Whamcloud - gitweb
LU-6245 libcfs: remove sysctl module handling
[fs/lustre-release.git] / lnet / klnds / socklnd / socklnd.c
index 360f905..15d9ed1 100644 (file)
@@ -1,6 +1,4 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
  * GPL HEADER START
  *
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  * GPL HEADER END
  */
 /*
- * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2015, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
 
 #include "socklnd.h"
 
-lnd_t the_ksocklnd = {
-        .lnd_type       = SOCKLND,
-        .lnd_startup    = ksocknal_startup,
-        .lnd_shutdown   = ksocknal_shutdown,
-        .lnd_ctl        = ksocknal_ctl,
-        .lnd_send       = ksocknal_send,
-        .lnd_recv       = ksocknal_recv,
-        .lnd_notify     = ksocknal_notify,
-        .lnd_accept     = ksocknal_accept,
-};
-
+static lnd_t                   the_ksocklnd;
 ksock_nal_data_t        ksocknal_data;
 
-ksock_interface_t *
+static ksock_interface_t *
 ksocknal_ip2iface(lnet_ni_t *ni, __u32 ip)
 {
         ksock_net_t       *net = ni->ni_data;
@@ -74,19 +64,19 @@ ksocknal_ip2iface(lnet_ni_t *ni, __u32 ip)
         return (NULL);
 }
 
-ksock_route_t *
+static ksock_route_t *
 ksocknal_create_route (__u32 ipaddr, int port)
 {
-        ksock_route_t *route;
+       ksock_route_t *route;
 
-        LIBCFS_ALLOC (route, sizeof (*route));
-        if (route == NULL)
-                return (NULL);
+       LIBCFS_ALLOC (route, sizeof (*route));
+       if (route == NULL)
+               return (NULL);
 
-        atomic_set (&route->ksnr_refcount, 1);
-        route->ksnr_peer = NULL;
-        route->ksnr_retry_interval = 0;         /* OK to connect at any time */
-        route->ksnr_ipaddr = ipaddr;
+       atomic_set (&route->ksnr_refcount, 1);
+       route->ksnr_peer = NULL;
+       route->ksnr_retry_interval = 0;         /* OK to connect at any time */
+       route->ksnr_ipaddr = ipaddr;
         route->ksnr_port = port;
         route->ksnr_scheduled = 0;
         route->ksnr_connecting = 0;
@@ -101,113 +91,114 @@ ksocknal_create_route (__u32 ipaddr, int port)
 void
 ksocknal_destroy_route (ksock_route_t *route)
 {
-        LASSERT (atomic_read(&route->ksnr_refcount) == 0);
+       LASSERT (atomic_read(&route->ksnr_refcount) == 0);
 
-        if (route->ksnr_peer != NULL)
-                ksocknal_peer_decref(route->ksnr_peer);
+       if (route->ksnr_peer != NULL)
+               ksocknal_peer_decref(route->ksnr_peer);
 
-        LIBCFS_FREE (route, sizeof (*route));
+       LIBCFS_FREE (route, sizeof (*route));
 }
 
-int
-ksocknal_create_peer (ksock_peer_t **peerp, lnet_ni_t *ni, lnet_process_id_t id)
+static int
+ksocknal_create_peer(ksock_peer_t **peerp, lnet_ni_t *ni, lnet_process_id_t id)
 {
-        ksock_net_t   *net = ni->ni_data;
-        ksock_peer_t  *peer;
+       int             cpt = lnet_cpt_of_nid(id.nid);
+       ksock_net_t     *net = ni->ni_data;
+       ksock_peer_t    *peer;
 
-        LASSERT (id.nid != LNET_NID_ANY);
-        LASSERT (id.pid != LNET_PID_ANY);
-        LASSERT (!in_interrupt());
+       LASSERT(id.nid != LNET_NID_ANY);
+       LASSERT(id.pid != LNET_PID_ANY);
+       LASSERT(!in_interrupt());
 
-        LIBCFS_ALLOC (peer, sizeof (*peer));
-        if (peer == NULL)
-                return -ENOMEM;
+       LIBCFS_CPT_ALLOC(peer, lnet_cpt_table(), cpt, sizeof(*peer));
+       if (peer == NULL)
+               return -ENOMEM;
 
-        memset (peer, 0, sizeof (*peer));       /* NULL pointers/clear flags etc */
-
-        peer->ksnp_ni = ni;
-        peer->ksnp_id = id;
-        atomic_set (&peer->ksnp_refcount, 1);   /* 1 ref for caller */
-        peer->ksnp_closing = 0;
-        peer->ksnp_accepting = 0;
-        peer->ksnp_zc_next_cookie = 1;
-        peer->ksnp_proto = NULL;
-        CFS_INIT_LIST_HEAD (&peer->ksnp_conns);
-        CFS_INIT_LIST_HEAD (&peer->ksnp_routes);
-        CFS_INIT_LIST_HEAD (&peer->ksnp_tx_queue);
-        CFS_INIT_LIST_HEAD (&peer->ksnp_zc_req_list);
-        spin_lock_init(&peer->ksnp_lock);
-
-        spin_lock_bh (&net->ksnn_lock);
-
-        if (net->ksnn_shutdown) {
-                spin_unlock_bh (&net->ksnn_lock);
-
-                LIBCFS_FREE(peer, sizeof(*peer));
-                CERROR("Can't create peer: network shutdown\n");
-                return -ESHUTDOWN;
-        }
+       peer->ksnp_ni = ni;
+       peer->ksnp_id = id;
+       atomic_set(&peer->ksnp_refcount, 1);    /* 1 ref for caller */
+       peer->ksnp_closing = 0;
+       peer->ksnp_accepting = 0;
+       peer->ksnp_proto = NULL;
+       peer->ksnp_last_alive = 0;
+       peer->ksnp_zc_next_cookie = SOCKNAL_KEEPALIVE_PING + 1;
 
-        net->ksnn_npeers++;
+       INIT_LIST_HEAD(&peer->ksnp_conns);
+       INIT_LIST_HEAD(&peer->ksnp_routes);
+       INIT_LIST_HEAD(&peer->ksnp_tx_queue);
+       INIT_LIST_HEAD(&peer->ksnp_zc_req_list);
+       spin_lock_init(&peer->ksnp_lock);
 
-        spin_unlock_bh (&net->ksnn_lock);
+       spin_lock_bh(&net->ksnn_lock);
 
-        *peerp = peer;
-        return 0;
+       if (net->ksnn_shutdown) {
+               spin_unlock_bh(&net->ksnn_lock);
+
+               LIBCFS_FREE(peer, sizeof(*peer));
+               CERROR("Can't create peer: network shutdown\n");
+               return -ESHUTDOWN;
+       }
+
+       net->ksnn_npeers++;
+
+       spin_unlock_bh(&net->ksnn_lock);
+
+       *peerp = peer;
+       return 0;
 }
 
 void
 ksocknal_destroy_peer (ksock_peer_t *peer)
 {
-        ksock_net_t    *net = peer->ksnp_ni->ni_data;
+       ksock_net_t    *net = peer->ksnp_ni->ni_data;
 
-        CDEBUG (D_NET, "peer %s %p deleted\n",
-                libcfs_id2str(peer->ksnp_id), peer);
+       CDEBUG (D_NET, "peer %s %p deleted\n",
+               libcfs_id2str(peer->ksnp_id), peer);
 
-        LASSERT (atomic_read (&peer->ksnp_refcount) == 0);
-        LASSERT (peer->ksnp_accepting == 0);
-        LASSERT (list_empty (&peer->ksnp_conns));
-        LASSERT (list_empty (&peer->ksnp_routes));
-        LASSERT (list_empty (&peer->ksnp_tx_queue));
-        LASSERT (list_empty (&peer->ksnp_zc_req_list));
+       LASSERT(atomic_read(&peer->ksnp_refcount) == 0);
+       LASSERT(peer->ksnp_accepting == 0);
+       LASSERT(list_empty(&peer->ksnp_conns));
+       LASSERT(list_empty(&peer->ksnp_routes));
+       LASSERT(list_empty(&peer->ksnp_tx_queue));
+       LASSERT(list_empty(&peer->ksnp_zc_req_list));
 
-        LIBCFS_FREE (peer, sizeof (*peer));
+       LIBCFS_FREE(peer, sizeof(*peer));
 
         /* NB a peer's connections and routes keep a reference on their peer
          * until they are destroyed, so we can be assured that _all_ state to
          * do with this peer has been cleaned up when its refcount drops to
          * zero. */
-        spin_lock_bh (&net->ksnn_lock);
-        net->ksnn_npeers--;
-        spin_unlock_bh (&net->ksnn_lock);
+       spin_lock_bh(&net->ksnn_lock);
+       net->ksnn_npeers--;
+       spin_unlock_bh(&net->ksnn_lock);
 }
 
 ksock_peer_t *
 ksocknal_find_peer_locked (lnet_ni_t *ni, lnet_process_id_t id)
 {
-        struct list_head *peer_list = ksocknal_nid2peerlist(id.nid);
-        struct list_head *tmp;
-        ksock_peer_t     *peer;
+       struct list_head *peer_list = ksocknal_nid2peerlist(id.nid);
+       struct list_head *tmp;
+       ksock_peer_t     *peer;
 
-        list_for_each (tmp, peer_list) {
+       list_for_each(tmp, peer_list) {
 
-                peer = list_entry (tmp, ksock_peer_t, ksnp_list);
+               peer = list_entry(tmp, ksock_peer_t, ksnp_list);
 
-                LASSERT (!peer->ksnp_closing);
+               LASSERT(!peer->ksnp_closing);
 
-                if (peer->ksnp_ni != ni)
-                        continue;
+               if (peer->ksnp_ni != ni)
+                       continue;
 
-                if (peer->ksnp_id.nid != id.nid ||
-                    peer->ksnp_id.pid != id.pid)
-                        continue;
+               if (peer->ksnp_id.nid != id.nid ||
+                   peer->ksnp_id.pid != id.pid)
+                       continue;
 
-                CDEBUG(D_NET, "got peer [%p] -> %s (%d)\n",
-                       peer, libcfs_id2str(id),
-                       atomic_read(&peer->ksnp_refcount));
-                return (peer);
-        }
-        return (NULL);
+               CDEBUG(D_NET, "got peer [%p] -> %s (%d)\n",
+                      peer, libcfs_id2str(id),
+                      atomic_read(&peer->ksnp_refcount));
+               return peer;
+       }
+       return NULL;
 }
 
 ksock_peer_t *
@@ -215,16 +206,16 @@ ksocknal_find_peer (lnet_ni_t *ni, lnet_process_id_t id)
 {
         ksock_peer_t     *peer;
 
-        read_lock (&ksocknal_data.ksnd_global_lock);
-        peer = ksocknal_find_peer_locked (ni, id);
-        if (peer != NULL)                       /* +1 ref for caller? */
-                ksocknal_peer_addref(peer);
-        read_unlock (&ksocknal_data.ksnd_global_lock);
+       read_lock(&ksocknal_data.ksnd_global_lock);
+       peer = ksocknal_find_peer_locked(ni, id);
+       if (peer != NULL)                       /* +1 ref for caller? */
+               ksocknal_peer_addref(peer);
+       read_unlock(&ksocknal_data.ksnd_global_lock);
 
         return (peer);
 }
 
-void
+static void
 ksocknal_unlink_peer_locked (ksock_peer_t *peer)
 {
         int                i;
@@ -245,42 +236,41 @@ ksocknal_unlink_peer_locked (ksock_peer_t *peer)
                 iface->ksni_npeers--;
         }
 
-        LASSERT (list_empty(&peer->ksnp_conns));
-        LASSERT (list_empty(&peer->ksnp_routes));
-        LASSERT (!peer->ksnp_closing);
-        peer->ksnp_closing = 1;
-        list_del (&peer->ksnp_list);
-        /* lose peerlist's ref */
-        ksocknal_peer_decref(peer);
+       LASSERT(list_empty(&peer->ksnp_conns));
+       LASSERT(list_empty(&peer->ksnp_routes));
+       LASSERT(!peer->ksnp_closing);
+       peer->ksnp_closing = 1;
+       list_del(&peer->ksnp_list);
+       /* lose peerlist's ref */
+       ksocknal_peer_decref(peer);
 }
 
-int
+static int
 ksocknal_get_peer_info (lnet_ni_t *ni, int index,
-                        lnet_process_id_t *id, __u32 *myip, __u32 *peer_ip, int *port,
-                        int *conn_count, int *share_count)
+                        lnet_process_id_t *id, __u32 *myip, __u32 *peer_ip,
+                        int *port, int *conn_count, int *share_count)
 {
-        ksock_peer_t      *peer;
-        struct list_head  *ptmp;
-        ksock_route_t     *route;
-        struct list_head  *rtmp;
-        int                i;
+       ksock_peer_t      *peer;
+       struct list_head  *ptmp;
+       ksock_route_t     *route;
+       struct list_head  *rtmp;
+       int                i;
         int                j;
-        int                rc = -ENOENT;
+       int                rc = -ENOENT;
 
-        read_lock (&ksocknal_data.ksnd_global_lock);
+       read_lock(&ksocknal_data.ksnd_global_lock);
 
-        for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
+       for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
+               list_for_each(ptmp, &ksocknal_data.ksnd_peers[i]) {
+                       peer = list_entry(ptmp, ksock_peer_t, ksnp_list);
 
-                list_for_each (ptmp, &ksocknal_data.ksnd_peers[i]) {
-                        peer = list_entry (ptmp, ksock_peer_t, ksnp_list);
+                       if (peer->ksnp_ni != ni)
+                               continue;
 
-                        if (peer->ksnp_ni != ni)
-                                continue;
-
-                        if (peer->ksnp_n_passive_ips == 0 &&
-                            list_empty(&peer->ksnp_routes)) {
-                                if (index-- > 0)
-                                        continue;
+                       if (peer->ksnp_n_passive_ips == 0 &&
+                           list_empty(&peer->ksnp_routes)) {
+                               if (index-- > 0)
+                                       continue;
 
                                 *id = peer->ksnp_id;
                                 *myip = 0;
@@ -292,9 +282,9 @@ ksocknal_get_peer_info (lnet_ni_t *ni, int index,
                                 goto out;
                         }
 
-                        for (j = 0; j < peer->ksnp_n_passive_ips; j++) {
-                                if (index-- > 0)
-                                        continue;
+                       for (j = 0; j < peer->ksnp_n_passive_ips; j++) {
+                               if (index-- > 0)
+                                       continue;
 
                                 *id = peer->ksnp_id;
                                 *myip = peer->ksnp_passive_ips[j];
@@ -306,53 +296,52 @@ ksocknal_get_peer_info (lnet_ni_t *ni, int index,
                                 goto out;
                         }
 
-                        list_for_each (rtmp, &peer->ksnp_routes) {
-                                if (index-- > 0)
-                                        continue;
-
-                                route = list_entry(rtmp, ksock_route_t,
-                                                   ksnr_list);
-
-                                *id = peer->ksnp_id;
-                                *myip = route->ksnr_myipaddr;
-                                *peer_ip = route->ksnr_ipaddr;
-                                *port = route->ksnr_port;
-                                *conn_count = route->ksnr_conn_count;
-                                *share_count = route->ksnr_share_count;
-                                rc = 0;
-                                goto out;
-                        }
-                }
-        }
- out:
-        read_unlock (&ksocknal_data.ksnd_global_lock);
-        return (rc);
+                       list_for_each(rtmp, &peer->ksnp_routes) {
+                               if (index-- > 0)
+                                       continue;
+
+                               route = list_entry(rtmp, ksock_route_t,
+                                                  ksnr_list);
+
+                               *id = peer->ksnp_id;
+                               *myip = route->ksnr_myipaddr;
+                               *peer_ip = route->ksnr_ipaddr;
+                               *port = route->ksnr_port;
+                               *conn_count = route->ksnr_conn_count;
+                               *share_count = route->ksnr_share_count;
+                               rc = 0;
+                               goto out;
+                       }
+               }
+       }
+out:
+       read_unlock(&ksocknal_data.ksnd_global_lock);
+       return rc;
 }
 
-void
+static void
 ksocknal_associate_route_conn_locked(ksock_route_t *route, ksock_conn_t *conn)
 {
-        ksock_peer_t      *peer = route->ksnr_peer;
-        int                type = conn->ksnc_type;
-        ksock_interface_t *iface;
-
-        conn->ksnc_route = route;
-        ksocknal_route_addref(route);
-
-        if (route->ksnr_myipaddr != conn->ksnc_myipaddr) {
-                if (route->ksnr_myipaddr == 0) {
-                        /* route wasn't bound locally yet (the initial route) */
-                        CDEBUG(D_NET, "Binding %s %u.%u.%u.%u to %u.%u.%u.%u\n",
-                               libcfs_id2str(peer->ksnp_id),
-                               HIPQUAD(route->ksnr_ipaddr),
-                               HIPQUAD(conn->ksnc_myipaddr));
-                } else {
-                        CDEBUG(D_NET, "Rebinding %s %u.%u.%u.%u from "
-                               "%u.%u.%u.%u to %u.%u.%u.%u\n",
-                               libcfs_id2str(peer->ksnp_id),
-                               HIPQUAD(route->ksnr_ipaddr),
-                               HIPQUAD(route->ksnr_myipaddr),
-                               HIPQUAD(conn->ksnc_myipaddr));
+       ksock_peer_t      *peer = route->ksnr_peer;
+       int                type = conn->ksnc_type;
+       ksock_interface_t *iface;
+
+       conn->ksnc_route = route;
+       ksocknal_route_addref(route);
+
+       if (route->ksnr_myipaddr != conn->ksnc_myipaddr) {
+               if (route->ksnr_myipaddr == 0) {
+                       /* route wasn't bound locally yet (the initial route) */
+                       CDEBUG(D_NET, "Binding %s %pI4h to %pI4h\n",
+                              libcfs_id2str(peer->ksnp_id),
+                              &route->ksnr_ipaddr,
+                              &conn->ksnc_myipaddr);
+               } else {
+                       CDEBUG(D_NET, "Rebinding %s %pI4h from %pI4h "
+                              "to %pI4h\n", libcfs_id2str(peer->ksnp_id),
+                              &route->ksnr_ipaddr,
+                              &route->ksnr_myipaddr,
+                              &conn->ksnc_myipaddr);
 
