Add slow disk diagnosis to ZED

[mirror_zfs.git] / cmd / zed / agents / zfs_diagnosis.c

/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or https://opensource.org/licenses/CDDL-1.0.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */

/*
 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
 * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
 * Copyright (c) 2016, Intel Corporation.
 * Copyright (c) 2023, Klara Inc.
 */

#include <stddef.h>
#include <string.h>
#include <libuutil.h>
#include <libzfs.h>
#include <sys/types.h>
#include <sys/time.h>
#include <sys/fs/zfs.h>
#include <sys/fm/protocol.h>
#include <sys/fm/fs/zfs.h>
#include <sys/zio.h>

#include "zfs_agents.h"
#include "fmd_api.h"

/*
 * Default values for the serd engine when processing checksum or io errors. The
 * semantics are N <events> in T <seconds>.
 */
#define	DEFAULT_CHECKSUM_N	10	/* events */
#define	DEFAULT_CHECKSUM_T	600	/* seconds */
#define	DEFAULT_IO_N		10	/* events */
#define	DEFAULT_IO_T		600	/* seconds */
#define	DEFAULT_SLOW_IO_N	10	/* events */
#define	DEFAULT_SLOW_IO_T	30	/* seconds */

#define	CASE_GC_TIMEOUT_SECS	43200	/* 12 hours */

/*
 * Our serd engines are named in the following format:
 *     'zfs_<pool_guid>_<vdev_guid>_{checksum,io,slow_io}'
 * This #define reserves enough space for two 64-bit hex values plus the
 * length of the longest string.
 */
#define	MAX_SERDLEN	(16 * 2 + sizeof ("zfs___checksum"))

/*
 * On-disk case structure.  This must maintain backwards compatibility with
 * previous versions of the DE.  By default, any members appended to the end
 * will be filled with zeros if they don't exist in a previous version.
 */
typedef struct zfs_case_data {
	uint64_t	zc_version;
	uint64_t	zc_ena;
	uint64_t	zc_pool_guid;
	uint64_t	zc_vdev_guid;
	int		zc_pool_state;
	char		zc_serd_checksum[MAX_SERDLEN];
	char		zc_serd_io[MAX_SERDLEN];
	char		zc_serd_slow_io[MAX_SERDLEN];
	int		zc_has_remove_timer;
} zfs_case_data_t;

/*
 * Time-of-day
 */
typedef struct er_timeval {
	uint64_t	ertv_sec;
	uint64_t	ertv_nsec;
} er_timeval_t;

/*
 * In-core case structure.
 */
typedef struct zfs_case {
	boolean_t	zc_present;
	uint32_t	zc_version;
	zfs_case_data_t	zc_data;
	fmd_case_t	*zc_case;
	uu_list_node_t	zc_node;
	id_t		zc_remove_timer;
	char		*zc_fru;
	er_timeval_t	zc_when;
} zfs_case_t;

#define	CASE_DATA			"data"
#define	CASE_FRU			"fru"
#define	CASE_DATA_VERSION_INITIAL	1
#define	CASE_DATA_VERSION_SERD		2

typedef struct zfs_de_stats {
	fmd_stat_t	old_drops;
	fmd_stat_t	dev_drops;
	fmd_stat_t	vdev_drops;
	fmd_stat_t	import_drops;
	fmd_stat_t	resource_drops;
} zfs_de_stats_t;

zfs_de_stats_t zfs_stats = {
	{ "old_drops", FMD_TYPE_UINT64, "ereports dropped (from before load)" },
	{ "dev_drops", FMD_TYPE_UINT64, "ereports dropped (dev during open)"},
	{ "vdev_drops", FMD_TYPE_UINT64, "ereports dropped (weird vdev types)"},
	{ "import_drops", FMD_TYPE_UINT64, "ereports dropped (during import)" },
	{ "resource_drops", FMD_TYPE_UINT64, "resource related ereports" }
};

/* wait 15 seconds after a removal */
static hrtime_t zfs_remove_timeout = SEC2NSEC(15);

uu_list_pool_t *zfs_case_pool;
uu_list_t *zfs_cases;

#define	ZFS_MAKE_RSRC(type)	\
    FM_RSRC_CLASS "." ZFS_ERROR_CLASS "." type
#define	ZFS_MAKE_EREPORT(type)	\
    FM_EREPORT_CLASS "." ZFS_ERROR_CLASS "." type

static void zfs_purge_cases(fmd_hdl_t *hdl);

/*
 * Write out the persistent representation of an active case.
 */
static void
zfs_case_serialize(zfs_case_t *zcp)
{
	zcp->zc_data.zc_version = CASE_DATA_VERSION_SERD;
}

/*
 * Read back the persistent representation of an active case.
 */
static zfs_case_t *
zfs_case_unserialize(fmd_hdl_t *hdl, fmd_case_t *cp)
{
	zfs_case_t *zcp;

	zcp = fmd_hdl_zalloc(hdl, sizeof (zfs_case_t), FMD_SLEEP);
	zcp->zc_case = cp;

	fmd_buf_read(hdl, cp, CASE_DATA, &zcp->zc_data,
	    sizeof (zcp->zc_data));

	if (zcp->zc_data.zc_version > CASE_DATA_VERSION_SERD) {
		fmd_hdl_free(hdl, zcp, sizeof (zfs_case_t));
		return (NULL);
	}

