[deliverable/linux.git] / drivers / staging / lustre / lnet / lnet / router.c

/*
 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
 *
 * Copyright (c) 2011, 2015, Intel Corporation.
 *
 *   This file is part of Portals
 *   http://sourceforge.net/projects/sandiaportals/
 *
 *   Portals is free software; you can redistribute it and/or
 *   modify it under the terms of version 2 of the GNU General Public
 *   License as published by the Free Software Foundation.
 *
 *   Portals is distributed in the hope that it will be useful,
 *   but WITHOUT ANY WARRANTY; without even the implied warranty of
 *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *   GNU General Public License for more details.
 *
 */

#define DEBUG_SUBSYSTEM S_LNET
#include "../../include/linux/lnet/lib-lnet.h"

#define LNET_NRB_TINY_MIN	512	/* min value for each CPT */
#define LNET_NRB_TINY		(LNET_NRB_TINY_MIN * 4)
#define LNET_NRB_SMALL_MIN	4096	/* min value for each CPT */
#define LNET_NRB_SMALL		(LNET_NRB_SMALL_MIN * 4)
#define LNET_NRB_SMALL_PAGES	1
#define LNET_NRB_LARGE_MIN	256	/* min value for each CPT */
#define LNET_NRB_LARGE		(LNET_NRB_LARGE_MIN * 4)
#define LNET_NRB_LARGE_PAGES   ((LNET_MTU + PAGE_SIZE - 1) >> \
				 PAGE_SHIFT)

static char *forwarding = "";
module_param(forwarding, charp, 0444);
MODULE_PARM_DESC(forwarding, "Explicitly enable/disable forwarding between networks");

static int tiny_router_buffers;
module_param(tiny_router_buffers, int, 0444);
MODULE_PARM_DESC(tiny_router_buffers, "# of 0 payload messages to buffer in the router");
static int small_router_buffers;
module_param(small_router_buffers, int, 0444);
MODULE_PARM_DESC(small_router_buffers, "# of small (1 page) messages to buffer in the router");
static int large_router_buffers;
module_param(large_router_buffers, int, 0444);
MODULE_PARM_DESC(large_router_buffers, "# of large messages to buffer in the router");
static int peer_buffer_credits;
module_param(peer_buffer_credits, int, 0444);
MODULE_PARM_DESC(peer_buffer_credits, "# router buffer credits per peer");

static int auto_down = 1;
module_param(auto_down, int, 0444);
MODULE_PARM_DESC(auto_down, "Automatically mark peers down on comms error");

int
lnet_peer_buffer_credits(lnet_ni_t *ni)
{
	/* NI option overrides LNet default */
	if (ni->ni_peerrtrcredits > 0)
		return ni->ni_peerrtrcredits;
	if (peer_buffer_credits > 0)
		return peer_buffer_credits;

	/*
	 * As an approximation, allow this peer the same number of router
	 * buffers as it is allowed outstanding sends
	 */
	return ni->ni_peertxcredits;
}

/* forward ref's */
static int lnet_router_checker(void *);

static int check_routers_before_use;
module_param(check_routers_before_use, int, 0444);
MODULE_PARM_DESC(check_routers_before_use, "Assume routers are down and ping them before use");

int avoid_asym_router_failure = 1;
module_param(avoid_asym_router_failure, int, 0644);
MODULE_PARM_DESC(avoid_asym_router_failure, "Avoid asymmetrical router failures (0 to disable)");

static int dead_router_check_interval = 60;
module_param(dead_router_check_interval, int, 0644);
MODULE_PARM_DESC(dead_router_check_interval, "Seconds between dead router health checks (<= 0 to disable)");

static int live_router_check_interval = 60;
module_param(live_router_check_interval, int, 0644);
MODULE_PARM_DESC(live_router_check_interval, "Seconds between live router health checks (<= 0 to disable)");

static int router_ping_timeout = 50;
module_param(router_ping_timeout, int, 0644);
MODULE_PARM_DESC(router_ping_timeout, "Seconds to wait for the reply to a router health query");

int
lnet_peers_start_down(void)
{
	return check_routers_before_use;
}

void
lnet_notify_locked(lnet_peer_t *lp, int notifylnd, int alive,
		   unsigned long when)
{
	if (time_before(when, lp->lp_timestamp)) { /* out of date information */
		CDEBUG(D_NET, "Out of date\n");
		return;
	}

	lp->lp_timestamp = when;		/* update timestamp */
	lp->lp_ping_deadline = 0;	       /* disable ping timeout */

	if (lp->lp_alive_count &&	  /* got old news */
	    (!lp->lp_alive) == (!alive)) {      /* new date for old news */
		CDEBUG(D_NET, "Old news\n");
		return;
	}

	/* Flag that notification is outstanding */

	lp->lp_alive_count++;
	lp->lp_alive = !(!alive);	       /* 1 bit! */
	lp->lp_notify = 1;
	lp->lp_notifylnd |= notifylnd;
	if (lp->lp_alive)
		lp->lp_ping_feats = LNET_PING_FEAT_INVAL; /* reset */

	CDEBUG(D_NET, "set %s %d\n", libcfs_nid2str(lp->lp_nid), alive);
}

static void
lnet_ni_notify_locked(lnet_ni_t *ni, lnet_peer_t *lp)
{
	int alive;
	int notifylnd;

	/*
	 * Notify only in 1 thread at any time to ensure ordered notification.
	 * NB individual events can be missed; the only guarantee is that you
	 * always get the most recent news
	 */
	if (lp->lp_notifying || !ni)
		return;

	lp->lp_notifying = 1;

	while (lp->lp_notify) {
		alive = lp->lp_alive;
		notifylnd = lp->lp_notifylnd;

		lp->lp_notifylnd = 0;
		lp->lp_notify    = 0;

		if (notifylnd && ni->ni_lnd->lnd_notify) {
			lnet_net_unlock(lp->lp_cpt);

			/*
			 * A new notification could happen now; I'll handle it
			 * when control returns to me
			 */
			ni->ni_lnd->lnd_notify(ni, lp->lp_nid, alive);

			lnet_net_lock(lp->lp_cpt);
		}
	}

	lp->lp_notifying = 0;
}

static void
lnet_rtr_addref_locked(lnet_peer_t *lp)
{
	LASSERT(lp->lp_refcount > 0);
	LASSERT(lp->lp_rtr_refcount >= 0);

	/* lnet_net_lock must be exclusively locked */
	lp->lp_rtr_refcount++;
	if (lp->lp_rtr_refcount == 1) {
		struct list_head *pos;

		/* a simple insertion sort */
		list_for_each_prev(pos, &the_lnet.ln_routers) {
			lnet_peer_t *rtr = list_entry(pos, lnet_peer_t,
						      lp_rtr_list);

			if (rtr->lp_nid < lp->lp_nid)
				break;
		}

		list_add(&lp->lp_rtr_list, pos);
		/* addref for the_lnet.ln_routers */
		lnet_peer_addref_locked(lp);
		the_lnet.ln_routers_version++;
	}
}

static void
lnet_rtr_decref_locked(lnet_peer_t *lp)
{
	LASSERT(lp->lp_refcount > 0);
	LASSERT(lp->lp_rtr_refcount > 0);