                         iface = ksocknal_ip2iface(route->ksnr_peer->ksnp_ni,
                                                   route->ksnr_myipaddr);
@@ -374,96 +363,96 @@ ksocknal_associate_route_conn_locked(ksock_route_t *route, ksock_conn_t *conn)
         route->ksnr_retry_interval = 0;
 }
 
-void
+static void
 ksocknal_add_route_locked (ksock_peer_t *peer, ksock_route_t *route)
 {
-        struct list_head  *tmp;
-        ksock_conn_t      *conn;
-        ksock_route_t     *route2;
-
-        LASSERT (!peer->ksnp_closing);
-        LASSERT (route->ksnr_peer == NULL);
-        LASSERT (!route->ksnr_scheduled);
-        LASSERT (!route->ksnr_connecting);
-        LASSERT (route->ksnr_connected == 0);
-
-        /* LASSERT(unique) */
-        list_for_each(tmp, &peer->ksnp_routes) {
-                route2 = list_entry(tmp, ksock_route_t, ksnr_list);
-
-                if (route2->ksnr_ipaddr == route->ksnr_ipaddr) {
-                        CERROR ("Duplicate route %s %u.%u.%u.%u\n",
-                                libcfs_id2str(peer->ksnp_id),
-                                HIPQUAD(route->ksnr_ipaddr));
-                        LBUG();
-                }
-        }
-
-        route->ksnr_peer = peer;
-        ksocknal_peer_addref(peer);
-        /* peer's routelist takes over my ref on 'route' */
-        list_add_tail(&route->ksnr_list, &peer->ksnp_routes);
-
-        list_for_each(tmp, &peer->ksnp_conns) {
-                conn = list_entry(tmp, ksock_conn_t, ksnc_list);
-
-                if (conn->ksnc_ipaddr != route->ksnr_ipaddr)
-                        continue;
-
-                ksocknal_associate_route_conn_locked(route, conn);
-                /* keep going (typed routes) */
-        }
+       struct list_head *tmp;
+       ksock_conn_t     *conn;
+       ksock_route_t    *route2;
+
+       LASSERT(!peer->ksnp_closing);
+       LASSERT(route->ksnr_peer == NULL);
+       LASSERT(!route->ksnr_scheduled);
+       LASSERT(!route->ksnr_connecting);
+       LASSERT(route->ksnr_connected == 0);
+
+       /* LASSERT(unique) */
+       list_for_each(tmp, &peer->ksnp_routes) {
+               route2 = list_entry(tmp, ksock_route_t, ksnr_list);
+
+               if (route2->ksnr_ipaddr == route->ksnr_ipaddr) {
+                       CERROR("Duplicate route %s %pI4h\n",
+                              libcfs_id2str(peer->ksnp_id),
+                              &route->ksnr_ipaddr);
+                       LBUG();
+               }
+       }
+
+       route->ksnr_peer = peer;
+       ksocknal_peer_addref(peer);
+       /* peer's routelist takes over my ref on 'route' */
+       list_add_tail(&route->ksnr_list, &peer->ksnp_routes);
+
+       list_for_each(tmp, &peer->ksnp_conns) {
+               conn = list_entry(tmp, ksock_conn_t, ksnc_list);
+
+               if (conn->ksnc_ipaddr != route->ksnr_ipaddr)
+                       continue;
+
+               ksocknal_associate_route_conn_locked(route, conn);
+               /* keep going (typed routes) */
+       }
 }
 
-void
+static void
 ksocknal_del_route_locked (ksock_route_t *route)
 {
-        ksock_peer_t      *peer = route->ksnr_peer;
-        ksock_interface_t *iface;
-        ksock_conn_t      *conn;
-        struct list_head  *ctmp;
-        struct list_head  *cnxt;
-
-        LASSERT (!route->ksnr_deleted);
-
-        /* Close associated conns */
-        list_for_each_safe (ctmp, cnxt, &peer->ksnp_conns) {
-                conn = list_entry(ctmp, ksock_conn_t, ksnc_list);
-
-                if (conn->ksnc_route != route)
-                        continue;
-
-                ksocknal_close_conn_locked (conn, 0);
-        }
-
-        if (route->ksnr_myipaddr != 0) {
-                iface = ksocknal_ip2iface(route->ksnr_peer->ksnp_ni,
-                                          route->ksnr_myipaddr);
-                if (iface != NULL)
-                        iface->ksni_nroutes--;
-        }
-
-        route->ksnr_deleted = 1;
-        list_del (&route->ksnr_list);
-        ksocknal_route_decref(route);             /* drop peer's ref */
-
-        if (list_empty (&peer->ksnp_routes) &&
-            list_empty (&peer->ksnp_conns)) {
-                /* I've just removed the last route to a peer with no active
-                 * connections */
-                ksocknal_unlink_peer_locked (peer);
-        }
+       ksock_peer_t      *peer = route->ksnr_peer;
+       ksock_interface_t *iface;
+       ksock_conn_t      *conn;
+       struct list_head  *ctmp;
+       struct list_head  *cnxt;
+
+       LASSERT(!route->ksnr_deleted);
+
+       /* Close associated conns */
+       list_for_each_safe(ctmp, cnxt, &peer->ksnp_conns) {
+               conn = list_entry(ctmp, ksock_conn_t, ksnc_list);
+
+               if (conn->ksnc_route != route)
+                       continue;
+
+               ksocknal_close_conn_locked(conn, 0);
+       }
+
+       if (route->ksnr_myipaddr != 0) {
+               iface = ksocknal_ip2iface(route->ksnr_peer->ksnp_ni,
+                                         route->ksnr_myipaddr);
+               if (iface != NULL)
+                       iface->ksni_nroutes--;
+       }
+
+       route->ksnr_deleted = 1;
+       list_del(&route->ksnr_list);
+       ksocknal_route_decref(route);           /* drop peer's ref */
+
+       if (list_empty(&peer->ksnp_routes) &&
+           list_empty(&peer->ksnp_conns)) {
+               /* I've just removed the last route to a peer with no active
+                * connections */
+               ksocknal_unlink_peer_locked(peer);
+       }
 }
 
 int
-ksocknal_add_peer (lnet_ni_t *ni, lnet_process_id_t id, __u32 ipaddr, int port)
+ksocknal_add_peer(lnet_ni_t *ni, lnet_process_id_t id, __u32 ipaddr, int port)
 {
-        struct list_head  *tmp;
-        ksock_peer_t      *peer;
-        ksock_peer_t      *peer2;
-        ksock_route_t     *route;
-        ksock_route_t     *route2;
-        int                rc;
+       struct list_head *tmp;
+       ksock_peer_t     *peer;
+       ksock_peer_t     *peer2;
+       ksock_route_t    *route;
+       ksock_route_t    *route2;
+       int               rc;
 
         if (id.nid == LNET_NID_ANY ||
             id.pid == LNET_PID_ANY)
@@ -480,240 +469,238 @@ ksocknal_add_peer (lnet_ni_t *ni, lnet_process_id_t id, __u32 ipaddr, int port)
                 return (-ENOMEM);
         }
 
-        write_lock_bh (&ksocknal_data.ksnd_global_lock);
+       write_lock_bh(&ksocknal_data.ksnd_global_lock);
 
         /* always called with a ref on ni, so shutdown can't have started */
         LASSERT (((ksock_net_t *) ni->ni_data)->ksnn_shutdown == 0);
 
-        peer2 = ksocknal_find_peer_locked (ni, id);
-        if (peer2 != NULL) {
-                ksocknal_peer_decref(peer);
-                peer = peer2;
-        } else {
-                /* peer table takes my ref on peer */
-                list_add_tail (&peer->ksnp_list,
-                               ksocknal_nid2peerlist (id.nid));
-        }
-
-        route2 = NULL;
-        list_for_each (tmp, &peer->ksnp_routes) {
-                route2 = list_entry(tmp, ksock_route_t, ksnr_list);
-
-                if (route2->ksnr_ipaddr == ipaddr)
-                        break;
-
-                route2 = NULL;
-        }
-        if (route2 == NULL) {
-                ksocknal_add_route_locked(peer, route);
-                route->ksnr_share_count++;
-        } else {
-                ksocknal_route_decref(route);
-                route2->ksnr_share_count++;
-        }
-
-        write_unlock_bh (&ksocknal_data.ksnd_global_lock);
-
-        return (0);
+       peer2 = ksocknal_find_peer_locked(ni, id);
+       if (peer2 != NULL) {
+               ksocknal_peer_decref(peer);
+               peer = peer2;
+       } else {
+               /* peer table takes my ref on peer */
+               list_add_tail(&peer->ksnp_list,
+                             ksocknal_nid2peerlist(id.nid));
+       }
+
+       route2 = NULL;
+       list_for_each(tmp, &peer->ksnp_routes) {
+               route2 = list_entry(tmp, ksock_route_t, ksnr_list);
+
+               if (route2->ksnr_ipaddr == ipaddr)
+                       break;
+
+               route2 = NULL;
+       }
+       if (route2 == NULL) {
+               ksocknal_add_route_locked(peer, route);
+               route->ksnr_share_count++;
+       } else {
+               ksocknal_route_decref(route);
+               route2->ksnr_share_count++;
+       }
+
+       write_unlock_bh(&ksocknal_data.ksnd_global_lock);
+
+       return 0;
 }
 
-void
+static void
 ksocknal_del_peer_locked (ksock_peer_t *peer, __u32 ip)
 {
-        ksock_conn_t     *conn;
-        ksock_route_t    *route;
-        struct list_head *tmp;
-        struct list_head *nxt;
-        int               nshared;
+       ksock_conn_t     *conn;
+       ksock_route_t    *route;
+       struct list_head *tmp;
+       struct list_head *nxt;
+       int               nshared;
 
-        LASSERT (!peer->ksnp_closing);
+       LASSERT(!peer->ksnp_closing);
 
-        /* Extra ref prevents peer disappearing until I'm done with it */
-        ksocknal_peer_addref(peer);
+       /* Extra ref prevents peer disappearing until I'm done with it */
+       ksocknal_peer_addref(peer);
 
-        list_for_each_safe (tmp, nxt, &peer->ksnp_routes) {
-                route = list_entry(tmp, ksock_route_t, ksnr_list);
+       list_for_each_safe(tmp, nxt, &peer->ksnp_routes) {
+               route = list_entry(tmp, ksock_route_t, ksnr_list);
 
-                /* no match */
-                if (!(ip == 0 || route->ksnr_ipaddr == ip))
-                        continue;
+               /* no match */
+               if (!(ip == 0 || route->ksnr_ipaddr == ip))
+                       continue;
 
-                route->ksnr_share_count = 0;
-                /* This deletes associated conns too */
-                ksocknal_del_route_locked (route);
-        }
+               route->ksnr_share_count = 0;
+               /* This deletes associated conns too */
+               ksocknal_del_route_locked(route);
+       }
 
-        nshared = 0;
-        list_for_each_safe (tmp, nxt, &peer->ksnp_routes) {
-                route = list_entry(tmp, ksock_route_t, ksnr_list);
-                nshared += route->ksnr_share_count;
-        }
+       nshared = 0;
+       list_for_each_safe(tmp, nxt, &peer->ksnp_routes) {
+               route = list_entry(tmp, ksock_route_t, ksnr_list);
+               nshared += route->ksnr_share_count;
+       }
 
-        if (nshared == 0) {
-                /* remove everything else if there are no explicit entries
-                 * left */
+       if (nshared == 0) {
+               /* remove everything else if there are no explicit entries
+                * left */
 
-                list_for_each_safe (tmp, nxt, &peer->ksnp_routes) {
-                        route = list_entry(tmp, ksock_route_t, ksnr_list);
+               list_for_each_safe(tmp, nxt, &peer->ksnp_routes) {
+                       route = list_entry(tmp, ksock_route_t, ksnr_list);
 
-                        /* we should only be removing auto-entries */
-                        LASSERT(route->ksnr_share_count == 0);
-                        ksocknal_del_route_locked (route);
-                }
+                       /* we should only be removing auto-entries */
+                       LASSERT(route->ksnr_share_count == 0);
+                       ksocknal_del_route_locked(route);
+               }
 
-                list_for_each_safe (tmp, nxt, &peer->ksnp_conns) {
-                        conn = list_entry(tmp, ksock_conn_t, ksnc_list);
+               list_for_each_safe(tmp, nxt, &peer->ksnp_conns) {
+                       conn = list_entry(tmp, ksock_conn_t, ksnc_list);
 
-                        ksocknal_close_conn_locked(conn, 0);
-                }
-        }
+                       ksocknal_close_conn_locked(conn, 0);
+               }
+       }
 
-        ksocknal_peer_decref(peer);
-        /* NB peer unlinks itself when last conn/route is removed */
+       ksocknal_peer_decref(peer);
+               /* NB peer unlinks itself when last conn/route is removed */
 }
 
-int
+static int
 ksocknal_del_peer (lnet_ni_t *ni, lnet_process_id_t id, __u32 ip)
 {
-        CFS_LIST_HEAD     (zombies);
-        struct list_head  *ptmp;
-        struct list_head  *pnxt;
-        ksock_peer_t      *peer;
-        int                lo;
-        int                hi;
-        int                i;
-        int                rc = -ENOENT;
+       struct list_head  zombies = LIST_HEAD_INIT(zombies);
+       struct list_head *ptmp;
+       struct list_head *pnxt;
+       ksock_peer_t     *peer;
+       int               lo;
+       int               hi;
+       int               i;
+       int               rc = -ENOENT;
 
-        write_lock_bh (&ksocknal_data.ksnd_global_lock);
+       write_lock_bh(&ksocknal_data.ksnd_global_lock);
 
-        if (id.nid != LNET_NID_ANY)
-                lo = hi = ksocknal_nid2peerlist(id.nid) - ksocknal_data.ksnd_peers;
-        else {
-                lo = 0;
-                hi = ksocknal_data.ksnd_peer_hash_size - 1;
-        }
+       if (id.nid != LNET_NID_ANY) {
+               hi = (int)(ksocknal_nid2peerlist(id.nid) -
+                          ksocknal_data.ksnd_peers);
+               lo = hi;
+       } else {
+               lo = 0;
+               hi = ksocknal_data.ksnd_peer_hash_size - 1;
+       }
 
-        for (i = lo; i <= hi; i++) {
-                list_for_each_safe (ptmp, pnxt, &ksocknal_data.ksnd_peers[i]) {
-                        peer = list_entry (ptmp, ksock_peer_t, ksnp_list);
+       for (i = lo; i <= hi; i++) {
+               list_for_each_safe(ptmp, pnxt,
+                                  &ksocknal_data.ksnd_peers[i]) {
+                       peer = list_entry(ptmp, ksock_peer_t, ksnp_list);
 
-                        if (peer->ksnp_ni != ni)
-                                continue;
+                       if (peer->ksnp_ni != ni)
+                               continue;
 
-                        if (!((id.nid == LNET_NID_ANY || peer->ksnp_id.nid == id.nid) &&
-                              (id.pid == LNET_PID_ANY || peer->ksnp_id.pid == id.pid)))
-                                continue;
+                       if (!((id.nid == LNET_NID_ANY ||
+                              peer->ksnp_id.nid == id.nid) &&
+                             (id.pid == LNET_PID_ANY ||
+                              peer->ksnp_id.pid == id.pid)))
+                               continue;
 
-                        ksocknal_peer_addref(peer);     /* a ref for me... */
+                       ksocknal_peer_addref(peer);     /* a ref for me... */
 
-                        ksocknal_del_peer_locked (peer, ip);
+                       ksocknal_del_peer_locked(peer, ip);
 
-                        if (peer->ksnp_closing && !list_empty(&peer->ksnp_tx_queue)) {
-                                LASSERT (list_empty(&peer->ksnp_conns));
-                                LASSERT (list_empty(&peer->ksnp_routes));
+                       if (peer->ksnp_closing &&
+                           !list_empty(&peer->ksnp_tx_queue)) {
+                               LASSERT(list_empty(&peer->ksnp_conns));
+                               LASSERT(list_empty(&peer->ksnp_routes));
 
-                                list_splice_init(&peer->ksnp_tx_queue, &zombies);
-                        }
+                               list_splice_init(&peer->ksnp_tx_queue,
+                                                &zombies);
+                       }
 
-                        ksocknal_peer_decref(peer);     /* ...till here */
+                       ksocknal_peer_decref(peer);     /* ...till here */
 
-                        rc = 0;                 /* matched! */
-                }
-        }
+                       rc = 0;                         /* matched! */
+               }
+       }
 
-        write_unlock_bh (&ksocknal_data.ksnd_global_lock);
+       write_unlock_bh(&ksocknal_data.ksnd_global_lock);
 
-        ksocknal_txlist_done(ni, &zombies, 1);
+       ksocknal_txlist_done(ni, &zombies, 1);
 
-        return (rc);
+       return rc;
 }
 
-ksock_conn_t *
+static ksock_conn_t *
 ksocknal_get_conn_by_idx (lnet_ni_t *ni, int index)
 {
-        ksock_peer_t      *peer;
-        struct list_head  *ptmp;
-        ksock_conn_t      *conn;
-        struct list_head  *ctmp;
-        int                i;
-
-        read_lock (&ksocknal_data.ksnd_global_lock);
-
-        for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
-                list_for_each (ptmp, &ksocknal_data.ksnd_peers[i]) {
-                        peer = list_entry (ptmp, ksock_peer_t, ksnp_list);
-
-                        LASSERT (!peer->ksnp_closing);
-
-                        if (peer->ksnp_ni != ni)
-                                continue;
-
-                        list_for_each (ctmp, &peer->ksnp_conns) {
-                                if (index-- > 0)
-                                        continue;
-
-                                conn = list_entry (ctmp, ksock_conn_t, ksnc_list);
-                                ksocknal_conn_addref(conn);
-                                read_unlock (&ksocknal_data.ksnd_global_lock);
-                                return (conn);
-                        }
-                }
-        }
-
-        read_unlock (&ksocknal_data.ksnd_global_lock);
-        return (NULL);
+       ksock_peer_t     *peer;
+       struct list_head *ptmp;
+       ksock_conn_t     *conn;
+       struct list_head *ctmp;
+       int               i;
+
+       read_lock(&ksocknal_data.ksnd_global_lock);
+
+       for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
+               list_for_each(ptmp, &ksocknal_data.ksnd_peers[i]) {
+                       peer = list_entry(ptmp, ksock_peer_t, ksnp_list);
+
+                       LASSERT(!peer->ksnp_closing);
+
+                       if (peer->ksnp_ni != ni)
+                               continue;
+
+                       list_for_each(ctmp, &peer->ksnp_conns) {
+                               if (index-- > 0)
+                                       continue;
+
+                               conn = list_entry(ctmp, ksock_conn_t,
+                                                 ksnc_list);
+                               ksocknal_conn_addref(conn);
+                               read_unlock(&ksocknal_data. \
+                                           ksnd_global_lock);
+                               return conn;
+                       }
+               }
+       }
+
+       read_unlock(&ksocknal_data.ksnd_global_lock);
+       return NULL;
 }
 