	/*
	 * fmd_buf_read() will have already zeroed out the remainder of the
	 * buffer, so we don't have to do anything special if the version
	 * doesn't include the SERD engine name.
	 */

	if (zcp->zc_data.zc_has_remove_timer)
		zcp->zc_remove_timer = fmd_timer_install(hdl, zcp,
		    NULL, zfs_remove_timeout);

	uu_list_node_init(zcp, &zcp->zc_node, zfs_case_pool);
	(void) uu_list_insert_before(zfs_cases, NULL, zcp);

	fmd_case_setspecific(hdl, cp, zcp);

	return (zcp);
}

/*
 * count other unique slow-io cases in a pool
 */
static uint_t
zfs_other_slow_cases(fmd_hdl_t *hdl, const zfs_case_data_t *zfs_case)
{
	zfs_case_t *zcp;
	uint_t cases = 0;
	static hrtime_t next_check = 0;

	/*
	 * Note that plumbing in some external GC would require adding locking,
	 * since most of this module code is not thread safe and assumes there
	 * is only one thread running against the module. So we perform GC here
	 * inline periodically so that future delay induced faults will be
	 * possible once the issue causing multiple vdev delays is resolved.
	 */
	if (gethrestime_sec() > next_check) {
		/* Periodically purge old SERD entries and stale cases */
		fmd_serd_gc(hdl);
		zfs_purge_cases(hdl);
		next_check = gethrestime_sec() + CASE_GC_TIMEOUT_SECS;
	}

	for (zcp = uu_list_first(zfs_cases); zcp != NULL;
	    zcp = uu_list_next(zfs_cases, zcp)) {
		if (zcp->zc_data.zc_pool_guid == zfs_case->zc_pool_guid &&
		    zcp->zc_data.zc_vdev_guid != zfs_case->zc_vdev_guid &&
		    zcp->zc_data.zc_serd_slow_io[0] != '\0' &&
		    fmd_serd_active(hdl, zcp->zc_data.zc_serd_slow_io)) {
			cases++;
		}
	}
	return (cases);
}

/*
 * Iterate over any active cases.  If any cases are associated with a pool or
 * vdev which is no longer present on the system, close the associated case.
 */
static void
zfs_mark_vdev(uint64_t pool_guid, nvlist_t *vd, er_timeval_t *loaded)
{
	uint64_t vdev_guid = 0;
	uint_t c, children;
	nvlist_t **child;
	zfs_case_t *zcp;

	(void) nvlist_lookup_uint64(vd, ZPOOL_CONFIG_GUID, &vdev_guid);

	/*
	 * Mark any cases associated with this (pool, vdev) pair.
	 */
	for (zcp = uu_list_first(zfs_cases); zcp != NULL;
	    zcp = uu_list_next(zfs_cases, zcp)) {
		if (zcp->zc_data.zc_pool_guid == pool_guid &&
		    zcp->zc_data.zc_vdev_guid == vdev_guid) {
			zcp->zc_present = B_TRUE;
			zcp->zc_when = *loaded;
		}
	}

	/*
	 * Iterate over all children.
	 */
	if (nvlist_lookup_nvlist_array(vd, ZPOOL_CONFIG_CHILDREN, &child,
	    &children) == 0) {
		for (c = 0; c < children; c++)
			zfs_mark_vdev(pool_guid, child[c], loaded);
	}

	if (nvlist_lookup_nvlist_array(vd, ZPOOL_CONFIG_L2CACHE, &child,
	    &children) == 0) {
		for (c = 0; c < children; c++)
			zfs_mark_vdev(pool_guid, child[c], loaded);
	}

	if (nvlist_lookup_nvlist_array(vd, ZPOOL_CONFIG_SPARES, &child,
	    &children) == 0) {
		for (c = 0; c < children; c++)
			zfs_mark_vdev(pool_guid, child[c], loaded);
	}
}

static int
zfs_mark_pool(zpool_handle_t *zhp, void *unused)
{
	(void) unused;
	zfs_case_t *zcp;
	uint64_t pool_guid;
	uint64_t *tod;
	er_timeval_t loaded = { 0 };
	nvlist_t *config, *vd;
	uint_t nelem = 0;
	int ret;

	pool_guid = zpool_get_prop_int(zhp, ZPOOL_PROP_GUID, NULL);
	/*
	 * Mark any cases associated with just this pool.
	 */
	for (zcp = uu_list_first(zfs_cases); zcp != NULL;
	    zcp = uu_list_next(zfs_cases, zcp)) {
		if (zcp->zc_data.zc_pool_guid == pool_guid &&
		    zcp->zc_data.zc_vdev_guid == 0)
			zcp->zc_present = B_TRUE;
	}

	if ((config = zpool_get_config(zhp, NULL)) == NULL) {
		zpool_close(zhp);
		return (-1);
	}