	/* lnet_net_lock must be exclusively locked */
	lp->lp_rtr_refcount--;
	if (!lp->lp_rtr_refcount) {
		LASSERT(list_empty(&lp->lp_routes));

		if (lp->lp_rcd) {
			list_add(&lp->lp_rcd->rcd_list,
				 &the_lnet.ln_rcd_deathrow);
			lp->lp_rcd = NULL;
		}

		list_del(&lp->lp_rtr_list);
		/* decref for the_lnet.ln_routers */
		lnet_peer_decref_locked(lp);
		the_lnet.ln_routers_version++;
	}
}

lnet_remotenet_t *
lnet_find_net_locked(__u32 net)
{
	lnet_remotenet_t *rnet;
	struct list_head *tmp;
	struct list_head *rn_list;

	LASSERT(!the_lnet.ln_shutdown);

	rn_list = lnet_net2rnethash(net);
	list_for_each(tmp, rn_list) {
		rnet = list_entry(tmp, lnet_remotenet_t, lrn_list);

		if (rnet->lrn_net == net)
			return rnet;
	}
	return NULL;
}

static void lnet_shuffle_seed(void)
{
	static int seeded;
	__u32 lnd_type, seed[2];
	struct timespec64 ts;
	lnet_ni_t *ni;
	struct list_head *tmp;

	if (seeded)
		return;

	cfs_get_random_bytes(seed, sizeof(seed));

	/*
	 * Nodes with small feet have little entropy
	 * the NID for this node gives the most entropy in the low bits
	 */
	list_for_each(tmp, &the_lnet.ln_nis) {
		ni = list_entry(tmp, lnet_ni_t, ni_list);
		lnd_type = LNET_NETTYP(LNET_NIDNET(ni->ni_nid));

		if (lnd_type != LOLND)
			seed[0] ^= (LNET_NIDADDR(ni->ni_nid) | lnd_type);
	}

	ktime_get_ts64(&ts);
	cfs_srand(ts.tv_sec ^ seed[0], ts.tv_nsec ^ seed[1]);
	seeded = 1;
}

/* NB expects LNET_LOCK held */
static void
lnet_add_route_to_rnet(lnet_remotenet_t *rnet, lnet_route_t *route)
{
	unsigned int len = 0;
	unsigned int offset = 0;
	struct list_head *e;

	lnet_shuffle_seed();

	list_for_each(e, &rnet->lrn_routes) {
		len++;
	}

	/* len+1 positions to add a new entry, also prevents division by 0 */
	offset = cfs_rand() % (len + 1);
	list_for_each(e, &rnet->lrn_routes) {
		if (!offset)
			break;
		offset--;
	}
	list_add(&route->lr_list, e);
	list_add(&route->lr_gwlist, &route->lr_gateway->lp_routes);

	the_lnet.ln_remote_nets_version++;
	lnet_rtr_addref_locked(route->lr_gateway);
}

int
lnet_add_route(__u32 net, __u32 hops, lnet_nid_t gateway,
	       unsigned int priority)
{
	struct list_head *e;
	lnet_remotenet_t *rnet;
	lnet_remotenet_t *rnet2;
	lnet_route_t *route;
	lnet_ni_t *ni;
	int add_route;
	int rc;

	CDEBUG(D_NET, "Add route: net %s hops %d priority %u gw %s\n",
	       libcfs_net2str(net), hops, priority, libcfs_nid2str(gateway));

	if (gateway == LNET_NID_ANY ||
	    LNET_NETTYP(LNET_NIDNET(gateway)) == LOLND ||
	    net == LNET_NIDNET(LNET_NID_ANY) ||
	    LNET_NETTYP(net) == LOLND ||
	    LNET_NIDNET(gateway) == net ||
	    (hops != LNET_UNDEFINED_HOPS && (hops < 1 || hops > 255)))
		return -EINVAL;

	if (lnet_islocalnet(net))	       /* it's a local network */
		return -EEXIST;

	/* Assume net, route, all new */
	LIBCFS_ALLOC(route, sizeof(*route));
	LIBCFS_ALLOC(rnet, sizeof(*rnet));
	if (!route || !rnet) {
		CERROR("Out of memory creating route %s %d %s\n",
		       libcfs_net2str(net), hops, libcfs_nid2str(gateway));
		if (route)
			LIBCFS_FREE(route, sizeof(*route));
		if (rnet)
			LIBCFS_FREE(rnet, sizeof(*rnet));
		return -ENOMEM;
	}

	INIT_LIST_HEAD(&rnet->lrn_routes);
	rnet->lrn_net = net;
	route->lr_hops = hops;
	route->lr_net = net;
	route->lr_priority = priority;

	lnet_net_lock(LNET_LOCK_EX);

	rc = lnet_nid2peer_locked(&route->lr_gateway, gateway, LNET_LOCK_EX);
	if (rc) {
		lnet_net_unlock(LNET_LOCK_EX);

		LIBCFS_FREE(route, sizeof(*route));
		LIBCFS_FREE(rnet, sizeof(*rnet));

		if (rc == -EHOSTUNREACH) /* gateway is not on a local net */
			return rc;	/* ignore the route entry */
		CERROR("Error %d creating route %s %d %s\n", rc,
		       libcfs_net2str(net), hops,
		       libcfs_nid2str(gateway));
		return rc;
	}

	LASSERT(!the_lnet.ln_shutdown);

	rnet2 = lnet_find_net_locked(net);
	if (!rnet2) {
		/* new network */
		list_add_tail(&rnet->lrn_list, lnet_net2rnethash(net));
		rnet2 = rnet;
	}

	/* Search for a duplicate route (it's a NOOP if it is) */
	add_route = 1;
	list_for_each(e, &rnet2->lrn_routes) {
		lnet_route_t *route2 = list_entry(e, lnet_route_t, lr_list);

		if (route2->lr_gateway == route->lr_gateway) {
			add_route = 0;
			break;
		}

		/* our lookups must be true */
		LASSERT(route2->lr_gateway->lp_nid != gateway);
	}

	if (add_route) {
		lnet_peer_addref_locked(route->lr_gateway); /* +1 for notify */
		lnet_add_route_to_rnet(rnet2, route);

		ni = route->lr_gateway->lp_ni;
		lnet_net_unlock(LNET_LOCK_EX);

		/* XXX Assume alive */
		if (ni->ni_lnd->lnd_notify)
			ni->ni_lnd->lnd_notify(ni, gateway, 1);

		lnet_net_lock(LNET_LOCK_EX);
	}

	/* -1 for notify or !add_route */
	lnet_peer_decref_locked(route->lr_gateway);
	lnet_net_unlock(LNET_LOCK_EX);
	rc = 0;

	if (!add_route) {
		rc = -EEXIST;
		LIBCFS_FREE(route, sizeof(*route));
	}

	if (rnet != rnet2)
		LIBCFS_FREE(rnet, sizeof(*rnet));

