6528e88224f12409a5f23060994eschrock * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
99653d4ee642c6528e88224f12409a5f23060994eschrock * or http://www.opensolaris.org/os/licensing.
99653d4ee642c6528e88224f12409a5f23060994eschrock * See the License for the specific language governing permissions
99653d4ee642c6528e88224f12409a5f23060994eschrock * and limitations under the License.
99653d4ee642c6528e88224f12409a5f23060994eschrock *
99653d4ee642c6528e88224f12409a5f23060994eschrock * When distributing Covered Code, include this CDDL HEADER in each
99653d4ee642c6528e88224f12409a5f23060994eschrock * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
99653d4ee642c6528e88224f12409a5f23060994eschrock * If applicable, add the following below this CDDL HEADER, with the
99653d4ee642c6528e88224f12409a5f23060994eschrock * fields enclosed by brackets "[]" replaced with your own identifying
99653d4ee642c6528e88224f12409a5f23060994eschrock * information: Portions Copyright [yyyy] [name of copyright owner]
99653d4ee642c6528e88224f12409a5f23060994eschrock *
99653d4ee642c6528e88224f12409a5f23060994eschrock * CDDL HEADER END
99653d4ee642c6528e88224f12409a5f23060994eschrock */
99653d4ee642c6528e88224f12409a5f23060994eschrock/*
cbf75e67acb6c32a2f4884f28a839d59f7988d37Stephen Hanson * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
99653d4ee642c6528e88224f12409a5f23060994eschrock * Use is subject to license terms.
99653d4ee642c6528e88224f12409a5f23060994eschrock */
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrock/*
99653d4ee642c6528e88224f12409a5f23060994eschrock * The ZFS retire agent is responsible for managing hot spares across all pools.
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * When we see a device fault or a device removal, we try to open the associated
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * pool and look for any hot spares. We iterate over any available hot spares
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * and attempt a 'zpool replace' for each one.
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock *
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * For vdevs diagnosed as faulty, the agent is also responsible for proactively
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * marking the vdev FAULTY (for I/O errors) or DEGRADED (for checksum errors).
99653d4ee642c6528e88224f12409a5f23060994eschrock */
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrock#include <fm/fmd_api.h>
99653d4ee642c6528e88224f12409a5f23060994eschrock#include <sys/fs/zfs.h>
99653d4ee642c6528e88224f12409a5f23060994eschrock#include <sys/fm/protocol.h>
99653d4ee642c6528e88224f12409a5f23060994eschrock#include <sys/fm/fs/zfs.h>
99653d4ee642c6528e88224f12409a5f23060994eschrock#include <libzfs.h>
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock#include <string.h>
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrock/*
99653d4ee642c6528e88224f12409a5f23060994eschrock * Find a pool with a matching GUID.
99653d4ee642c6528e88224f12409a5f23060994eschrock */
99653d4ee642c6528e88224f12409a5f23060994eschrocktypedef struct find_cbdata {
99653d4ee642c6528e88224f12409a5f23060994eschrock uint64_t cb_guid;
99653d4ee642c6528e88224f12409a5f23060994eschrock zpool_handle_t *cb_zhp;
99653d4ee642c6528e88224f12409a5f23060994eschrock} find_cbdata_t;
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrockstatic int
99653d4ee642c6528e88224f12409a5f23060994eschrockfind_pool(zpool_handle_t *zhp, void *data)
99653d4ee642c6528e88224f12409a5f23060994eschrock{
99653d4ee642c6528e88224f12409a5f23060994eschrock find_cbdata_t *cbp = data;
99653d4ee642c6528e88224f12409a5f23060994eschrock
990b4856d0eaada6f8140335733a1b1771ed2746lling if (cbp->cb_guid ==
990b4856d0eaada6f8140335733a1b1771ed2746lling zpool_get_prop_int(zhp, ZPOOL_PROP_GUID, NULL)) {
99653d4ee642c6528e88224f12409a5f23060994eschrock cbp->cb_zhp = zhp;
99653d4ee642c6528e88224f12409a5f23060994eschrock return (1);
99653d4ee642c6528e88224f12409a5f23060994eschrock }
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrock zpool_close(zhp);
99653d4ee642c6528e88224f12409a5f23060994eschrock return (0);
99653d4ee642c6528e88224f12409a5f23060994eschrock}
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrock/*
99653d4ee642c6528e88224f12409a5f23060994eschrock * Find a vdev within a tree with a matching GUID.
