xref: /illumos-gate/usr/src/lib/fm/topo/modules/common/ses/ses.c (revision 6efb64ca)

/*
 * 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 http://www.opensolaris.org/os/licensing.
 * 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 2009 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

#include <alloca.h>
#include <dirent.h>
#include <devid.h>
#include <fm/libdiskstatus.h>
#include <inttypes.h>
#include <pthread.h>
#include <strings.h>
#include <unistd.h>
#include <sys/dkio.h>
#include <sys/fm/protocol.h>
#include <sys/scsi/scsi_types.h>

#include "disk.h"
#include "ses.h"

#define	SES_VERSION	1

#define	SES_SNAP_FREQ		1000	/* in milliseconds */

#define	SES_STATUS_UNAVAIL(s)	\
	((s) == SES_ESC_UNSUPPORTED || (s) >= SES_ESC_NOT_INSTALLED)

/*
 * Because multiple SES targets can be part of a single chassis, we construct
 * our own hierarchy that takes this into account.  These SES targets may refer
 * to the same devices (multiple paths) or to different devices (managing
 * different portions of the space).  We arrange things into a
 * ses_enum_enclosure_t, which contains a set of ses targets, and a list of all
 * nodes found so far.
 */
typedef struct ses_alt_node {
	topo_list_t		san_link;
	ses_node_t		*san_node;
} ses_alt_node_t;

typedef struct ses_enum_node {
	topo_list_t		sen_link;
	ses_node_t		*sen_node;
	topo_list_t		sen_alt_nodes;
	uint64_t		sen_type;
	uint64_t		sen_instance;
	ses_enum_target_t	*sen_target;
} ses_enum_node_t;

typedef struct ses_enum_chassis {
	topo_list_t		sec_link;
	topo_list_t		sec_nodes;
	topo_list_t		sec_targets;
	const char		*sec_csn;
	ses_node_t		*sec_enclosure;
	ses_enum_target_t	*sec_target;
	topo_instance_t		sec_instance;
	boolean_t		sec_hasdev;
	boolean_t		sec_internal;
} ses_enum_chassis_t;

typedef struct ses_enum_data {
	topo_list_t		sed_disks;
	topo_list_t		sed_chassis;
	ses_enum_chassis_t	*sed_current;
	ses_enum_target_t	*sed_target;
	int			sed_errno;
	char			*sed_name;
	topo_mod_t		*sed_mod;
	topo_instance_t		sed_instance;
} ses_enum_data_t;

static int ses_present(topo_mod_t *, tnode_t *, topo_version_t, nvlist_t *,
    nvlist_t **);
static int ses_contains(topo_mod_t *, tnode_t *, topo_version_t, nvlist_t *,
    nvlist_t **);

static const topo_method_t ses_component_methods[] = {
	{ TOPO_METH_PRESENT, TOPO_METH_PRESENT_DESC,
	    TOPO_METH_PRESENT_VERSION0, TOPO_STABILITY_INTERNAL, ses_present },
	{ TOPO_METH_FAC_ENUM, TOPO_METH_FAC_ENUM_DESC, 0,
	    TOPO_STABILITY_INTERNAL, ses_node_enum_facility },
	{ NULL }
};

static const topo_method_t ses_bay_methods[] = {
	{ TOPO_METH_FAC_ENUM, TOPO_METH_FAC_ENUM_DESC, 0,
	    TOPO_STABILITY_INTERNAL, ses_node_enum_facility },
	{ NULL }
};

static const topo_method_t ses_enclosure_methods[] = {
	{ TOPO_METH_CONTAINS, TOPO_METH_CONTAINS_DESC,
	    TOPO_METH_CONTAINS_VERSION, TOPO_STABILITY_INTERNAL, ses_contains },
	{ TOPO_METH_FAC_ENUM, TOPO_METH_FAC_ENUM_DESC, 0,
	    TOPO_STABILITY_INTERNAL, ses_enc_enum_facility },
	{ NULL }
};

static void
ses_target_free(topo_mod_t *mod, ses_enum_target_t *stp)
{
	if (--stp->set_refcount == 0) {
		ses_snap_rele(stp->set_snap);
		ses_close(stp->set_target);
		topo_mod_strfree(mod, stp->set_devpath);
		topo_mod_free(mod, stp, sizeof (ses_enum_target_t));
	}
}

static void
ses_data_free(ses_enum_data_t *sdp)
{
	topo_mod_t *mod = sdp->sed_mod;
	ses_enum_chassis_t *cp;
	ses_enum_node_t *np;
	ses_enum_target_t *tp;
	ses_alt_node_t *ap;

	while ((cp = topo_list_next(&sdp->sed_chassis)) != NULL) {
		topo_list_delete(&sdp->sed_chassis, cp);

		while ((np = topo_list_next(&cp->sec_nodes)) != NULL) {
			while ((ap = topo_list_next(&np->sen_alt_nodes)) !=
			    NULL) {
				topo_list_delete(&np->sen_alt_nodes, ap);
				topo_mod_free(mod, ap, sizeof (ses_alt_node_t));
			}
			topo_list_delete(&cp->sec_nodes, np);
			topo_mod_free(mod, np, sizeof (ses_enum_node_t));
		}

		while ((tp = topo_list_next(&cp->sec_targets)) != NULL) {
			topo_list_delete(&cp->sec_targets, tp);
			ses_target_free(mod, tp);
		}

		topo_mod_free(mod, cp, sizeof (ses_enum_chassis_t));
	}

	disk_list_free(mod, &sdp->sed_disks);
	topo_mod_free(mod, sdp, sizeof (ses_enum_data_t));
}