-ksock_sched_t *
-ksocknal_choose_scheduler_locked (unsigned int irq)
+static ksock_sched_t *
+ksocknal_choose_scheduler_locked(unsigned int cpt)
 {
-        ksock_sched_t    *sched;
-        ksock_irqinfo_t  *info;
-        int               i;
-
-        LASSERT (irq < NR_IRQS);
-        info = &ksocknal_data.ksnd_irqinfo[irq];
-
-        if (irq != 0 &&                         /* hardware NIC */
-            info->ksni_valid) {                 /* already set up */
-                return (&ksocknal_data.ksnd_schedulers[info->ksni_sched]);
-        }
-
-        /* software NIC (irq == 0) || not associated with a scheduler yet.
-         * Choose the CPU with the fewest connections... */
-        sched = &ksocknal_data.ksnd_schedulers[0];
-        for (i = 1; i < ksocknal_data.ksnd_nschedulers; i++)
-                if (sched->kss_nconns >
-                    ksocknal_data.ksnd_schedulers[i].kss_nconns)
-                        sched = &ksocknal_data.ksnd_schedulers[i];
-
-        if (irq != 0) {                         /* Hardware NIC */
-                info->ksni_valid = 1;
-                info->ksni_sched = sched - ksocknal_data.ksnd_schedulers;
-
-                /* no overflow... */
-                LASSERT (info->ksni_sched == sched - ksocknal_data.ksnd_schedulers);
-        }
-
-        return (sched);
+       struct ksock_sched_info *info = ksocknal_data.ksnd_sched_info[cpt];
+       ksock_sched_t           *sched;
+       int                     i;
+
+       LASSERT(info->ksi_nthreads > 0);
+
+       sched = &info->ksi_scheds[0];
+       /*
+        * NB: it's safe so far, but info->ksi_nthreads could be changed
+        * at runtime when we have dynamic LNet configuration, then we
+        * need to take care of this.
+        */
+       for (i = 1; i < info->ksi_nthreads; i++) {
+               if (sched->kss_nconns > info->ksi_scheds[i].kss_nconns)
+                       sched = &info->ksi_scheds[i];
+       }
+
+       return sched;
 }
 
-int
+static int
 ksocknal_local_ipvec (lnet_ni_t *ni, __u32 *ipaddrs)
 {
         ksock_net_t       *net = ni->ni_data;
         int                i;
         int                nip;
 
-        read_lock (&ksocknal_data.ksnd_global_lock);
+       read_lock(&ksocknal_data.ksnd_global_lock);
 
         nip = net->ksnn_ninterfaces;
         LASSERT (nip <= LNET_MAX_INTERFACES);
 
-        /* Only offer interfaces for additional connections if I have 
+       /* Only offer interfaces for additional connections if I have
          * more than one. */
         if (nip < 2) {
-                read_unlock (&ksocknal_data.ksnd_global_lock);
+               read_unlock(&ksocknal_data.ksnd_global_lock);
                 return 0;
         }
 
@@ -722,11 +709,11 @@ ksocknal_local_ipvec (lnet_ni_t *ni, __u32 *ipaddrs)
                 LASSERT (ipaddrs[i] != 0);
         }
 
-        read_unlock (&ksocknal_data.ksnd_global_lock);
+       read_unlock(&ksocknal_data.ksnd_global_lock);
         return (nip);
 }
 
-int
+static int
 ksocknal_match_peerip (ksock_interface_t *iface, __u32 *ips, int nips)
 {
         int   best_netmatch = 0;
@@ -758,10 +745,10 @@ ksocknal_match_peerip (ksock_interface_t *iface, __u32 *ips, int nips)
         return (best);
 }
 
-int
+static int
 ksocknal_select_ips(ksock_peer_t *peer, __u32 *peerips, int n_peerips)
 {
-        rwlock_t           *global_lock = &ksocknal_data.ksnd_global_lock;
+       rwlock_t                *global_lock = &ksocknal_data.ksnd_global_lock;
         ksock_net_t        *net = peer->ksnp_ni->ni_data;
         ksock_interface_t  *iface;
         ksock_interface_t  *best_iface;
@@ -783,12 +770,12 @@ ksocknal_select_ips(ksock_peer_t *peer, __u32 *peerips, int n_peerips)
         /* Also note that I'm not going to return more than n_peerips
          * interfaces, even if I have more myself */
 
-        write_lock_bh (global_lock);
+       write_lock_bh(global_lock);
 
         LASSERT (n_peerips <= LNET_MAX_INTERFACES);
         LASSERT (net->ksnn_ninterfaces <= LNET_MAX_INTERFACES);
 
-        /* Only match interfaces for additional connections 
+       /* Only match interfaces for additional connections
          * if I have > 1 interface */
         n_ips = (net->ksnn_ninterfaces < 2) ? 0 :
                 MIN(n_peerips, net->ksnn_ninterfaces);
@@ -843,14 +830,14 @@ ksocknal_select_ips(ksock_peer_t *peer, __u32 *peerips, int n_peerips)
                                 best_npeers = iface->ksni_npeers;
                         }
 
+                       LASSERT(best_iface != NULL);
+
                         best_iface->ksni_npeers++;
                         ip = best_iface->ksni_ipaddr;
                         peer->ksnp_passive_ips[i] = ip;
                         peer->ksnp_n_passive_ips = i+1;
                 }
 
-                LASSERT (best_iface != NULL);
-
                 /* mark the best matching peer IP used */
                 j = ksocknal_match_peerip(best_iface, peerips, n_peerips);
                 peerips[j] = 0;
@@ -859,40 +846,40 @@ ksocknal_select_ips(ksock_peer_t *peer, __u32 *peerips, int n_peerips)
         /* Overwrite input peer IP addresses */
         memcpy(peerips, peer->ksnp_passive_ips, n_ips * sizeof(*peerips));
 
-        write_unlock_bh (global_lock);
+       write_unlock_bh(global_lock);
 
         return (n_ips);
 }
 
-void
+static void
 ksocknal_create_routes(ksock_peer_t *peer, int port,
                        __u32 *peer_ipaddrs, int npeer_ipaddrs)
 {
-        ksock_route_t      *newroute = NULL;
-        rwlock_t           *global_lock = &ksocknal_data.ksnd_global_lock;
-        lnet_ni_t          *ni = peer->ksnp_ni;
-        ksock_net_t        *net = ni->ni_data;
-        struct list_head   *rtmp;
-        ksock_route_t      *route;
-        ksock_interface_t  *iface;
-        ksock_interface_t  *best_iface;
-        int                 best_netmatch;
-        int                 this_netmatch;
-        int                 best_nroutes;
-        int                 i;
-        int                 j;
+       ksock_route_t           *newroute = NULL;
+       rwlock_t                *global_lock = &ksocknal_data.ksnd_global_lock;
+       lnet_ni_t               *ni = peer->ksnp_ni;
+       ksock_net_t             *net = ni->ni_data;
+       struct list_head        *rtmp;
+       ksock_route_t           *route;
+       ksock_interface_t       *iface;
+       ksock_interface_t       *best_iface;
+       int                     best_netmatch;
+       int                     this_netmatch;
+       int                     best_nroutes;
+       int                     i;
+       int                     j;
 
         /* CAVEAT EMPTOR: We do all our interface matching with an
          * exclusive hold of global lock at IRQ priority.  We're only
          * expecting to be dealing with small numbers of interfaces, so the
          * O(n**3)-ness here shouldn't matter */
 
-        write_lock_bh (global_lock);
+       write_lock_bh(global_lock);
 
         if (net->ksnn_ninterfaces < 2) {
-                /* Only create additional connections 
+               /* Only create additional connections
                  * if I have > 1 interface */
-                write_unlock_bh (global_lock);
+               write_unlock_bh(global_lock);
                 return;
         }
 
@@ -902,13 +889,13 @@ ksocknal_create_routes(ksock_peer_t *peer, int port,
                 if (newroute != NULL) {
                         newroute->ksnr_ipaddr = peer_ipaddrs[i];
                 } else {
-                        write_unlock_bh (global_lock);
+                       write_unlock_bh(global_lock);
 
                         newroute = ksocknal_create_route(peer_ipaddrs[i], port);
                         if (newroute == NULL)
                                 return;
 
-                        write_lock_bh (global_lock);
+                       write_lock_bh(global_lock);
                 }
 
                 if (peer->ksnp_closing) {
@@ -916,40 +903,41 @@ ksocknal_create_routes(ksock_peer_t *peer, int port,
                         break;
                 }
 
-                /* Already got a route? */
-                route = NULL;
-                list_for_each(rtmp, &peer->ksnp_routes) {
-                        route = list_entry(rtmp, ksock_route_t, ksnr_list);
+               /* Already got a route? */
+               route = NULL;
+               list_for_each(rtmp, &peer->ksnp_routes) {
+                       route = list_entry(rtmp, ksock_route_t, ksnr_list);
 
-                        if (route->ksnr_ipaddr == newroute->ksnr_ipaddr)
-                                break;
+                       if (route->ksnr_ipaddr == newroute->ksnr_ipaddr)
+                               break;
 
-                        route = NULL;
-                }
-                if (route != NULL)
-                        continue;
+                       route = NULL;
+               }
+               if (route != NULL)
+                       continue;
 
-                best_iface = NULL;
-                best_nroutes = 0;
-                best_netmatch = 0;
+               best_iface = NULL;
+               best_nroutes = 0;
+               best_netmatch = 0;
 
-                LASSERT (net->ksnn_ninterfaces <= LNET_MAX_INTERFACES);
+               LASSERT(net->ksnn_ninterfaces <= LNET_MAX_INTERFACES);
 
-                /* Select interface to connect from */
-                for (j = 0; j < net->ksnn_ninterfaces; j++) {
-                        iface = &net->ksnn_interfaces[j];
+               /* Select interface to connect from */
+               for (j = 0; j < net->ksnn_ninterfaces; j++) {
+                       iface = &net->ksnn_interfaces[j];
 
-                        /* Using this interface already? */
-                        list_for_each(rtmp, &peer->ksnp_routes) {
-                                route = list_entry(rtmp, ksock_route_t, ksnr_list);
+                       /* Using this interface already? */
+                       list_for_each(rtmp, &peer->ksnp_routes) {
+                               route = list_entry(rtmp, ksock_route_t,
+                                                  ksnr_list);
 
-                                if (route->ksnr_myipaddr == iface->ksni_ipaddr)
-                                        break;
+                               if (route->ksnr_myipaddr == iface->ksni_ipaddr)
+                                       break;
 
-                                route = NULL;
-                        }
-                        if (route != NULL)
-                                continue;
+                               route = NULL;
+                       }
+                       if (route != NULL)
+                               continue;
 
                         this_netmatch = (((iface->ksni_ipaddr ^
                                            newroute->ksnr_ipaddr) &
@@ -976,64 +964,62 @@ ksocknal_create_routes(ksock_peer_t *peer, int port,
                 newroute = NULL;
         }
 
-        write_unlock_bh (global_lock);
+       write_unlock_bh(global_lock);
         if (newroute != NULL)
                 ksocknal_route_decref(newroute);
 }
 
 int
-ksocknal_accept (lnet_ni_t *ni, cfs_socket_t *sock)
+ksocknal_accept(lnet_ni_t *ni, struct socket *sock)
 {
-        ksock_connreq_t    *cr;
-        int                 rc;
-        __u32               peer_ip;
-        int                 peer_port;
-
-        rc = libcfs_sock_getaddr(sock, 1, &peer_ip, &peer_port);
-        LASSERT (rc == 0);                      /* we succeeded before */
-
-        LIBCFS_ALLOC(cr, sizeof(*cr));
-        if (cr == NULL) {
-                LCONSOLE_ERROR_MSG(0x12f, "Dropping connection request from "
-                                   "%u.%u.%u.%u: memory exhausted\n",
-                                   HIPQUAD(peer_ip));
-                return -ENOMEM;
-        }
+       ksock_connreq_t *cr;
+       int              rc;
+       __u32            peer_ip;
+       int              peer_port;
 
-        lnet_ni_addref(ni);
-        cr->ksncr_ni   = ni;
-        cr->ksncr_sock = sock;
+       rc = lnet_sock_getaddr(sock, true, &peer_ip, &peer_port);
+       LASSERT(rc == 0);               /* we succeeded before */
 
-        spin_lock_bh (&ksocknal_data.ksnd_connd_lock);
+       LIBCFS_ALLOC(cr, sizeof(*cr));
+       if (cr == NULL) {
+               LCONSOLE_ERROR_MSG(0x12f, "Dropping connection request from "
+                                  "%pI4h: memory exhausted\n", &peer_ip);
+               return -ENOMEM;
+       }
 
-        list_add_tail(&cr->ksncr_list, &ksocknal_data.ksnd_connd_connreqs);
-        cfs_waitq_signal(&ksocknal_data.ksnd_connd_waitq);
+       lnet_ni_addref(ni);
+       cr->ksncr_ni   = ni;
+       cr->ksncr_sock = sock;
 
-        spin_unlock_bh (&ksocknal_data.ksnd_connd_lock);
-        return 0;
+       spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
+
+       list_add_tail(&cr->ksncr_list, &ksocknal_data.ksnd_connd_connreqs);
+       wake_up(&ksocknal_data.ksnd_connd_waitq);
+
+       spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
+       return 0;
 }
 
-int
+static int
 ksocknal_connecting (ksock_peer_t *peer, __u32 ipaddr)
 {
-        ksock_route_t   *route;
+       ksock_route_t *route;
 
-        list_for_each_entry (route, &peer->ksnp_routes, ksnr_list) {
-
-                if (route->ksnr_ipaddr == ipaddr)
-                        return route->ksnr_connecting;
-        }
-        return 0;
+       list_for_each_entry(route, &peer->ksnp_routes, ksnr_list) {
+               if (route->ksnr_ipaddr == ipaddr)
+                       return route->ksnr_connecting;
+       }
+       return 0;
 }
 
 int
-ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
-                      cfs_socket_t *sock, int type)
+ksocknal_create_conn(lnet_ni_t *ni, ksock_route_t *route,
+                    struct socket *sock, int type)
 {
-        rwlock_t          *global_lock = &ksocknal_data.ksnd_global_lock;
-        CFS_LIST_HEAD     (zombies);
-        lnet_process_id_t  peerid;
-        struct list_head  *tmp;
+       rwlock_t                *global_lock = &ksocknal_data.ksnd_global_lock;
+       struct list_head        zombies = LIST_HEAD_INIT(zombies);
+       lnet_process_id_t       peerid;
+       struct list_head        *tmp;
         __u64              incarnation;
         ksock_conn_t      *conn;
         ksock_conn_t      *conn2;
@@ -1041,8 +1027,9 @@ ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
         ksock_peer_t      *peer2;
         ksock_sched_t     *sched;
         ksock_hello_msg_t *hello;
-        unsigned int       irq;
+       int                cpt;
         ksock_tx_t        *tx;
+        ksock_tx_t        *txtmp;
         int                rc;
         int                active;
         char              *warn = NULL;
@@ -1051,37 +1038,33 @@ ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
 
         LASSERT (active == (type != SOCKLND_CONN_NONE));
 
-        irq = ksocknal_lib_sock_irq (sock);
-
         LIBCFS_ALLOC(conn, sizeof(*conn));
         if (conn == NULL) {
                 rc = -ENOMEM;
                 goto failed_0;
         }
 
-        memset (conn, 0, sizeof (*conn));
         conn->ksnc_peer = NULL;
         conn->ksnc_route = NULL;
         conn->ksnc_sock = sock;
-        /* 2 ref, 1 for conn, another extra ref prevents socket
-         * being closed before establishment of connection */
-        atomic_set (&conn->ksnc_sock_refcount, 2);
-        conn->ksnc_type = type;
-        ksocknal_lib_save_callback(sock, conn);
-        atomic_set (&conn->ksnc_conn_refcount, 1); /* 1 ref for me */
-
-        conn->ksnc_zc_capable = ksocknal_lib_zc_capable(sock);
-        conn->ksnc_rx_ready = 0;
-        conn->ksnc_rx_scheduled = 0;
-
-        CFS_INIT_LIST_HEAD (&conn->ksnc_tx_queue);
-        conn->ksnc_tx_ready = 0;
-        conn->ksnc_tx_scheduled = 0;
-        conn->ksnc_tx_mono = NULL;
-        atomic_set (&conn->ksnc_tx_nob, 0);
-
-        LIBCFS_ALLOC(hello, offsetof(ksock_hello_msg_t,
-                                     kshm_ips[LNET_MAX_INTERFACES]));
+       /* 2 ref, 1 for conn, another extra ref prevents socket
+        * being closed before establishment of connection */
+       atomic_set (&conn->ksnc_sock_refcount, 2);
+       conn->ksnc_type = type;
+       ksocknal_lib_save_callback(sock, conn);
+       atomic_set (&conn->ksnc_conn_refcount, 1); /* 1 ref for me */
+
+       conn->ksnc_rx_ready = 0;
+       conn->ksnc_rx_scheduled = 0;
+
+       INIT_LIST_HEAD(&conn->ksnc_tx_queue);
+       conn->ksnc_tx_ready = 0;
+       conn->ksnc_tx_scheduled = 0;
+       conn->ksnc_tx_carrier = NULL;
+       atomic_set (&conn->ksnc_tx_nob, 0);
+
+       LIBCFS_ALLOC(hello, offsetof(ksock_hello_msg_t,
+                                    kshm_ips[LNET_MAX_INTERFACES]));
         if (hello == NULL) {
                 rc = -ENOMEM;
                 goto failed_1;
@@ -1105,15 +1088,17 @@ ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
                 hello->kshm_nips = ksocknal_local_ipvec(ni, hello->kshm_ips);
                 peerid = peer->ksnp_id;
 