99653d4ee642c6528e88224f12409a5f23060994eschrock */
99653d4ee642c6528e88224f12409a5f23060994eschrockstatic nvlist_t *
99653d4ee642c6528e88224f12409a5f23060994eschrockfind_vdev(nvlist_t *nv, uint64_t search)
99653d4ee642c6528e88224f12409a5f23060994eschrock{
99653d4ee642c6528e88224f12409a5f23060994eschrock uint64_t guid;
99653d4ee642c6528e88224f12409a5f23060994eschrock nvlist_t **child;
99653d4ee642c6528e88224f12409a5f23060994eschrock uint_t c, children;
99653d4ee642c6528e88224f12409a5f23060994eschrock nvlist_t *ret;
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) == 0 &&
99653d4ee642c6528e88224f12409a5f23060994eschrock guid == search)
99653d4ee642c6528e88224f12409a5f23060994eschrock return (nv);
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrock if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
99653d4ee642c6528e88224f12409a5f23060994eschrock &child, &children) != 0)
99653d4ee642c6528e88224f12409a5f23060994eschrock return (NULL);
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrock for (c = 0; c < children; c++) {
99653d4ee642c6528e88224f12409a5f23060994eschrock if ((ret = find_vdev(child[c], search)) != NULL)
99653d4ee642c6528e88224f12409a5f23060994eschrock return (ret);
99653d4ee642c6528e88224f12409a5f23060994eschrock }
99653d4ee642c6528e88224f12409a5f23060994eschrock
c5904d138f3bdf0762dbf452a43d5a5c387ea6a8eschrock if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
c5904d138f3bdf0762dbf452a43d5a5c387ea6a8eschrock &child, &children) != 0)
c5904d138f3bdf0762dbf452a43d5a5c387ea6a8eschrock return (NULL);
c5904d138f3bdf0762dbf452a43d5a5c387ea6a8eschrock
c5904d138f3bdf0762dbf452a43d5a5c387ea6a8eschrock for (c = 0; c < children; c++) {
c5904d138f3bdf0762dbf452a43d5a5c387ea6a8eschrock if ((ret = find_vdev(child[c], search)) != NULL)
c5904d138f3bdf0762dbf452a43d5a5c387ea6a8eschrock return (ret);
c5904d138f3bdf0762dbf452a43d5a5c387ea6a8eschrock }
c5904d138f3bdf0762dbf452a43d5a5c387ea6a8eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrock return (NULL);
99653d4ee642c6528e88224f12409a5f23060994eschrock}
99653d4ee642c6528e88224f12409a5f23060994eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock/*
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * Given a (pool, vdev) GUID pair, find the matching pool and vdev.
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock */
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrockstatic zpool_handle_t *
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrockfind_by_guid(libzfs_handle_t *zhdl, uint64_t pool_guid, uint64_t vdev_guid,
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock nvlist_t **vdevp)
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock{
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock find_cbdata_t cb;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock zpool_handle_t *zhp;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock nvlist_t *config, *nvroot;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock /*
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * Find the corresponding pool and make sure the vdev still exists.
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock */
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock cb.cb_guid = pool_guid;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (zpool_iter(zhdl, find_pool, &cb) != 1)
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock return (NULL);
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock zhp = cb.cb_zhp;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock config = zpool_get_config(zhp, NULL);
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock &nvroot) != 0) {
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock zpool_close(zhp);
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock return (NULL);
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock }
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if ((*vdevp = find_vdev(nvroot, vdev_guid)) == NULL) {
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock zpool_close(zhp);
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock return (NULL);
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock }
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock return (zhp);
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock}
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock/*
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * Given a vdev, attempt to replace it with every known spare until one
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * succeeds.