/*
 * For enclosure nodes, we have a special contains method.  By default, the hc
 * walker will compare the node name and instance number to determine if an
 * FMRI matches.  For enclosures where the enumeration order is impossible to
 * predict, we instead use the chassis-id as a unique identifier, and ignore
 * the instance number.
 */
static int
fmri_contains(topo_mod_t *mod, nvlist_t *nv1, nvlist_t *nv2)
{
	uint8_t v1, v2;
	nvlist_t **hcp1, **hcp2;
	int err, i;
	uint_t nhcp1, nhcp2;
	nvlist_t *a1, *a2;
	char *c1, *c2;
	int mindepth;

	if (nvlist_lookup_uint8(nv1, FM_VERSION, &v1) != 0 ||
	    nvlist_lookup_uint8(nv2, FM_VERSION, &v2) != 0 ||
	    v1 > FM_HC_SCHEME_VERSION || v2 > FM_HC_SCHEME_VERSION)
		return (topo_mod_seterrno(mod, EMOD_FMRI_VERSION));

	err = nvlist_lookup_nvlist_array(nv1, FM_FMRI_HC_LIST, &hcp1, &nhcp1);
	err |= nvlist_lookup_nvlist_array(nv2, FM_FMRI_HC_LIST, &hcp2, &nhcp2);
	if (err != 0)
		return (topo_mod_seterrno(mod, EMOD_FMRI_NVL));

	/*
	 * If the chassis-id doesn't match, then these FMRIs are not
	 * equivalent.  If one of the FMRIs doesn't have a chassis ID, then we
	 * have no choice but to fall back to the instance ID.
	 */
	if (nvlist_lookup_nvlist(nv1, FM_FMRI_AUTHORITY, &a1) == 0 &&
	    nvlist_lookup_nvlist(nv2, FM_FMRI_AUTHORITY, &a2) == 0 &&
	    nvlist_lookup_string(a1, FM_FMRI_AUTH_CHASSIS, &c1) == 0 &&
	    nvlist_lookup_string(a2, FM_FMRI_AUTH_CHASSIS, &c2) == 0) {
		if (strcmp(c1, c2) != 0)
			return (0);

		mindepth = 1;
	} else {
		mindepth = 0;
	}

	if (nhcp2 < nhcp1)
		return (0);

	for (i = 0; i < nhcp1; i++) {
		char *nm1 = NULL;
		char *nm2 = NULL;
		char *id1 = NULL;
		char *id2 = NULL;

		(void) nvlist_lookup_string(hcp1[i], FM_FMRI_HC_NAME, &nm1);
		(void) nvlist_lookup_string(hcp2[i], FM_FMRI_HC_NAME, &nm2);
		(void) nvlist_lookup_string(hcp1[i], FM_FMRI_HC_ID, &id1);
		(void) nvlist_lookup_string(hcp2[i], FM_FMRI_HC_ID, &id2);
		if (nm1 == NULL || nm2 == NULL || id1 == NULL || id2 == NULL)
			return (topo_mod_seterrno(mod, EMOD_FMRI_NVL));

		if (strcmp(nm1, nm2) == 0 &&
		    (i < mindepth || strcmp(id1, id2) == 0))
			continue;

		return (0);
	}

	return (1);
}

/*ARGSUSED*/
static int
ses_contains(topo_mod_t *mod, tnode_t *tn, topo_version_t version,
    nvlist_t *in, nvlist_t **out)
{
	int ret;
	nvlist_t *nv1, *nv2;

	if (version > TOPO_METH_CONTAINS_VERSION)
		return (topo_mod_seterrno(mod, EMOD_VER_NEW));

	if (nvlist_lookup_nvlist(in, TOPO_METH_FMRI_ARG_FMRI, &nv1) != 0 ||
	    nvlist_lookup_nvlist(in, TOPO_METH_FMRI_ARG_SUBFMRI, &nv2) != 0)
		return (topo_mod_seterrno(mod, EMOD_METHOD_INVAL));

	ret = fmri_contains(mod, nv1, nv2);
	if (ret < 0)
		return (-1);

	if (topo_mod_nvalloc(mod, out, NV_UNIQUE_NAME) == 0) {
		if (nvlist_add_uint32(*out, TOPO_METH_CONTAINS_RET,
		    ret) == 0)
			return (0);
		else
			nvlist_free(*out);
	}

	return (-1);

}

/*
 * Return a current instance of the node.  This is somewhat complicated because
 * we need to take a new snapshot in order to get the new data, but we don't
 * want to be constantly taking SES snapshots if the consumer is going to do a
 * series of queries.  So we adopt the strategy of assuming that the SES state
 * is not going to be rapidly changing, and limit our snapshot frequency to
 * some defined bounds.
 */
ses_node_t *
ses_node_lock(topo_mod_t *mod, tnode_t *tn)
{
	struct timeval tv;
	ses_enum_target_t *tp = topo_node_getspecific(tn);
	uint64_t prev, now;
	ses_snap_t *snap;
	int err;
	uint64_t nodeid;
	ses_node_t *np;

	if (tp == NULL) {
		(void) topo_mod_seterrno(mod, EMOD_METHOD_NOTSUP);
		return (NULL);
	}

	(void) pthread_mutex_lock(&tp->set_lock);

	/*
	 * Determine if we need to take a new snapshot.
	 */
	if (gettimeofday(&tv, NULL) != 0) {
		tv.tv_sec = time(NULL);
		tv.tv_usec = 0;
	}

	now = tv.tv_sec * 1000 + tv.tv_usec / 1000;
	prev = tp->set_snaptime.tv_sec * 1000 +
	    tp->set_snaptime.tv_usec / 1000;

	if (now - prev > SES_SNAP_FREQ &&
	    (snap = ses_snap_new(tp->set_target)) != NULL) {
		if (ses_snap_generation(snap) !=
		    ses_snap_generation(tp->set_snap)) {
			/*
			 * If we find ourselves in this situation, we're in
			 * trouble.  The generation count has changed, which
			 * indicates that our current topology is out of date.
			 * But we need to consult the new topology in order to
			 * determine presence at this moment in time.  We can't
			 * go back and change the topo snapshot in situ, so
			 * we'll just have to fail the call in this unlikely
			 * scenario.
			 */
			ses_snap_rele(snap);
			(void) topo_mod_seterrno(mod, EMOD_METHOD_NOTSUP);
			(void) pthread_mutex_unlock(&tp->set_lock);
			return (NULL);
		} else {
			ses_snap_rele(tp->set_snap);
			tp->set_snap = snap;
		}
		tp->set_snaptime = tv;
	}

	snap = tp->set_snap;

	verify(topo_prop_get_uint64(tn, TOPO_PGROUP_SES,
	    TOPO_PROP_NODE_ID, &nodeid, &err) == 0);
	verify((np = ses_node_lookup(snap, nodeid)) != NULL);

	return (np);
}

/*ARGSUSED*/
void
ses_node_unlock(topo_mod_t *mod, tnode_t *tn)
{
	ses_enum_target_t *tp = topo_node_getspecific(tn);

	verify(tp != NULL);