-                write_lock_bh(global_lock);
+               write_lock_bh(global_lock);
                 conn->ksnc_proto = peer->ksnp_proto;
-                write_unlock_bh(global_lock);
+               write_unlock_bh(global_lock);
 
                 if (conn->ksnc_proto == NULL) {
-                        conn->ksnc_proto = &ksocknal_protocol_v2x;
+                         conn->ksnc_proto = &ksocknal_protocol_v3x;
 #if SOCKNAL_VERSION_DEBUG
-                        if (*ksocknal_tunables.ksnd_protocol != 2)
-                                conn->ksnc_proto = &ksocknal_protocol_v1x;
+                         if (*ksocknal_tunables.ksnd_protocol == 2)
+                                 conn->ksnc_proto = &ksocknal_protocol_v2x;
+                         else if (*ksocknal_tunables.ksnd_protocol == 1)
+                                 conn->ksnc_proto = &ksocknal_protocol_v1x;
 #endif
                 }
 
@@ -1136,29 +1121,31 @@ ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
         LASSERT (conn->ksnc_proto != NULL);
         LASSERT (peerid.nid != LNET_NID_ANY);
 
+       cpt = lnet_cpt_of_nid(peerid.nid);
+
         if (active) {
                 ksocknal_peer_addref(peer);
-                write_lock_bh (global_lock);
+               write_lock_bh(global_lock);
         } else {
                 rc = ksocknal_create_peer(&peer, ni, peerid);
                 if (rc != 0)
                         goto failed_1;
 
-                write_lock_bh (global_lock);
+               write_lock_bh(global_lock);
 
                 /* called with a ref on ni, so shutdown can't have started */
                 LASSERT (((ksock_net_t *) ni->ni_data)->ksnn_shutdown == 0);
 
-                peer2 = ksocknal_find_peer_locked(ni, peerid);
-                if (peer2 == NULL) {
-                        /* NB this puts an "empty" peer in the peer
-                         * table (which takes my ref) */
-                        list_add_tail(&peer->ksnp_list,
-                                      ksocknal_nid2peerlist(peerid.nid));
-                } else {
-                        ksocknal_peer_decref(peer);
-                        peer = peer2;
-                }
+               peer2 = ksocknal_find_peer_locked(ni, peerid);
+               if (peer2 == NULL) {
+                       /* NB this puts an "empty" peer in the peer
+                        * table (which takes my ref) */
+                       list_add_tail(&peer->ksnp_list,
+                                     ksocknal_nid2peerlist(peerid.nid));
+               } else {
+                       ksocknal_peer_decref(peer);
+                       peer = peer2;
+               }
 
                 /* +1 ref for me */
                 ksocknal_peer_addref(peer);
@@ -1182,16 +1169,16 @@ ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
                 goto failed_2;
         }
 
-        if (peer->ksnp_proto == NULL) {
-                /* Never connected before.
-                 * NB recv_hello may have returned EPROTO to signal my peer
-                 * wants a different protocol than the one I asked for.
-                 */
-                LASSERT (list_empty(&peer->ksnp_conns));
+       if (peer->ksnp_proto == NULL) {
+               /* Never connected before.
+                * NB recv_hello may have returned EPROTO to signal my peer
+                * wants a different protocol than the one I asked for.
+                */
+               LASSERT(list_empty(&peer->ksnp_conns));
 
-                peer->ksnp_proto = conn->ksnc_proto;
-                peer->ksnp_incarnation = incarnation;
-        }
+               peer->ksnp_proto = conn->ksnc_proto;
+               peer->ksnp_incarnation = incarnation;
+       }
 
         if (peer->ksnp_proto != conn->ksnc_proto ||
             peer->ksnp_incarnation != incarnation) {
@@ -1219,11 +1206,11 @@ ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
                 goto failed_2;
         }
 
-        /* Refuse to duplicate an existing connection, unless this is a
-         * loopback connection */
-        if (conn->ksnc_ipaddr != conn->ksnc_myipaddr) {
-                list_for_each(tmp, &peer->ksnp_conns) {
-                        conn2 = list_entry(tmp, ksock_conn_t, ksnc_list);
+       /* Refuse to duplicate an existing connection, unless this is a
+        * loopback connection */
+       if (conn->ksnc_ipaddr != conn->ksnc_myipaddr) {
+               list_for_each(tmp, &peer->ksnp_conns) {
+                       conn2 = list_entry(tmp, ksock_conn_t, ksnc_list);
 
                         if (conn2->ksnc_ipaddr != conn->ksnc_ipaddr ||
                             conn2->ksnc_myipaddr != conn->ksnc_myipaddr ||
@@ -1246,73 +1233,74 @@ ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
          * code below probably isn't going to work. */
         if (active &&
             route->ksnr_ipaddr != conn->ksnc_ipaddr) {
-                CERROR("Route %s %u.%u.%u.%u connected to %u.%u.%u.%u\n",
+               CERROR("Route %s %pI4h connected to %pI4h\n",
                        libcfs_id2str(peer->ksnp_id),
-                       HIPQUAD(route->ksnr_ipaddr),
-                       HIPQUAD(conn->ksnc_ipaddr));
+                      &route->ksnr_ipaddr,
+                      &conn->ksnc_ipaddr);
         }
 
-        /* Search for a route corresponding to the new connection and
-         * create an association.  This allows incoming connections created
-         * by routes in my peer to match my own route entries so I don't
-         * continually create duplicate routes. */
-        list_for_each (tmp, &peer->ksnp_routes) {
-                route = list_entry(tmp, ksock_route_t, ksnr_list);
+       /* Search for a route corresponding to the new connection and
+        * create an association.  This allows incoming connections created
+        * by routes in my peer to match my own route entries so I don't
+        * continually create duplicate routes. */
+       list_for_each(tmp, &peer->ksnp_routes) {
+               route = list_entry(tmp, ksock_route_t, ksnr_list);
 
-                if (route->ksnr_ipaddr != conn->ksnc_ipaddr)
-                        continue;
+               if (route->ksnr_ipaddr != conn->ksnc_ipaddr)
+                       continue;
 
-                ksocknal_associate_route_conn_locked(route, conn);
-                break;
-        }
+               ksocknal_associate_route_conn_locked(route, conn);
+               break;
+       }
 
         conn->ksnc_peer = peer;                 /* conn takes my ref on peer */
         peer->ksnp_last_alive = cfs_time_current();
+        peer->ksnp_send_keepalive = 0;
         peer->ksnp_error = 0;
 
-        sched = ksocknal_choose_scheduler_locked (irq);
+       sched = ksocknal_choose_scheduler_locked(cpt);
         sched->kss_nconns++;
         conn->ksnc_scheduler = sched;
 
-        /* Set the deadline for the outgoing HELLO to drain */
-        conn->ksnc_tx_bufnob = SOCK_WMEM_QUEUED(sock);
-        conn->ksnc_tx_deadline = cfs_time_shift(*ksocknal_tunables.ksnd_timeout);
-        mb();       /* order with adding to peer's conn list */
+       conn->ksnc_tx_last_post = cfs_time_current();
+       /* Set the deadline for the outgoing HELLO to drain */
+       conn->ksnc_tx_bufnob = sock->sk->sk_wmem_queued;
+       conn->ksnc_tx_deadline = cfs_time_shift(*ksocknal_tunables.ksnd_timeout);
+       smp_mb();   /* order with adding to peer's conn list */
 
-        list_add (&conn->ksnc_list, &peer->ksnp_conns);
-        ksocknal_conn_addref(conn);
+       list_add(&conn->ksnc_list, &peer->ksnp_conns);
+       ksocknal_conn_addref(conn);
 
-        ksocknal_new_packet(conn, 0);
+       ksocknal_new_packet(conn, 0);
 
-        /* Take all the packets blocking for a connection.
-         * NB, it might be nicer to share these blocked packets among any
-         * other connections that are becoming established. */
-        while (!list_empty (&peer->ksnp_tx_queue)) {
-                tx = list_entry (peer->ksnp_tx_queue.next,
-                                 ksock_tx_t, tx_list);
+        conn->ksnc_zc_capable = ksocknal_lib_zc_capable(conn);
 
-                list_del (&tx->tx_list);
-                ksocknal_queue_tx_locked (tx, conn);
-        }
+       /* Take packets blocking for this connection. */
+       list_for_each_entry_safe(tx, txtmp, &peer->ksnp_tx_queue, tx_list) {
+               if (conn->ksnc_proto->pro_match_tx(conn, tx, tx->tx_nonblk) ==
+                   SOCKNAL_MATCH_NO)
+                       continue;
+
+               list_del(&tx->tx_list);
+               ksocknal_queue_tx_locked(tx, conn);
+       }
 
-        write_unlock_bh (global_lock);
+       write_unlock_bh(global_lock);
 
         /* We've now got a new connection.  Any errors from here on are just
          * like "normal" comms errors and we close the connection normally.
          * NB (a) we still have to send the reply HELLO for passive
-         *        connections, 
+        *        connections,
          *    (b) normal I/O on the conn is blocked until I setup and call the
          *        socket callbacks.
          */
 
-        ksocknal_lib_bind_irq (irq);
-
-        CDEBUG(D_NET, "New conn %s p %d.x %u.%u.%u.%u -> %u.%u.%u.%u/%d"
-               " incarnation:"LPD64" sched[%d]/%d\n",
-               libcfs_id2str(peerid), conn->ksnc_proto->pro_version,
-               HIPQUAD(conn->ksnc_myipaddr), HIPQUAD(conn->ksnc_ipaddr),
-               conn->ksnc_port, incarnation,
-               (int)(conn->ksnc_scheduler - ksocknal_data.ksnd_schedulers), irq);
+       CDEBUG(D_NET, "New conn %s p %d.x %pI4h -> %pI4h/%d"
+              " incarnation:%lld sched[%d:%d]\n",
+              libcfs_id2str(peerid), conn->ksnc_proto->pro_version,
+              &conn->ksnc_myipaddr, &conn->ksnc_ipaddr,
+              conn->ksnc_port, incarnation, cpt,
+              (int)(sched - &sched->kss_info->ksi_scheds[0]));
 
         if (active) {
                 /* additional routes after interface exchange? */
@@ -1335,7 +1323,7 @@ ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
         if (rc == 0)
                 rc = ksocknal_lib_setup_sock(sock);
 
-        write_lock_bh(global_lock);
+       write_lock_bh(global_lock);
 
         /* NB my callbacks block while I hold ksnd_global_lock */
         ksocknal_lib_set_callback(sock, conn);
@@ -1343,12 +1331,15 @@ ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
         if (!active)
                 peer->ksnp_accepting--;
 
-        write_unlock_bh(global_lock);
+       write_unlock_bh(global_lock);
 
         if (rc != 0) {
-                write_lock_bh(global_lock);
-                ksocknal_close_conn_locked(conn, rc);
-                write_unlock_bh(global_lock);
+               write_lock_bh(global_lock);
+                if (!conn->ksnc_closing) {
+                        /* could be closed by another thread */
+                        ksocknal_close_conn_locked(conn, rc);
+                }
+               write_unlock_bh(global_lock);
         } else if (ksocknal_connsock_addref(conn) == 0) {
                 /* Allow I/O to proceed. */
                 ksocknal_read_callback(conn);
@@ -1360,16 +1351,16 @@ ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
         ksocknal_conn_decref(conn);
         return rc;
 
- failed_2:
-        if (!peer->ksnp_closing &&
-            list_empty (&peer->ksnp_conns) &&
-            list_empty (&peer->ksnp_routes)) {
-                list_add(&zombies, &peer->ksnp_tx_queue);
-                list_del_init(&peer->ksnp_tx_queue);
-                ksocknal_unlink_peer_locked(peer);
-        }
+failed_2:
+       if (!peer->ksnp_closing &&
+           list_empty(&peer->ksnp_conns) &&
+           list_empty(&peer->ksnp_routes)) {
+               list_add(&zombies, &peer->ksnp_tx_queue);
+               list_del_init(&peer->ksnp_tx_queue);
+               ksocknal_unlink_peer_locked(peer);
+       }
 
-        write_unlock_bh (global_lock);
+       write_unlock_bh(global_lock);
 
         if (warn != NULL) {
                 if (rc < 0)
@@ -1382,16 +1373,16 @@ ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
 
         if (!active) {
                 if (rc > 0) {
-                        /* Request retry by replying with CONN_NONE 
+                       /* Request retry by replying with CONN_NONE
                          * ksnc_proto has been set already */
                         conn->ksnc_type = SOCKLND_CONN_NONE;
                         hello->kshm_nips = 0;
                         ksocknal_send_hello(ni, conn, peerid.nid, hello);
                 }
 
-                write_lock_bh(global_lock);
+               write_lock_bh(global_lock);
                 peer->ksnp_accepting--;
-                write_unlock_bh(global_lock);
+               write_unlock_bh(global_lock);
         }
 
         ksocknal_txlist_done(ni, &zombies, 1);
@@ -1402,11 +1393,11 @@ ksocknal_create_conn (lnet_ni_t *ni, ksock_route_t *route,
                 LIBCFS_FREE(hello, offsetof(ksock_hello_msg_t,
                                             kshm_ips[LNET_MAX_INTERFACES]));
 
-        LIBCFS_FREE (conn, sizeof(*conn));
+       LIBCFS_FREE(conn, sizeof(*conn));
 
- failed_0:
-        libcfs_sock_release(sock);
-        return rc;
+failed_0:
+       sock_release(sock);
+       return rc;
 }
 
 void
@@ -1418,130 +1409,150 @@ ksocknal_close_conn_locked (ksock_conn_t *conn, int error)
         ksock_peer_t      *peer = conn->ksnc_peer;
         ksock_route_t     *route;
         ksock_conn_t      *conn2;
-        struct list_head  *tmp;
+       struct list_head  *tmp;
 
-        LASSERT (peer->ksnp_error == 0);
-        LASSERT (!conn->ksnc_closing);
-        conn->ksnc_closing = 1;
+       LASSERT(peer->ksnp_error == 0);
+       LASSERT(!conn->ksnc_closing);
+       conn->ksnc_closing = 1;
 
-        /* ksnd_deathrow_conns takes over peer's ref */
-        list_del (&conn->ksnc_list);
+       /* ksnd_deathrow_conns takes over peer's ref */
+       list_del(&conn->ksnc_list);
 
-        route = conn->ksnc_route;
-        if (route != NULL) {
-                /* dissociate conn from route... */
-                LASSERT (!route->ksnr_deleted);
-                LASSERT ((route->ksnr_connected & (1 << conn->ksnc_type)) != 0);
+       route = conn->ksnc_route;
+       if (route != NULL) {
+               /* dissociate conn from route... */
+               LASSERT(!route->ksnr_deleted);
+               LASSERT((route->ksnr_connected & (1 << conn->ksnc_type)) != 0);
 
-                conn2 = NULL;
-                list_for_each(tmp, &peer->ksnp_conns) {
-                        conn2 = list_entry(tmp, ksock_conn_t, ksnc_list);
+               conn2 = NULL;
+               list_for_each(tmp, &peer->ksnp_conns) {
+                       conn2 = list_entry(tmp, ksock_conn_t, ksnc_list);
 
-                        if (conn2->ksnc_route == route &&
-                            conn2->ksnc_type == conn->ksnc_type)
-                                break;
+                       if (conn2->ksnc_route == route &&
+                           conn2->ksnc_type == conn->ksnc_type)
+                               break;
 
-                        conn2 = NULL;
-                }
-                if (conn2 == NULL)
-                        route->ksnr_connected &= ~(1 << conn->ksnc_type);
+                       conn2 = NULL;
+               }
+               if (conn2 == NULL)
+                       route->ksnr_connected &= ~(1 << conn->ksnc_type);
 
-                conn->ksnc_route = NULL;
+               conn->ksnc_route = NULL;
 
-#if 0           /* irrelevent with only eager routes */
-                list_del (&route->ksnr_list);   /* make route least favourite */
-                list_add_tail (&route->ksnr_list, &peer->ksnp_routes);
+#if 0          /* irrelevent with only eager routes */
+               /* make route least favourite */
+               list_del(&route->ksnr_list);
+               list_add_tail(&route->ksnr_list, &peer->ksnp_routes);
 #endif
-                ksocknal_route_decref(route);     /* drop conn's ref on route */
-        }
+               ksocknal_route_decref(route);   /* drop conn's ref on route */
+       }
 
-        if (list_empty (&peer->ksnp_conns)) {
-                /* No more connections to this peer */
+       if (list_empty(&peer->ksnp_conns)) {
+               /* No more connections to this peer */
 
-                peer->ksnp_proto = NULL;        /* renegotiate protocol version */
-                peer->ksnp_error = error;       /* stash last conn close reason */
+               if (!list_empty(&peer->ksnp_tx_queue)) {
+                               ksock_tx_t *tx;
 
-                if (list_empty (&peer->ksnp_routes)) {
-                        /* I've just closed last conn belonging to a
-                         * peer with no routes to it */
-                        ksocknal_unlink_peer_locked (peer);
-                }
-        }
+                       LASSERT(conn->ksnc_proto == &ksocknal_protocol_v3x);
+
+                       /* throw them to the last connection...,
+                        * these TXs will be send to /dev/null by scheduler */
+                       list_for_each_entry(tx, &peer->ksnp_tx_queue,
+                                           tx_list)
+                               ksocknal_tx_prep(conn, tx);
 