	/* indicate to startup the router checker if configured */
	wake_up(&the_lnet.ln_rc_waitq);

	return rc;
}

int
lnet_check_routes(void)
{
	lnet_remotenet_t *rnet;
	lnet_route_t *route;
	lnet_route_t *route2;
	struct list_head *e1;
	struct list_head *e2;
	int cpt;
	struct list_head *rn_list;
	int i;

	cpt = lnet_net_lock_current();

	for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) {
		rn_list = &the_lnet.ln_remote_nets_hash[i];
		list_for_each(e1, rn_list) {
			rnet = list_entry(e1, lnet_remotenet_t, lrn_list);

			route2 = NULL;
			list_for_each(e2, &rnet->lrn_routes) {
				lnet_nid_t nid1;
				lnet_nid_t nid2;
				int net;

				route = list_entry(e2, lnet_route_t, lr_list);

				if (!route2) {
					route2 = route;
					continue;
				}

				if (route->lr_gateway->lp_ni ==
				    route2->lr_gateway->lp_ni)
					continue;

				nid1 = route->lr_gateway->lp_nid;
				nid2 = route2->lr_gateway->lp_nid;
				net = rnet->lrn_net;

				lnet_net_unlock(cpt);

				CERROR("Routes to %s via %s and %s not supported\n",
				       libcfs_net2str(net),
				       libcfs_nid2str(nid1),
				       libcfs_nid2str(nid2));
				return -EINVAL;
			}
		}
	}

	lnet_net_unlock(cpt);
	return 0;
}

int
lnet_del_route(__u32 net, lnet_nid_t gw_nid)
{
	struct lnet_peer *gateway;
	lnet_remotenet_t *rnet;
	lnet_route_t *route;
	struct list_head *e1;
	struct list_head *e2;
	int rc = -ENOENT;
	struct list_head *rn_list;
	int idx = 0;

	CDEBUG(D_NET, "Del route: net %s : gw %s\n",
	       libcfs_net2str(net), libcfs_nid2str(gw_nid));

	/*
	 * NB Caller may specify either all routes via the given gateway
	 * or a specific route entry actual NIDs)
	 */
	lnet_net_lock(LNET_LOCK_EX);
	if (net == LNET_NIDNET(LNET_NID_ANY))
		rn_list = &the_lnet.ln_remote_nets_hash[0];
	else
		rn_list = lnet_net2rnethash(net);

 again:
	list_for_each(e1, rn_list) {
		rnet = list_entry(e1, lnet_remotenet_t, lrn_list);

		if (!(net == LNET_NIDNET(LNET_NID_ANY) ||
		      net == rnet->lrn_net))
			continue;

		list_for_each(e2, &rnet->lrn_routes) {
			route = list_entry(e2, lnet_route_t, lr_list);

			gateway = route->lr_gateway;
			if (!(gw_nid == LNET_NID_ANY ||
			      gw_nid == gateway->lp_nid))
				continue;

			list_del(&route->lr_list);
			list_del(&route->lr_gwlist);
			the_lnet.ln_remote_nets_version++;

			if (list_empty(&rnet->lrn_routes))
				list_del(&rnet->lrn_list);
			else
				rnet = NULL;

			lnet_rtr_decref_locked(gateway);
			lnet_peer_decref_locked(gateway);

			lnet_net_unlock(LNET_LOCK_EX);

			LIBCFS_FREE(route, sizeof(*route));

			if (rnet)
				LIBCFS_FREE(rnet, sizeof(*rnet));

			rc = 0;
			lnet_net_lock(LNET_LOCK_EX);
			goto again;
		}
	}

	if (net == LNET_NIDNET(LNET_NID_ANY) &&
	    ++idx < LNET_REMOTE_NETS_HASH_SIZE) {
		rn_list = &the_lnet.ln_remote_nets_hash[idx];
		goto again;
	}
	lnet_net_unlock(LNET_LOCK_EX);

	return rc;
}

void
lnet_destroy_routes(void)
{
	lnet_del_route(LNET_NIDNET(LNET_NID_ANY), LNET_NID_ANY);
}

int lnet_get_rtr_pool_cfg(int idx, struct lnet_ioctl_pool_cfg *pool_cfg)
{
	int i, rc = -ENOENT, j;

	if (!the_lnet.ln_rtrpools)
		return rc;

	for (i = 0; i < LNET_NRBPOOLS; i++) {
		lnet_rtrbufpool_t *rbp;

		lnet_net_lock(LNET_LOCK_EX);
		cfs_percpt_for_each(rbp, j, the_lnet.ln_rtrpools) {
			if (i++ != idx)
				continue;

			pool_cfg->pl_pools[i].pl_npages = rbp[i].rbp_npages;
			pool_cfg->pl_pools[i].pl_nbuffers = rbp[i].rbp_nbuffers;
			pool_cfg->pl_pools[i].pl_credits = rbp[i].rbp_credits;
			pool_cfg->pl_pools[i].pl_mincredits = rbp[i].rbp_mincredits;
			rc = 0;
			break;
		}
		lnet_net_unlock(LNET_LOCK_EX);
	}

	lnet_net_lock(LNET_LOCK_EX);
	pool_cfg->pl_routing = the_lnet.ln_routing;
	lnet_net_unlock(LNET_LOCK_EX);

	return rc;
}

int
lnet_get_route(int idx, __u32 *net, __u32 *hops,
	       lnet_nid_t *gateway, __u32 *alive, __u32 *priority)
{
	struct list_head *e1;
	struct list_head *e2;
	lnet_remotenet_t *rnet;
	lnet_route_t *route;
	int cpt;
	int i;
	struct list_head *rn_list;

	cpt = lnet_net_lock_current();

	for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) {
		rn_list = &the_lnet.ln_remote_nets_hash[i];
		list_for_each(e1, rn_list) {
			rnet = list_entry(e1, lnet_remotenet_t, lrn_list);

			list_for_each(e2, &rnet->lrn_routes) {
				route = list_entry(e2, lnet_route_t, lr_list);

				if (!idx--) {
					*net      = rnet->lrn_net;
					*hops     = route->lr_hops;
					*priority = route->lr_priority;
					*gateway  = route->lr_gateway->lp_nid;
					*alive = lnet_is_route_alive(route);
					lnet_net_unlock(cpt);
					return 0;
				}
			}
		}
	}

	lnet_net_unlock(cpt);
	return -ENOENT;
}

void
lnet_swap_pinginfo(lnet_ping_info_t *info)
{
	int i;
	lnet_ni_status_t *stat;

	__swab32s(&info->pi_magic);
	__swab32s(&info->pi_features);
	__swab32s(&info->pi_pid);
	__swab32s(&info->pi_nnis);
	for (i = 0; i < info->pi_nnis && i < LNET_MAX_RTR_NIS; i++) {
		stat = &info->pi_ni[i];
		__swab64s(&stat->ns_nid);
		__swab32s(&stat->ns_status);
	}
}

/**
 * parse router-checker pinginfo, record number of down NIs for remote
 * networks on that router.
 */
static void
lnet_parse_rc_info(lnet_rc_data_t *rcd)
{
	lnet_ping_info_t *info = rcd->rcd_pinginfo;
	struct lnet_peer *gw = rcd->rcd_gateway;
	lnet_route_t *rte;

	if (!gw->lp_alive)
		return;

	if (info->pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
		lnet_swap_pinginfo(info);