	(void) pthread_mutex_unlock(&tp->set_lock);
}

/*
 * Determine if the element is present.
 */
/*ARGSUSED*/
static int
ses_present(topo_mod_t *mod, tnode_t *tn, topo_version_t version,
    nvlist_t *in, nvlist_t **out)
{
	boolean_t present;
	ses_node_t *np;
	nvlist_t *props, *nvl;
	uint64_t status;

	if ((np = ses_node_lock(mod, tn)) == NULL)
		return (-1);

	verify((props = ses_node_props(np)) != NULL);
	verify(nvlist_lookup_uint64(props,
	    SES_PROP_STATUS_CODE, &status) == 0);

	ses_node_unlock(mod, tn);

	present = (status != SES_ESC_NOT_INSTALLED);

	if (topo_mod_nvalloc(mod, &nvl, NV_UNIQUE_NAME) != 0)
		return (topo_mod_seterrno(mod, EMOD_FMRI_NVL));

	if (nvlist_add_uint32(nvl, TOPO_METH_PRESENT_RET,
	    present) != 0) {
		nvlist_free(nvl);
		return (topo_mod_seterrno(mod, EMOD_FMRI_NVL));
	}

	*out = nvl;

	return (0);
}

/*
 * Sets standard properties for a ses node (enclosure or bay).  This includes
 * setting the FRU to be the same as the resource, as well as setting the
 * authority information.
 */
static int
ses_set_standard_props(topo_mod_t *mod, tnode_t *tn, nvlist_t *auth,
    uint64_t nodeid, const char *path)
{
	int err;
	char *product, *chassis;
	nvlist_t *fmri;
	topo_pgroup_info_t pgi;

	/*
	 * Set the authority explicitly if specified.
	 */
	if (auth) {
		verify(nvlist_lookup_string(auth, FM_FMRI_AUTH_PRODUCT,
		    &product) == 0);
		verify(nvlist_lookup_string(auth, FM_FMRI_AUTH_CHASSIS,
		    &chassis) == 0);
		if (topo_prop_set_string(tn, FM_FMRI_AUTHORITY,
		    FM_FMRI_AUTH_PRODUCT, TOPO_PROP_IMMUTABLE, product,
		    &err) != 0 ||
		    topo_prop_set_string(tn, FM_FMRI_AUTHORITY,
		    FM_FMRI_AUTH_CHASSIS, TOPO_PROP_IMMUTABLE, chassis,
		    &err) != 0 ||
		    topo_prop_set_string(tn, FM_FMRI_AUTHORITY,
		    FM_FMRI_AUTH_SERVER, TOPO_PROP_IMMUTABLE, "",
		    &err) != 0) {
			topo_mod_dprintf(mod, "failed to add authority "
			    "properties: %s\n", topo_strerror(err));
			return (topo_mod_seterrno(mod, err));
		}
	}

	/*
	 * Copy the resource and set that as the FRU.
	 */
	if (topo_node_resource(tn, &fmri, &err) != 0) {
		topo_mod_dprintf(mod,
		    "topo_node_resource() failed : %s\n",
		    topo_strerror(err));
		return (topo_mod_seterrno(mod, err));
	}

	if (topo_node_fru_set(tn, fmri, 0, &err) != 0) {
		topo_mod_dprintf(mod,
		    "topo_node_fru_set() failed : %s\n",
		    topo_strerror(err));
		nvlist_free(fmri);
		return (topo_mod_seterrno(mod, err));
	}

	nvlist_free(fmri);

	/*
	 * Set the SES-specific properties so that consumers can query
	 * additional information about the particular SES element.
	 */
	pgi.tpi_name = TOPO_PGROUP_SES;
	pgi.tpi_namestab = TOPO_STABILITY_PRIVATE;
	pgi.tpi_datastab = TOPO_STABILITY_PRIVATE;
	pgi.tpi_version = TOPO_VERSION;
	if (topo_pgroup_create(tn, &pgi, &err) != 0) {
		topo_mod_dprintf(mod, "failed to create propgroup "
		    "%s: %s\n", TOPO_PGROUP_SES, topo_strerror(err));
		return (-1);
	}

	if (topo_prop_set_uint64(tn, TOPO_PGROUP_SES,
	    TOPO_PROP_NODE_ID, TOPO_PROP_IMMUTABLE,
	    nodeid, &err) != 0) {
		topo_mod_dprintf(mod,
		    "failed to create property %s: %s\n",
		    TOPO_PROP_NODE_ID, topo_strerror(err));
		return (-1);
	}

	if (topo_prop_set_string(tn, TOPO_PGROUP_SES,
	    TOPO_PROP_TARGET_PATH, TOPO_PROP_IMMUTABLE,
	    path, &err) != 0) {
		topo_mod_dprintf(mod,
		    "failed to create property %s: %s\n",
		    TOPO_PROP_TARGET_PATH, topo_strerror(err));
		return (-1);
	}

	return (0);
}

/*
 * Callback to add a disk to a given bay.  We first check the status-code to
 * determine if a disk is present, ignoring those that aren't in an appropriate
 * state.  We then scan the parent bay node's SAS address array to determine
 * possible attached SAS addresses.  We create a disk node if the disk is not
 * SAS or the SES target does not support the necessary pages for this; if we
 * find the SAS address, we create a disk node and also correlate it with
 * the corresponding Solaris device node to fill in the rest of the data.
 */
static int
ses_create_disk(ses_enum_data_t *sdp, tnode_t *pnode, nvlist_t *props)
{
	topo_mod_t *mod = sdp->sed_mod;
	uint64_t status;
	nvlist_t **sas;
	uint_t s, nsas;
	char **paths;
	int err;

	/*
	 * Skip devices that are not in a present (and possibly damaged) state.
	 */
	if (nvlist_lookup_uint64(props, SES_PROP_STATUS_CODE, &status) != 0)
		return (0);

	if (status != SES_ESC_OK &&
	    status != SES_ESC_CRITICAL &&
	    status != SES_ESC_NONCRITICAL &&
	    status != SES_ESC_UNRECOVERABLE &&
	    status != SES_ESC_NO_ACCESS)
		return (0);

	topo_mod_dprintf(mod, "found attached disk");