-        spin_lock_bh (&ksocknal_data.ksnd_reaper_lock);
+                       spin_lock_bh(&conn->ksnc_scheduler->kss_lock);
+                       list_splice_init(&peer->ksnp_tx_queue,
+                                        &conn->ksnc_tx_queue);
+                       spin_unlock_bh(&conn->ksnc_scheduler->kss_lock);
+               }
 
-        list_add_tail (&conn->ksnc_list, &ksocknal_data.ksnd_deathrow_conns);
-        cfs_waitq_signal (&ksocknal_data.ksnd_reaper_waitq);
+               /* renegotiate protocol version */
+               peer->ksnp_proto = NULL;
+               /* stash last conn close reason */
+               peer->ksnp_error = error;
 
-        spin_unlock_bh (&ksocknal_data.ksnd_reaper_lock);
+               if (list_empty(&peer->ksnp_routes)) {
+                       /* I've just closed last conn belonging to a
+                        * peer with no routes to it */
+                       ksocknal_unlink_peer_locked(peer);
+               }
+       }
+
+       spin_lock_bh(&ksocknal_data.ksnd_reaper_lock);
+
+       list_add_tail(&conn->ksnc_list,
+                     &ksocknal_data.ksnd_deathrow_conns);
+       wake_up(&ksocknal_data.ksnd_reaper_waitq);
+
+       spin_unlock_bh(&ksocknal_data.ksnd_reaper_lock);
 }
 
 void
 ksocknal_peer_failed (ksock_peer_t *peer)
 {
-        time_t    last_alive = 0;
-        int       notify = 0;
-
-        /* There has been a connection failure or comms error; but I'll only
-         * tell LNET I think the peer is dead if it's to another kernel and
-         * there are no connections or connection attempts in existance. */
-
-        read_lock (&ksocknal_data.ksnd_global_lock);
-
-        if ((peer->ksnp_id.pid & LNET_PID_USERFLAG) == 0 &&
-            list_empty(&peer->ksnp_conns) &&
-            peer->ksnp_accepting == 0 &&
-            ksocknal_find_connecting_route_locked(peer) == NULL) {
-                notify = 1;
-                last_alive = cfs_time_current_sec() -
-                        cfs_duration_sec(cfs_time_current() -
-                                         peer->ksnp_last_alive);
-        }
+        int        notify = 0;
+        cfs_time_t last_alive = 0;
+
+       /* There has been a connection failure or comms error; but I'll only
+        * tell LNET I think the peer is dead if it's to another kernel and
+        * there are no connections or connection attempts in existence. */
+
+       read_lock(&ksocknal_data.ksnd_global_lock);
 
-        read_unlock (&ksocknal_data.ksnd_global_lock);
+       if ((peer->ksnp_id.pid & LNET_PID_USERFLAG) == 0 &&
+            list_empty(&peer->ksnp_conns) &&
+            peer->ksnp_accepting == 0 &&
+            ksocknal_find_connecting_route_locked(peer) == NULL) {
+               notify = 1;
+               last_alive = peer->ksnp_last_alive;
+       }
 
-        if (notify)
-                lnet_notify (peer->ksnp_ni, peer->ksnp_id.nid, 0,
-                             last_alive);
+       read_unlock(&ksocknal_data.ksnd_global_lock);
+
+       if (notify)
+               lnet_notify(peer->ksnp_ni, peer->ksnp_id.nid, 0,
+                           last_alive);
 }
 
 void
 ksocknal_finalize_zcreq(ksock_conn_t *conn)
 {
-        ksock_peer_t     *peer = conn->ksnc_peer;
-        ksock_tx_t       *tx;
-        ksock_tx_t       *tmp;
-        CFS_LIST_HEAD    (zlist);
+       ksock_peer_t     *peer = conn->ksnc_peer;
+       ksock_tx_t       *tx;
+       ksock_tx_t       *tmp;
+       struct list_head  zlist = LIST_HEAD_INIT(zlist);
 
-        /* NB safe to finalize TXs because closing of socket will
-         * abort all buffered data */
-        LASSERT (conn->ksnc_sock == NULL);
+       /* NB safe to finalize TXs because closing of socket will
+        * abort all buffered data */
+       LASSERT(conn->ksnc_sock == NULL);
 
-        spin_lock(&peer->ksnp_lock);
+       spin_lock(&peer->ksnp_lock);
 
-        list_for_each_entry_safe(tx, tmp, &peer->ksnp_zc_req_list, tx_zc_list) {
-                if (tx->tx_conn != conn)
-                        continue;
+       list_for_each_entry_safe(tx, tmp, &peer->ksnp_zc_req_list, tx_zc_list) {
+               if (tx->tx_conn != conn)
+                       continue;
 
-                LASSERT (tx->tx_msg.ksm_zc_req_cookie != 0);
+               LASSERT(tx->tx_msg.ksm_zc_cookies[0] != 0);
 
-                tx->tx_msg.ksm_zc_req_cookie = 0;
-                list_del(&tx->tx_zc_list);
-                list_add(&tx->tx_zc_list, &zlist);
-        }
+               tx->tx_msg.ksm_zc_cookies[0] = 0;
+               tx->tx_zc_aborted = 1;  /* mark it as not-acked */
+               list_del(&tx->tx_zc_list);
+               list_add(&tx->tx_zc_list, &zlist);
+       }
 
-        spin_unlock(&peer->ksnp_lock);
+       spin_unlock(&peer->ksnp_lock);
 
-        while (!list_empty(&zlist)) {
-                tx = list_entry(zlist.next, ksock_tx_t, tx_zc_list);
+       while (!list_empty(&zlist)) {
+               tx = list_entry(zlist.next, ksock_tx_t, tx_zc_list);
 
-                list_del(&tx->tx_zc_list);
-                ksocknal_tx_decref(tx);
-        }
+               list_del(&tx->tx_zc_list);
+               ksocknal_tx_decref(tx);
+       }
 }
 
 void
-ksocknal_terminate_conn (ksock_conn_t *conn)
+ksocknal_terminate_conn(ksock_conn_t *conn)
 {
         /* This gets called by the reaper (guaranteed thread context) to
          * disengage the socket from its callbacks and close it.
@@ -1554,26 +1565,26 @@ ksocknal_terminate_conn (ksock_conn_t *conn)
         LASSERT(conn->ksnc_closing);
 
         /* wake up the scheduler to "send" all remaining packets to /dev/null */
-        spin_lock_bh (&sched->kss_lock);
+       spin_lock_bh(&sched->kss_lock);
 
         /* a closing conn is always ready to tx */
         conn->ksnc_tx_ready = 1;
 
         if (!conn->ksnc_tx_scheduled &&
-            !list_empty(&conn->ksnc_tx_queue)){
-                list_add_tail (&conn->ksnc_tx_list,
+           !list_empty(&conn->ksnc_tx_queue)) {
+               list_add_tail(&conn->ksnc_tx_list,
                                &sched->kss_tx_conns);
                 conn->ksnc_tx_scheduled = 1;
                 /* extra ref for scheduler */
-                ksocknal_conn_addref(conn);
+               ksocknal_conn_addref(conn);
 
-                cfs_waitq_signal (&sched->kss_waitq);
-        }
+               wake_up (&sched->kss_waitq);
+       }
 
-        spin_unlock_bh (&sched->kss_lock);
+       spin_unlock_bh(&sched->kss_lock);
 
-        /* serialise with callbacks */
-        write_lock_bh (&ksocknal_data.ksnd_global_lock);
+       /* serialise with callbacks */
+       write_lock_bh(&ksocknal_data.ksnd_global_lock);
 
         ksocknal_lib_reset_callback(conn->ksnc_sock, conn);
 
@@ -1583,12 +1594,12 @@ ksocknal_terminate_conn (ksock_conn_t *conn)
 
         if (peer->ksnp_error != 0) {
                 /* peer's last conn closed in error */
-                LASSERT (list_empty (&peer->ksnp_conns));
+               LASSERT(list_empty(&peer->ksnp_conns));
                 failed = 1;
                 peer->ksnp_error = 0;     /* avoid multiple notifications */
         }
 
-        write_unlock_bh (&ksocknal_data.ksnd_global_lock);
+       write_unlock_bh(&ksocknal_data.ksnd_global_lock);
 
         if (failed)
                 ksocknal_peer_failed(peer);
@@ -1604,63 +1615,71 @@ ksocknal_terminate_conn (ksock_conn_t *conn)
 void
 ksocknal_queue_zombie_conn (ksock_conn_t *conn)
 {
-        /* Queue the conn for the reaper to destroy */
+       /* Queue the conn for the reaper to destroy */
 
-        LASSERT (atomic_read(&conn->ksnc_conn_refcount) == 0);
-        spin_lock_bh (&ksocknal_data.ksnd_reaper_lock);
+       LASSERT(atomic_read(&conn->ksnc_conn_refcount) == 0);
+       spin_lock_bh(&ksocknal_data.ksnd_reaper_lock);
 
-        list_add_tail(&conn->ksnc_list, &ksocknal_data.ksnd_zombie_conns);
-        cfs_waitq_signal(&ksocknal_data.ksnd_reaper_waitq);
+       list_add_tail(&conn->ksnc_list, &ksocknal_data.ksnd_zombie_conns);
+       wake_up(&ksocknal_data.ksnd_reaper_waitq);
 
-        spin_unlock_bh (&ksocknal_data.ksnd_reaper_lock);
+       spin_unlock_bh(&ksocknal_data.ksnd_reaper_lock);
 }
 
 void
 ksocknal_destroy_conn (ksock_conn_t *conn)
 {
-        /* Final coup-de-grace of the reaper */
-        CDEBUG (D_NET, "connection %p\n", conn);
+       cfs_time_t      last_rcv;
+
+       /* Final coup-de-grace of the reaper */
+       CDEBUG (D_NET, "connection %p\n", conn);
 
-        LASSERT (atomic_read (&conn->ksnc_conn_refcount) == 0);
-        LASSERT (atomic_read (&conn->ksnc_sock_refcount) == 0);
-        LASSERT (conn->ksnc_sock == NULL);
-        LASSERT (conn->ksnc_route == NULL);
-        LASSERT (!conn->ksnc_tx_scheduled);
-        LASSERT (!conn->ksnc_rx_scheduled);
-        LASSERT (list_empty(&conn->ksnc_tx_queue));
+       LASSERT (atomic_read (&conn->ksnc_conn_refcount) == 0);
+       LASSERT (atomic_read (&conn->ksnc_sock_refcount) == 0);
+       LASSERT (conn->ksnc_sock == NULL);
+       LASSERT (conn->ksnc_route == NULL);
+       LASSERT (!conn->ksnc_tx_scheduled);
+       LASSERT (!conn->ksnc_rx_scheduled);
+       LASSERT(list_empty(&conn->ksnc_tx_queue));
 
         /* complete current receive if any */
         switch (conn->ksnc_rx_state) {
         case SOCKNAL_RX_LNET_PAYLOAD:
-                CERROR("Completing partial receive from %s"
-                       ", ip %d.%d.%d.%d:%d, with error\n",
-                       libcfs_id2str(conn->ksnc_peer->ksnp_id),
-                       HIPQUAD(conn->ksnc_ipaddr), conn->ksnc_port);
+                last_rcv = conn->ksnc_rx_deadline -
+                           cfs_time_seconds(*ksocknal_tunables.ksnd_timeout);
+               CERROR("Completing partial receive from %s[%d], "
+                      "ip %pI4h:%d, with error, wanted: %d, left: %d, "
+                       "last alive is %ld secs ago\n",
+                       libcfs_id2str(conn->ksnc_peer->ksnp_id), conn->ksnc_type,
+                      &conn->ksnc_ipaddr, conn->ksnc_port,
+                       conn->ksnc_rx_nob_wanted, conn->ksnc_rx_nob_left,
+                       cfs_duration_sec(cfs_time_sub(cfs_time_current(),
+                                        last_rcv)));
                 lnet_finalize (conn->ksnc_peer->ksnp_ni,
                                conn->ksnc_cookie, -EIO);
                 break;
         case SOCKNAL_RX_LNET_HEADER:
                 if (conn->ksnc_rx_started)
-                        CERROR("Incomplete receive of lnet header from %s"
-                               ", ip %d.%d.%d.%d:%d, with error, protocol: %d.x.\n",
+                       CERROR("Incomplete receive of lnet header from %s, "
+                              "ip %pI4h:%d, with error, protocol: %d.x.\n",
                                libcfs_id2str(conn->ksnc_peer->ksnp_id),
-                               HIPQUAD(conn->ksnc_ipaddr), conn->ksnc_port,
+                              &conn->ksnc_ipaddr, conn->ksnc_port,
                                conn->ksnc_proto->pro_version);
                 break;
         case SOCKNAL_RX_KSM_HEADER:
                 if (conn->ksnc_rx_started)
-                        CERROR("Incomplete receive of ksock message from %s"
-                               ", ip %d.%d.%d.%d:%d, with error, protocol: %d.x.\n",
+                       CERROR("Incomplete receive of ksock message from %s, "
+                              "ip %pI4h:%d, with error, protocol: %d.x.\n",
                                libcfs_id2str(conn->ksnc_peer->ksnp_id),
-                               HIPQUAD(conn->ksnc_ipaddr), conn->ksnc_port,
+                              &conn->ksnc_ipaddr, conn->ksnc_port,
                                conn->ksnc_proto->pro_version);
                 break;
         case SOCKNAL_RX_SLOP:
                 if (conn->ksnc_rx_started)
-                        CERROR("Incomplete receive of slops from %s"
-                               ", ip %d.%d.%d.%d:%d, with error\n",
+                       CERROR("Incomplete receive of slops from %s, "
+                              "ip %pI4h:%d, with error\n",
                                libcfs_id2str(conn->ksnc_peer->ksnp_id),
-                               HIPQUAD(conn->ksnc_ipaddr), conn->ksnc_port);
+                              &conn->ksnc_ipaddr, conn->ksnc_port);
                break;
         default:
                 LBUG ();
@@ -1676,12 +1695,12 @@ int
 ksocknal_close_peer_conns_locked (ksock_peer_t *peer, __u32 ipaddr, int why)
 {
         ksock_conn_t       *conn;
-        struct list_head   *ctmp;
-        struct list_head   *cnxt;
+       struct list_head         *ctmp;
+       struct list_head         *cnxt;
         int                 count = 0;
 
-        list_for_each_safe (ctmp, cnxt, &peer->ksnp_conns) {
-                conn = list_entry (ctmp, ksock_conn_t, ksnc_list);
+       list_for_each_safe(ctmp, cnxt, &peer->ksnp_conns) {
+               conn = list_entry(ctmp, ksock_conn_t, ksnc_list);
 
                 if (ipaddr == 0 ||
                     conn->ksnc_ipaddr == ipaddr) {
@@ -1700,11 +1719,11 @@ ksocknal_close_conn_and_siblings (ksock_conn_t *conn, int why)
         __u32             ipaddr = conn->ksnc_ipaddr;
         int               count;
 
-        write_lock_bh (&ksocknal_data.ksnd_global_lock);
+       write_lock_bh(&ksocknal_data.ksnd_global_lock);
 
         count = ksocknal_close_peer_conns_locked (peer, ipaddr, why);
 
-        write_unlock_bh (&ksocknal_data.ksnd_global_lock);
+       write_unlock_bh(&ksocknal_data.ksnd_global_lock);
 
         return (count);
 }
@@ -1713,26 +1732,26 @@ int
 ksocknal_close_matching_conns (lnet_process_id_t id, __u32 ipaddr)
 {
         ksock_peer_t       *peer;
-        struct list_head   *ptmp;
-        struct list_head   *pnxt;
+       struct list_head         *ptmp;
+       struct list_head         *pnxt;
         int                 lo;
         int                 hi;
         int                 i;
         int                 count = 0;
 
-        write_lock_bh (&ksocknal_data.ksnd_global_lock);
+       write_lock_bh(&ksocknal_data.ksnd_global_lock);
 
         if (id.nid != LNET_NID_ANY)
-                lo = hi = ksocknal_nid2peerlist(id.nid) - ksocknal_data.ksnd_peers;
+                lo = hi = (int)(ksocknal_nid2peerlist(id.nid) - ksocknal_data.ksnd_peers);
         else {
                 lo = 0;
                 hi = ksocknal_data.ksnd_peer_hash_size - 1;
         }
 
         for (i = lo; i <= hi; i++) {
-                list_for_each_safe (ptmp, pnxt, &ksocknal_data.ksnd_peers[i]) {
+               list_for_each_safe(ptmp, pnxt, &ksocknal_data.ksnd_peers[i]) {
 
-                        peer = list_entry (ptmp, ksock_peer_t, ksnp_list);
+                       peer = list_entry(ptmp, ksock_peer_t, ksnp_list);
 
                         if (!((id.nid == LNET_NID_ANY || id.nid == peer->ksnp_id.nid) &&
                               (id.pid == LNET_PID_ANY || id.pid == peer->ksnp_id.pid)))
@@ -1742,7 +1761,7 @@ ksocknal_close_matching_conns (lnet_process_id_t id, __u32 ipaddr)
                 }
         }
 
-        write_unlock_bh (&ksocknal_data.ksnd_global_lock);
+       write_unlock_bh(&ksocknal_data.ksnd_global_lock);
 
         /* wildcards always succeed */
         if (id.nid == LNET_NID_ANY || id.pid == LNET_PID_ANY || ipaddr == 0)
@@ -1756,7 +1775,10 @@ ksocknal_notify (lnet_ni_t *ni, lnet_nid_t gw_nid, int alive)
 {
         /* The router is telling me she's been notified of a change in
          * gateway state.... */
-        lnet_process_id_t  id = {.nid = gw_nid, .pid = LNET_PID_ANY};
+        lnet_process_id_t  id = {0};
+
+        id.nid = gw_nid;
+        id.pid = LNET_PID_ANY;
 