	/* NB always racing with network! */
	if (info->pi_magic != LNET_PROTO_PING_MAGIC) {
		CDEBUG(D_NET, "%s: Unexpected magic %08x\n",
		       libcfs_nid2str(gw->lp_nid), info->pi_magic);
		gw->lp_ping_feats = LNET_PING_FEAT_INVAL;
		return;
	}

	gw->lp_ping_feats = info->pi_features;
	if (!(gw->lp_ping_feats & LNET_PING_FEAT_MASK)) {
		CDEBUG(D_NET, "%s: Unexpected features 0x%x\n",
		       libcfs_nid2str(gw->lp_nid), gw->lp_ping_feats);
		return; /* nothing I can understand */
	}

	if (!(gw->lp_ping_feats & LNET_PING_FEAT_NI_STATUS))
		return; /* can't carry NI status info */

	list_for_each_entry(rte, &gw->lp_routes, lr_gwlist) {
		int down = 0;
		int up = 0;
		int i;

		if (gw->lp_ping_feats & LNET_PING_FEAT_RTE_DISABLED) {
			rte->lr_downis = 1;
			continue;
		}

		for (i = 0; i < info->pi_nnis && i < LNET_MAX_RTR_NIS; i++) {
			lnet_ni_status_t *stat = &info->pi_ni[i];
			lnet_nid_t nid = stat->ns_nid;

			if (nid == LNET_NID_ANY) {
				CDEBUG(D_NET, "%s: unexpected LNET_NID_ANY\n",
				       libcfs_nid2str(gw->lp_nid));
				gw->lp_ping_feats = LNET_PING_FEAT_INVAL;
				return;
			}

			if (LNET_NETTYP(LNET_NIDNET(nid)) == LOLND)
				continue;

			if (stat->ns_status == LNET_NI_STATUS_DOWN) {
				down++;
				continue;
			}

			if (stat->ns_status == LNET_NI_STATUS_UP) {
				if (LNET_NIDNET(nid) == rte->lr_net) {
					up = 1;
					break;
				}
				continue;
			}

			CDEBUG(D_NET, "%s: Unexpected status 0x%x\n",
			       libcfs_nid2str(gw->lp_nid), stat->ns_status);
			gw->lp_ping_feats = LNET_PING_FEAT_INVAL;
			return;
		}

		if (up) { /* ignore downed NIs if NI for dest network is up */
			rte->lr_downis = 0;
			continue;
		}
		/**
		 * if @down is zero and this route is single-hop, it means
		 * we can't find NI for target network
		 */
		if (!down && rte->lr_hops == 1)
			down = 1;

		rte->lr_downis = down;
	}
}

static void
lnet_router_checker_event(lnet_event_t *event)
{
	lnet_rc_data_t *rcd = event->md.user_ptr;
	struct lnet_peer *lp;

	LASSERT(rcd);

	if (event->unlinked) {
		LNetInvalidateHandle(&rcd->rcd_mdh);
		return;
	}

	LASSERT(event->type == LNET_EVENT_SEND ||
		event->type == LNET_EVENT_REPLY);

	lp = rcd->rcd_gateway;
	LASSERT(lp);

	/*
	 * NB: it's called with holding lnet_res_lock, we have a few
	 * places need to hold both locks at the same time, please take
	 * care of lock ordering
	 */
	lnet_net_lock(lp->lp_cpt);
	if (!lnet_isrouter(lp) || lp->lp_rcd != rcd) {
		/* ignore if no longer a router or rcd is replaced */
		goto out;
	}

	if (event->type == LNET_EVENT_SEND) {
		lp->lp_ping_notsent = 0;
		if (!event->status)
			goto out;
	}

	/* LNET_EVENT_REPLY */
	/*
	 * A successful REPLY means the router is up.  If _any_ comms
	 * to the router fail I assume it's down (this will happen if
	 * we ping alive routers to try to detect router death before
	 * apps get burned).
	 */
	lnet_notify_locked(lp, 1, !event->status, cfs_time_current());

	/*
	 * The router checker will wake up very shortly and do the
	 * actual notification.
	 * XXX If 'lp' stops being a router before then, it will still
	 * have the notification pending!!!
	 */
	if (avoid_asym_router_failure && !event->status)
		lnet_parse_rc_info(rcd);

 out:
	lnet_net_unlock(lp->lp_cpt);
}

static void
lnet_wait_known_routerstate(void)
{
	lnet_peer_t *rtr;
	struct list_head *entry;
	int all_known;

	LASSERT(the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING);

	for (;;) {
		int cpt = lnet_net_lock_current();

		all_known = 1;
		list_for_each(entry, &the_lnet.ln_routers) {
			rtr = list_entry(entry, lnet_peer_t, lp_rtr_list);

			if (!rtr->lp_alive_count) {
				all_known = 0;
				break;
			}
		}

		lnet_net_unlock(cpt);

		if (all_known)
			return;

		set_current_state(TASK_UNINTERRUPTIBLE);
		schedule_timeout(cfs_time_seconds(1));
	}
}

void
lnet_router_ni_update_locked(lnet_peer_t *gw, __u32 net)
{
	lnet_route_t *rte;

	if ((gw->lp_ping_feats & LNET_PING_FEAT_NI_STATUS)) {
		list_for_each_entry(rte, &gw->lp_routes, lr_gwlist) {
			if (rte->lr_net == net) {
				rte->lr_downis = 0;
				break;
			}
		}
	}
}

static void
lnet_update_ni_status_locked(void)
{
	lnet_ni_t *ni;
	time64_t now;
	int timeout;

	LASSERT(the_lnet.ln_routing);

	timeout = router_ping_timeout +
		  max(live_router_check_interval, dead_router_check_interval);

	now = ktime_get_real_seconds();
	list_for_each_entry(ni, &the_lnet.ln_nis, ni_list) {
		if (ni->ni_lnd->lnd_type == LOLND)
			continue;

		if (now < ni->ni_last_alive + timeout)
			continue;

		lnet_ni_lock(ni);
		/* re-check with lock */
		if (now < ni->ni_last_alive + timeout) {
			lnet_ni_unlock(ni);
			continue;
		}

		LASSERT(ni->ni_status);

		if (ni->ni_status->ns_status != LNET_NI_STATUS_DOWN) {
			CDEBUG(D_NET, "NI(%s:%d) status changed to down\n",
			       libcfs_nid2str(ni->ni_nid), timeout);
			/*
			 * NB: so far, this is the only place to set
			 * NI status to "down"
			 */
			ni->ni_status->ns_status = LNET_NI_STATUS_DOWN;
		}
		lnet_ni_unlock(ni);
	}
}

static void
lnet_destroy_rc_data(lnet_rc_data_t *rcd)
{
	LASSERT(list_empty(&rcd->rcd_list));
	/* detached from network */
	LASSERT(LNetHandleIsInvalid(rcd->rcd_mdh));

	if (rcd->rcd_gateway) {
		int cpt = rcd->rcd_gateway->lp_cpt;

		lnet_net_lock(cpt);
		lnet_peer_decref_locked(rcd->rcd_gateway);
		lnet_net_unlock(cpt);
	}

	if (rcd->rcd_pinginfo)
		LIBCFS_FREE(rcd->rcd_pinginfo, LNET_PINGINFO_SIZE);

