X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lnet%2Flnet%2Frouter.c;h=fb60c56d0fb40f56caff1bbda4effc5d8e288dd1;hp=fb7f905d06aba17adbf744b0d06ab00dd010099c;hb=2e07619477;hpb=00a2932b0aa7fe7584b5f46456d2ffaa57672716;ds=sidebyside diff --git a/lnet/lnet/router.c b/lnet/lnet/router.c index fb7f905..fb60c56 100644 --- a/lnet/lnet/router.c +++ b/lnet/lnet/router.c @@ -78,6 +78,14 @@ int avoid_asym_router_failure = 1; module_param(avoid_asym_router_failure, int, 0644); MODULE_PARM_DESC(avoid_asym_router_failure, "Avoid asymmetrical router failures (0 to disable)"); +int dead_router_check_interval = INT_MIN; +module_param(dead_router_check_interval, int, 0444); +MODULE_PARM_DESC(dead_router_check_interval, "(DEPRECATED - Use alive_router_check_interval)"); + +int live_router_check_interval = INT_MIN; +module_param(live_router_check_interval, int, 0444); +MODULE_PARM_DESC(live_router_check_interval, "(DEPRECATED - Use alive_router_check_interval)"); + int alive_router_check_interval = 60; module_param(alive_router_check_interval, int, 0644); MODULE_PARM_DESC(alive_router_check_interval, "Seconds between live router health checks (<= 0 to disable)"); @@ -110,6 +118,11 @@ module_param_call(router_sensitivity_percentage, rtr_sensitivity_set, param_get_ MODULE_PARM_DESC(router_sensitivity_percentage, "How healthy a gateway should be to be used in percent"); +static void lnet_add_route_to_rnet(struct lnet_remotenet *rnet, + struct lnet_route *route); +static void lnet_del_route_from_rnet(lnet_nid_t gw_nid, struct list_head *route_list, + struct list_head *zombies); + static int rtr_sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp) { @@ -142,30 +155,89 @@ rtr_sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp) } void +lnet_move_route(struct lnet_route *route, struct lnet_peer *lp, + struct list_head *rt_list) +{ + struct lnet_remotenet *rnet; + struct list_head zombies; + struct list_head *l; + + INIT_LIST_HEAD(&zombies); + + if (rt_list) + l = rt_list; + else + l = &zombies; + + rnet = lnet_find_rnet_locked(route->lr_net); + LASSERT(rnet); + + CDEBUG(D_NET, "deleting route %s->%s\n", + libcfs_net2str(route->lr_net), + libcfs_nid2str(route->lr_nid)); + + /* + * use the gateway's lp_primary_nid to delete the route as the + * lr_nid can be a constituent NID of the peer + */ + lnet_del_route_from_rnet(route->lr_gateway->lp_primary_nid, + &rnet->lrn_routes, l); + + if (lp) { + route = list_first_entry(l, struct lnet_route, + lr_list); + route->lr_gateway = lp; + lnet_add_route_to_rnet(rnet, route); + } else { + while (!list_empty(l) && !rt_list) { + route = list_first_entry(l, struct lnet_route, + lr_list); + list_del(&route->lr_list); + LIBCFS_FREE(route, sizeof(*route)); + } + } +} + +void lnet_rtr_transfer_to_peer(struct lnet_peer *src, struct lnet_peer *target) { struct lnet_route *route; + struct lnet_route *tmp, *tmp2; lnet_net_lock(LNET_LOCK_EX); - target->lp_rtr_refcount += src->lp_rtr_refcount; - /* move the list of queued messages to the new peer */ + CDEBUG(D_NET, "transfering routes from %s -> %s\n", + libcfs_nid2str(src->lp_primary_nid), + libcfs_nid2str(target->lp_primary_nid)); + list_for_each_entry(route, &src->lp_routes, lr_gwlist) { + CDEBUG(D_NET, "%s: %s->%s\n", libcfs_nid2str(src->lp_primary_nid), + libcfs_net2str(route->lr_net), + libcfs_nid2str(route->lr_nid)); + } list_splice_init(&src->lp_rtrq, &target->lp_rtrq); - /* move all the routes that reference the peer */ - list_splice_init(&src->lp_routes, &target->lp_routes); - /* update all the routes to point to the new