         CDEBUG (D_NET, "gw %s %s\n", libcfs_nid2str(gw_nid),
                 alive ? "up" : "down");
@@ -1772,28 +1794,93 @@ ksocknal_notify (lnet_ni_t *ni, lnet_nid_t gw_nid, int alive)
 }
 
 void
+ksocknal_query (lnet_ni_t *ni, lnet_nid_t nid, cfs_time_t *when)
+{
+        int                connect = 1;
+        cfs_time_t         last_alive = 0;
+        cfs_time_t         now = cfs_time_current();
+        ksock_peer_t      *peer = NULL;
+       rwlock_t                *glock = &ksocknal_data.ksnd_global_lock;
+       lnet_process_id_t  id = {
+               .nid = nid,
+               .pid = LNET_PID_LUSTRE,
+       };
+
+       read_lock(glock);
+
+        peer = ksocknal_find_peer_locked(ni, id);
+        if (peer != NULL) {
+               struct list_head       *tmp;
+                ksock_conn_t     *conn;
+                int               bufnob;
+
+               list_for_each(tmp, &peer->ksnp_conns) {
+                       conn = list_entry(tmp, ksock_conn_t, ksnc_list);
+                       bufnob = conn->ksnc_sock->sk->sk_wmem_queued;
+
+                        if (bufnob < conn->ksnc_tx_bufnob) {
+                                /* something got ACKed */
+                                conn->ksnc_tx_deadline =
+                                        cfs_time_shift(*ksocknal_tunables.ksnd_timeout);
+                                peer->ksnp_last_alive = now;
+                                conn->ksnc_tx_bufnob = bufnob;
+                        }
+                }
+
+                last_alive = peer->ksnp_last_alive;
+                if (ksocknal_find_connectable_route_locked(peer) == NULL)
+                        connect = 0;
+        }
+
+       read_unlock(glock);
+
+        if (last_alive != 0)
+                *when = last_alive;
+
+        CDEBUG(D_NET, "Peer %s %p, alive %ld secs ago, connect %d\n",
+               libcfs_nid2str(nid), peer,
+               last_alive ? cfs_duration_sec(now - last_alive) : -1,
+               connect);
+
+        if (!connect)
+                return;
+
+        ksocknal_add_peer(ni, id, LNET_NIDADDR(nid), lnet_acceptor_port());
+
+       write_lock_bh(glock);
+
+        peer = ksocknal_find_peer_locked(ni, id);
+        if (peer != NULL)
+                ksocknal_launch_all_connections_locked(peer);
+
+       write_unlock_bh(glock);
+        return;
+}
+
+static void
 ksocknal_push_peer (ksock_peer_t *peer)
 {
         int               index;
         int               i;
-        struct list_head *tmp;
+       struct list_head       *tmp;
         ksock_conn_t     *conn;
 
         for (index = 0; ; index++) {
-                read_lock (&ksocknal_data.ksnd_global_lock);
+               read_lock(&ksocknal_data.ksnd_global_lock);
 
                 i = 0;
                 conn = NULL;
 
-                list_for_each (tmp, &peer->ksnp_conns) {
+               list_for_each(tmp, &peer->ksnp_conns) {
                         if (i++ == index) {
-                                conn = list_entry (tmp, ksock_conn_t, ksnc_list);
+                               conn = list_entry(tmp, ksock_conn_t,
+                                                       ksnc_list);
                                 ksocknal_conn_addref(conn);
                                 break;
                         }
                 }
 
-                read_unlock (&ksocknal_data.ksnd_global_lock);
+               read_unlock(&ksocknal_data.ksnd_global_lock);
 
                 if (conn == NULL)
                         break;
@@ -1803,56 +1890,56 @@ ksocknal_push_peer (ksock_peer_t *peer)
         }
 }
 
-int
+static int
 ksocknal_push (lnet_ni_t *ni, lnet_process_id_t id)
 {
-        ksock_peer_t      *peer;
-        struct list_head  *tmp;
-        int                index;
-        int                i;
-        int                j;
-        int                rc = -ENOENT;
-
-        for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
-                for (j = 0; ; j++) {
-                        read_lock (&ksocknal_data.ksnd_global_lock);
-
-                        index = 0;
-                        peer = NULL;
-
-                        list_for_each (tmp, &ksocknal_data.ksnd_peers[i]) {
-                                peer = list_entry(tmp, ksock_peer_t,
-                                                  ksnp_list);
-
-                                if (!((id.nid == LNET_NID_ANY ||
-                                       id.nid == peer->ksnp_id.nid) &&
-                                      (id.pid == LNET_PID_ANY ||
-                                       id.pid == peer->ksnp_id.pid))) {
-                                        peer = NULL;
-                                        continue;
-                                }
-
-                                if (index++ == j) {
-                                        ksocknal_peer_addref(peer);
-                                        break;
-                                }
-                        }
-
-                        read_unlock (&ksocknal_data.ksnd_global_lock);
-
-                        if (peer != NULL) {
-                                rc = 0;
-                                ksocknal_push_peer (peer);
-                                ksocknal_peer_decref(peer);
-                        }
-                }
-
-        }
-
-        return (rc);
+       struct list_head *start;
+       struct list_head *end;
+       struct list_head *tmp;
+       int               rc = -ENOENT;
+       unsigned int      hsize = ksocknal_data.ksnd_peer_hash_size;
+
+       if (id.nid == LNET_NID_ANY) {
+               start = &ksocknal_data.ksnd_peers[0];
+               end = &ksocknal_data.ksnd_peers[hsize - 1];
+       } else {
+               start = end = ksocknal_nid2peerlist(id.nid);
+       }
+
+       for (tmp = start; tmp <= end; tmp++) {
+               int     peer_off; /* searching offset in peer hash table */
+
+               for (peer_off = 0; ; peer_off++) {
+                       ksock_peer_t *peer;
+                       int           i = 0;
+
+                       read_lock(&ksocknal_data.ksnd_global_lock);
+                       list_for_each_entry(peer, tmp, ksnp_list) {
+                               if (!((id.nid == LNET_NID_ANY ||
+                                      id.nid == peer->ksnp_id.nid) &&
+                                     (id.pid == LNET_PID_ANY ||
+                                      id.pid == peer->ksnp_id.pid)))
+                                       continue;
+
+                               if (i++ == peer_off) {
+                                       ksocknal_peer_addref(peer);
+                                       break;
+                               }
+                       }
+                       read_unlock(&ksocknal_data.ksnd_global_lock);
+
+                       if (i == 0) /* no match */
+                               break;
+
+                       rc = 0;
+                       ksocknal_push_peer(peer);
+                       ksocknal_peer_decref(peer);
+               }
+       }
+       return rc;
 }
 
-int
+static int
 ksocknal_add_interface(lnet_ni_t *ni, __u32 ipaddress, __u32 netmask)
 {
         ksock_net_t       *net = ni->ni_data;
@@ -1860,16 +1947,16 @@ ksocknal_add_interface(lnet_ni_t *ni, __u32 ipaddress, __u32 netmask)
         int                rc;
         int                i;
         int                j;
-        struct list_head  *ptmp;
+       struct list_head        *ptmp;
         ksock_peer_t      *peer;
-        struct list_head  *rtmp;
+       struct list_head        *rtmp;
         ksock_route_t     *route;
 
         if (ipaddress == 0 ||
             netmask == 0)
                 return (-EINVAL);
 
-        write_lock_bh (&ksocknal_data.ksnd_global_lock);
+       write_lock_bh(&ksocknal_data.ksnd_global_lock);
 
         iface = ksocknal_ip2iface(ni, ipaddress);
         if (iface != NULL) {
@@ -1886,15 +1973,18 @@ ksocknal_add_interface(lnet_ni_t *ni, __u32 ipaddress, __u32 netmask)
                 iface->ksni_npeers = 0;
 
                 for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
-                        list_for_each(ptmp, &ksocknal_data.ksnd_peers[i]) {
-                                peer = list_entry(ptmp, ksock_peer_t, ksnp_list);
+                       list_for_each(ptmp, &ksocknal_data.ksnd_peers[i]) {
+                               peer = list_entry(ptmp, ksock_peer_t,
+                                                      ksnp_list);
 
                                 for (j = 0; j < peer->ksnp_n_passive_ips; j++)
                                         if (peer->ksnp_passive_ips[j] == ipaddress)
                                                 iface->ksni_npeers++;
 
-                                list_for_each(rtmp, &peer->ksnp_routes) {
-                                        route = list_entry(rtmp, ksock_route_t, ksnr_list);
+                               list_for_each(rtmp, &peer->ksnp_routes) {
+                                       route = list_entry(rtmp,
+                                                               ksock_route_t,
+                                                               ksnr_list);
 
                                         if (route->ksnr_myipaddr == ipaddress)
                                                 iface->ksni_nroutes++;
@@ -1906,16 +1996,16 @@ ksocknal_add_interface(lnet_ni_t *ni, __u32 ipaddress, __u32 netmask)
                 /* NB only new connections will pay attention to the new interface! */
         }
 
-        write_unlock_bh (&ksocknal_data.ksnd_global_lock);
+       write_unlock_bh(&ksocknal_data.ksnd_global_lock);
 
         return (rc);
 }
 
-void
+static void
 ksocknal_peer_del_interface_locked(ksock_peer_t *peer, __u32 ipaddr)
 {
-        struct list_head   *tmp;
-        struct list_head   *nxt;
+       struct list_head         *tmp;
+       struct list_head         *nxt;
         ksock_route_t      *route;
         ksock_conn_t       *conn;
         int                 i;
@@ -1930,8 +2020,8 @@ ksocknal_peer_del_interface_locked(ksock_peer_t *peer, __u32 ipaddr)
                         break;
                 }
 
-        list_for_each_safe(tmp, nxt, &peer->ksnp_routes) {
-                route = list_entry (tmp, ksock_route_t, ksnr_list);
+       list_for_each_safe(tmp, nxt, &peer->ksnp_routes) {
+               route = list_entry(tmp, ksock_route_t, ksnr_list);
 
                 if (route->ksnr_myipaddr != ipaddr)
                         continue;
@@ -1944,27 +2034,27 @@ ksocknal_peer_del_interface_locked(ksock_peer_t *peer, __u32 ipaddr)
                 }
         }
 
-        list_for_each_safe(tmp, nxt, &peer->ksnp_conns) {
-                conn = list_entry(tmp, ksock_conn_t, ksnc_list);
+       list_for_each_safe(tmp, nxt, &peer->ksnp_conns) {
+               conn = list_entry(tmp, ksock_conn_t, ksnc_list);
 
                 if (conn->ksnc_myipaddr == ipaddr)
                         ksocknal_close_conn_locked (conn, 0);
         }
 }
 
-int
+static int
 ksocknal_del_interface(lnet_ni_t *ni, __u32 ipaddress)
 {
         ksock_net_t       *net = ni->ni_data;
         int                rc = -ENOENT;
-        struct list_head  *tmp;
-        struct list_head  *nxt;
+       struct list_head        *tmp;
+       struct list_head        *nxt;
         ksock_peer_t      *peer;
         __u32              this_ip;
         int                i;
         int                j;
 
-        write_lock_bh (&ksocknal_data.ksnd_global_lock);
+       write_lock_bh(&ksocknal_data.ksnd_global_lock);
 
         for (i = 0; i < net->ksnn_ninterfaces; i++) {
                 this_ip = net->ksnn_interfaces[i].ksni_ipaddr;
@@ -1982,8 +2072,10 @@ ksocknal_del_interface(lnet_ni_t *ni, __u32 ipaddress)
                 net->ksnn_ninterfaces--;
 
                 for (j = 0; j < ksocknal_data.ksnd_peer_hash_size; j++) {
-                        list_for_each_safe(tmp, nxt, &ksocknal_data.ksnd_peers[j]) {
-                                peer = list_entry(tmp, ksock_peer_t, ksnp_list);
+                       list_for_each_safe(tmp, nxt,
+                                               &ksocknal_data.ksnd_peers[j]) {
+                               peer = list_entry(tmp, ksock_peer_t,
+                                                      ksnp_list);
 
                                 if (peer->ksnp_ni != ni)
                                         continue;
@@ -1993,7 +2085,7 @@ ksocknal_del_interface(lnet_ni_t *ni, __u32 ipaddress)
                 }
         }
 
-        write_unlock_bh (&ksocknal_data.ksnd_global_lock);
+       write_unlock_bh(&ksocknal_data.ksnd_global_lock);
 
         return (rc);
 }
@@ -2001,6 +2093,7 @@ ksocknal_del_interface(lnet_ni_t *ni, __u32 ipaddress)
 int
 ksocknal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg)
 {
+       lnet_process_id_t id = {0};
         struct libcfs_ioctl_data *data = arg;
         int rc;
 
@@ -2009,10 +2102,9 @@ ksocknal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg)
                 ksock_net_t       *net = ni->ni_data;
                 ksock_interface_t *iface;
 
-                read_lock (&ksocknal_data.ksnd_global_lock);
+               read_lock(&ksocknal_data.ksnd_global_lock);
 
-                if (data->ioc_count < 0 ||
-                    data->ioc_count >= net->ksnn_ninterfaces) {
+                if (data->ioc_count >= (__u32)net->ksnn_ninterfaces) {
                         rc = -ENOENT;
                 } else {
                         rc = 0;
@@ -2024,7 +2116,7 @@ ksocknal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg)
                         data->ioc_u32[3] = iface->ksni_nroutes;
                 }
 
-                read_unlock (&ksocknal_data.ksnd_global_lock);
+               read_unlock(&ksocknal_data.ksnd_global_lock);
                 return rc;
         }
 
@@ -2038,7 +2130,6 @@ ksocknal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg)
                                               data->ioc_u32[0]); /* IP address */
 
         case IOC_LIBCFS_GET_PEER: {
-                lnet_process_id_t id = {0,};
                 __u32            myip = 0;
                 __u32            ip = 0;
                 int              port = 0;
@@ -2061,19 +2152,19 @@ ksocknal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg)
                 return 0;
         }
 
-        case IOC_LIBCFS_ADD_PEER: {
-                lnet_process_id_t  id = {.nid = data->ioc_nid,
-                                         .pid = LUSTRE_SRV_LNET_PID};
+        case IOC_LIBCFS_ADD_PEER:
+                id.nid = data->ioc_nid;
+               id.pid = LNET_PID_LUSTRE;
                 return ksocknal_add_peer (ni, id,
                                           data->ioc_u32[0], /* IP */
                                           data->ioc_u32[1]); /* port */
-        }
-        case IOC_LIBCFS_DEL_PEER: {
-                lnet_process_id_t  id = {.nid = data->ioc_nid,
-                                         .pid = LNET_PID_ANY};
+
+        case IOC_LIBCFS_DEL_PEER:
+                id.nid = data->ioc_nid;
+                id.pid = LNET_PID_ANY;
                 return ksocknal_del_peer (ni, id,
                                           data->ioc_u32[0]); /* IP */
-        }
+
         case IOC_LIBCFS_GET_CONN: {
                 int           txmem;
                 int           rxmem;
@@ -2092,21 +2183,19 @@ ksocknal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg)
                 data->ioc_u32[1] = conn->ksnc_port;
                 data->ioc_u32[2] = conn->ksnc_myipaddr;
                 data->ioc_u32[3] = conn->ksnc_type;
-                data->ioc_u32[4] = conn->ksnc_scheduler -
-                                   ksocknal_data.ksnd_schedulers;
+               data->ioc_u32[4] = conn->ksnc_scheduler->kss_info->ksi_cpt;
                 data->ioc_u32[5] = rxmem;
                 data->ioc_u32[6] = conn->ksnc_peer->ksnp_id.pid;
                 ksocknal_conn_decref(conn);
                 return 0;
         }
 
-        case IOC_LIBCFS_CLOSE_CONNECTION: {
-                lnet_process_id_t  id = {.nid = data->ioc_nid,
-                                        .pid = LNET_PID_ANY};
-
+        case IOC_LIBCFS_CLOSE_CONNECTION:
+                id.nid = data->ioc_nid;
+                id.pid = LNET_PID_ANY;
                 return ksocknal_close_matching_conns (id,
                                                       data->ioc_u32[0]);
-        }
+
         case IOC_LIBCFS_REGISTER_MYNID:
                 /* Ignore if this is a noop */
                 if (data->ioc_nid == ni->ni_nid)
@@ -2117,60 +2206,71 @@ ksocknal_ctl(lnet_ni_t *ni, unsigned int cmd, void *arg)
                        libcfs_nid2str(ni->ni_nid));
                 return -EINVAL;
 
-        case IOC_LIBCFS_PUSH_CONNECTION: {
-                lnet_process_id_t  id = {.nid = data->ioc_nid,
-                                        .pid = LNET_PID_ANY};
-
+        case IOC_LIBCFS_PUSH_CONNECTION:
+                id.nid = data->ioc_nid;
+                id.pid = LNET_PID_ANY;
                 return ksocknal_push(ni, id);
-        }
+
         default:
                 return -EINVAL;
         }
         /* not reached */
 }
 
-void
+static void
 ksocknal_free_buffers (void)
 {
-        LASSERT (atomic_read(&ksocknal_data.ksnd_nactive_txs) == 0);
-
-        if (ksocknal_data.ksnd_schedulers != NULL)
-                LIBCFS_FREE (ksocknal_data.ksnd_schedulers,
-                             sizeof (ksock_sched_t) * ksocknal_data.ksnd_nschedulers);
+       LASSERT (atomic_read(&ksocknal_data.ksnd_nactive_txs) == 0);
+
+       if (ksocknal_data.ksnd_sched_info != NULL) {
+               struct ksock_sched_info *info;
+               int                     i;
+
+               cfs_percpt_for_each(info, i, ksocknal_data.ksnd_sched_info) {
+                       if (info->ksi_scheds != NULL) {
+                               LIBCFS_FREE(info->ksi_scheds,
+                                           info->ksi_nthreads_max *
+                                           sizeof(info->ksi_scheds[0]));
+                       }
+               }
+               cfs_percpt_free(ksocknal_data.ksnd_sched_info);
+       }
 
         LIBCFS_FREE (ksocknal_data.ksnd_peers,
-                     sizeof (struct list_head) *
+                    sizeof(struct list_head) *
                      ksocknal_data.ksnd_peer_hash_size);
 
-        spin_lock(&ksocknal_data.ksnd_tx_lock);
+       spin_lock(&ksocknal_data.ksnd_tx_lock);
 