	LIBCFS_FREE(rcd, sizeof(*rcd));
}

static lnet_rc_data_t *
lnet_create_rc_data_locked(lnet_peer_t *gateway)
{
	lnet_rc_data_t *rcd = NULL;
	lnet_ping_info_t *pi;
	int rc;
	int i;

	lnet_net_unlock(gateway->lp_cpt);

	LIBCFS_ALLOC(rcd, sizeof(*rcd));
	if (!rcd)
		goto out;

	LNetInvalidateHandle(&rcd->rcd_mdh);
	INIT_LIST_HEAD(&rcd->rcd_list);

	LIBCFS_ALLOC(pi, LNET_PINGINFO_SIZE);
	if (!pi)
		goto out;

	for (i = 0; i < LNET_MAX_RTR_NIS; i++) {
		pi->pi_ni[i].ns_nid = LNET_NID_ANY;
		pi->pi_ni[i].ns_status = LNET_NI_STATUS_INVALID;
	}
	rcd->rcd_pinginfo = pi;

	LASSERT(!LNetHandleIsInvalid(the_lnet.ln_rc_eqh));
	rc = LNetMDBind((lnet_md_t){.start     = pi,
				    .user_ptr  = rcd,
				    .length    = LNET_PINGINFO_SIZE,
				    .threshold = LNET_MD_THRESH_INF,
				    .options   = LNET_MD_TRUNCATE,
				    .eq_handle = the_lnet.ln_rc_eqh},
			LNET_UNLINK,
			&rcd->rcd_mdh);
	if (rc < 0) {
		CERROR("Can't bind MD: %d\n", rc);
		goto out;
	}
	LASSERT(!rc);

	lnet_net_lock(gateway->lp_cpt);
	/* router table changed or someone has created rcd for this gateway */
	if (!lnet_isrouter(gateway) || gateway->lp_rcd) {
		lnet_net_unlock(gateway->lp_cpt);
		goto out;
	}

	lnet_peer_addref_locked(gateway);
	rcd->rcd_gateway = gateway;
	gateway->lp_rcd = rcd;
	gateway->lp_ping_notsent = 0;

	return rcd;

 out:
	if (rcd) {
		if (!LNetHandleIsInvalid(rcd->rcd_mdh)) {
			rc = LNetMDUnlink(rcd->rcd_mdh);
			LASSERT(!rc);
		}
		lnet_destroy_rc_data(rcd);
	}

	lnet_net_lock(gateway->lp_cpt);
	return gateway->lp_rcd;
}

static int
lnet_router_check_interval(lnet_peer_t *rtr)
{
	int secs;

	secs = rtr->lp_alive ? live_router_check_interval :
			       dead_router_check_interval;
	if (secs < 0)
		secs = 0;

	return secs;
}

static void
lnet_ping_router_locked(lnet_peer_t *rtr)
{
	lnet_rc_data_t *rcd = NULL;
	unsigned long now = cfs_time_current();
	int secs;

	lnet_peer_addref_locked(rtr);

	if (rtr->lp_ping_deadline && /* ping timed out? */
	    cfs_time_after(now, rtr->lp_ping_deadline))
		lnet_notify_locked(rtr, 1, 0, now);

	/* Run any outstanding notifications */
	lnet_ni_notify_locked(rtr->lp_ni, rtr);

	if (!lnet_isrouter(rtr) ||
	    the_lnet.ln_rc_state != LNET_RC_STATE_RUNNING) {
		/* router table changed or router checker is shutting down */
		lnet_peer_decref_locked(rtr);
		return;
	}

	rcd = rtr->lp_rcd ?
	      rtr->lp_rcd : lnet_create_rc_data_locked(rtr);

	if (!rcd)
		return;

	secs = lnet_router_check_interval(rtr);

	CDEBUG(D_NET,
	       "rtr %s %d: deadline %lu ping_notsent %d alive %d alive_count %d lp_ping_timestamp %lu\n",
	       libcfs_nid2str(rtr->lp_nid), secs,
	       rtr->lp_ping_deadline, rtr->lp_ping_notsent,
	       rtr->lp_alive, rtr->lp_alive_count, rtr->lp_ping_timestamp);

	if (secs && !rtr->lp_ping_notsent &&
	    cfs_time_after(now, cfs_time_add(rtr->lp_ping_timestamp,
					     cfs_time_seconds(secs)))) {
		int rc;
		lnet_process_id_t id;
		lnet_handle_md_t mdh;

		id.nid = rtr->lp_nid;
		id.pid = LNET_PID_LUSTRE;
		CDEBUG(D_NET, "Check: %s\n", libcfs_id2str(id));

		rtr->lp_ping_notsent   = 1;
		rtr->lp_ping_timestamp = now;

		mdh = rcd->rcd_mdh;

		if (!rtr->lp_ping_deadline) {
			rtr->lp_ping_deadline =
				cfs_time_shift(router_ping_timeout);
		}

		lnet_net_unlock(rtr->lp_cpt);

		rc = LNetGet(LNET_NID_ANY, mdh, id, LNET_RESERVED_PORTAL,
			     LNET_PROTO_PING_MATCHBITS, 0);

		lnet_net_lock(rtr->lp_cpt);
		if (rc)
			rtr->lp_ping_notsent = 0; /* no event pending */
	}

	lnet_peer_decref_locked(rtr);
}

int
lnet_router_checker_start(void)
{
	struct task_struct *task;
	int rc;
	int eqsz = 0;

	LASSERT(the_lnet.ln_rc_state == LNET_RC_STATE_SHUTDOWN);

	if (check_routers_before_use &&
	    dead_router_check_interval <= 0) {
		LCONSOLE_ERROR_MSG(0x10a, "'dead_router_check_interval' must be set if 'check_routers_before_use' is set\n");
		return -EINVAL;
	}

	sema_init(&the_lnet.ln_rc_signal, 0);

	rc = LNetEQAlloc(0, lnet_router_checker_event, &the_lnet.ln_rc_eqh);
	if (rc) {
		CERROR("Can't allocate EQ(%d): %d\n", eqsz, rc);
		return -ENOMEM;
	}

	the_lnet.ln_rc_state = LNET_RC_STATE_RUNNING;
	task = kthread_run(lnet_router_checker, NULL, "router_checker");
	if (IS_ERR(task)) {
		rc = PTR_ERR(task);
		CERROR("Can't start router checker thread: %d\n", rc);
		/* block until event callback signals exit */
		down(&the_lnet.ln_rc_signal);
		rc = LNetEQFree(the_lnet.ln_rc_eqh);
		LASSERT(!rc);
		the_lnet.ln_rc_state = LNET_RC_STATE_SHUTDOWN;
		return -ENOMEM;
	}

	if (check_routers_before_use) {
		/*
		 * Note that a helpful side-effect of pinging all known routers
		 * at startup is that it makes them drop stale connections they
		 * may have to a previous instance of me.
		 */
		lnet_wait_known_routerstate();
	}

	return 0;
}

void
lnet_router_checker_stop(void)
{
	int rc;

	if (the_lnet.ln_rc_state == LNET_RC_STATE_SHUTDOWN)
		return;

	LASSERT(the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING);
	the_lnet.ln_rc_state = LNET_RC_STATE_STOPPING;
	/* wakeup the RC thread if it's sleeping */
	wake_up(&the_lnet.ln_rc_waitq);