peer */ - list_for_each_entry(route, &target->lp_routes, lr_gwlist) - route->lr_gateway = target; - /* remove the old peer from the ln_routers list */ - list_del_init(&src->lp_rtr_list); - /* add the new peer to the ln_routers list */ + list_for_each_entry_safe(route, tmp, &src->lp_routes, lr_gwlist) { + struct lnet_route *r2; + bool present = false; + list_for_each_entry_safe(r2, tmp2, &target->lp_routes, lr_gwlist) { + if (route->lr_net == r2->lr_net) { + if (route->lr_priority >= r2->lr_priority) + present = true; + else if (route->lr_hops >= r2->lr_hops) + present = true; + else + lnet_move_route(r2, NULL, NULL); + } + } + if (present) + lnet_move_route(route, NULL, NULL); + else + lnet_move_route(route, target, NULL); + } + if (list_empty(&target->lp_rtr_list)) { lnet_peer_addref_locked(target); list_add_tail(&target->lp_rtr_list, &the_lnet.ln_routers); } - /* reset the ref count on the old peer and decrement its ref count */ - src->lp_rtr_refcount = 0; - lnet_peer_decref_locked(src); - /* update the router version */ + the_lnet.ln_routers_version++; lnet_net_unlock(LNET_LOCK_EX); } @@ -177,7 +249,8 @@ lnet_peers_start_down(void) } /* - * A net is alive if at least one gateway NI on the network is alive. + * The peer_net of a gateway is alive if at least one of the peer_ni's on + * that peer_net is alive. */ static bool lnet_is_gateway_net_alive(struct lnet_peer_net *lpn) @@ -200,6 +273,9 @@ bool lnet_is_gateway_alive(struct lnet_peer *gw) { struct lnet_peer_net *lpn; + if (!gw->lp_alive) + return false; + list_for_each_entry(lpn, &gw->lp_peer_nets, lpn_peer_nets) { if (!lnet_is_gateway_net_alive(lpn)) return false; @@ -220,39 +296,56 @@ bool lnet_is_route_alive(struct lnet_route *route) struct lnet_peer *gw = route->lr_gateway; struct lnet_peer_net *llpn; struct lnet_peer_net *rlpn; - bool route_alive; + + /* If the gateway is down then all routes are considered down */ + if (!gw->lp_alive) + return false; /* - * check the gateway's interfaces on the route rnet to make sure - * that the gateway is viable. + * if discovery is disabled then rely on the cached aliveness + * information. This is handicapped information which we log when + * we receive the discovery ping response. The most uptodate + * aliveness information can only be obtained when discovery is + * enabled. + */ + if (lnet_is_discovery_disabled(gw)) + return route->lr_alive; + + /* + * check the gateway's interfaces on the local network */ llpn = lnet_peer_get_net_locked(gw, route->lr_lnet); if (!llpn) return false; - route_alive = lnet_is_gateway_net_alive(llpn); + if (!lnet_is_gateway_net_alive(llpn)) + return false; - if (avoid_asym_router_failure) { + /* + * For single hop routes avoid_asym_router_failure dictates + * that the remote net must exist on the gateway. For multi-hop + * routes the next-hop will not have the remote net. + */ + if (avoid_asym_router_failure && + (route->lr_hops == 1 || route->lr_hops == LNET_UNDEFINED_HOPS)) { rlpn = lnet_peer_get_net_locked(gw, route->lr_net); if (!rlpn) return false; - route_alive = route_alive && - lnet_is_gateway_net_alive(rlpn); + if (!lnet_is_gateway_net_alive(rlpn)) + return false; } - if (!route_alive) - return route_alive; - spin_lock(&gw->lp_lock); if (!