	(void) nvlist_lookup_uint64_array(config, ZPOOL_CONFIG_LOADED_TIME,
	    &tod, &nelem);
	if (nelem == 2) {
		loaded.ertv_sec = tod[0];
		loaded.ertv_nsec = tod[1];
		for (zcp = uu_list_first(zfs_cases); zcp != NULL;
		    zcp = uu_list_next(zfs_cases, zcp)) {
			if (zcp->zc_data.zc_pool_guid == pool_guid &&
			    zcp->zc_data.zc_vdev_guid == 0) {
				zcp->zc_when = loaded;
			}
		}
	}

	ret = nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &vd);
	if (ret) {
		zpool_close(zhp);
		return (-1);
	}

	zfs_mark_vdev(pool_guid, vd, &loaded);

	zpool_close(zhp);

	return (0);
}

struct load_time_arg {
	uint64_t lt_guid;
	er_timeval_t *lt_time;
	boolean_t lt_found;
};

static int
zpool_find_load_time(zpool_handle_t *zhp, void *arg)
{
	struct load_time_arg *lta = arg;
	uint64_t pool_guid;
	uint64_t *tod;
	nvlist_t *config;
	uint_t nelem;

	if (lta->lt_found) {
		zpool_close(zhp);
		return (0);
	}

	pool_guid = zpool_get_prop_int(zhp, ZPOOL_PROP_GUID, NULL);
	if (pool_guid != lta->lt_guid) {
		zpool_close(zhp);
		return (0);
	}

	if ((config = zpool_get_config(zhp, NULL)) == NULL) {
		zpool_close(zhp);
		return (-1);
	}

	if (nvlist_lookup_uint64_array(config, ZPOOL_CONFIG_LOADED_TIME,
	    &tod, &nelem) == 0 && nelem == 2) {
		lta->lt_found = B_TRUE;
		lta->lt_time->ertv_sec = tod[0];
		lta->lt_time->ertv_nsec = tod[1];
	}

	zpool_close(zhp);

	return (0);
}

static void
zfs_purge_cases(fmd_hdl_t *hdl)
{
	zfs_case_t *zcp;
	uu_list_walk_t *walk;
	libzfs_handle_t *zhdl = fmd_hdl_getspecific(hdl);

	/*
	 * There is no way to open a pool by GUID, or lookup a vdev by GUID.  No
	 * matter what we do, we're going to have to stomach an O(vdevs * cases)
	 * algorithm.  In reality, both quantities are likely so small that
	 * neither will matter. Given that iterating over pools is more
	 * expensive than iterating over the in-memory case list, we opt for a
	 * 'present' flag in each case that starts off cleared.  We then iterate
	 * over all pools, marking those that are still present, and removing
	 * those that aren't found.
	 *
	 * Note that we could also construct an FMRI and rely on
	 * fmd_nvl_fmri_present(), but this would end up doing the same search.
	 */

	/*
	 * Mark the cases as not present.
	 */
	for (zcp = uu_list_first(zfs_cases); zcp != NULL;
	    zcp = uu_list_next(zfs_cases, zcp))
		zcp->zc_present = B_FALSE;

	/*
	 * Iterate over all pools and mark the pools and vdevs found.  If this
	 * fails (most probably because we're out of memory), then don't close
	 * any of the cases and we cannot be sure they are accurate.
	 */
	if (zpool_iter(zhdl, zfs_mark_pool, NULL) != 0)
		return;

	/*
	 * Remove those cases which were not found.
	 */
	walk = uu_list_walk_start(zfs_cases, UU_WALK_ROBUST);
	while ((zcp = uu_list_walk_next(walk)) != NULL) {
		if (!zcp->zc_present)
			fmd_case_close(hdl, zcp->zc_case);
	}
	uu_list_walk_end(walk);
}

/*
 * Construct the name of a serd engine given the pool/vdev GUID and type (io or
 * checksum).
 */
static void
zfs_serd_name(char *buf, uint64_t pool_guid, uint64_t vdev_guid,
    const char *type)
{
	(void) snprintf(buf, MAX_SERDLEN, "zfs_%llx_%llx_%s",
	    (long long unsigned int)pool_guid,
	    (long long unsigned int)vdev_guid, type);
}

static void
zfs_case_retire(fmd_hdl_t *hdl, zfs_case_t *zcp)
{
	fmd_hdl_debug(hdl, "retiring case");

	fmd_case_close(hdl, zcp->zc_case);
}

/*
 * Solve a given ZFS case.  This first checks to make sure the diagnosis is
 * still valid, as well as cleaning up any pending timer associated with the
 * case.
 */
static void
zfs_case_solve(fmd_hdl_t *hdl, zfs_case_t *zcp, const char *faultname)
{
	nvlist_t *detector, *fault;
	boolean_t serialize;
	nvlist_t *fru = NULL;
	fmd_hdl_debug(hdl, "solving fault '%s'", faultname);

	/*
	 * Construct the detector from the case data.  The detector is in the
	 * ZFS scheme, and is either the pool or the vdev, depending on whether
	 * this is a vdev or pool fault.
	 */
	detector = fmd_nvl_alloc(hdl, FMD_SLEEP);