	/* block until event callback signals exit */
	down(&the_lnet.ln_rc_signal);
	LASSERT(the_lnet.ln_rc_state == LNET_RC_STATE_SHUTDOWN);

	rc = LNetEQFree(the_lnet.ln_rc_eqh);
	LASSERT(!rc);
}

static void
lnet_prune_rc_data(int wait_unlink)
{
	lnet_rc_data_t *rcd;
	lnet_rc_data_t *tmp;
	lnet_peer_t *lp;
	struct list_head head;
	int i = 2;

	if (likely(the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING &&
		   list_empty(&the_lnet.ln_rcd_deathrow) &&
		   list_empty(&the_lnet.ln_rcd_zombie)))
		return;

	INIT_LIST_HEAD(&head);

	lnet_net_lock(LNET_LOCK_EX);

	if (the_lnet.ln_rc_state != LNET_RC_STATE_RUNNING) {
		/* router checker is stopping, prune all */
		list_for_each_entry(lp, &the_lnet.ln_routers,
				    lp_rtr_list) {
			if (!lp->lp_rcd)
				continue;

			LASSERT(list_empty(&lp->lp_rcd->rcd_list));
			list_add(&lp->lp_rcd->rcd_list,
				 &the_lnet.ln_rcd_deathrow);
			lp->lp_rcd = NULL;
		}
	}

	/* unlink all RCDs on deathrow list */
	list_splice_init(&the_lnet.ln_rcd_deathrow, &head);

	if (!list_empty(&head)) {
		lnet_net_unlock(LNET_LOCK_EX);

		list_for_each_entry(rcd, &head, rcd_list)
			LNetMDUnlink(rcd->rcd_mdh);

		lnet_net_lock(LNET_LOCK_EX);
	}

	list_splice_init(&head, &the_lnet.ln_rcd_zombie);

	/* release all zombie RCDs */
	while (!list_empty(&the_lnet.ln_rcd_zombie)) {
		list_for_each_entry_safe(rcd, tmp, &the_lnet.ln_rcd_zombie,
					 rcd_list) {
			if (LNetHandleIsInvalid(rcd->rcd_mdh))
				list_move(&rcd->rcd_list, &head);
		}

		wait_unlink = wait_unlink &&
			      !list_empty(&the_lnet.ln_rcd_zombie);

		lnet_net_unlock(LNET_LOCK_EX);

		while (!list_empty(&head)) {
			rcd = list_entry(head.next,
					 lnet_rc_data_t, rcd_list);
			list_del_init(&rcd->rcd_list);
			lnet_destroy_rc_data(rcd);
		}

		if (!wait_unlink)
			return;

		i++;
		CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET,
		       "Waiting for rc buffers to unlink\n");
		set_current_state(TASK_UNINTERRUPTIBLE);
		schedule_timeout(cfs_time_seconds(1) / 4);

		lnet_net_lock(LNET_LOCK_EX);
	}

	lnet_net_unlock(LNET_LOCK_EX);
}

/*
 * This function is called to check if the RC should block indefinitely.
 * It's called from lnet_router_checker() as well as being passed to
 * wait_event_interruptible() to avoid the lost wake_up problem.
 *
 * When it's called from wait_event_interruptible() it is necessary to
 * also not sleep if the rc state is not running to avoid a deadlock
 * when the system is shutting down
 */
static inline bool
lnet_router_checker_active(void)
{
	if (the_lnet.ln_rc_state != LNET_RC_STATE_RUNNING)
		return true;

	/*
	 * Router Checker thread needs to run when routing is enabled in
	 * order to call lnet_update_ni_status_locked()
	 */
	if (the_lnet.ln_routing)
		return true;

	return !list_empty(&the_lnet.ln_routers) &&
		(live_router_check_interval > 0 ||
		 dead_router_check_interval > 0);
}

static int
lnet_router_checker(void *arg)
{
	lnet_peer_t *rtr;
	struct list_head *entry;

	cfs_block_allsigs();

	while (the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING) {
		__u64 version;
		int cpt;
		int cpt2;

		cpt = lnet_net_lock_current();
rescan:
		version = the_lnet.ln_routers_version;

		list_for_each(entry, &the_lnet.ln_routers) {
			rtr = list_entry(entry, lnet_peer_t, lp_rtr_list);

			cpt2 = lnet_cpt_of_nid_locked(rtr->lp_nid);
			if (cpt != cpt2) {
				lnet_net_unlock(cpt);
				cpt = cpt2;
				lnet_net_lock(cpt);
				/* the routers list has changed */
				if (version != the_lnet.ln_routers_version)
					goto rescan;
			}

			lnet_ping_router_locked(rtr);

			/* NB dropped lock */
			if (version != the_lnet.ln_routers_version) {
				/* the routers list has changed */
				goto rescan;
			}
		}

		if (the_lnet.ln_routing)
			lnet_update_ni_status_locked();

		lnet_net_unlock(cpt);

		lnet_prune_rc_data(0); /* don't wait for UNLINK */

		/*
		 * Call schedule_timeout() here always adds 1 to load average
		 * because kernel counts # active tasks as nr_running
		 * + nr_uninterruptible.
		 */
		/*
		 * if there are any routes then wakeup every second.  If
		 * there are no routes then sleep indefinitely until woken
		 * up by a user adding a route
		 */
		if (!lnet_router_checker_active())
			wait_event_interruptible(the_lnet.ln_rc_waitq,
						 lnet_router_checker_active());
		else
			wait_event_interruptible_timeout(the_lnet.ln_rc_waitq,
							 false,
							 cfs_time_seconds(1));
	}

	lnet_prune_rc_data(1); /* wait for UNLINK */

	the_lnet.ln_rc_state = LNET_RC_STATE_SHUTDOWN;
	up(&the_lnet.ln_rc_signal);
	/* The unlink event callback will signal final completion */
	return 0;
}

void
lnet_destroy_rtrbuf(lnet_rtrbuf_t *rb, int npages)
{
	int sz = offsetof(lnet_rtrbuf_t, rb_kiov[npages]);

	while (--npages >= 0)
		__free_page(rb->rb_kiov[npages].kiov_page);

	LIBCFS_FREE(rb, sz);
}

static lnet_rtrbuf_t *
lnet_new_rtrbuf(lnet_rtrbufpool_t *rbp, int cpt)
{
	int npages = rbp->rbp_npages;
	int sz = offsetof(lnet_rtrbuf_t, rb_kiov[npages]);
	struct page *page;
	lnet_rtrbuf_t *rb;
	int i;

	LIBCFS_CPT_ALLOC(rb, lnet_cpt_table(), cpt, sz);
	if (!rb)
		return NULL;

	rb->rb_pool = rbp;

	for (i = 0; i < npages; i++) {
		page = alloc_pages_node(
				cfs_cpt_spread_node(lnet_cpt_table(), cpt),
				GFP_KERNEL | __GFP_ZERO, 0);
		if (!page) {
			while (--i >= 0)
				__free_page(rb->rb_kiov[i].kiov_page);