	/*
	 * Create the disk range.
	 */
	if (topo_node_range_create(mod, pnode, DISK, 0, 0) != 0) {
		topo_mod_dprintf(mod,
		    "topo_node_create_range() failed: %s",
		    topo_mod_errmsg(mod));
		return (-1);
	}

	/*
	 * Look through all SAS addresses and attempt to correlate them to a
	 * known Solaris device.  If we don't find a matching node, then we
	 * don't enumerate the disk node.
	 */
	if (nvlist_lookup_nvlist_array(props, SES_SAS_PROP_PHYS,
	    &sas, &nsas) != 0)
		return (0);

	if (topo_prop_get_string_array(pnode, TOPO_PGROUP_SES,
	    TOPO_PROP_SAS_ADDR, &paths, &nsas, &err) != 0)
		return (0);

	err = 0;

	for (s = 0; s < nsas; s++) {
		if (disk_declare_addr(mod, pnode,
		    &sdp->sed_disks, paths[s]) != 0 &&
		    topo_mod_errno(mod) != EMOD_NODE_BOUND) {
			err = -1;
			break;
		}
	}

	for (s = 0; s < nsas; s++)
		topo_mod_free(mod, paths[s], strlen(paths[s]) + 1);
	topo_mod_free(mod, paths, nsas * sizeof (char *));

	return (err);
}

static int
ses_add_bay_props(topo_mod_t *mod, tnode_t *tn, ses_enum_node_t *snp)
{
	ses_alt_node_t *ap;
	ses_node_t *np;
	nvlist_t *props;

	nvlist_t **phys;
	uint_t i, j, n_phys, all_phys = 0;
	char **paths;
	uint64_t addr;
	size_t len;
	int terr, err = -1;

	for (ap = topo_list_next(&snp->sen_alt_nodes); ap != NULL;
	    ap = topo_list_next(ap)) {
		np = ap->san_node;
		props = ses_node_props(np);

		if (nvlist_lookup_nvlist_array(props, SES_SAS_PROP_PHYS,
		    &phys, &n_phys) != 0)
			continue;

		all_phys += n_phys;
	}

	if (all_phys == 0)
		return (0);

	if ((paths = topo_mod_zalloc(mod, all_phys * sizeof (char *))) == NULL)
		return (-1);

	for (i = 0, ap = topo_list_next(&snp->sen_alt_nodes); ap != NULL;
	    ap = topo_list_next(ap)) {
		np = ap->san_node;
		props = ses_node_props(np);

		if (nvlist_lookup_nvlist_array(props, SES_SAS_PROP_PHYS,
		    &phys, &n_phys) != 0)
			continue;

		for (j = 0; j < n_phys; j++) {
			if (nvlist_lookup_uint64(phys[j], SES_SAS_PROP_ADDR,
			    &addr) != 0)
				continue;

			len = snprintf(NULL, 0, "%016llx", addr) + 1;
			if ((paths[i] = topo_mod_alloc(mod, len)) == NULL)
				goto error;

			(void) snprintf(paths[i], len, "%016llx", addr);

			++i;
		}
	}

	err = topo_prop_set_string_array(tn, TOPO_PGROUP_SES,
	    TOPO_PROP_SAS_ADDR, TOPO_PROP_IMMUTABLE,
	    (const char **)paths, i, &terr);
	if (err != 0)
		err = topo_mod_seterrno(mod, terr);

error:
	for (i = 0; i < all_phys && paths[i] != NULL; i++)
		topo_mod_free(mod, paths[i], strlen(paths[i]) + 1);
	topo_mod_free(mod, paths, all_phys * sizeof (char *));

	return (err);
}

/*
 * Callback to create a basic node (bay, psu, fan, or controller).
 */
static int
ses_create_generic(ses_enum_data_t *sdp, ses_enum_node_t *snp,
    tnode_t *pnode, const char *nodename, const char *labelname)
{
	ses_node_t *np = snp->sen_node;
	ses_node_t *parent;
	uint64_t instance = snp->sen_instance;
	topo_mod_t *mod = sdp->sed_mod;
	nvlist_t *props, *aprops;
	nvlist_t *auth = NULL, *fmri = NULL;
	tnode_t *tn;
	char label[128];
	int err;
	char *part = NULL, *serial = NULL, *revision = NULL;
	char *desc;
	boolean_t report;

	props = ses_node_props(np);

	(void) nvlist_lookup_string(props, LIBSES_PROP_PART, &part);
	(void) nvlist_lookup_string(props, LIBSES_PROP_SERIAL, &serial);

	topo_mod_dprintf(mod, "adding %s %llu", nodename, instance);

	/*
	 * Create the node.  The interesting information is all copied from the
	 * parent enclosure node, so there is not much to do.
	 */
	if ((auth = topo_mod_auth(mod, pnode)) == NULL)
		goto error;

	/*
	 * We want to report revision information for the controller nodes, but
	 * we do not get per-element revision information.  However, we do have
	 * revision information for the entire enclosure, and we can use the
	 * 'reported-via' property to know that this controller corresponds to
	 * the given revision information.  This means we cannot get revision
	 * information for targets we are not explicitly connected to, but
	 * there is little we can do about the situation.
	 */
	if (strcmp(nodename, CONTROLLER) == 0 &&
	    nvlist_lookup_boolean_value(props, SES_PROP_REPORT, &report) == 0 &&
	    report) {
		for (parent = ses_node_parent(np); parent != NULL;
		    parent = ses_node_parent(parent)) {
			if (ses_node_type(parent) == SES_NODE_ENCLOSURE) {
				(void) nvlist_lookup_string(
				    ses_node_props(parent),
				    SES_EN_PROP_REV, &revision);
				break;
			}
		}
	}

	if ((fmri = topo_mod_hcfmri(mod, pnode, FM_HC_SCHEME_VERSION,
	    nodename, (topo_instance_t)instance, NULL, auth, part, revision,
	    serial)) == NULL) {
		topo_mod_dprintf(mod, "topo_mod_hcfmri() failed: %s",
		    topo_mod_errmsg(mod));
		goto error;
	}

	if ((tn = topo_node_bind(mod, pnode, nodename,
	    instance, fmri)) == NULL) {
		topo_mod_dprintf(mod, "topo_node_bind() failed: %s",
		    topo_mod_errmsg(mod));
		goto error;
	}