	(void) nvlist_add_uint8(detector, FM_VERSION, ZFS_SCHEME_VERSION0);
	(void) nvlist_add_string(detector, FM_FMRI_SCHEME, FM_FMRI_SCHEME_ZFS);
	(void) nvlist_add_uint64(detector, FM_FMRI_ZFS_POOL,
	    zcp->zc_data.zc_pool_guid);
	if (zcp->zc_data.zc_vdev_guid != 0) {
		(void) nvlist_add_uint64(detector, FM_FMRI_ZFS_VDEV,
		    zcp->zc_data.zc_vdev_guid);
	}

	fault = fmd_nvl_create_fault(hdl, faultname, 100, detector,
	    fru, detector);
	fmd_case_add_suspect(hdl, zcp->zc_case, fault);

	nvlist_free(fru);

	fmd_case_solve(hdl, zcp->zc_case);

	serialize = B_FALSE;
	if (zcp->zc_data.zc_has_remove_timer) {
		fmd_timer_remove(hdl, zcp->zc_remove_timer);
		zcp->zc_data.zc_has_remove_timer = 0;
		serialize = B_TRUE;
	}
	if (serialize)
		zfs_case_serialize(zcp);

	nvlist_free(detector);
}

static boolean_t
timeval_earlier(er_timeval_t *a, er_timeval_t *b)
{
	return (a->ertv_sec < b->ertv_sec ||
	    (a->ertv_sec == b->ertv_sec && a->ertv_nsec < b->ertv_nsec));
}

static void
zfs_ereport_when(fmd_hdl_t *hdl, nvlist_t *nvl, er_timeval_t *when)
{
	(void) hdl;
	int64_t *tod;
	uint_t	nelem;

	if (nvlist_lookup_int64_array(nvl, FM_EREPORT_TIME, &tod,
	    &nelem) == 0 && nelem == 2) {
		when->ertv_sec = tod[0];
		when->ertv_nsec = tod[1];
	} else {
		when->ertv_sec = when->ertv_nsec = UINT64_MAX;
	}
}

/*
 * Main fmd entry point.
 */
static void
zfs_fm_recv(fmd_hdl_t *hdl, fmd_event_t *ep, nvlist_t *nvl, const char *class)
{
	zfs_case_t *zcp, *dcp;
	int32_t pool_state;
	uint64_t ena, pool_guid, vdev_guid;
	uint64_t checksum_n, checksum_t;
	uint64_t io_n, io_t;
	er_timeval_t pool_load;
	er_timeval_t er_when;
	nvlist_t *detector;
	boolean_t pool_found = B_FALSE;
	boolean_t isresource;
	const char *type;

	/*
	 * We subscribe to notifications for vdev or pool removal.  In these
	 * cases, there may be cases that no longer apply.  Purge any cases
	 * that no longer apply.
	 */
	if (fmd_nvl_class_match(hdl, nvl, "sysevent.fs.zfs.*")) {
		fmd_hdl_debug(hdl, "purging orphaned cases from %s",
		    strrchr(class, '.') + 1);
		zfs_purge_cases(hdl);
		zfs_stats.resource_drops.fmds_value.ui64++;
		return;
	}

	isresource = fmd_nvl_class_match(hdl, nvl, "resource.fs.zfs.*");

	if (isresource) {
		/*
		 * For resources, we don't have a normal payload.
		 */
		if (nvlist_lookup_uint64(nvl, FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID,
		    &vdev_guid) != 0)
			pool_state = SPA_LOAD_OPEN;
		else
			pool_state = SPA_LOAD_NONE;
		detector = NULL;
	} else {
		(void) nvlist_lookup_nvlist(nvl,
		    FM_EREPORT_DETECTOR, &detector);
		(void) nvlist_lookup_int32(nvl,
		    FM_EREPORT_PAYLOAD_ZFS_POOL_CONTEXT, &pool_state);
	}

	/*
	 * We also ignore all ereports generated during an import of a pool,
	 * since the only possible fault (.pool) would result in import failure,
	 * and hence no persistent fault.  Some day we may want to do something
	 * with these ereports, so we continue generating them internally.
	 */
	if (pool_state == SPA_LOAD_IMPORT) {
		zfs_stats.import_drops.fmds_value.ui64++;
		fmd_hdl_debug(hdl, "ignoring '%s' during import", class);
		return;
	}

	/*
	 * Device I/O errors are ignored during pool open.
	 */
	if (pool_state == SPA_LOAD_OPEN &&
	    (fmd_nvl_class_match(hdl, nvl,
	    ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_CHECKSUM)) ||
	    fmd_nvl_class_match(hdl, nvl,
	    ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_IO)) ||
	    fmd_nvl_class_match(hdl, nvl,
	    ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_PROBE_FAILURE)))) {
		fmd_hdl_debug(hdl, "ignoring '%s' during pool open", class);
		zfs_stats.dev_drops.fmds_value.ui64++;
		return;
	}

	/*
	 * We ignore ereports for anything except disks and files.
	 */
	if (nvlist_lookup_string(nvl, FM_EREPORT_PAYLOAD_ZFS_VDEV_TYPE,
	    &type) == 0) {
		if (strcmp(type, VDEV_TYPE_DISK) != 0 &&
		    strcmp(type, VDEV_TYPE_FILE) != 0) {
			zfs_stats.vdev_drops.fmds_value.ui64++;
			return;
		}
	}

	/*
	 * Determine if this ereport corresponds to an open case.
	 * Each vdev or pool can have a single case.
	 */
	(void) nvlist_lookup_uint64(nvl,
	    FM_EREPORT_PAYLOAD_ZFS_POOL_GUID, &pool_guid);
	if (nvlist_lookup_uint64(nvl,
	    FM_EREPORT_PAYLOAD_ZFS_VDEV_GUID, &vdev_guid) != 0)
		vdev_guid = 0;
	if (nvlist_lookup_uint64(nvl, FM_EREPORT_ENA, &ena) != 0)
		ena = 0;

	zfs_ereport_when(hdl, nvl, &er_when);

	for (zcp = uu_list_first(zfs_cases); zcp != NULL;
	    zcp = uu_list_next(zfs_cases, zcp)) {
		if (zcp->zc_data.zc_pool_guid == pool_guid) {
			pool_found = B_TRUE;
			pool_load = zcp->zc_when;
		}
		if (zcp->zc_data.zc_vdev_guid == vdev_guid)
			break;
	}