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock */
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrockstatic void
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrockreplace_with_spare(zpool_handle_t *zhp, nvlist_t *vdev)
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock{
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock nvlist_t *config, *nvroot, *replacement;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock nvlist_t **spares;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock uint_t s, nspares;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock char *dev_name;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock config = zpool_get_config(zhp, NULL);
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock &nvroot) != 0)
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock return;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock /*
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * Find out if there are any hot spares available in the pool.
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock */
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock &spares, &nspares) != 0)
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock return;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (nvlist_alloc(&replacement, NV_UNIQUE_NAME, 0) != 0)
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock return;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (nvlist_add_string(replacement, ZPOOL_CONFIG_TYPE,
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock VDEV_TYPE_ROOT) != 0) {
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock nvlist_free(replacement);
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock return;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock }
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
88ecc943b4eb72f7c4fbbd8435997b85ef171fc3George Wilson dev_name = zpool_vdev_name(NULL, zhp, vdev, B_FALSE);
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock /*
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * Try to replace each spare, ending when we successfully
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * replace it.
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock */
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock for (s = 0; s < nspares; s++) {
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock char *spare_name;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (nvlist_lookup_string(spares[s], ZPOOL_CONFIG_PATH,
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock &spare_name) != 0)
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock continue;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (nvlist_add_nvlist_array(replacement,
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock ZPOOL_CONFIG_CHILDREN, &spares[s], 1) != 0)
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock continue;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (zpool_vdev_attach(zhp, dev_name, spare_name,
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock replacement, B_TRUE) == 0)
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock break;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock }
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock free(dev_name);
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock nvlist_free(replacement);
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock}
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrock/*ARGSUSED*/
99653d4ee642c6528e88224f12409a5f23060994eschrockstatic void
99653d4ee642c6528e88224f12409a5f23060994eschrockzfs_retire_recv(fmd_hdl_t *hdl, fmd_event_t *ep, nvlist_t *nvl,
99653d4ee642c6528e88224f12409a5f23060994eschrock const char *class)
99653d4ee642c6528e88224f12409a5f23060994eschrock{
99653d4ee642c6528e88224f12409a5f23060994eschrock uint64_t pool_guid, vdev_guid;
99653d4ee642c6528e88224f12409a5f23060994eschrock zpool_handle_t *zhp;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock nvlist_t *resource, *fault;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock nvlist_t **faults;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock uint_t f, nfaults;
99653d4ee642c6528e88224f12409a5f23060994eschrock libzfs_handle_t *zhdl = fmd_hdl_getspecific(hdl);
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock boolean_t fault_device, degrade_device;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock boolean_t is_repair;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock char *scheme;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock nvlist_t *vdev;
25c6ff4b77fcddf4097ce78a8277275ca603b46cstephh char *uuid;
25c6ff4b77fcddf4097ce78a8277275ca603b46cstephh int repair_done = 0;
cbf75e67acb6c32a2f4884f28a839d59f7988d37Stephen Hanson boolean_t retire;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock /*
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * If this is a resource notifying us of device removal, then simply
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * check for an available spare and continue.
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock */
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (strcmp(class, "resource.fs.zfs.removed") == 0) {
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (nvlist_lookup_uint64(nvl, FM_EREPORT_PAYLOAD_ZFS_POOL_GUID,
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock &pool_guid) != 0 ||
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock nvlist_lookup_uint64(nvl, FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID,
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock &vdev_guid) != 0)
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock return;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if ((zhp = find_by_guid(zhdl, pool_guid, vdev_guid,
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock &vdev)) == NULL)
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock return;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (fmd_prop_get_int32(hdl, "spare_on_remove"))
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock replace_with_spare(zhp, vdev);
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock zpool_close(zhp);
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock return;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock }
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
cbf75e67acb6c32a2f4884f28a839d59f7988d37Stephen Hanson if (strcmp(class, FM_LIST_RESOLVED_CLASS) == 0)
cbf75e67acb6c32a2f4884f28a839d59f7988d37Stephen Hanson return;
cbf75e67acb6c32a2f4884f28a839d59f7988d37Stephen Hanson
25c6ff4b77fcddf4097ce78a8277275ca603b46cstephh if (strcmp(class, FM_LIST_REPAIRED_CLASS) == 0)
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock is_repair = B_TRUE;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock else
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock is_repair = B_FALSE;
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrock /*
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * We subscribe to zfs faults as well as all repair events.