	/*
	 * For the node label, we look for the following in order:
	 *
	 * 	<ses-description>
	 * 	<ses-class-description> <instance>
	 * 	<default-type-label> <instance>
	 */
	if (nvlist_lookup_string(props, SES_PROP_DESCRIPTION, &desc) != 0 ||
	    desc[0] == '\0') {
		parent = ses_node_parent(np);
		aprops = ses_node_props(parent);
		if (nvlist_lookup_string(aprops, SES_PROP_CLASS_DESCRIPTION,
		    &desc) == 0 && desc[0] != '\0')
			labelname = desc;
		(void) snprintf(label, sizeof (label), "%s %llu", desc,
		    instance);
		desc = label;
	}

	if (topo_node_label_set(tn, desc, &err) != 0)
		goto error;

	if (ses_set_standard_props(mod, tn, NULL, ses_node_id(np),
	    snp->sen_target->set_devpath) != 0)
		goto error;

	if (strcmp(nodename, "bay") == 0) {
		if (ses_add_bay_props(mod, tn, snp) != 0)
			goto error;

		if (ses_create_disk(sdp, tn, props) != 0)
			goto error;

		if (topo_method_register(mod, tn, ses_bay_methods) != 0) {
			topo_mod_dprintf(mod,
			    "topo_method_register() failed: %s",
			    topo_mod_errmsg(mod));
			goto error;
		}
	} else {
		/*
		 * Only fan, psu, and controller nodes have a 'present' method.
		 * Bay nodes are always present, and disk nodes are present by
		 * virtue of being enumerated.
		 */
		if (topo_method_register(mod, tn, ses_component_methods) != 0) {
			topo_mod_dprintf(mod,
			    "topo_method_register() failed: %s",
			    topo_mod_errmsg(mod));
			goto error;
		}

	}

	snp->sen_target->set_refcount++;
	topo_node_setspecific(tn, snp->sen_target);

	nvlist_free(auth);
	nvlist_free(fmri);
	return (0);

error:
	nvlist_free(auth);
	nvlist_free(fmri);
	return (-1);
}

/*
 * Instantiate any children of a given type.
 */
static int
ses_create_children(ses_enum_data_t *sdp, tnode_t *pnode, uint64_t type,
    const char *nodename, const char *defaultlabel, ses_enum_chassis_t *cp,
    boolean_t dorange)
{
	topo_mod_t *mod = sdp->sed_mod;
	boolean_t found;
	uint64_t max;
	ses_enum_node_t *snp;

	/*
	 * First go through and count how many matching nodes we have.
	 */
	max = 0;
	found = B_FALSE;
	for (snp = topo_list_next(&cp->sec_nodes); snp != NULL;
	    snp = topo_list_next(snp)) {
		if (snp->sen_type == type) {
			found = B_TRUE;
			if (snp->sen_instance > max)
				max = snp->sen_instance;
		}
	}

	/*
	 * No enclosure should export both DEVICE and ARRAY_DEVICE elements.
	 * Since we map both of these to 'disk', if an enclosure does this, we
	 * just ignore the array elements.
	 */
	if (!found ||
	    (type == SES_ET_ARRAY_DEVICE && cp->sec_hasdev))
		return (0);

	topo_mod_dprintf(mod, "%s: creating %llu %s nodes",
	    cp->sec_csn, max, nodename);

	if (dorange && topo_node_range_create(mod, pnode,
	    nodename, 0, max) != 0) {
		topo_mod_dprintf(mod,
		    "topo_node_create_range() failed: %s",
		    topo_mod_errmsg(mod));
		return (-1);
	}

	for (snp = topo_list_next(&cp->sec_nodes); snp != NULL;
	    snp = topo_list_next(snp)) {
		if (snp->sen_type == type) {
			if (ses_create_generic(sdp, snp, pnode,
			    nodename, defaultlabel) != 0)
				return (-1);
		}
	}

	return (0);
}

/*
 * Instantiate a new chassis instance in the topology.
 */
static int
ses_create_chassis(ses_enum_data_t *sdp, tnode_t *pnode, ses_enum_chassis_t *cp)
{
	topo_mod_t *mod = sdp->sed_mod;
	nvlist_t *props;
	char *raw_manufacturer, *raw_model, *raw_revision;
	char *manufacturer = NULL, *model = NULL, *product = NULL;
	char *revision = NULL;
	char *serial;
	char **paths;
	size_t prodlen;
	tnode_t *tn;
	nvlist_t *fmri = NULL, *auth = NULL;
	int ret = -1;
	ses_enum_node_t *snp;
	ses_enum_target_t *stp;
	int i, err;

	/*
	 * Ignore any internal enclosures.
	 */
	if (cp->sec_internal)
		return (0);

	/*
	 * Check to see if there are any devices presennt in the chassis.  If
	 * not, ignore the chassis alltogether.  This is most useful for
	 * ignoring internal HBAs that present a SES target but don't actually
	 * manage any of the devices.
	 */
	for (snp = topo_list_next(&cp->sec_nodes); snp != NULL;
	    snp = topo_list_next(snp)) {
		if (snp->sen_type == SES_ET_DEVICE ||
		    snp->sen_type == SES_ET_ARRAY_DEVICE)
			break;
	}

	if (snp == NULL)
		return (0);

	props = ses_node_props(cp->sec_enclosure);

	/*
	 * We use the following property mappings:
	 *
	 * 	manufacturer		vendor-id
	 * 	model			product-id
	 * 	serial-number		libses-chassis-serial
	 */
	verify(nvlist_lookup_string(props, SES_EN_PROP_VID,
	    &raw_manufacturer) == 0);
	verify(nvlist_lookup_string(props, SES_EN_PROP_PID, &raw_model) == 0);
	verify(nvlist_lookup_string(props, SES_EN_PROP_REV,
	    &raw_revision) == 0);
	verify(nvlist_lookup_string(props, LIBSES_EN_PROP_CSN, &serial) == 0);