(gw->lp_state & LNET_PEER_ROUTER_ENABLED)) { + spin_unlock(&gw->lp_lock); if (gw->lp_rtr_refcount > 0) CERROR("peer %s is being used as a gateway but routing feature is not turned on\n", libcfs_nid2str(gw->lp_primary_nid)); - route_alive = false; + return false; } spin_unlock(&gw->lp_lock); - return route_alive; + return true; } void @@ -280,38 +373,189 @@ lnet_consolidate_routes_locked(struct lnet_peer *orig_lp, lnet_net_unlock(LNET_LOCK_EX); } } +} + +static inline void +lnet_check_route_inconsistency(struct lnet_route *route) +{ + if (!route->lr_single_hop && + (route->lr_hops == 1 || route->lr_hops == LNET_UNDEFINED_HOPS)) { + CWARN("route %s->%s is detected to be multi-hop but hop count is set to %d\n", + libcfs_net2str(route->lr_net), + libcfs_nid2str(route->lr_gateway->lp_primary_nid), + (int) route->lr_hops); + } +} + +static void +lnet_set_route_hop_type(struct lnet_peer *gw, struct lnet_route *route) +{ + struct lnet_peer_net *lpn; + bool single_hop = false; + + list_for_each_entry(lpn, &gw->lp_peer_nets, lpn_peer_nets) { + if (route->lr_net == lpn->lpn_net_id) { + single_hop = true; + break; + } + } + route->lr_single_hop = single_hop; + lnet_check_route_inconsistency(route); +} + +/* Must hold net_lock/EX */ +static inline void +lnet_set_route_aliveness(struct lnet_route *route, bool alive) +{ + /* Log when there's a state change */ + if (route->lr_alive != alive) { + CERROR("route to %s through %s has gone from %s to %s\n", + libcfs_net2str(route->lr_net), + libcfs_nid2str(route->lr_gateway->lp_primary_nid), + (route->lr_alive) ? "up" : "down", + alive ? "up" : "down"); + route->lr_alive = alive; + } +} + +/* Must hold net_lock/EX */ +void +lnet_router_discovery_ping_reply(struct lnet_peer *lp) +{ + struct lnet_ping_buffer *pbuf = lp->lp_data; + struct lnet_peer_net *llpn; + struct lnet_route *route; + bool single_hop = false; + bool net_up = false; + unsigned lp_state; + __u32 net; + int i; + + + spin_lock(&lp->lp_lock); + lp_state = lp->lp_state; + + /* only handle replies if discovery is disabled. */ + if (!lnet_is_discovery_disabled_locked(lp)) { + spin_unlock(&lp->lp_lock); + return; + } + + spin_unlock(&lp->lp_lock); + + if (lp_state & LNET_PEER_PING_FAILED || + pbuf->pb_info.pi_features & LNET_PING_FEAT_RTE_DISABLED) { + CDEBUG(D_NET, "Set routes down for gw %s because %s %d\n", + libcfs_nid2str(lp->lp_primary_nid), + lp_state & LNET_PEER_PING_FAILED ? "ping failed" : + "route feature is disabled", lp->lp_ping_error); + /* If the ping failed or the peer has routing disabled then + * mark the routes served by this peer down + */ + list_for_each_entry(route, &lp->lp_routes, lr_gwlist) + lnet_set_route_aliveness(route, false); + return; + } + + CDEBUG(D_NET, "Discovery is disabled. Processing reply for gw: %s:%d\n", + libcfs_nid2str(lp->lp_primary_nid), pbuf->pb_info.pi_nnis); + + /* + * examine the ping response to determine if the routes on that + * gateway should be declared alive. + * The route is alive if: + * 1. local network to reach the route is alive and + * 2. route is single hop, avoid_async_router_failure is set and + * there exists at least one NI on the route's remote net + */ + list_for_each_entry(route, &lp->lp_routes, lr_gwlist) { + llpn = lnet_peer_get_net_locked(lp, route->lr_lnet); + if (!llpn) { + lnet_set_route_aliveness(route, false); + continue; + } + + if (!lnet_is_gateway_net_alive(llpn)) { + lnet_set_route_aliveness(route, false); + continue; + } + + single_hop = net_up = false; + for (i = 1; i < pbuf->pb_info.pi_nnis; i++) { + net = LNET_NIDNET(pbuf->pb_info.pi_ni[i].ns_nid); + + if (route->lr_net == net) { + single_hop = true; + if (pbuf->pb_info.pi_ni[i].ns_status == + LNET_NI_STATUS_UP) { + net_up = true; + break; + } + } + } + + route->lr_single_hop = single_hop; + if (avoid_asym_router_failure && + (route->lr_hops == 1 || + route->lr_hops == LNET_UNDEFINED_HOPS)) + lnet_set_route_aliveness(route, net_up); + else + lnet_set_route_aliveness(route, true); + /* + * warn that the route is configured as single-hop but it + * really is multi-hop as far as we can tell. + */ + lnet_check_route_inconsistency(route); + } } void lnet_router_discovery_complete(struct lnet_peer *lp) { struct lnet_peer_ni *lpni = NULL; + struct lnet_route *route; spin_lock(&lp->lp_lock); lp->lp_state &= ~LNET_PEER_RTR_DISCOVERY; + lp->lp_state |= LNET_PEER_RTR_DISCOVERED; + lp->lp_alive = lp->lp_dc_error == 0; spin_unlock(&lp->lp_lock); - /* - * Router discovery successful? All peer information would've been - * updated already. No need to do any more processing - */ - if (!lp->lp_dc_error) + /* ping replies are being handled when discovery is disabled */ + if (lnet_is_discovery_disabled_locked(lp)) return; + + if (!lp->lp_dc_error) { + /* + * mark single-hop routes. If the remote net is not configured on + * the gateway we assume this is intentional and we mark the + * gateway as multi-hop + */ + list_for_each_entry(route, &lp->lp_routes, lr_gwlist) { + lnet_set_route_aliveness(route, true); + lnet_set_route_hop_type(lp, route); + } + + return; + } + /* - * discovery failed? then we need to set the status of each lpni - * to DOWN. It will be updated the next time we discover the - * router. For router peer NIs not on local networks, we never send - * messages directly to them, so their health will always remain - * at maximum. We can only tell if they are up or down from the - * status returned in the PING response. If we fail to get that - * status in our scheduled router discovery, then we'll assume - * it's down until we're told otherwise. + * We do not send messages directly to the remote interfaces + * of an LNet router. As such, we rely on the PING response + * to determine the up/down status of these interfaces. If + * a PING response is not receieved, or some other problem with + * discovery occurs that prevents us from getting this status, + * we assume all interfaces are down until we're able to + * determine otherwise. */ CDEBUG(D_NET, "%s: Router discovery failed %d\n", libcfs_nid2str(lp->lp_primary_nid), lp->lp_dc_error); while ((lpni = lnet_get_next_peer_ni_locked(lp, NULL, lpni)) != NULL) lpni->lpni_ns_status = LNET_NI_STATUS_DOWN; + + list_for_each_entry(route, &lp->lp_routes, lr_gwlist) + lnet_set_route_aliveness(route, false); } static void @@ -380,15 +624,15 @@ static void lnet_shuffle_seed(void) add_device_randomness(&ni->ni_nid, sizeof(ni->ni_nid)); seeded = 1; - return; } /* NB expects LNET_LOCK held */ static void lnet_add_route_to_rnet(struct lnet_remotenet *rnet, struct lnet_route *route) { - unsigned int len = 0; + struct lnet_peer_net *lpn; unsigned int offset = 0; + unsigned int len = 0; struct list_head *e; lnet_shuffle_seed(); @@ -401,7 +645,7 @@ lnet_add_route_to_rnet(struct lnet_remotenet *rnet, struct lnet_route *route) * different nodes are using the same list of routers, they end up * preferring different routers. */ - offset = cfs_rand() % (len + 1); + offset = prandom_u32_max(len + 1); list_for_each(e, &rnet->lrn_routes) { if (offset == 0) break; @@ -412,7 +656,10 @@ lnet_add_route_to_rnet(struct lnet_remotenet *rnet, struct lnet_route *route) * force a router check on the gateway to make sure the route is * alive */ - route->lr_gateway->lp_rtrcheck_timestamp = 0; + list_for_each_entry(lpn, &route->lr_gateway->lp_peer_nets, + lpn_peer_nets) { + lpn->lpn_rtrcheck_timestamp = 0; + } the_lnet.ln_remote_nets_version++; @@ -440,8 +687,8 @@ lnet_add_route(__u32 net, __u32 hops, lnet_nid_t gateway, libcfs_net2str(net), hops, priority, libcfs_nid2str(gateway)); if (gateway == LNET_NID_ANY || - LNET_NETTYP(LNET_NIDNET(gateway)) == LOLND || - net == LNET_NIDNET(LNET_NID_ANY) || + gateway == LNET_NID_LO_0 || + net == LNET_NET_ANY || LNET_NETTYP(net) == LOLND || LNET_NIDNET(gateway) == net || (hops != LNET_UNDEFINED_HOPS && (hops < 1 || hops > 255))) @@ -451,6 +698,13 @@ lnet_add_route(__u32 net, __u32 hops, lnet_nid_t gateway, if (lnet_islocalnet(net)) return -EEXIST; + if (!lnet_islocalnet(LNET_NIDNET(gateway))) { + CERROR("Cannot add route with gateway %s. There is no local interface configured on LNet %s\n", + libcfs_nid2str(gateway), + libcfs_net2str(LNET_NIDNET(gateway))); + return -EHOSTUNREACH; + } + /* Assume net, route, all new */ LIBCFS_ALLOC(route, sizeof(*route)); LIBCFS_ALLOC(rnet, sizeof(*rnet)); @@ -531,6 +785,9 @@ lnet_add_route(__u32 net, __u32 hops, lnet_nid_t gateway, if (add_route) { gw->lp_health_sensitivity = sensitivity; lnet_add_route_to_rnet(rnet2, route); + if (lnet_peer_discovery_disabled) + CWARN("Consider turning discovery on to enable full " + "Multi-Rail routing functionality\n"); } /* @@ -550,12 +807,12 @@ lnet_add_route(__u32 net, __u32 hops, lnet_nid_t gateway, LIBCFS_FREE(rnet, sizeof(*rnet)); /* kick start the monitor thread to handle the added route */ - wake_up(&the_lnet.ln_mt_waitq); + complete(&the_lnet.ln_mt_wait_complete); return rc; } -static void +void lnet_del_route_from_rnet(lnet_nid_t gw_nid, struct list_head *route_list, struct list_head *zombies) { @@ -588,19 +845,16 @@ lnet_del_route_from_rnet(lnet_nid_t gw_nid, struct list_head *route_list, int lnet_del_route(__u32 net, lnet_nid_t gw_nid) { - struct list_head rnet_zombies; + LIST_HEAD(rnet_zombies); struct lnet_remotenet *rnet; struct lnet_remotenet *tmp; struct list_head *rn_list; struct lnet_peer_ni *lpni; struct lnet_route *route; - struct list_head zombies; - struct lnet_peer *lp; + LIST_HEAD(zombies); + struct lnet_peer *lp = NULL; int i = 0; - INIT_LIST_HEAD(&rnet_zombies); - INIT_LIST_HEAD(&zombies); - CDEBUG(D_NET, "Del route: net %s : gw %s\n", libcfs_net2str(net), libcfs_nid2str(gw_nid)); @@ -617,7 +871,7 @@ lnet_del_route(__u32 net, lnet_nid_t gw_nid) lnet_peer_ni_decref_locked(lpni); } - if (net != LNET_NIDNET(LNET_NID_ANY)) { + if (net != LNET_NET_ANY) { rnet = lnet_find_rnet_locked(net); if (!rnet) { lnet_net_unlock(LNET_LOCK_EX); @@ -642,6 +896,18 @@ lnet_del_route(__u32 net, lnet_nid_t gw_nid) } delete_zombies: + /* + * check if there are any routes remaining on the gateway + * If there are no more routes make sure to set the peer's + * lp_disc_net_id to 0 (invalid), in case we add more routes in + * the future on that gateway, then we start our discovery process + * from scratch + */ + if (lpni) { + if (list_empty(&lp->lp_routes)) + lp->lp_disc_net_id = 0; + } + lnet_net_unlock(LNET_LOCK_EX); while (!list_empty(&zombies)) { @@ -663,7 +929,7 @@ delete_zombies: void lnet_destroy_routes (void) { - lnet_del_route(LNET_NIDNET(LNET_NID_ANY), LNET_NID_ANY); + lnet_del_route(LNET_NET_ANY, LNET_NID_ANY); } int lnet_get_rtr_pool_cfg(int cpt, struct lnet_ioctl_pool_cfg *pool_cfg) @@ -700,7 +966,7 @@ int lnet_get_rtr_pool_cfg(int cpt, struct lnet_ioctl_pool_cfg *pool_cfg) int lnet_get_route(int idx, __u32 *net, __u32 *hops, - lnet_nid_t *gateway, __u32 *alive, __u32 *priority, __u32 *sensitivity) + lnet_nid_t *gateway, __u32 *flags, __u32 *priority, __u32 *sensitivity) { struct lnet_remotenet *rnet; struct list_head *rn_list; @@ -728,7 +994,14 @@ lnet_get_route(int idx, __u32 *net, __u32 *hops, *priority = route->lr_priority; *sensitivity = route->lr_gateway-> lp_health_sensitivity; - *alive = lnet_is_route_alive(route); + if (lnet_is_route_alive(route)) + *flags |= LNET_RT_ALIVE; + else + *flags &= ~LNET_RT_ALIVE; + if (route->lr_single_hop) + *flags &= ~LNET_RT_MULTI_HOP; + else + *flags |= LNET_RT_MULTI_HOP; lnet_net_unlock(cpt); return 0; } @@ -759,7 +1032,7 @@ lnet_wait_known_routerstate(void) spin_lock(&rtr->lp_lock); - if ((rtr->lp_state & LNET_PEER_DISCOVERED) == 0) { + if ((rtr->lp_state & LNET_PEER_RTR_DISCOVERED) == 0) { all_known = 0; spin_unlock(&rtr->lp_lock); break; @@ -772,15 +1045,29 @@ lnet_wait_known_routerstate(void) if (all_known) return; - set_current_state(TASK_UNINTERRUPTIBLE); - schedule_timeout(cfs_time_seconds(1)); + schedule_timeout_uninterruptible(cfs_time_seconds(1)); } } -static void +static inline bool +lnet_net_set_status_locked(struct lnet_net *net, __u32 status) +{ + struct lnet_ni *ni; + bool update = false; + + list_for_each_entry(ni, &net->net_ni_list, ni_netlist) + if (lnet_ni_set_status(ni, status)) + update = true; + + return update; +} + +static bool lnet_update_ni_status_locked(void) { - struct lnet_ni *ni = NULL; + struct lnet_net *net; + struct lnet_ni *ni; + bool push = false; time64_t now; time64_t timeout; @@ -789,31 +1076,48 @@ lnet_update_ni_status_locked(void) timeout = router_ping_timeout + alive_router_check_interval; now = ktime_get_real_seconds(); - while ((ni = lnet_get_next_ni_locked(NULL, ni))) { - if (ni->ni_net->net_lnd->lnd_type == LOLND) + list_for_each_entry(net, &the_lnet.ln_nets, net_list) { + if (net->net_lnd->lnd_type == LOLND) continue; - if (now < ni->ni_last_alive + timeout) - continue; + if (now < net->net_last_alive + timeout) + goto check_ni_fatal; - lnet_ni_lock(ni); + spin_lock(&net->net_lock); /* re-check with lock */ - if (now < ni->ni_last_alive + timeout) { - lnet_ni_unlock(ni); - continue; + if (now < net->net_last_alive + timeout) { + spin_unlock(&net->net_lock); + goto check_ni_fatal; } + spin_unlock(&net->net_lock); - LASSERT(ni->ni_status != NULL); + /* + * if the net didn't receive any traffic for past the + * timeout on any of its constituent NIs, then mark all + * the NIs down. + */ + if (lnet_net_set_status_locked(net, LNET_NI_STATUS_DOWN)) { + push = true; + continue; + } - if (ni->ni_status->ns_status != LNET_NI_STATUS_DOWN) { - CDEBUG(D_NET, "NI(%s:%lld) status changed to down\n", - libcfs_nid2str(ni->ni_nid), timeout); - /* NB: so far, this is the only place to set - * NI status to "down" */ - ni->ni_status->ns_status = LNET_NI_STATUS_DOWN; +check_ni_fatal: + list_for_each_entry(ni, &net->net_ni_list, ni_netlist) { + /* lnet_ni_set_status() will perform the same check of + * ni_status while holding the ni lock. We can safely + * check ni_status without that lock because it is only + * written to under net_lock/EX and our caller is + * holding a net lock. + */ + if (atomic_read(&ni->ni_fatal_error_on) && + ni->ni_status && + ni->ni_status->ns_status != LNET_NI_STATUS_DOWN && + lnet_ni_set_status(ni, LNET_NI_STATUS_DOWN)) + push = true; } - lnet_ni_unlock(ni); } + + return push; } void lnet_wait_router_start(void) @@ -830,8 +1134,7 @@ void lnet_wait_router_start(void) * This function is called from the monitor thread to check if there are * any active routers that need to be checked. */ -inline bool -lnet_router_checker_active(void) +bool lnet_router_checker_active(void) { /* Router Checker thread needs to run when routing is enabled in * order to call lnet_update_ni_status_locked() */ @@ -845,10 +1148,15 @@ lnet_router_checker_active(void) void lnet_check_routers(void) { + struct lnet_peer_net *first_lpn = NULL; + struct lnet_peer_net *lpn; struct lnet_peer_ni *lpni; struct list_head *entry; struct lnet_peer *rtr; + bool push = false; + bool found_lpn; __u64 version; + __u32 net_id; time64_t now; int cpt; int rc; @@ -869,8 +1177,31 @@ rescan: * interfaces could be down and in that case they would be * undergoing recovery separately from this discovery. */ - if (now - rtr->lp_rtrcheck_timestamp < - alive_router_check_interval) + /* find next peer net which is also local */ + net_id = rtr->lp_disc_net_id; + do { + lpn = lnet_get_next_peer_net_locked(rtr, net_id); + if (!lpn) { + CERROR("gateway %s has no networks\n", + libcfs_nid2str(rtr->lp_primary_nid)); + break; + } + if (first_lpn == lpn) + break; + if (!first_lpn) + first_lpn = lpn; + found_lpn = lnet_islocalnet_locked(lpn->lpn_net_id); + net_id = lpn->lpn_net_id; + } while (!found_lpn); + + if (!found_lpn || !lpn) { + CERROR("no local network found for gateway %s\n", + libcfs_nid2str(rtr->lp_primary_nid)); + continue; + } + + if (now - lpn->lpn_rtrcheck_timestamp < + alive_router_check_interval / lnet_current_net_count) continue; /* @@ -882,9 +1213,10 @@ rescan: spin_unlock(&rtr->lp_lock); continue; } - /* make sure we actively discover the router */ + /* make sure we fully discover the router */ rtr->lp_state &= ~LNET_PEER_NIDS_UPTODATE; - rtr->lp_state |= LNET_PEER_RTR_DISCOVERY; + rtr->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH | + LNET_PEER_RTR_DISCOVERY; spin_unlock(&rtr->lp_lock); /* find the peer_ni associated with the primary NID */ @@ -896,6 +1228,9 @@ rescan: } lnet_peer_ni_addref_locked(lpni); + /* specify the net to use */ + rtr->lp_disc_net_id = lpn->lpn_net_id; + /* discover the router */ CDEBUG(D_NET, "discover %s, cpt = %d\n", libcfs_nid2str(lpni->lpni_nid), cpt); @@ -905,7 +1240,7 @@ rescan: lnet_peer_ni_decref_locked(lpni); if (!rc) - rtr->lp_rtrcheck_timestamp = now; + lpn->lpn_rtrcheck_timestamp = now; else CERROR("Failed to discover router %s\n", libcfs_nid2str(rtr->lp_primary_nid)); @@ -918,9 +1253,13 @@ rescan: } if (the_lnet.ln_routing) - lnet_update_ni_status_locked(); + push = lnet_update_ni_status_locked(); lnet_net_unlock(cpt); + + /* if the status of the ni changed update the peers */ + if (push) + lnet_push_update_to_peers(1); } void @@ -929,7 +1268,7 @@ lnet_destroy_rtrbuf(struct lnet_rtrbuf *rb, int npages) int sz = offsetof(struct lnet_rtrbuf, rb_kiov[npages]); while (--npages >= 0) - __free_page(rb->rb_kiov[npages].kiov_page); + __free_page(rb->rb_kiov[npages].bv_page); LIBCFS_FREE(rb, sz); } @@ -954,15 +1293,15 @@ lnet_new_rtrbuf(struct lnet_rtrbufpool *rbp, int cpt) GFP_KERNEL | __GFP_ZERO); if (page == NULL) { while (--i >= 0) - __free_page(rb->rb_kiov[i].kiov_page); + __free_page(rb->rb_kiov[i].bv_page); LIBCFS_FREE(rb, sz); return NULL; } - rb->rb_kiov[i].kiov_len = PAGE_SIZE; - rb->rb_kiov[i].kiov_offset = 0; - rb->rb_kiov[i].kiov_page = page; + rb->rb_kiov[i].bv_len = PAGE_SIZE; + rb->rb_kiov[i].bv_offset = 0; + rb->rb_kiov[i].bv_page = page; } return rb; @@ -973,13 +1312,11 @@ lnet_rtrpool_free_bufs(struct lnet_rtrbufpool *rbp, int cpt) { int npages = rbp->rbp_npages; struct lnet_rtrbuf *rb; - struct list_head tmp; + LIST_HEAD(tmp); if (rbp->rbp_nbuffers == 0) /* not initialized or already freed */ return; - INIT_LIST_HEAD(&tmp); - lnet_net_lock(cpt); list_splice_init(&rbp->rbp_msgs, &tmp); lnet_drop_routed_msgs_locked(&tmp, cpt); @@ -1000,7 +1337,7 @@ lnet_rtrpool_free_bufs(struct lnet_rtrbufpool *rbp, int cpt) static int lnet_rtrpool_adjust_bufs(struct lnet_rtrbufpool *rbp, int nbufs, int cpt) { - struct list_head rb_list; + LIST_HEAD(rb_list); struct lnet_rtrbuf *rb; int num_rb; int num_buffers = 0; @@ -1028,8 +1365,6 @@ lnet_rtrpool_adjust_bufs(struct lnet_rtrbufpool *rbp, int nbufs, int cpt) rbp->rbp_req_nbuffers = nbufs; lnet_net_unlock(cpt); - INIT_LIST_HEAD(&rb_list); - /* allocate the buffers on a local list first. If all buffers are * allocated successfully then join this list to the rbp buffer * list. If not then free all allocated buffers. */ @@ -1236,7 +1571,7 @@ lnet_rtrpools_alloc(int im_a_router) lnet_net_lock(LNET_LOCK_EX); the_lnet.ln_routing = 1; lnet_net_unlock(LNET_LOCK_EX); - wake_up(&the_lnet.ln_mt_waitq); + complete(&the_lnet.ln_mt_wait_complete); return 0; failed: @@ -1328,6 +1663,10 @@ lnet_rtrpools_enable(void) ~LNET_PING_FEAT_RTE_DISABLED; lnet_net_unlock(LNET_LOCK_EX); + if (lnet_peer_discovery_disabled) + CWARN("Consider turning discovery on to enable full " + "Multi-Rail routing functionality\n"); + return rc; } @@ -1368,15 +1707,16 @@ lnet_notify(struct lnet_ni *ni, lnet_nid_t nid, bool alive, bool reset, time64_t when) { struct lnet_peer_ni *lpni = NULL; + struct lnet_route *route; + struct lnet_peer *lp; time64_t now = ktime_get_seconds(); int cpt; - LASSERT (!in_interrupt ()); + LASSERT(!in_interrupt()); - CDEBUG (D_NET, "%s notifying %s: %s\n", - (ni == NULL) ? "userspace" : libcfs_nid2str(ni->ni_nid), - libcfs_nid2str(nid), - alive ? "up" : "down"); + CDEBUG(D_NET, "%s notifying %s: %s\n", + (ni == NULL) ? "userspace" : libcfs_nid2str(ni->ni_nid), + libcfs_nid2str(nid), alive ? "up" : "down"); if (ni != NULL && LNET_NIDNET(ni->ni_nid) != LNET_NIDNET(nid)) { @@ -1418,17 +1758,55 @@ lnet_notify(struct lnet_ni *ni, lnet_nid_t nid, bool alive, bool reset, } if (alive) { - if (reset) - lnet_set_healthv(&lpni->lpni_healthv, - LNET_MAX_HEALTH_VALUE); - else - lnet_inc_healthv(&lpni->lpni_healthv); - } else { - lnet_handle_remote_failure_locked(lpni); + if (reset) { + lpni->lpni_ns_status = LNET_NI_STATUS_UP; + lnet_set_lpni_healthv_locked(lpni, + LNET_MAX_HEALTH_VALUE); + } else { + __u32 sensitivity = lpni->lpni_peer_net-> + lpn_peer->lp_health_sensitivity; + + lnet_inc_lpni_healthv_locked(lpni, + (sensitivity) ? sensitivity : + lnet_health_sensitivity); + } + } else if (reset) { + lpni->lpni_ns_status = LNET_NI_STATUS_DOWN; } /* recalculate aliveness */ alive = lnet_is_peer_ni_alive(lpni); + + lp = lpni->lpni_peer_net->lpn_peer; + /* If this is an LNet router then update route aliveness */ + if (lp->lp_rtr_refcount) { + if (reset) + /* reset flag indicates gateway peer went up or down */ + lp->lp_alive = alive; + + /* If discovery is disabled, locally or on the gateway, then + * any routes using lpni as next-hop need to be updated + * + * NB: We can get many notifications while a route is down, so + * we try and avoid the expensive net_lock/EX here for the + * common case of receiving duplicate lnet_notify() calls (i.e. + * only grab EX lock when we actually need to update the route + * aliveness). + */ + if (lnet_is_discovery_disabled(lp)) { + list_for_each_entry(route, &lp->lp_routes, lr_gwlist) { + if (route->lr_nid == lpni->lpni_nid && + route->lr_alive != alive) { + lnet_net_unlock(0); + lnet_net_lock(LNET_LOCK_EX); + lnet_set_route_aliveness(route, alive); + lnet_net_unlock(LNET_LOCK_EX); + lnet_net_lock(0); + } + } + } + } + lnet_net_unlock(0); if (ni != NULL && !alive)