99653d4ee642c6528e88224f12409a5f23060994eschrock */
99653d4ee642c6528e88224f12409a5f23060994eschrock if (nvlist_lookup_nvlist_array(nvl, FM_SUSPECT_FAULT_LIST,
99653d4ee642c6528e88224f12409a5f23060994eschrock &faults, &nfaults) != 0)
99653d4ee642c6528e88224f12409a5f23060994eschrock return;
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrock for (f = 0; f < nfaults; f++) {
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock fault = faults[f];
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock fault_device = B_FALSE;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock degrade_device = B_FALSE;
99653d4ee642c6528e88224f12409a5f23060994eschrock
cbf75e67acb6c32a2f4884f28a839d59f7988d37Stephen Hanson if (nvlist_lookup_boolean_value(fault, FM_SUSPECT_RETIRE,
cbf75e67acb6c32a2f4884f28a839d59f7988d37Stephen Hanson &retire) == 0 && retire == 0)
cbf75e67acb6c32a2f4884f28a839d59f7988d37Stephen Hanson continue;
cbf75e67acb6c32a2f4884f28a839d59f7988d37Stephen Hanson
99653d4ee642c6528e88224f12409a5f23060994eschrock /*
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * While we subscribe to fault.fs.zfs.*, we only take action
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * for faults targeting a specific vdev (open failure or SERD
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * failure).
99653d4ee642c6528e88224f12409a5f23060994eschrock */
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (fmd_nvl_class_match(hdl, fault, "fault.fs.zfs.vdev.io"))
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock fault_device = B_TRUE;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock else if (fmd_nvl_class_match(hdl, fault,
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock "fault.fs.zfs.vdev.checksum"))
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock degrade_device = B_TRUE;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock else if (fmd_nvl_class_match(hdl, fault, "fault.fs.zfs.device"))
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock fault_device = B_FALSE;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock else
99653d4ee642c6528e88224f12409a5f23060994eschrock continue;
99653d4ee642c6528e88224f12409a5f23060994eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (nvlist_lookup_nvlist(fault, FM_FAULT_RESOURCE,
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock &resource) != 0 ||
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock nvlist_lookup_string(resource, FM_FMRI_SCHEME,
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock &scheme) != 0)
99653d4ee642c6528e88224f12409a5f23060994eschrock continue;
99653d4ee642c6528e88224f12409a5f23060994eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (strcmp(scheme, FM_FMRI_SCHEME_ZFS) != 0)
99653d4ee642c6528e88224f12409a5f23060994eschrock continue;
99653d4ee642c6528e88224f12409a5f23060994eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (nvlist_lookup_uint64(resource, FM_FMRI_ZFS_POOL,
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock &pool_guid) != 0 ||
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock nvlist_lookup_uint64(resource, FM_FMRI_ZFS_VDEV,
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock &vdev_guid) != 0)
99653d4ee642c6528e88224f12409a5f23060994eschrock continue;
99653d4ee642c6528e88224f12409a5f23060994eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if ((zhp = find_by_guid(zhdl, pool_guid, vdev_guid,
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock &vdev)) == NULL)
99653d4ee642c6528e88224f12409a5f23060994eschrock continue;
99653d4ee642c6528e88224f12409a5f23060994eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock /*
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * If this is a repair event, then mark the vdev as repaired and
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * continue.
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock */
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (is_repair) {
25c6ff4b77fcddf4097ce78a8277275ca603b46cstephh repair_done = 1;
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock (void) zpool_vdev_clear(zhp, vdev_guid);
99653d4ee642c6528e88224f12409a5f23060994eschrock zpool_close(zhp);
99653d4ee642c6528e88224f12409a5f23060994eschrock continue;
99653d4ee642c6528e88224f12409a5f23060994eschrock }
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrock /*
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * Actively fault the device if needed.