	/*
	 * Avoid falsely accusing a pool of being faulty.  Do so by
	 * not replaying ereports that were generated prior to the
	 * current import.  If the failure that generated them was
	 * transient because the device was actually removed but we
	 * didn't receive the normal asynchronous notification, we
	 * don't want to mark it as faulted and potentially panic. If
	 * there is still a problem we'd expect not to be able to
	 * import the pool, or that new ereports will be generated
	 * once the pool is used.
	 */
	if (pool_found && timeval_earlier(&er_when, &pool_load)) {
		fmd_hdl_debug(hdl, "ignoring pool %llx, "
		    "ereport time %lld.%lld, pool load time = %lld.%lld",
		    pool_guid, er_when.ertv_sec, er_when.ertv_nsec,
		    pool_load.ertv_sec, pool_load.ertv_nsec);
		zfs_stats.old_drops.fmds_value.ui64++;
		return;
	}

	if (!pool_found) {
		/*
		 * Haven't yet seen this pool, but same situation
		 * may apply.
		 */
		libzfs_handle_t *zhdl = fmd_hdl_getspecific(hdl);
		struct load_time_arg la;

		la.lt_guid = pool_guid;
		la.lt_time = &pool_load;
		la.lt_found = B_FALSE;

		if (zhdl != NULL &&
		    zpool_iter(zhdl, zpool_find_load_time, &la) == 0 &&
		    la.lt_found == B_TRUE) {
			pool_found = B_TRUE;

			if (timeval_earlier(&er_when, &pool_load)) {
				fmd_hdl_debug(hdl, "ignoring pool %llx, "
				    "ereport time %lld.%lld, "
				    "pool load time = %lld.%lld",
				    pool_guid, er_when.ertv_sec,
				    er_when.ertv_nsec, pool_load.ertv_sec,
				    pool_load.ertv_nsec);
				zfs_stats.old_drops.fmds_value.ui64++;
				return;
			}
		}
	}

	if (zcp == NULL) {
		fmd_case_t *cs;
		zfs_case_data_t data = { 0 };

		/*
		 * If this is one of our 'fake' resource ereports, and there is
		 * no case open, simply discard it.
		 */
		if (isresource) {
			zfs_stats.resource_drops.fmds_value.ui64++;
			fmd_hdl_debug(hdl, "discarding '%s for vdev %llu",
			    class, vdev_guid);
			return;
		}

		/*
		 * Skip tracking some ereports
		 */
		if (strcmp(class,
		    ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_DATA)) == 0 ||
		    strcmp(class,
		    ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE)) == 0) {
			zfs_stats.resource_drops.fmds_value.ui64++;
			return;
		}

		/*
		 * Open a new case.
		 */
		cs = fmd_case_open(hdl, NULL);

		fmd_hdl_debug(hdl, "opening case for vdev %llu due to '%s'",
		    vdev_guid, class);

		/*
		 * Initialize the case buffer.  To commonize code, we actually
		 * create the buffer with existing data, and then call
		 * zfs_case_unserialize() to instantiate the in-core structure.
		 */
		fmd_buf_create(hdl, cs, CASE_DATA, sizeof (zfs_case_data_t));

		data.zc_version = CASE_DATA_VERSION_SERD;
		data.zc_ena = ena;
		data.zc_pool_guid = pool_guid;
		data.zc_vdev_guid = vdev_guid;
		data.zc_pool_state = (int)pool_state;

		fmd_buf_write(hdl, cs, CASE_DATA, &data, sizeof (data));

		zcp = zfs_case_unserialize(hdl, cs);
		assert(zcp != NULL);
		if (pool_found)
			zcp->zc_when = pool_load;
	}

	if (isresource) {
		fmd_hdl_debug(hdl, "resource event '%s'", class);

		if (fmd_nvl_class_match(hdl, nvl,
		    ZFS_MAKE_RSRC(FM_RESOURCE_AUTOREPLACE))) {
			/*
			 * The 'resource.fs.zfs.autoreplace' event indicates
			 * that the pool was loaded with the 'autoreplace'
			 * property set.  In this case, any pending device
			 * failures should be ignored, as the asynchronous
			 * autoreplace handling will take care of them.
			 */
			fmd_case_close(hdl, zcp->zc_case);
		} else if (fmd_nvl_class_match(hdl, nvl,
		    ZFS_MAKE_RSRC(FM_RESOURCE_REMOVED))) {
			/*
			 * The 'resource.fs.zfs.removed' event indicates that
			 * device removal was detected, and the device was
			 * closed asynchronously.  If this is the case, we
			 * assume that any recent I/O errors were due to the
			 * device removal, not any fault of the device itself.
			 * We reset the SERD engine, and cancel any pending
			 * timers.
			 */
			if (zcp->zc_data.zc_has_remove_timer) {
				fmd_timer_remove(hdl, zcp->zc_remove_timer);
				zcp->zc_data.zc_has_remove_timer = 0;
				zfs_case_serialize(zcp);
			}
			if (zcp->zc_data.zc_serd_io[0] != '\0')
				fmd_serd_reset(hdl, zcp->zc_data.zc_serd_io);
			if (zcp->zc_data.zc_serd_checksum[0] != '\0')
				fmd_serd_reset(hdl,
				    zcp->zc_data.zc_serd_checksum);
			if (zcp->zc_data.zc_serd_slow_io[0] != '\0')
				fmd_serd_reset(hdl,
				    zcp->zc_data.zc_serd_slow_io);
		} else if (fmd_nvl_class_match(hdl, nvl,
		    ZFS_MAKE_RSRC(FM_RESOURCE_STATECHANGE))) {
			uint64_t state = 0;