	/*
	 * To construct the authority information, we 'clean' each string by
	 * removing any offensive characters and trimmming whitespace.  For the
	 * 'product-id', we use a concatenation of 'manufacturer-model'.  We
	 * also take the numerical serial number and convert it to a string.
	 */
	if ((manufacturer = disk_auth_clean(mod, raw_manufacturer)) == NULL ||
	    (model = disk_auth_clean(mod, raw_model)) == NULL ||
	    (revision = disk_auth_clean(mod, raw_revision)) == NULL) {
		goto error;
	}

	prodlen = strlen(manufacturer) + strlen(model) + 2;
	if ((product = topo_mod_alloc(mod, prodlen)) == NULL)
		goto error;

	(void) snprintf(product, prodlen, "%s-%s", manufacturer, model);

	/*
	 * Construct the topo node and bind it to our parent.
	 */
	if (topo_mod_nvalloc(mod, &auth, NV_UNIQUE_NAME) != 0)
		goto error;

	if (nvlist_add_string(auth, FM_FMRI_AUTH_PRODUCT, product) != 0 ||
	    nvlist_add_string(auth, FM_FMRI_AUTH_CHASSIS, serial) != 0) {
		(void) topo_mod_seterrno(mod, EMOD_NVL_INVAL);
		goto error;
	}

	/*
	 * We pass NULL for the parent FMRI because there is no resource
	 * associated with it.  For the toplevel enclosure, we leave the
	 * serial/part/revision portions empty, which are reserved for
	 * individual components within the chassis.
	 */
	if ((fmri = topo_mod_hcfmri(mod, NULL, FM_HC_SCHEME_VERSION,
	    SES_ENCLOSURE, cp->sec_instance, NULL, auth,
	    model, revision, serial)) == NULL) {
		topo_mod_dprintf(mod, "topo_mod_hcfmri() failed: %s",
		    topo_mod_errmsg(mod));
		goto error;
	}

	if ((tn = topo_node_bind(mod, pnode, SES_ENCLOSURE,
	    cp->sec_instance, fmri)) == NULL) {
		topo_mod_dprintf(mod, "topo_node_bind() failed: %s",
		    topo_mod_errmsg(mod));
		goto error;
	}

	if (topo_method_register(mod, tn, ses_enclosure_methods) != 0) {
		topo_mod_dprintf(mod,
		    "topo_method_register() failed: %s",
		    topo_mod_errmsg(mod));
		goto error;
	}

	if (ses_set_standard_props(mod, tn, auth,
	    ses_node_id(cp->sec_enclosure), cp->sec_target->set_devpath) != 0)
		goto error;

	/*
	 * For enclosures, we want to include all possible targets (for upgrade
	 * purposes).
	 */
	for (i = 0, stp = topo_list_next(&cp->sec_targets); stp != NULL;
	    stp = topo_list_next(stp), i++)
		;

	verify(i != 0);
	paths = alloca(i * sizeof (char *));

	for (i = 0, stp = topo_list_next(&cp->sec_targets); stp != NULL;
	    stp = topo_list_next(stp), i++)
		paths[i] = stp->set_devpath;


	if (topo_prop_set_string_array(tn, TOPO_PGROUP_SES,
	    TOPO_PROP_PATHS, TOPO_PROP_IMMUTABLE, (const char **)paths,
	    i, &err) != 0) {
		topo_mod_dprintf(mod,
		    "failed to create property %s: %s\n",
		    TOPO_PROP_PATHS, topo_strerror(err));
		goto error;
	}

	/*
	 * Create the nodes for power supplies, fans, and devices.
	 */
	if (ses_create_children(sdp, tn, SES_ET_POWER_SUPPLY,
	    PSU, "PSU", cp, B_TRUE) != 0 ||
	    ses_create_children(sdp, tn, SES_ET_COOLING,
	    FAN, "FAN", cp, B_TRUE) != 0 ||
	    ses_create_children(sdp, tn, SES_ET_ESC_ELECTRONICS,
	    CONTROLLER, "CONTROLLER", cp, B_TRUE) != 0 ||
	    ses_create_children(sdp, tn, SES_ET_DEVICE,
	    BAY, "BAY", cp, B_TRUE) != 0 ||
	    ses_create_children(sdp, tn, SES_ET_ARRAY_DEVICE,
	    BAY, "BAY", cp, B_TRUE) != 0)
		goto error;

	cp->sec_target->set_refcount++;
	topo_node_setspecific(tn, cp->sec_target);

	ret = 0;
error:
	topo_mod_strfree(mod, manufacturer);
	topo_mod_strfree(mod, model);
	topo_mod_strfree(mod, revision);
	topo_mod_strfree(mod, product);

	nvlist_free(fmri);
	nvlist_free(auth);
	return (ret);
}

/*
 * Create a bay node explicitly enumerated via XML.
 */
static int
ses_create_bays(ses_enum_data_t *sdp, tnode_t *pnode)
{
	topo_mod_t *mod = sdp->sed_mod;
	ses_enum_chassis_t *cp;

	/*
	 * Iterate over chassis looking for an internal enclosure.  This
	 * property is set via a vendor-specific plugin, and there should only
	 * ever be a single internal chassis in a system.
	 */
	for (cp = topo_list_next(&sdp->sed_chassis); cp != NULL;
	    cp = topo_list_next(cp)) {
		if (cp->sec_internal)
			break;
	}

	if (cp == NULL) {
		topo_mod_dprintf(mod, "failed to find internal chassis\n");
		return (-1);
	}

	if (ses_create_children(sdp, pnode, SES_ET_DEVICE,
	    BAY, "BAY", cp, B_FALSE) != 0 ||
	    ses_create_children(sdp, pnode, SES_ET_ARRAY_DEVICE,
	    BAY, "BAY", cp, B_FALSE) != 0)
		return (-1);

	return (0);
}
/*
 * Gather nodes from the current SES target into our chassis list, merging the
 * results if necessary.
 */
static ses_walk_action_t
ses_enum_gather(ses_node_t *np, void *data)
{
	nvlist_t *props = ses_node_props(np);
	ses_enum_data_t *sdp = data;
	topo_mod_t *mod = sdp->sed_mod;
	ses_enum_chassis_t *cp;
	ses_enum_node_t *snp;
	ses_alt_node_t *sap;
	char *csn;
	uint64_t instance, type;
	uint64_t prevstatus, status;
	boolean_t report, internal, ident;

	if (ses_node_type(np) == SES_NODE_ENCLOSURE) {
		/*
		 * If we have already identified the chassis for this target,
		 * then this is a secondary enclosure and we should ignore it,
		 * along with the rest of the tree (since this is depth-first).
		 */
		if (sdp->sed_current != NULL)
			return (SES_WALK_ACTION_TERMINATE);