99653d4ee642c6528e88224f12409a5f23060994eschrock */
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (fault_device)
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock (void) zpool_vdev_fault(zhp, vdev_guid);
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock if (degrade_device)
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock (void) zpool_vdev_degrade(zhp, vdev_guid);
99653d4ee642c6528e88224f12409a5f23060994eschrock
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock /*
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock * Attempt to substitute a hot spare.
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock */
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock replace_with_spare(zhp, vdev);
99653d4ee642c6528e88224f12409a5f23060994eschrock zpool_close(zhp);
99653d4ee642c6528e88224f12409a5f23060994eschrock }
25c6ff4b77fcddf4097ce78a8277275ca603b46cstephh
25c6ff4b77fcddf4097ce78a8277275ca603b46cstephh if (strcmp(class, FM_LIST_REPAIRED_CLASS) == 0 && repair_done &&
25c6ff4b77fcddf4097ce78a8277275ca603b46cstephh nvlist_lookup_string(nvl, FM_SUSPECT_UUID, &uuid) == 0)
25c6ff4b77fcddf4097ce78a8277275ca603b46cstephh fmd_case_uuresolved(hdl, uuid);
99653d4ee642c6528e88224f12409a5f23060994eschrock}
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrockstatic const fmd_hdl_ops_t fmd_ops = {
99653d4ee642c6528e88224f12409a5f23060994eschrock zfs_retire_recv, /* fmdo_recv */
99653d4ee642c6528e88224f12409a5f23060994eschrock NULL, /* fmdo_timeout */
99653d4ee642c6528e88224f12409a5f23060994eschrock NULL, /* fmdo_close */
99653d4ee642c6528e88224f12409a5f23060994eschrock NULL, /* fmdo_stats */
99653d4ee642c6528e88224f12409a5f23060994eschrock NULL, /* fmdo_gc */
99653d4ee642c6528e88224f12409a5f23060994eschrock};
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrockstatic const fmd_prop_t fmd_props[] = {
3d7072f8bd27709dba14f6fe336f149d25d9e207eschrock { "spare_on_remove", FMD_TYPE_BOOL, "true" },
99653d4ee642c6528e88224f12409a5f23060994eschrock { NULL, 0, NULL }
99653d4ee642c6528e88224f12409a5f23060994eschrock};
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrockstatic const fmd_hdl_info_t fmd_info = {
99653d4ee642c6528e88224f12409a5f23060994eschrock "ZFS Retire Agent", "1.0", &fmd_ops, fmd_props
99653d4ee642c6528e88224f12409a5f23060994eschrock};
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrockvoid
99653d4ee642c6528e88224f12409a5f23060994eschrock_fmd_init(fmd_hdl_t *hdl)
99653d4ee642c6528e88224f12409a5f23060994eschrock{
99653d4ee642c6528e88224f12409a5f23060994eschrock libzfs_handle_t *zhdl;
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrock if ((zhdl = libzfs_init()) == NULL)
99653d4ee642c6528e88224f12409a5f23060994eschrock return;
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrock if (fmd_hdl_register(hdl, FMD_API_VERSION, &fmd_info) != 0) {
99653d4ee642c6528e88224f12409a5f23060994eschrock libzfs_fini(zhdl);
99653d4ee642c6528e88224f12409a5f23060994eschrock return;
99653d4ee642c6528e88224f12409a5f23060994eschrock }
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrock fmd_hdl_setspecific(hdl, zhdl);
99653d4ee642c6528e88224f12409a5f23060994eschrock}
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrockvoid
99653d4ee642c6528e88224f12409a5f23060994eschrock_fmd_fini(fmd_hdl_t *hdl)
99653d4ee642c6528e88224f12409a5f23060994eschrock{
99653d4ee642c6528e88224f12409a5f23060994eschrock libzfs_handle_t *zhdl = fmd_hdl_getspecific(hdl);
99653d4ee642c6528e88224f12409a5f23060994eschrock
99653d4ee642c6528e88224f12409a5f23060994eschrock if (zhdl != NULL)
99653d4ee642c6528e88224f12409a5f23060994eschrock libzfs_fini(zhdl);
99653d4ee642c6528e88224f12409a5f23060994eschrock}