			if (zcp != NULL &&
			    nvlist_lookup_uint64(nvl,
			    FM_EREPORT_PAYLOAD_ZFS_VDEV_STATE, &state) == 0 &&
			    state == VDEV_STATE_HEALTHY) {
				fmd_hdl_debug(hdl, "closing case after a "
				    "device statechange to healthy");
				fmd_case_close(hdl, zcp->zc_case);
			}
		}
		zfs_stats.resource_drops.fmds_value.ui64++;
		return;
	}

	/*
	 * Associate the ereport with this case.
	 */
	fmd_case_add_ereport(hdl, zcp->zc_case, ep);

	/*
	 * Don't do anything else if this case is already solved.
	 */
	if (fmd_case_solved(hdl, zcp->zc_case))
		return;

	if (vdev_guid)
		fmd_hdl_debug(hdl, "error event '%s', vdev %llu", class,
		    vdev_guid);
	else
		fmd_hdl_debug(hdl, "error event '%s'", class);

	/*
	 * Determine if we should solve the case and generate a fault.  We solve
	 * a case if:
	 *
	 * 	a. A pool failed to open (ereport.fs.zfs.pool)
	 * 	b. A device failed to open (ereport.fs.zfs.pool) while a pool
	 *	   was up and running.
	 *
	 * We may see a series of ereports associated with a pool open, all
	 * chained together by the same ENA.  If the pool open succeeds, then
	 * we'll see no further ereports.  To detect when a pool open has
	 * succeeded, we associate a timer with the event.  When it expires, we
	 * close the case.
	 */
	if (fmd_nvl_class_match(hdl, nvl,
	    ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_POOL))) {
		/*
		 * Pool level fault.  Before solving the case, go through and
		 * close any open device cases that may be pending.
		 */
		for (dcp = uu_list_first(zfs_cases); dcp != NULL;
		    dcp = uu_list_next(zfs_cases, dcp)) {
			if (dcp->zc_data.zc_pool_guid ==
			    zcp->zc_data.zc_pool_guid &&
			    dcp->zc_data.zc_vdev_guid != 0)
				fmd_case_close(hdl, dcp->zc_case);
		}

		zfs_case_solve(hdl, zcp, "fault.fs.zfs.pool");
	} else if (fmd_nvl_class_match(hdl, nvl,
	    ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_LOG_REPLAY))) {
		/*
		 * Pool level fault for reading the intent logs.
		 */
		zfs_case_solve(hdl, zcp, "fault.fs.zfs.log_replay");
	} else if (fmd_nvl_class_match(hdl, nvl, "ereport.fs.zfs.vdev.*")) {
		/*
		 * Device fault.
		 */
		zfs_case_solve(hdl, zcp, "fault.fs.zfs.device");
	} else if (fmd_nvl_class_match(hdl, nvl,
	    ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_IO)) ||
	    fmd_nvl_class_match(hdl, nvl,
	    ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_CHECKSUM)) ||
	    fmd_nvl_class_match(hdl, nvl,
	    ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_IO_FAILURE)) ||
	    fmd_nvl_class_match(hdl, nvl,
	    ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_DELAY)) ||
	    fmd_nvl_class_match(hdl, nvl,
	    ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_PROBE_FAILURE))) {
		const char *failmode = NULL;
		boolean_t checkremove = B_FALSE;
		uint32_t pri = 0;
		int32_t flags = 0;

		/*
		 * If this is a checksum or I/O error, then toss it into the
		 * appropriate SERD engine and check to see if it has fired.
		 * Ideally, we want to do something more sophisticated,
		 * (persistent errors for a single data block, etc).  For now,
		 * a single SERD engine is sufficient.
		 */
		if (fmd_nvl_class_match(hdl, nvl,
		    ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_IO))) {
			if (zcp->zc_data.zc_serd_io[0] == '\0') {
				if (nvlist_lookup_uint64(nvl,
				    FM_EREPORT_PAYLOAD_ZFS_VDEV_IO_N,
				    &io_n) != 0) {
					io_n = DEFAULT_IO_N;
				}
				if (nvlist_lookup_uint64(nvl,
				    FM_EREPORT_PAYLOAD_ZFS_VDEV_IO_T,
				    &io_t) != 0) {
					io_t = DEFAULT_IO_T;
				}
				zfs_serd_name(zcp->zc_data.zc_serd_io,
				    pool_guid, vdev_guid, "io");
				fmd_serd_create(hdl, zcp->zc_data.zc_serd_io,
				    io_n,
				    SEC2NSEC(io_t));
				zfs_case_serialize(zcp);
			}
			if (fmd_serd_record(hdl, zcp->zc_data.zc_serd_io, ep))
				checkremove = B_TRUE;
		} else if (fmd_nvl_class_match(hdl, nvl,
		    ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_DELAY))) {
			uint64_t slow_io_n, slow_io_t;