-        if (!list_empty(&ksocknal_data.ksnd_idle_noop_txs)) {
-                struct list_head  zlist;
-                ksock_tx_t       *tx;
+       if (!list_empty(&ksocknal_data.ksnd_idle_noop_txs)) {
+               struct list_head        zlist;
+               ksock_tx_t      *tx;
 
-                list_add(&zlist, &ksocknal_data.ksnd_idle_noop_txs);
-                list_del_init(&ksocknal_data.ksnd_idle_noop_txs);
-                spin_unlock(&ksocknal_data.ksnd_tx_lock);
+               list_add(&zlist, &ksocknal_data.ksnd_idle_noop_txs);
+               list_del_init(&ksocknal_data.ksnd_idle_noop_txs);
+               spin_unlock(&ksocknal_data.ksnd_tx_lock);
 
-                while(!list_empty(&zlist)) {
-                        tx = list_entry(zlist.next, ksock_tx_t, tx_list);
-                        list_del(&tx->tx_list);
-                        LIBCFS_FREE(tx, tx->tx_desc_size);
-                }
-        } else {
-                spin_unlock(&ksocknal_data.ksnd_tx_lock);
-        }
+               while (!list_empty(&zlist)) {
+                       tx = list_entry(zlist.next, ksock_tx_t, tx_list);
+                       list_del(&tx->tx_list);
+                       LIBCFS_FREE(tx, tx->tx_desc_size);
+               }
+       } else {
+               spin_unlock(&ksocknal_data.ksnd_tx_lock);
+       }
 }
 
-void
-ksocknal_base_shutdown (void)
+static void
+ksocknal_base_shutdown(void)
 {
-        ksock_sched_t *sched;
-        int            i;
+       struct ksock_sched_info *info;
+       ksock_sched_t           *sched;
+       int                     i;
+       int                     j;
 
-        CDEBUG(D_MALLOC, "before NAL cleanup: kmem %d\n",
-               atomic_read (&libcfs_kmemory));
-        LASSERT (ksocknal_data.ksnd_nnets == 0);
+       CDEBUG(D_MALLOC, "before NAL cleanup: kmem %d\n",
+              atomic_read (&libcfs_kmemory));
+       LASSERT (ksocknal_data.ksnd_nnets == 0);
 
         switch (ksocknal_data.ksnd_init) {
         default:
@@ -2180,47 +2280,67 @@ ksocknal_base_shutdown (void)
         case SOCKNAL_INIT_DATA:
                 LASSERT (ksocknal_data.ksnd_peers != NULL);
                 for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
-                        LASSERT (list_empty (&ksocknal_data.ksnd_peers[i]));
+                       LASSERT(list_empty(&ksocknal_data.ksnd_peers[i]));
                 }
-                LASSERT (list_empty (&ksocknal_data.ksnd_enomem_conns));
-                LASSERT (list_empty (&ksocknal_data.ksnd_zombie_conns));
-                LASSERT (list_empty (&ksocknal_data.ksnd_connd_connreqs));
-                LASSERT (list_empty (&ksocknal_data.ksnd_connd_routes));
-
-                if (ksocknal_data.ksnd_schedulers != NULL)
-                        for (i = 0; i < ksocknal_data.ksnd_nschedulers; i++) {
-                                ksock_sched_t *kss =
-                                        &ksocknal_data.ksnd_schedulers[i];
-
-                                LASSERT (list_empty (&kss->kss_tx_conns));
-                                LASSERT (list_empty (&kss->kss_rx_conns));
-                                LASSERT (list_empty (&kss->kss_zombie_noop_txs));
-                                LASSERT (kss->kss_nconns == 0);
-                        }
-
-                /* flag threads to terminate; wake and wait for them to die */
-                ksocknal_data.ksnd_shuttingdown = 1;
-                cfs_waitq_broadcast (&ksocknal_data.ksnd_connd_waitq);
-                cfs_waitq_broadcast (&ksocknal_data.ksnd_reaper_waitq);
-
-                if (ksocknal_data.ksnd_schedulers != NULL)
-                        for (i = 0; i < ksocknal_data.ksnd_nschedulers; i++) {
-                                sched = &ksocknal_data.ksnd_schedulers[i];
-                                cfs_waitq_broadcast(&sched->kss_waitq);
-                        }
 
-                i = 4;
-                read_lock (&ksocknal_data.ksnd_global_lock);
-                while (ksocknal_data.ksnd_nthreads != 0) {
-                        i++;
-                        CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET, /* power of 2? */
-                               "waiting for %d threads to terminate\n",
-                                ksocknal_data.ksnd_nthreads);
-                        read_unlock (&ksocknal_data.ksnd_global_lock);
-                        cfs_pause(cfs_time_seconds(1));
-                        read_lock (&ksocknal_data.ksnd_global_lock);
-                }
-                read_unlock (&ksocknal_data.ksnd_global_lock);
+               LASSERT(list_empty(&ksocknal_data.ksnd_nets));
+               LASSERT(list_empty(&ksocknal_data.ksnd_enomem_conns));
+               LASSERT(list_empty(&ksocknal_data.ksnd_zombie_conns));
+               LASSERT(list_empty(&ksocknal_data.ksnd_connd_connreqs));
+               LASSERT(list_empty(&ksocknal_data.ksnd_connd_routes));
+
+               if (ksocknal_data.ksnd_sched_info != NULL) {
+                       cfs_percpt_for_each(info, i,
+                                           ksocknal_data.ksnd_sched_info) {
+                               if (info->ksi_scheds == NULL)
+                                       continue;
+
+                               for (j = 0; j < info->ksi_nthreads_max; j++) {
+
+                                       sched = &info->ksi_scheds[j];
+                                       LASSERT(list_empty(&sched->\
+                                                              kss_tx_conns));
+                                       LASSERT(list_empty(&sched->\
+                                                              kss_rx_conns));
+                                       LASSERT(list_empty(&sched-> \
+                                                 kss_zombie_noop_txs));
+                                       LASSERT(sched->kss_nconns == 0);
+                               }
+                       }
+               }
+
+               /* flag threads to terminate; wake and wait for them to die */
+               ksocknal_data.ksnd_shuttingdown = 1;
+               wake_up_all(&ksocknal_data.ksnd_connd_waitq);
+               wake_up_all(&ksocknal_data.ksnd_reaper_waitq);
+
+               if (ksocknal_data.ksnd_sched_info != NULL) {
+                       cfs_percpt_for_each(info, i,
+                                           ksocknal_data.ksnd_sched_info) {
+                               if (info->ksi_scheds == NULL)
+                                       continue;
+
+                               for (j = 0; j < info->ksi_nthreads_max; j++) {
+                                       sched = &info->ksi_scheds[j];
+                                       wake_up_all(&sched->kss_waitq);
+                               }
+                       }
+               }
+
+               i = 4;
+               read_lock(&ksocknal_data.ksnd_global_lock);
+               while (ksocknal_data.ksnd_nthreads != 0) {
+                       i++;
+                       /* power of 2? */
+                       CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET,
+                               "waiting for %d threads to terminate\n",
+                               ksocknal_data.ksnd_nthreads);
+                       read_unlock(&ksocknal_data.ksnd_global_lock);
+                       set_current_state(TASK_UNINTERRUPTIBLE);
+                       schedule_timeout(cfs_time_seconds(1));
+                       read_lock(&ksocknal_data.ksnd_global_lock);
+               }
+               read_unlock(&ksocknal_data.ksnd_global_lock);
 
                 ksocknal_free_buffers();
 
@@ -2228,33 +2348,32 @@ ksocknal_base_shutdown (void)
                 break;
         }
 
-        CDEBUG(D_MALLOC, "after NAL cleanup: kmem %d\n",
-               atomic_read (&libcfs_kmemory));
+       CDEBUG(D_MALLOC, "after NAL cleanup: kmem %d\n",
+              atomic_read (&libcfs_kmemory));
 
-        PORTAL_MODULE_UNUSE;
+       module_put(THIS_MODULE);
 }
 
-
-__u64
-ksocknal_new_incarnation (void)
+static __u64 ksocknal_new_incarnation(void)
 {
-        struct timeval tv;
+       struct timeval tv;
 
-        /* The incarnation number is the time this module loaded and it
-         * identifies this particular instance of the socknal.  Hopefully
-         * we won't be able to reboot more frequently than 1MHz for the
-         * forseeable future :) */
+       /* The incarnation number is the time this module loaded and it
+        * identifies this particular instance of the socknal.  Hopefully
+        * we won't be able to reboot more frequently than 1MHz for the
+        * forseeable future :) */
 
-        do_gettimeofday(&tv);
+       do_gettimeofday(&tv);
 
-        return (((__u64)tv.tv_sec) * 1000000) + tv.tv_usec;
+       return (((__u64)tv.tv_sec) * 1000000) + tv.tv_usec;
 }
 
-int
-ksocknal_base_startup (void)
+static int
+ksocknal_base_startup(void)
 {
-        int               rc;
-        int               i;
+       struct ksock_sched_info *info;
+       int                     rc;
+       int                     i;
 
         LASSERT (ksocknal_data.ksnd_init == SOCKNAL_INIT_NOTHING);
         LASSERT (ksocknal_data.ksnd_nnets == 0);
@@ -2262,77 +2381,110 @@ ksocknal_base_startup (void)
         memset (&ksocknal_data, 0, sizeof (ksocknal_data)); /* zero pointers */
 
         ksocknal_data.ksnd_peer_hash_size = SOCKNAL_PEER_HASH_SIZE;
-        LIBCFS_ALLOC (ksocknal_data.ksnd_peers,
-                      sizeof (struct list_head) * ksocknal_data.ksnd_peer_hash_size);
+       LIBCFS_ALLOC(ksocknal_data.ksnd_peers,
+                    sizeof(struct list_head) *
+                    ksocknal_data.ksnd_peer_hash_size);
         if (ksocknal_data.ksnd_peers == NULL)
                 return -ENOMEM;
 
         for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++)
-                CFS_INIT_LIST_HEAD(&ksocknal_data.ksnd_peers[i]);
-
-        rwlock_init(&ksocknal_data.ksnd_global_lock);
-
-        spin_lock_init (&ksocknal_data.ksnd_reaper_lock);
-        CFS_INIT_LIST_HEAD (&ksocknal_data.ksnd_enomem_conns);
-        CFS_INIT_LIST_HEAD (&ksocknal_data.ksnd_zombie_conns);
-        CFS_INIT_LIST_HEAD (&ksocknal_data.ksnd_deathrow_conns);
-        cfs_waitq_init(&ksocknal_data.ksnd_reaper_waitq);
-
-        spin_lock_init (&ksocknal_data.ksnd_connd_lock);
-        CFS_INIT_LIST_HEAD (&ksocknal_data.ksnd_connd_connreqs);
-        CFS_INIT_LIST_HEAD (&ksocknal_data.ksnd_connd_routes);
-        cfs_waitq_init(&ksocknal_data.ksnd_connd_waitq);
-
-        spin_lock_init (&ksocknal_data.ksnd_tx_lock);
-        CFS_INIT_LIST_HEAD (&ksocknal_data.ksnd_idle_noop_txs);
-
-        /* NB memset above zeros whole of ksocknal_data, including
-         * ksocknal_data.ksnd_irqinfo[all].ksni_valid */
-
-        /* flag lists/ptrs/locks initialised */
-        ksocknal_data.ksnd_init = SOCKNAL_INIT_DATA;
-        PORTAL_MODULE_USE;
-
-        ksocknal_data.ksnd_nschedulers = ksocknal_nsched();
-        LIBCFS_ALLOC(ksocknal_data.ksnd_schedulers,
-                     sizeof(ksock_sched_t) * ksocknal_data.ksnd_nschedulers);
-        if (ksocknal_data.ksnd_schedulers == NULL)
-                goto failed;
-
-        for (i = 0; i < ksocknal_data.ksnd_nschedulers; i++) {
-                ksock_sched_t *kss = &ksocknal_data.ksnd_schedulers[i];
-
-                spin_lock_init (&kss->kss_lock);
-                CFS_INIT_LIST_HEAD (&kss->kss_rx_conns);
-                CFS_INIT_LIST_HEAD (&kss->kss_tx_conns);
-                CFS_INIT_LIST_HEAD (&kss->kss_zombie_noop_txs);
-                cfs_waitq_init (&kss->kss_waitq);
-        }
-
-        for (i = 0; i < ksocknal_data.ksnd_nschedulers; i++) {
-                rc = ksocknal_thread_start (ksocknal_scheduler,
-                                            &ksocknal_data.ksnd_schedulers[i]);
-                if (rc != 0) {
-                        CERROR("Can't spawn socknal scheduler[%d]: %d\n",
-                               i, rc);
-                        goto failed;
-                }
-        }
-
+               INIT_LIST_HEAD(&ksocknal_data.ksnd_peers[i]);
+
+       rwlock_init(&ksocknal_data.ksnd_global_lock);
+       INIT_LIST_HEAD(&ksocknal_data.ksnd_nets);
+
+       spin_lock_init(&ksocknal_data.ksnd_reaper_lock);
+       INIT_LIST_HEAD(&ksocknal_data.ksnd_enomem_conns);
+       INIT_LIST_HEAD(&ksocknal_data.ksnd_zombie_conns);
+       INIT_LIST_HEAD(&ksocknal_data.ksnd_deathrow_conns);
+       init_waitqueue_head(&ksocknal_data.ksnd_reaper_waitq);
+
+       spin_lock_init(&ksocknal_data.ksnd_connd_lock);
+       INIT_LIST_HEAD(&ksocknal_data.ksnd_connd_connreqs);
+       INIT_LIST_HEAD(&ksocknal_data.ksnd_connd_routes);
+       init_waitqueue_head(&ksocknal_data.ksnd_connd_waitq);
+
+       spin_lock_init(&ksocknal_data.ksnd_tx_lock);
+       INIT_LIST_HEAD(&ksocknal_data.ksnd_idle_noop_txs);
+
+       /* NB memset above zeros whole of ksocknal_data */
+
+       /* flag lists/ptrs/locks initialised */
+       ksocknal_data.ksnd_init = SOCKNAL_INIT_DATA;
+       try_module_get(THIS_MODULE);
+
+       ksocknal_data.ksnd_sched_info = cfs_percpt_alloc(lnet_cpt_table(),
+                                                        sizeof(*info));
+       if (ksocknal_data.ksnd_sched_info == NULL)
+               goto failed;
+
+       cfs_percpt_for_each(info, i, ksocknal_data.ksnd_sched_info) {
+               ksock_sched_t   *sched;
+               int             nthrs;
+
+               nthrs = cfs_cpt_weight(lnet_cpt_table(), i);
+               if (*ksocknal_tunables.ksnd_nscheds > 0) {
+                       nthrs = min(nthrs, *ksocknal_tunables.ksnd_nscheds);
+               } else {
+                       /* max to half of CPUs, assume another half should be
+                        * reserved for upper layer modules */
+                       nthrs = min(max(SOCKNAL_NSCHEDS, nthrs >> 1), nthrs);
+               }
+
+               info->ksi_nthreads_max = nthrs;
+               info->ksi_cpt = i;
+
+               LIBCFS_CPT_ALLOC(info->ksi_scheds, lnet_cpt_table(), i,
+                                info->ksi_nthreads_max * sizeof(*sched));
+               if (info->ksi_scheds == NULL)
+                       goto failed;
+
+               for (; nthrs > 0; nthrs--) {
+                       sched = &info->ksi_scheds[nthrs - 1];
+
+                       sched->kss_info = info;
+                       spin_lock_init(&sched->kss_lock);
+                       INIT_LIST_HEAD(&sched->kss_rx_conns);
+                       INIT_LIST_HEAD(&sched->kss_tx_conns);
+                       INIT_LIST_HEAD(&sched->kss_zombie_noop_txs);
+                       init_waitqueue_head(&sched->kss_waitq);
+               }
+        }
+
+        ksocknal_data.ksnd_connd_starting         = 0;
+        ksocknal_data.ksnd_connd_failed_stamp     = 0;
+        ksocknal_data.ksnd_connd_starting_stamp   = cfs_time_current_sec();
         /* must have at least 2 connds to remain responsive to accepts while
          * connecting */
-        if (*ksocknal_tunables.ksnd_nconnds < 2)
-                *ksocknal_tunables.ksnd_nconnds = 2;
+        if (*ksocknal_tunables.ksnd_nconnds < SOCKNAL_CONND_RESV + 1)
+                *ksocknal_tunables.ksnd_nconnds = SOCKNAL_CONND_RESV + 1;
+
+        if (*ksocknal_tunables.ksnd_nconnds_max <
+            *ksocknal_tunables.ksnd_nconnds) {
+                ksocknal_tunables.ksnd_nconnds_max =
+                        ksocknal_tunables.ksnd_nconnds;
+        }
 
         for (i = 0; i < *ksocknal_tunables.ksnd_nconnds; i++) {
-                rc = ksocknal_thread_start (ksocknal_connd, (void *)((long)i));
-                if (rc != 0) {
+               char name[16];
+               spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
+               ksocknal_data.ksnd_connd_starting++;
+               spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
+
+
+               snprintf(name, sizeof(name), "socknal_cd%02d", i);
+               rc = ksocknal_thread_start(ksocknal_connd,
+                                          (void *)((ulong_ptr_t)i), name);
+               if (rc != 0) {
+                       spin_lock_bh(&ksocknal_data.ksnd_connd_lock);
+                       ksocknal_data.ksnd_connd_starting--;
+                       spin_unlock_bh(&ksocknal_data.ksnd_connd_lock);
                         CERROR("Can't spawn socknal connd: %d\n", rc);
                         goto failed;
                 }
         }
 
-        rc = ksocknal_thread_start (ksocknal_reaper, NULL);
+       rc = ksocknal_thread_start(ksocknal_reaper, NULL, "socknal_reaper");
         if (rc != 0) {
                 CERROR ("Can't spawn socknal reaper: %d\n", rc);
                 goto failed;
@@ -2348,18 +2500,18 @@ ksocknal_base_startup (void)
         return -ENETDOWN;
 }
 