			LIBCFS_FREE(rb, sz);
			return NULL;
		}

		rb->rb_kiov[i].kiov_len = PAGE_SIZE;
		rb->rb_kiov[i].kiov_offset = 0;
		rb->rb_kiov[i].kiov_page = page;
	}

	return rb;
}

static void
lnet_rtrpool_free_bufs(lnet_rtrbufpool_t *rbp, int cpt)
{
	int npages = rbp->rbp_npages;
	struct list_head tmp;
	lnet_rtrbuf_t *rb;
	lnet_rtrbuf_t *temp;

	if (!rbp->rbp_nbuffers) /* not initialized or already freed */
		return;

	INIT_LIST_HEAD(&tmp);

	lnet_net_lock(cpt);
	lnet_drop_routed_msgs_locked(&rbp->rbp_msgs, cpt);
	list_splice_init(&rbp->rbp_bufs, &tmp);
	rbp->rbp_req_nbuffers = 0;
	rbp->rbp_nbuffers = 0;
	rbp->rbp_credits = 0;
	rbp->rbp_mincredits = 0;
	lnet_net_unlock(cpt);

	/* Free buffers on the free list. */
	list_for_each_entry_safe(rb, temp, &tmp, rb_list) {
		list_del(&rb->rb_list);
		lnet_destroy_rtrbuf(rb, npages);
	}
}

static int
lnet_rtrpool_adjust_bufs(lnet_rtrbufpool_t *rbp, int nbufs, int cpt)
{
	struct list_head rb_list;
	lnet_rtrbuf_t *rb;
	int num_rb;
	int num_buffers = 0;
	int old_req_nbufs;
	int npages = rbp->rbp_npages;

	lnet_net_lock(cpt);
	/*
	 * If we are called for less buffers than already in the pool, we
	 * just lower the req_nbuffers number and excess buffers will be
	 * thrown away as they are returned to the free list.  Credits
	 * then get adjusted as well.
	 * If we already have enough buffers allocated to serve the
	 * increase requested, then we can treat that the same way as we
	 * do the decrease.
	 */
	num_rb = nbufs - rbp->rbp_nbuffers;
	if (nbufs <= rbp->rbp_req_nbuffers || num_rb <= 0) {
		rbp->rbp_req_nbuffers = nbufs;
		lnet_net_unlock(cpt);
		return 0;
	}
	/*
	 * store the older value of rbp_req_nbuffers and then set it to
	 * the new request to prevent lnet_return_rx_credits_locked() from
	 * freeing buffers that we need to keep around
	 */
	old_req_nbufs = rbp->rbp_req_nbuffers;
	rbp->rbp_req_nbuffers = nbufs;
	lnet_net_unlock(cpt);

	INIT_LIST_HEAD(&rb_list);

	/*
	 * allocate the buffers on a local list first.  If all buffers are
	 * allocated successfully then join this list to the rbp buffer
	 * list. If not then free all allocated buffers.
	 */
	while (num_rb-- > 0) {
		rb = lnet_new_rtrbuf(rbp, cpt);
		if (!rb) {
			CERROR("Failed to allocate %d route bufs of %d pages\n",
			       nbufs, npages);

			lnet_net_lock(cpt);
			rbp->rbp_req_nbuffers = old_req_nbufs;
			lnet_net_unlock(cpt);

			goto failed;
		}

		list_add(&rb->rb_list, &rb_list);
		num_buffers++;
	}

	lnet_net_lock(cpt);

	list_splice_tail(&rb_list, &rbp->rbp_bufs);
	rbp->rbp_nbuffers += num_buffers;
	rbp->rbp_credits += num_buffers;
	rbp->rbp_mincredits = rbp->rbp_credits;
	/*
	 * We need to schedule blocked msg using the newly
	 * added buffers.
	 */
	while (!list_empty(&rbp->rbp_bufs) &&
	       !list_empty(&rbp->rbp_msgs))
		lnet_schedule_blocked_locked(rbp);

	lnet_net_unlock(cpt);

	return 0;

failed:
	while (!list_empty(&rb_list)) {
		rb = list_entry(rb_list.next, lnet_rtrbuf_t, rb_list);
		list_del(&rb->rb_list);
		lnet_destroy_rtrbuf(rb, npages);
	}

	return -ENOMEM;
}

static void
lnet_rtrpool_init(lnet_rtrbufpool_t *rbp, int npages)
{
	INIT_LIST_HEAD(&rbp->rbp_msgs);
	INIT_LIST_HEAD(&rbp->rbp_bufs);

	rbp->rbp_npages = npages;
	rbp->rbp_credits = 0;
	rbp->rbp_mincredits = 0;
}

void
lnet_rtrpools_free(int keep_pools)
{
	lnet_rtrbufpool_t *rtrp;
	int i;

	if (!the_lnet.ln_rtrpools) /* uninitialized or freed */
		return;

	cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
		lnet_rtrpool_free_bufs(&rtrp[LNET_TINY_BUF_IDX], i);
		lnet_rtrpool_free_bufs(&rtrp[LNET_SMALL_BUF_IDX], i);
		lnet_rtrpool_free_bufs(&rtrp[LNET_LARGE_BUF_IDX], i);
	}

	if (!keep_pools) {
		cfs_percpt_free(the_lnet.ln_rtrpools);
		the_lnet.ln_rtrpools = NULL;
	}
}

static int
lnet_nrb_tiny_calculate(void)
{
	int nrbs = LNET_NRB_TINY;

	if (tiny_router_buffers < 0) {
		LCONSOLE_ERROR_MSG(0x10c,
				   "tiny_router_buffers=%d invalid when routing enabled\n",
				   tiny_router_buffers);
		return -EINVAL;
	}

	if (tiny_router_buffers > 0)
		nrbs = tiny_router_buffers;

	nrbs /= LNET_CPT_NUMBER;
	return max(nrbs, LNET_NRB_TINY_MIN);
}

static int
lnet_nrb_small_calculate(void)
{
	int nrbs = LNET_NRB_SMALL;

	if (small_router_buffers < 0) {
		LCONSOLE_ERROR_MSG(0x10c,
				   "small_router_buffers=%d invalid when routing enabled\n",
				   small_router_buffers);
		return -EINVAL;
	}

	if (small_router_buffers > 0)
		nrbs = small_router_buffers;

	nrbs /= LNET_CPT_NUMBER;
	return max(nrbs, LNET_NRB_SMALL_MIN);
}

static int
lnet_nrb_large_calculate(void)
{
	int nrbs = LNET_NRB_LARGE;

	if (large_router_buffers < 0) {
		LCONSOLE_ERROR_MSG(0x10c,
				   "large_router_buffers=%d invalid when routing enabled\n",
				   large_router_buffers);
		return -EINVAL;
	}

	if (large_router_buffers > 0)
		nrbs = large_router_buffers;

	nrbs /= LNET_CPT_NUMBER;
	return max(nrbs, LNET_NRB_LARGE_MIN);
}

int
lnet_rtrpools_alloc(int im_a_router)
{
	lnet_rtrbufpool_t *rtrp;
	int nrb_tiny;
	int nrb_small;
	int nrb_large;
	int rc;
	int i;