			/*
			 * Create a slow io SERD engine when the VDEV has the
			 * 'vdev_slow_io_n' and 'vdev_slow_io_n' properties.
			 */
			if (zcp->zc_data.zc_serd_slow_io[0] == '\0' &&
			    nvlist_lookup_uint64(nvl,
			    FM_EREPORT_PAYLOAD_ZFS_VDEV_SLOW_IO_N,
			    &slow_io_n) == 0 &&
			    nvlist_lookup_uint64(nvl,
			    FM_EREPORT_PAYLOAD_ZFS_VDEV_SLOW_IO_T,
			    &slow_io_t) == 0) {
				zfs_serd_name(zcp->zc_data.zc_serd_slow_io,
				    pool_guid, vdev_guid, "slow_io");
				fmd_serd_create(hdl,
				    zcp->zc_data.zc_serd_slow_io,
				    slow_io_n,
				    SEC2NSEC(slow_io_t));
				zfs_case_serialize(zcp);
			}
			/* Pass event to SERD engine and see if this triggers */
			if (zcp->zc_data.zc_serd_slow_io[0] != '\0' &&
			    fmd_serd_record(hdl, zcp->zc_data.zc_serd_slow_io,
			    ep)) {
				/*
				 * Ignore a slow io diagnosis when other
				 * VDEVs in the pool show signs of being slow.
				 */
				if (zfs_other_slow_cases(hdl, &zcp->zc_data)) {
					zfs_case_retire(hdl, zcp);
					fmd_hdl_debug(hdl, "pool %llu has "
					    "multiple slow io cases -- skip "
					    "degrading vdev %llu",
					    (u_longlong_t)
					    zcp->zc_data.zc_pool_guid,
					    (u_longlong_t)
					    zcp->zc_data.zc_vdev_guid);
				} else {
					zfs_case_solve(hdl, zcp,
					    "fault.fs.zfs.vdev.slow_io");
				}
			}
		} else if (fmd_nvl_class_match(hdl, nvl,
		    ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_CHECKSUM))) {
			/*
			 * We ignore ereports for checksum errors generated by
			 * scrub/resilver I/O to avoid potentially further
			 * degrading the pool while it's being repaired.
			 */
			if (((nvlist_lookup_uint32(nvl,
			    FM_EREPORT_PAYLOAD_ZFS_ZIO_PRIORITY, &pri) == 0) &&
			    (pri == ZIO_PRIORITY_SCRUB ||
			    pri == ZIO_PRIORITY_REBUILD)) ||
			    ((nvlist_lookup_int32(nvl,
			    FM_EREPORT_PAYLOAD_ZFS_ZIO_FLAGS, &flags) == 0) &&
			    (flags & (ZIO_FLAG_SCRUB | ZIO_FLAG_RESILVER)))) {
				fmd_hdl_debug(hdl, "ignoring '%s' for "
				    "scrub/resilver I/O", class);
				return;
			}

			if (zcp->zc_data.zc_serd_checksum[0] == '\0') {
				if (nvlist_lookup_uint64(nvl,
				    FM_EREPORT_PAYLOAD_ZFS_VDEV_CKSUM_N,
				    &checksum_n) != 0) {
					checksum_n = DEFAULT_CHECKSUM_N;
				}
				if (nvlist_lookup_uint64(nvl,
				    FM_EREPORT_PAYLOAD_ZFS_VDEV_CKSUM_T,
				    &checksum_t) != 0) {
					checksum_t = DEFAULT_CHECKSUM_T;
				}

				zfs_serd_name(zcp->zc_data.zc_serd_checksum,
				    pool_guid, vdev_guid, "checksum");
				fmd_serd_create(hdl,
				    zcp->zc_data.zc_serd_checksum,
				    checksum_n,
				    SEC2NSEC(checksum_t));
				zfs_case_serialize(zcp);
			}
			if (fmd_serd_record(hdl,
			    zcp->zc_data.zc_serd_checksum, ep)) {
				zfs_case_solve(hdl, zcp,
				    "fault.fs.zfs.vdev.checksum");
			}
		} else if (fmd_nvl_class_match(hdl, nvl,
		    ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_IO_FAILURE)) &&
		    (nvlist_lookup_string(nvl,
		    FM_EREPORT_PAYLOAD_ZFS_POOL_FAILMODE, &failmode) == 0) &&
		    failmode != NULL) {
			if (strncmp(failmode, FM_EREPORT_FAILMODE_CONTINUE,
			    strlen(FM_EREPORT_FAILMODE_CONTINUE)) == 0) {
				zfs_case_solve(hdl, zcp,
				    "fault.fs.zfs.io_failure_continue");
			} else if (strncmp(failmode, FM_EREPORT_FAILMODE_WAIT,
			    strlen(FM_EREPORT_FAILMODE_WAIT)) == 0) {
				zfs_case_solve(hdl, zcp,
				    "fault.fs.zfs.io_failure_wait");
			}
		} else if (fmd_nvl_class_match(hdl, nvl,
		    ZFS_MAKE_EREPORT(FM_EREPORT_ZFS_PROBE_FAILURE))) {
#ifndef __linux__
			/* This causes an unexpected fault diagnosis on linux */
			checkremove = B_TRUE;
#endif
		}