		/*
		 * Go through the list of chassis we have seen so far and see
		 * if this serial number matches one of the known values.
		 */
		if (nvlist_lookup_string(props, LIBSES_EN_PROP_CSN,
		    &csn) != 0)
			return (SES_WALK_ACTION_TERMINATE);

		for (cp = topo_list_next(&sdp->sed_chassis); cp != NULL;
		    cp = topo_list_next(cp)) {
			if (strcmp(cp->sec_csn, csn) == 0) {
				topo_mod_dprintf(mod, "%s: part of already "
				    "known chassis %s", sdp->sed_name, csn);
				break;
			}
		}

		if (cp == NULL) {
			topo_mod_dprintf(mod, "%s: creating chassis %s",
			    sdp->sed_name, csn);

			if ((cp = topo_mod_zalloc(mod,
			    sizeof (ses_enum_chassis_t))) == NULL)
				goto error;

			if (nvlist_lookup_boolean_value(props,
			    LIBSES_EN_PROP_INTERNAL, &internal) == 0)
				cp->sec_internal = internal;

			cp->sec_csn = csn;
			cp->sec_enclosure = np;
			cp->sec_target = sdp->sed_target;
			cp->sec_instance = sdp->sed_instance++;
			topo_list_append(&sdp->sed_chassis, cp);
		} else {
			/*
			 * For the enclosure node, it is possible to have
			 * multiple targets, only one of which support an
			 * enclosure element descriptor.  We assume that this
			 * is the one that is responsible for managing the
			 * enclosure itself, so we prefer one with the
			 * SES_PROP_IDENT property (which is only present for a
			 * target that has an enclosure element descriptor).
			 */
			if (nvlist_lookup_boolean_value(props, SES_PROP_IDENT,
			    &ident) == 0) {
				topo_mod_dprintf(mod,
				    "overriding enclosure node");
				cp->sec_enclosure = np;
				cp->sec_target = sdp->sed_target;
			}
		}

		topo_list_append(&cp->sec_targets, sdp->sed_target);
		sdp->sed_current = cp;

	} else if (ses_node_type(np) == SES_NODE_ELEMENT) {
		/*
		 * If we haven't yet seen an enclosure node and identified the
		 * current chassis, something is very wrong; bail out.
		 */
		if (sdp->sed_current == NULL)
			return (SES_WALK_ACTION_TERMINATE);

		/*
		 * If this isn't one of the element types we care about, then
		 * ignore it.
		 */
		verify(nvlist_lookup_uint64(props, SES_PROP_ELEMENT_TYPE,
		    &type) == 0);
		if (type != SES_ET_DEVICE &&
		    type != SES_ET_ARRAY_DEVICE &&
		    type != SES_ET_COOLING &&
		    type != SES_ET_POWER_SUPPLY &&
		    type != SES_ET_ESC_ELECTRONICS)
			return (SES_WALK_ACTION_CONTINUE);

		/*
		 * Get the current instance number and see if we already know
		 * about this element.  If so, it means we have multiple paths
		 * to the same elements, and we should ignore the current path.
		 */
		verify(nvlist_lookup_uint64(props, SES_PROP_ELEMENT_CLASS_INDEX,
		    &instance) == 0);
		if (type == SES_ET_DEVICE || type == SES_ET_ARRAY_DEVICE)
			(void) nvlist_lookup_uint64(props, SES_PROP_BAY_NUMBER,
			    &instance);

		cp = sdp->sed_current;

		for (snp = topo_list_next(&cp->sec_nodes); snp != NULL;
		    snp = topo_list_next(snp)) {
			if (snp->sen_type == type &&
			    snp->sen_instance == instance)
				break;
		}

		/*
		 * We prefer the new element under the following circumstances:
		 *
		 * - The currently known element's status is unknown or not
		 *   available, but the new element has a known status.  This
		 *   occurs if a given element is only available through a
		 *   particular target.
		 *
		 * - This is an ESC_ELECTRONICS element, and the 'reported-via'
		 *   property is set.  This allows us to get reliable firmware
		 *   revision information from the enclosure node.
		 */
		if (snp != NULL) {
			if (nvlist_lookup_uint64(
			    ses_node_props(snp->sen_node),
			    SES_PROP_STATUS_CODE, &prevstatus) != 0)
				prevstatus = SES_ESC_UNSUPPORTED;
			if (nvlist_lookup_uint64(
			    props, SES_PROP_STATUS_CODE, &status) != 0)
				status = SES_ESC_UNSUPPORTED;
			if (nvlist_lookup_boolean_value(
			    props, SES_PROP_REPORT, &report) != 0)
				report = B_FALSE;

			if ((SES_STATUS_UNAVAIL(prevstatus) &&
			    !SES_STATUS_UNAVAIL(status)) ||
			    (type == SES_ET_ESC_ELECTRONICS &&
			    report)) {
				snp->sen_node = np;
				snp->sen_target = sdp->sed_target;
			}

			if ((sap = topo_mod_zalloc(mod,
			    sizeof (ses_alt_node_t))) == NULL)
				goto error;

			sap->san_node = np;
			topo_list_append(&snp->sen_alt_nodes, sap);

			return (SES_WALK_ACTION_CONTINUE);
		}

		if ((snp = topo_mod_zalloc(mod,
		    sizeof (ses_enum_node_t))) == NULL)
			goto error;

		if ((sap = topo_mod_zalloc(mod,
		    sizeof (ses_alt_node_t))) == NULL) {
			topo_mod_free(mod, snp, sizeof (ses_enum_node_t));
			goto error;
		}

		topo_mod_dprintf(mod, "%s: adding node (%llu, %llu)",
		    sdp->sed_name, type, instance);
		snp->sen_node = np;
		snp->sen_type = type;
		snp->sen_instance = instance;
		snp->sen_target = sdp->sed_target;
		sap->san_node = np;
		topo_list_append(&snp->sen_alt_nodes, sap);
		topo_list_append(&cp->sec_nodes, snp);

		if (type == SES_ET_DEVICE)
			cp->sec_hasdev = B_TRUE;
	}

	return (SES_WALK_ACTION_CONTINUE);

error:
	sdp->sed_errno = -1;
	return (SES_WALK_ACTION_TERMINATE);
}

static int
ses_process_dir(const char *dirpath, ses_enum_data_t *sdp)
{
	topo_mod_t *mod = sdp->sed_mod;
	DIR *dir;
	struct dirent *dp;
	char path[PATH_MAX];
	ses_enum_target_t *stp;
	int err = -1;