	if (!strcmp(forwarding, "")) {
		/* not set either way */
		if (!im_a_router)
			return 0;
	} else if (!strcmp(forwarding, "disabled")) {
		/* explicitly disabled */
		return 0;
	} else if (!strcmp(forwarding, "enabled")) {
		/* explicitly enabled */
	} else {
		LCONSOLE_ERROR_MSG(0x10b, "'forwarding' not set to either 'enabled' or 'disabled'\n");
		return -EINVAL;
	}

	nrb_tiny = lnet_nrb_tiny_calculate();
	if (nrb_tiny < 0)
		return -EINVAL;

	nrb_small = lnet_nrb_small_calculate();
	if (nrb_small < 0)
		return -EINVAL;

	nrb_large = lnet_nrb_large_calculate();
	if (nrb_large < 0)
		return -EINVAL;

	the_lnet.ln_rtrpools = cfs_percpt_alloc(lnet_cpt_table(),
						LNET_NRBPOOLS *
						sizeof(lnet_rtrbufpool_t));
	if (!the_lnet.ln_rtrpools) {
		LCONSOLE_ERROR_MSG(0x10c,
				   "Failed to initialize router buffe pool\n");
		return -ENOMEM;
	}

	cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
		lnet_rtrpool_init(&rtrp[LNET_TINY_BUF_IDX], 0);
		rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_TINY_BUF_IDX],
					      nrb_tiny, i);
		if (rc)
			goto failed;

		lnet_rtrpool_init(&rtrp[LNET_SMALL_BUF_IDX],
				  LNET_NRB_SMALL_PAGES);
		rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_SMALL_BUF_IDX],
					      nrb_small, i);
		if (rc)
			goto failed;

		lnet_rtrpool_init(&rtrp[LNET_LARGE_BUF_IDX],
				  LNET_NRB_LARGE_PAGES);
		rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_LARGE_BUF_IDX],
					      nrb_large, i);
		if (rc)
			goto failed;
	}

	lnet_net_lock(LNET_LOCK_EX);
	the_lnet.ln_routing = 1;
	lnet_net_unlock(LNET_LOCK_EX);

	return 0;

 failed:
	lnet_rtrpools_free(0);
	return rc;
}

static int
lnet_rtrpools_adjust_helper(int tiny, int small, int large)
{
	int nrb = 0;
	int rc = 0;
	int i;
	lnet_rtrbufpool_t *rtrp;

	/*
	 * If the provided values for each buffer pool are different than the
	 * configured values, we need to take action.
	 */
	if (tiny >= 0) {
		tiny_router_buffers = tiny;
		nrb = lnet_nrb_tiny_calculate();
		cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
			rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_TINY_BUF_IDX],
						      nrb, i);
			if (rc)
				return rc;
		}
	}
	if (small >= 0) {
		small_router_buffers = small;
		nrb = lnet_nrb_small_calculate();
		cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
			rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_SMALL_BUF_IDX],
						      nrb, i);
			if (rc)
				return rc;
		}
	}
	if (large >= 0) {
		large_router_buffers = large;
		nrb = lnet_nrb_large_calculate();
		cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
			rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_LARGE_BUF_IDX],
						      nrb, i);
			if (rc)
				return rc;
		}
	}

	return 0;
}

int
lnet_rtrpools_adjust(int tiny, int small, int large)
{
	/*
	 * this function doesn't revert the changes if adding new buffers
	 * failed.  It's up to the user space caller to revert the
	 * changes.
	 */
	if (!the_lnet.ln_routing)
		return 0;

	return lnet_rtrpools_adjust_helper(tiny, small, large);
}

int
lnet_rtrpools_enable(void)
{
	int rc;

	if (the_lnet.ln_routing)
		return 0;

	if (!the_lnet.ln_rtrpools)
		/*
		 * If routing is turned off, and we have never
		 * initialized the pools before, just call the
		 * standard buffer pool allocation routine as
		 * if we are just configuring this for the first
		 * time.
		 */
		return lnet_rtrpools_alloc(1);

	rc = lnet_rtrpools_adjust_helper(0, 0, 0);
	if (rc)
		return rc;

	lnet_net_lock(LNET_LOCK_EX);
	the_lnet.ln_routing = 1;

	the_lnet.ln_ping_info->pi_features &= ~LNET_PING_FEAT_RTE_DISABLED;
	lnet_net_unlock(LNET_LOCK_EX);

	return 0;
}

void
lnet_rtrpools_disable(void)
{
	if (!the_lnet.ln_routing)
		return;

	lnet_net_lock(LNET_LOCK_EX);
	the_lnet.ln_routing = 0;
	the_lnet.ln_ping_info->pi_features |= LNET_PING_FEAT_RTE_DISABLED;

	tiny_router_buffers = 0;
	small_router_buffers = 0;
	large_router_buffers = 0;
	lnet_net_unlock(LNET_LOCK_EX);
	lnet_rtrpools_free(1);
}

int
lnet_notify(lnet_ni_t *ni, lnet_nid_t nid, int alive, unsigned long when)
{
	struct lnet_peer *lp = NULL;
	unsigned long now = cfs_time_current();
	int cpt = lnet_cpt_of_nid(nid);

	LASSERT(!in_interrupt());

	CDEBUG(D_NET, "%s notifying %s: %s\n",
	       !ni ? "userspace" : libcfs_nid2str(ni->ni_nid),
	       libcfs_nid2str(nid),
	       alive ? "up" : "down");

	if (ni &&
	    LNET_NIDNET(ni->ni_nid) != LNET_NIDNET(nid)) {
		CWARN("Ignoring notification of %s %s by %s (different net)\n",
		      libcfs_nid2str(nid), alive ? "birth" : "death",
		      libcfs_nid2str(ni->ni_nid));
		return -EINVAL;
	}

	/* can't do predictions... */
	if (cfs_time_after(when, now)) {
		CWARN("Ignoring prediction from %s of %s %s %ld seconds in the future\n",
		      !ni ? "userspace" : libcfs_nid2str(ni->ni_nid),
		      libcfs_nid2str(nid), alive ? "up" : "down",
		      cfs_duration_sec(cfs_time_sub(when, now)));
		return -EINVAL;
	}

	if (ni && !alive &&	     /* LND telling me she's down */
	    !auto_down) {		       /* auto-down disabled */
		CDEBUG(D_NET, "Auto-down disabled\n");
		return 0;
	}

	lnet_net_lock(cpt);

	if (the_lnet.ln_shutdown) {
		lnet_net_unlock(cpt);
		return -ESHUTDOWN;
	}

	lp = lnet_find_peer_locked(the_lnet.ln_peer_tables[cpt], nid);
	if (!lp) {
		/* nid not found */
		lnet_net_unlock(cpt);
		CDEBUG(D_NET, "%s not found\n", libcfs_nid2str(nid));
		return 0;
	}

	/*
	 * We can't fully trust LND on reporting exact peer last_alive
	 * if he notifies us about dead peer. For example ksocklnd can
	 * call us with when == _time_when_the_node_was_booted_ if
	 * no connections were successfully established
	 */
	if (ni && !alive && when < lp->lp_last_alive)
		when = lp->lp_last_alive;

	lnet_notify_locked(lp, !ni, alive, when);

	if (ni)
		lnet_ni_notify_locked(ni, lp);

	lnet_peer_decref_locked(lp);

	lnet_net_unlock(cpt);
	return 0;
}
EXPORT_SYMBOL(lnet_notify);