-void
+static void
 ksocknal_debug_peerhash (lnet_ni_t *ni)
 {
-        ksock_peer_t     *peer = NULL;
-        struct list_head *tmp;
-        int               i;
+       ksock_peer_t    *peer = NULL;
+       struct list_head        *tmp;
+       int             i;
 
-        read_lock (&ksocknal_data.ksnd_global_lock);
+       read_lock(&ksocknal_data.ksnd_global_lock);
 
         for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
-                list_for_each (tmp, &ksocknal_data.ksnd_peers[i]) {
-                        peer = list_entry (tmp, ksock_peer_t, ksnp_list);
+               list_for_each(tmp, &ksocknal_data.ksnd_peers[i]) {
+                       peer = list_entry(tmp, ksock_peer_t, ksnp_list);
 
                         if (peer->ksnp_ni == ni) break;
 
@@ -2371,35 +2523,35 @@ ksocknal_debug_peerhash (lnet_ni_t *ni)
                 ksock_route_t *route;
                 ksock_conn_t  *conn;
 
-                CWARN ("Active peer on shutdown: %s, ref %d, scnt %d, "
-                       "closing %d, accepting %d, err %d, zcookie "LPU64", "
-                       "txq %d, zc_req %d\n", libcfs_id2str(peer->ksnp_id),
-                       atomic_read(&peer->ksnp_refcount),
-                       peer->ksnp_sharecount, peer->ksnp_closing,
-                       peer->ksnp_accepting, peer->ksnp_error,
-                       peer->ksnp_zc_next_cookie,
-                       !list_empty(&peer->ksnp_tx_queue),
-                       !list_empty(&peer->ksnp_zc_req_list));
-
-                list_for_each (tmp, &peer->ksnp_routes) {
-                        route = list_entry(tmp, ksock_route_t, ksnr_list);
-                        CWARN ("Route: ref %d, schd %d, conn %d, cnted %d, "
-                               "del %d\n", atomic_read(&route->ksnr_refcount),
-                               route->ksnr_scheduled, route->ksnr_connecting,
-                               route->ksnr_connected, route->ksnr_deleted);
-                }
-
-                list_for_each (tmp, &peer->ksnp_conns) {
-                        conn = list_entry(tmp, ksock_conn_t, ksnc_list);
-                        CWARN ("Conn: ref %d, sref %d, t %d, c %d\n",
-                               atomic_read(&conn->ksnc_conn_refcount),
-                               atomic_read(&conn->ksnc_sock_refcount),
-                               conn->ksnc_type, conn->ksnc_closing);
-                }
-        }
-
-        read_unlock (&ksocknal_data.ksnd_global_lock);
-        return;
+               CWARN ("Active peer on shutdown: %s, ref %d, scnt %d, "
+                      "closing %d, accepting %d, err %d, zcookie "LPU64", "
+                      "txq %d, zc_req %d\n", libcfs_id2str(peer->ksnp_id),
+                      atomic_read(&peer->ksnp_refcount),
+                      peer->ksnp_sharecount, peer->ksnp_closing,
+                      peer->ksnp_accepting, peer->ksnp_error,
+                      peer->ksnp_zc_next_cookie,
+                      !list_empty(&peer->ksnp_tx_queue),
+                      !list_empty(&peer->ksnp_zc_req_list));
+
+               list_for_each(tmp, &peer->ksnp_routes) {
+                       route = list_entry(tmp, ksock_route_t, ksnr_list);
+                       CWARN ("Route: ref %d, schd %d, conn %d, cnted %d, "
+                              "del %d\n", atomic_read(&route->ksnr_refcount),
+                              route->ksnr_scheduled, route->ksnr_connecting,
+                              route->ksnr_connected, route->ksnr_deleted);
+               }
+
+               list_for_each(tmp, &peer->ksnp_conns) {
+                       conn = list_entry(tmp, ksock_conn_t, ksnc_list);
+                       CWARN ("Conn: ref %d, sref %d, t %d, c %d\n",
+                              atomic_read(&conn->ksnc_conn_refcount),
+                              atomic_read(&conn->ksnc_sock_refcount),
+                              conn->ksnc_type, conn->ksnc_closing);
+               }
+       }
+
+       read_unlock(&ksocknal_data.ksnd_global_lock);
+       return;
 }
 
 void
@@ -2407,50 +2559,54 @@ ksocknal_shutdown (lnet_ni_t *ni)
 {
         ksock_net_t      *net = ni->ni_data;
         int               i;
-        lnet_process_id_t  anyid = {.nid = LNET_NID_ANY,
-                                   .pid = LNET_PID_ANY};
+        lnet_process_id_t anyid = {0};
+
+        anyid.nid =  LNET_NID_ANY;
+        anyid.pid =  LNET_PID_ANY;
 
         LASSERT(ksocknal_data.ksnd_init == SOCKNAL_INIT_ALL);
         LASSERT(ksocknal_data.ksnd_nnets > 0);
 
-        spin_lock_bh (&net->ksnn_lock);
-        net->ksnn_shutdown = 1;                 /* prevent new peers */
-        spin_unlock_bh (&net->ksnn_lock);
+       spin_lock_bh(&net->ksnn_lock);
+       net->ksnn_shutdown = 1;                 /* prevent new peers */
+       spin_unlock_bh(&net->ksnn_lock);
 
-        /* Delete all peers */
-        ksocknal_del_peer(ni, anyid, 0);
+       /* Delete all peers */
+       ksocknal_del_peer(ni, anyid, 0);
 
-        /* Wait for all peer state to clean up */
-        i = 2;
-        spin_lock_bh (&net->ksnn_lock);
-        while (net->ksnn_npeers != 0) {
-                spin_unlock_bh (&net->ksnn_lock);
+       /* Wait for all peer state to clean up */
+       i = 2;
+       spin_lock_bh(&net->ksnn_lock);
+       while (net->ksnn_npeers != 0) {
+               spin_unlock_bh(&net->ksnn_lock);
 
-                i++;
-                CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET, /* power of 2? */
-                       "waiting for %d peers to disconnect\n",
-                       net->ksnn_npeers);
-                cfs_pause(cfs_time_seconds(1));
+               i++;
+               CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET, /* power of 2? */
+                      "waiting for %d peers to disconnect\n",
+                      net->ksnn_npeers);
+               set_current_state(TASK_UNINTERRUPTIBLE);
+               schedule_timeout(cfs_time_seconds(1));
 
-                ksocknal_debug_peerhash(ni);
+               ksocknal_debug_peerhash(ni);
 
-                spin_lock_bh (&net->ksnn_lock);
-        }
-        spin_unlock_bh (&net->ksnn_lock);
+               spin_lock_bh(&net->ksnn_lock);
+       }
+       spin_unlock_bh(&net->ksnn_lock);
 
         for (i = 0; i < net->ksnn_ninterfaces; i++) {
                 LASSERT (net->ksnn_interfaces[i].ksni_npeers == 0);
                 LASSERT (net->ksnn_interfaces[i].ksni_nroutes == 0);
         }
 
-        LIBCFS_FREE(net, sizeof(*net));
+       list_del(&net->ksnn_list);
+       LIBCFS_FREE(net, sizeof(*net));
 
         ksocknal_data.ksnd_nnets--;
         if (ksocknal_data.ksnd_nnets == 0)
                 ksocknal_base_shutdown();
 }
 
-int
+static int
 ksocknal_enumerate_interfaces(ksock_net_t *net)
 {
         char      **names;
@@ -2459,7 +2615,7 @@ ksocknal_enumerate_interfaces(ksock_net_t *net)
         int         rc;
         int         n;
 
-        n = libcfs_ipif_enumerate(&names);
+       n = lnet_ipif_enumerate(&names);
         if (n <= 0) {
                 CERROR("Can't enumerate interfaces: %d\n", n);
                 return n;
@@ -2473,7 +2629,7 @@ ksocknal_enumerate_interfaces(ksock_net_t *net)
                 if (!strcmp(names[i], "lo")) /* skip the loopback IF */
                         continue;
 
-                rc = libcfs_ipif_query(names[i], &up, &ip, &mask);
+               rc = lnet_ipif_query(names[i], &up, &ip, &mask);
                 if (rc != 0) {
                         CWARN("Can't get interface %s info: %d\n",
                               names[i], rc);
@@ -2494,10 +2650,12 @@ ksocknal_enumerate_interfaces(ksock_net_t *net)
 
                 net->ksnn_interfaces[j].ksni_ipaddr = ip;
                 net->ksnn_interfaces[j].ksni_netmask = mask;
+               strlcpy(net->ksnn_interfaces[j].ksni_name,
+                       names[i], sizeof(net->ksnn_interfaces[j].ksni_name));
                 j++;
         }
 
-        libcfs_ipif_free_enumeration(names, n);
+       lnet_ipif_free_enumeration(names, n);
 
         if (j == 0)
                 CERROR("Can't find any usable interfaces\n");
@@ -2505,6 +2663,120 @@ ksocknal_enumerate_interfaces(ksock_net_t *net)
         return j;
 }
 
+static int
+ksocknal_search_new_ipif(ksock_net_t *net)
+{
+       int     new_ipif = 0;
+       int     i;
+
+       for (i = 0; i < net->ksnn_ninterfaces; i++) {
+               char            *ifnam = &net->ksnn_interfaces[i].ksni_name[0];
+               char            *colon = strchr(ifnam, ':');
+               int             found  = 0;
+               ksock_net_t     *tmp;
+               int             j;
+
+               if (colon != NULL) /* ignore alias device */
+                       *colon = 0;
+
+               list_for_each_entry(tmp, &ksocknal_data.ksnd_nets,
+                                       ksnn_list) {
+                       for (j = 0; !found && j < tmp->ksnn_ninterfaces; j++) {
+                               char *ifnam2 = &tmp->ksnn_interfaces[j].\
+                                            ksni_name[0];
+                               char *colon2 = strchr(ifnam2, ':');
+
+                               if (colon2 != NULL)
+                                       *colon2 = 0;
+
+                               found = strcmp(ifnam, ifnam2) == 0;
+                               if (colon2 != NULL)
+                                       *colon2 = ':';
+                       }
+                       if (found)
+                               break;
+               }
+
+               new_ipif += !found;
+               if (colon != NULL)
+                       *colon = ':';
+       }
+
+       return new_ipif;
+}
+
+static int
+ksocknal_start_schedulers(struct ksock_sched_info *info)
+{
+       int     nthrs;
+       int     rc = 0;
+       int     i;
+
+       if (info->ksi_nthreads == 0) {
+               if (*ksocknal_tunables.ksnd_nscheds > 0) {
+                       nthrs = info->ksi_nthreads_max;
+               } else {
+                       nthrs = cfs_cpt_weight(lnet_cpt_table(),
+                                              info->ksi_cpt);
+                       nthrs = min(max(SOCKNAL_NSCHEDS, nthrs >> 1), nthrs);
+                       nthrs = min(SOCKNAL_NSCHEDS_HIGH, nthrs);
+               }
+               nthrs = min(nthrs, info->ksi_nthreads_max);
+       } else {
+               LASSERT(info->ksi_nthreads <= info->ksi_nthreads_max);
+               /* increase two threads if there is new interface */
+               nthrs = min(2, info->ksi_nthreads_max - info->ksi_nthreads);
+       }
+
+       for (i = 0; i < nthrs; i++) {
+               long            id;
+               char            name[20];
+               ksock_sched_t   *sched;
+               id = KSOCK_THREAD_ID(info->ksi_cpt, info->ksi_nthreads + i);
+               sched = &info->ksi_scheds[KSOCK_THREAD_SID(id)];
+               snprintf(name, sizeof(name), "socknal_sd%02d_%02d",
+                        info->ksi_cpt, (int)(sched - &info->ksi_scheds[0]));
+
+               rc = ksocknal_thread_start(ksocknal_scheduler,
+                                          (void *)id, name);
+               if (rc == 0)
+                       continue;
+
+               CERROR("Can't spawn thread %d for scheduler[%d]: %d\n",
+                      info->ksi_cpt, info->ksi_nthreads + i, rc);
+               break;
+       }
+
+       info->ksi_nthreads += i;
+       return rc;
+}
+
+static int
+ksocknal_net_start_threads(ksock_net_t *net, __u32 *cpts, int ncpts)
+{
+       int     newif = ksocknal_search_new_ipif(net);
+       int     rc;
+       int     i;
+
+       LASSERT(ncpts > 0 && ncpts <= cfs_cpt_number(lnet_cpt_table()));
+
+       for (i = 0; i < ncpts; i++) {
+               struct ksock_sched_info *info;
+               int cpt = (cpts == NULL) ? i : cpts[i];
+
+               LASSERT(cpt < cfs_cpt_number(lnet_cpt_table()));
+               info = ksocknal_data.ksnd_sched_info[cpt];
+
+               if (!newif && info->ksi_nthreads > 0)
+                       continue;
+
+               rc = ksocknal_start_schedulers(info);
+               if (rc != 0)
+                       return rc;
+       }
+       return 0;
+}
+
 int
 ksocknal_startup (lnet_ni_t *ni)
 {
@@ -2524,12 +2796,13 @@ ksocknal_startup (lnet_ni_t *ni)
         if (net == NULL)
                 goto fail_0;
 
-        memset(net, 0, sizeof(*net));
-        spin_lock_init(&net->ksnn_lock);
+       spin_lock_init(&net->ksnn_lock);
         net->ksnn_incarnation = ksocknal_new_incarnation();
         ni->ni_data = net;
-        ni->ni_maxtxcredits = *ksocknal_tunables.ksnd_credits;
-        ni->ni_peertxcredits = *ksocknal_tunables.ksnd_peercredits;
+        ni->ni_peertimeout    = *ksocknal_tunables.ksnd_peertimeout;
+        ni->ni_maxtxcredits   = *ksocknal_tunables.ksnd_credits;
+        ni->ni_peertxcredits  = *ksocknal_tunables.ksnd_peertxcredits;
+        ni->ni_peerrtrcredits = *ksocknal_tunables.ksnd_peerrtrcredits;
 
         if (ni->ni_interfaces[0] == NULL) {
                 rc = ksocknal_enumerate_interfaces(net);
@@ -2544,8 +2817,7 @@ ksocknal_startup (lnet_ni_t *ni)
                         if (ni->ni_interfaces[i] == NULL)
                                 break;
 
-                        rc = libcfs_ipif_query(
-                                ni->ni_interfaces[i], &up,
+                       rc = lnet_ipif_query(ni->ni_interfaces[i], &up,
                                 &net->ksnn_interfaces[i].ksni_ipaddr,
                                 &net->ksnn_interfaces[i].ksni_netmask);
 
@@ -2560,12 +2832,22 @@ ksocknal_startup (lnet_ni_t *ni)
                                        ni->ni_interfaces[i]);
                                 goto fail_1;
                         }
-                }
-                net->ksnn_ninterfaces = i;
-        }
 
-        ni->ni_nid = LNET_MKNID(LNET_NIDNET(ni->ni_nid),
-                                net->ksnn_interfaces[0].ksni_ipaddr);
+                       strlcpy(net->ksnn_interfaces[i].ksni_name,
+                               ni->ni_interfaces[i],
+                               sizeof(net->ksnn_interfaces[i].ksni_name));
+               }
+               net->ksnn_ninterfaces = i;
+       }
+
+       /* call it before add it to ksocknal_data.ksnd_nets */
+       rc = ksocknal_net_start_threads(net, ni->ni_cpts, ni->ni_ncpts);
+       if (rc != 0)
+               goto fail_1;
+
+       ni->ni_nid = LNET_MKNID(LNET_NIDNET(ni->ni_nid),
+                               net->ksnn_interfaces[0].ksni_ipaddr);
+       list_add(&net->ksnn_list, &ksocknal_data.ksnd_nets);
 
         ksocknal_data.ksnd_nnets++;
 
@@ -2581,32 +2863,43 @@ ksocknal_startup (lnet_ni_t *ni)
 }
 
 
-void __exit
-ksocknal_module_fini (void)
+static void __exit ksocklnd_exit(void)
 {
-        lnet_unregister_lnd(&the_ksocklnd);
-        ksocknal_lib_tunables_fini();
+       lnet_unregister_lnd(&the_ksocklnd);
 }
 
-int __init
-ksocknal_module_init (void)
+static int __init ksocklnd_init(void)
 {
-        int    rc;
-
-        /* check ksnr_connected/connecting field large enough */
-        CLASSERT(SOCKLND_CONN_NTYPES <= 4);
-
-        rc = ksocknal_lib_tunables_init();
-        if (rc != 0)
-                return rc;
-
-        lnet_register_lnd(&the_ksocklnd);
-
-        return 0;
+       int rc;
+
+       /* check ksnr_connected/connecting field large enough */
+       CLASSERT(SOCKLND_CONN_NTYPES <= 4);
+       CLASSERT(SOCKLND_CONN_ACK == SOCKLND_CONN_BULK_IN);
+
+       /* initialize the_ksocklnd */
+       the_ksocklnd.lnd_type     = SOCKLND;
+       the_ksocklnd.lnd_startup  = ksocknal_startup;
+       the_ksocklnd.lnd_shutdown = ksocknal_shutdown;
+       the_ksocklnd.lnd_ctl      = ksocknal_ctl;
+       the_ksocklnd.lnd_send     = ksocknal_send;
+       the_ksocklnd.lnd_recv     = ksocknal_recv;
+       the_ksocklnd.lnd_notify   = ksocknal_notify;
+       the_ksocklnd.lnd_query    = ksocknal_query;
+       the_ksocklnd.lnd_accept   = ksocknal_accept;
+
+       rc = ksocknal_tunables_init();
+       if (rc != 0)
+               return rc;
+
+       lnet_register_lnd(&the_ksocklnd);
+
+       return 0;
 }
 
-MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
-MODULE_DESCRIPTION("Kernel TCP Socket LND v2.0.0");
+MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
+MODULE_DESCRIPTION("TCP Socket LNet Network Driver");
+MODULE_VERSION("2.8.0");
 MODULE_LICENSE("GPL");
 
-cfs_module(ksocknal, "2.0.0", ksocknal_module_init, ksocknal_module_fini);
+module_init(ksocklnd_init);
+module_exit(ksocklnd_exit);