	/*
	 * Open the SES target directory and iterate over any available
	 * targets.
	 */
	if ((dir = opendir(dirpath)) == NULL) {
		/*
		 * If the SES target directory does not exist, then return as if
		 * there are no active targets.
		 */
		topo_mod_dprintf(mod, "failed to open ses "
		    "directory '%s'", dirpath);
		return (0);
	}

	while ((dp = readdir(dir)) != NULL) {
		if (strcmp(dp->d_name, ".") == 0 ||
		    strcmp(dp->d_name, "..") == 0)
			continue;

		/*
		 * Create a new target instance and take a snapshot.
		 */
		if ((stp = topo_mod_zalloc(mod,
		    sizeof (ses_enum_target_t))) == NULL)
			goto error;

		(void) pthread_mutex_init(&stp->set_lock, NULL);

		(void) snprintf(path, sizeof (path), "%s/%s", dirpath,
		    dp->d_name);

		/*
		 * We keep track of the SES device path and export it on a
		 * per-node basis to allow higher level software to get to the
		 * corresponding SES state.
		 */
		if ((stp->set_devpath = topo_mod_strdup(mod, path)) == NULL) {
			topo_mod_free(mod, stp, sizeof (ses_enum_target_t));
			goto error;
		}

		if ((stp->set_target =
		    ses_open(LIBSES_VERSION, path)) == NULL) {
			topo_mod_dprintf(mod, "failed to open ses target "
			    "'%s': %s", dp->d_name, ses_errmsg());
			topo_mod_strfree(mod, stp->set_devpath);
			topo_mod_free(mod, stp, sizeof (ses_enum_target_t));
			continue;
		}

		stp->set_refcount = 1;
		sdp->sed_target = stp;
		stp->set_snap = ses_snap_hold(stp->set_target);
		if (gettimeofday(&stp->set_snaptime, NULL) != 0)
			stp->set_snaptime.tv_sec = time(NULL);

		/*
		 * Enumerate over all SES elements and merge them into the
		 * correct ses_enum_chassis_t.
		 */
		sdp->sed_current = NULL;
		sdp->sed_errno = 0;
		sdp->sed_name = dp->d_name;
		(void) ses_walk(stp->set_snap, ses_enum_gather, sdp);

		if (sdp->sed_errno != 0)
			goto error;
	}

	err = 0;
error:
	closedir(dir);
	return (err);
}

static void
ses_release(topo_mod_t *mod, tnode_t *tn)
{
	ses_enum_target_t *stp;

	if ((stp = topo_node_getspecific(tn)) != NULL)
		ses_target_free(mod, stp);
}

/*ARGSUSED*/
static int
ses_enum(topo_mod_t *mod, tnode_t *rnode, const char *name,
    topo_instance_t min, topo_instance_t max, void *arg, void *notused)
{
	ses_enum_chassis_t *cp;
	ses_enum_data_t *data;

	/*
	 * Check to make sure we're being invoked sensibly, and that we're not
	 * being invoked as part of a post-processing step.
	 */
	if (strcmp(name, SES_ENCLOSURE) != 0 && strcmp(name, BAY) != 0)
		return (0);

	/*
	 * If this is the first time we've called our enumeration method, then
	 * gather information about any available enclosures.
	 */
	if ((data = topo_mod_getspecific(mod)) == NULL) {
		if ((data = topo_mod_zalloc(mod, sizeof (ses_enum_data_t))) ==
		    NULL)
			return (-1);

		data->sed_mod = mod;
		topo_mod_setspecific(mod, data);

		if (disk_list_gather(mod, &data->sed_disks) != 0)
			goto error;

		/*
		 * We search both the ses(7D) and sgen(7D) locations, so we are
		 * independent of any particular driver class bindings.
		 */
		if (ses_process_dir("/dev/es", data) != 0 ||
		    ses_process_dir("/dev/scsi/ses", data) != 0)
			goto error;
	}

	if (strcmp(name, SES_ENCLOSURE) == 0) {
		/*
		 * This is a request to enumerate external enclosures.  Go
		 * through all the targets and create chassis nodes where
		 * necessary.
		 */
		for (cp = topo_list_next(&data->sed_chassis); cp != NULL;
		    cp = topo_list_next(cp)) {
			if (ses_create_chassis(data, rnode, cp) != 0)
				goto error;
		}
	} else {
		/*
		 * This is a request to enumerate a specific bay underneath the
		 * root chassis (for internal disks).
		 */
		if (ses_create_bays(data, rnode) != 0)
			goto error;
	}

	/*
	 * This is a bit of a kludge.  In order to allow internal disks to be
	 * enumerated and share snapshot-specific information with the external
	 * enclosure enumeration, we rely on the fact that we will be invoked
	 * for the 'ses-enclosure' node last.
	 */
	if (strcmp(name, SES_ENCLOSURE) == 0) {
		ses_data_free(data);
		topo_mod_setspecific(mod, NULL);
	}
	return (0);

error:
	ses_data_free(data);
	topo_mod_setspecific(mod, NULL);
	return (-1);
}

static const topo_modops_t ses_ops =
	{ ses_enum, ses_release };

static topo_modinfo_t ses_info =
	{ SES_ENCLOSURE, FM_FMRI_SCHEME_HC, SES_VERSION, &ses_ops };

/*ARGSUSED*/
int
_topo_init(topo_mod_t *mod, topo_version_t version)
{
	if (getenv("TOPOSESDEBUG") != NULL)
		topo_mod_setdebug(mod);

	topo_mod_dprintf(mod, "initializing %s enumerator\n",
	    SES_ENCLOSURE);

	return (topo_mod_register(mod, &ses_info, TOPO_VERSION));
}

void
_topo_fini(topo_mod_t *mod)
{
	topo_mod_unregister(mod);
}