		/*
		 * Because I/O errors may be due to device removal, we postpone
		 * any diagnosis until we're sure that we aren't about to
		 * receive a 'resource.fs.zfs.removed' event.
		 */
		if (checkremove) {
			if (zcp->zc_data.zc_has_remove_timer)
				fmd_timer_remove(hdl, zcp->zc_remove_timer);
			zcp->zc_remove_timer = fmd_timer_install(hdl, zcp, NULL,
			    zfs_remove_timeout);
			if (!zcp->zc_data.zc_has_remove_timer) {
				zcp->zc_data.zc_has_remove_timer = 1;
				zfs_case_serialize(zcp);
			}
		}
	}
}

/*
 * The timeout is fired when we diagnosed an I/O error, and it was not due to
 * device removal (which would cause the timeout to be cancelled).
 */
static void
zfs_fm_timeout(fmd_hdl_t *hdl, id_t id, void *data)
{
	zfs_case_t *zcp = data;

	if (id == zcp->zc_remove_timer)
		zfs_case_solve(hdl, zcp, "fault.fs.zfs.vdev.io");
}

/*
 * The specified case has been closed and any case-specific
 * data structures should be deallocated.
 */
static void
zfs_fm_close(fmd_hdl_t *hdl, fmd_case_t *cs)
{
	zfs_case_t *zcp = fmd_case_getspecific(hdl, cs);

	if (zcp->zc_data.zc_serd_checksum[0] != '\0')
		fmd_serd_destroy(hdl, zcp->zc_data.zc_serd_checksum);
	if (zcp->zc_data.zc_serd_io[0] != '\0')
		fmd_serd_destroy(hdl, zcp->zc_data.zc_serd_io);
	if (zcp->zc_data.zc_serd_slow_io[0] != '\0')
		fmd_serd_destroy(hdl, zcp->zc_data.zc_serd_slow_io);
	if (zcp->zc_data.zc_has_remove_timer)
		fmd_timer_remove(hdl, zcp->zc_remove_timer);

	uu_list_remove(zfs_cases, zcp);
	uu_list_node_fini(zcp, &zcp->zc_node, zfs_case_pool);
	fmd_hdl_free(hdl, zcp, sizeof (zfs_case_t));
}

static const fmd_hdl_ops_t fmd_ops = {
	zfs_fm_recv,	/* fmdo_recv */
	zfs_fm_timeout,	/* fmdo_timeout */
	zfs_fm_close,	/* fmdo_close */
	NULL,		/* fmdo_stats */
	NULL,	/* fmdo_gc */
};

static const fmd_prop_t fmd_props[] = {
	{ NULL, 0, NULL }
};

static const fmd_hdl_info_t fmd_info = {
	"ZFS Diagnosis Engine", "1.0", &fmd_ops, fmd_props
};

void
_zfs_diagnosis_init(fmd_hdl_t *hdl)
{
	libzfs_handle_t *zhdl;

	if ((zhdl = libzfs_init()) == NULL)
		return;

	if ((zfs_case_pool = uu_list_pool_create("zfs_case_pool",
	    sizeof (zfs_case_t), offsetof(zfs_case_t, zc_node),
	    NULL, UU_LIST_POOL_DEBUG)) == NULL) {
		libzfs_fini(zhdl);
		return;
	}

	if ((zfs_cases = uu_list_create(zfs_case_pool, NULL,
	    UU_LIST_DEBUG)) == NULL) {
		uu_list_pool_destroy(zfs_case_pool);
		libzfs_fini(zhdl);
		return;
	}

	if (fmd_hdl_register(hdl, FMD_API_VERSION, &fmd_info) != 0) {
		uu_list_destroy(zfs_cases);
		uu_list_pool_destroy(zfs_case_pool);
		libzfs_fini(zhdl);
		return;
	}

	fmd_hdl_setspecific(hdl, zhdl);

	(void) fmd_stat_create(hdl, FMD_STAT_NOALLOC, sizeof (zfs_stats) /
	    sizeof (fmd_stat_t), (fmd_stat_t *)&zfs_stats);
}

void
_zfs_diagnosis_fini(fmd_hdl_t *hdl)
{
	zfs_case_t *zcp;
	uu_list_walk_t *walk;
	libzfs_handle_t *zhdl;

	/*
	 * Remove all active cases.
	 */
	walk = uu_list_walk_start(zfs_cases, UU_WALK_ROBUST);
	while ((zcp = uu_list_walk_next(walk)) != NULL) {
		fmd_hdl_debug(hdl, "removing case ena %llu",
		    (long long unsigned)zcp->zc_data.zc_ena);
		uu_list_remove(zfs_cases, zcp);
		uu_list_node_fini(zcp, &zcp->zc_node, zfs_case_pool);
		fmd_hdl_free(hdl, zcp, sizeof (zfs_case_t));
	}
	uu_list_walk_end(walk);

	uu_list_destroy(zfs_cases);
	uu_list_pool_destroy(zfs_case_pool);

	zhdl = fmd_hdl_getspecific(hdl);
	libzfs_fini(zhdl);
}