2 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
4 * Copyright (c) 2011, 2017, Intel Corporation.
6 * This file is part of Lustre, https://wiki.whamcloud.com/
8 * Portals is free software; you can redistribute it and/or
9 * modify it under the terms of version 2 of the GNU General Public
10 * License as published by the Free Software Foundation.
12 * Portals is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with Portals; if not, write to the Free Software
19 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 #define DEBUG_SUBSYSTEM S_LNET
25 #include <linux/random.h>
26 #include <lnet/lib-lnet.h>
28 #define LNET_NRB_TINY_MIN 512 /* min value for each CPT */
29 #define LNET_NRB_TINY (LNET_NRB_TINY_MIN * 4)
30 #define LNET_NRB_SMALL_MIN 4096 /* min value for each CPT */
31 #define LNET_NRB_SMALL (LNET_NRB_SMALL_MIN * 4)
32 #define LNET_NRB_SMALL_PAGES 1
33 #define LNET_NRB_LARGE_MIN 256 /* min value for each CPT */
34 #define LNET_NRB_LARGE (LNET_NRB_LARGE_MIN * 4)
35 #define LNET_NRB_LARGE_PAGES ((LNET_MTU + PAGE_SIZE - 1) >> \
38 static char *forwarding = "";
39 module_param(forwarding, charp, 0444);
40 MODULE_PARM_DESC(forwarding, "Explicitly enable/disable forwarding between networks");
42 static int tiny_router_buffers;
43 module_param(tiny_router_buffers, int, 0444);
44 MODULE_PARM_DESC(tiny_router_buffers, "# of 0 payload messages to buffer in the router");
45 static int small_router_buffers;
46 module_param(small_router_buffers, int, 0444);
47 MODULE_PARM_DESC(small_router_buffers, "# of small (1 page) messages to buffer in the router");
48 static int large_router_buffers;
49 module_param(large_router_buffers, int, 0444);
50 MODULE_PARM_DESC(large_router_buffers, "# of large messages to buffer in the router");
51 static int peer_buffer_credits;
52 module_param(peer_buffer_credits, int, 0444);
53 MODULE_PARM_DESC(peer_buffer_credits, "# router buffer credits per peer");
55 static int auto_down = 1;
56 module_param(auto_down, int, 0444);
57 MODULE_PARM_DESC(auto_down, "Automatically mark peers down on comms error");
60 lnet_peer_buffer_credits(struct lnet_net *net)
62 /* NI option overrides LNet default */
63 if (net->net_tunables.lct_peer_rtr_credits > 0)
64 return net->net_tunables.lct_peer_rtr_credits;
65 if (peer_buffer_credits > 0)
66 return peer_buffer_credits;
68 /* As an approximation, allow this peer the same number of router
69 * buffers as it is allowed outstanding sends */
70 return net->net_tunables.lct_peer_tx_credits;
73 static int check_routers_before_use;
74 module_param(check_routers_before_use, int, 0444);
75 MODULE_PARM_DESC(check_routers_before_use, "Assume routers are down and ping them before use");
77 int avoid_asym_router_failure = 1;
78 module_param(avoid_asym_router_failure, int, 0644);
79 MODULE_PARM_DESC(avoid_asym_router_failure, "Avoid asymmetrical router failures (0 to disable)");
81 int dead_router_check_interval = INT_MIN;
82 module_param(dead_router_check_interval, int, 0444);
83 MODULE_PARM_DESC(dead_router_check_interval, "(DEPRECATED - Use alive_router_check_interval)");
85 int live_router_check_interval = INT_MIN;
86 module_param(live_router_check_interval, int, 0444);
87 MODULE_PARM_DESC(live_router_check_interval, "(DEPRECATED - Use alive_router_check_interval)");
89 int alive_router_check_interval = 60;
90 module_param(alive_router_check_interval, int, 0644);
91 MODULE_PARM_DESC(alive_router_check_interval, "Seconds between live router health checks (<= 0 to disable)");
93 static int router_ping_timeout = 50;
94 module_param(router_ping_timeout, int, 0644);
95 MODULE_PARM_DESC(router_ping_timeout, "Seconds to wait for the reply to a router health query");
98 * A value between 0 and 100. 0 meaning that even if router's interfaces
99 * have the worse health still consider the gateway usable.
100 * 100 means that at least one interface on the route's remote net is 100%
101 * healthy to consider the route alive.
102 * The default is set to 100 to ensure we maintain the original behavior.
104 unsigned int router_sensitivity_percentage = 100;
105 static int rtr_sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp);
106 static struct kernel_param_ops param_ops_rtr_sensitivity = {
107 .set = rtr_sensitivity_set,
108 .get = param_get_int,
110 #define param_check_rtr_sensitivity(name, p) \
111 __param_check(name, p, int)
112 #ifdef HAVE_KERNEL_PARAM_OPS
113 module_param(router_sensitivity_percentage, rtr_sensitivity, S_IRUGO|S_IWUSR);
115 module_param_call(router_sensitivity_percentage, rtr_sensitivity_set, param_get_int,
116 &router_sensitivity_percentage, S_IRUGO|S_IWUSR);
118 MODULE_PARM_DESC(router_sensitivity_percentage,
119 "How healthy a gateway should be to be used in percent");
121 static void lnet_add_route_to_rnet(struct lnet_remotenet *rnet,
122 struct lnet_route *route);
123 static void lnet_del_route_from_rnet(lnet_nid_t gw_nid, struct list_head *route_list,
124 struct list_head *zombies);
127 rtr_sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
130 unsigned *sen = (unsigned *)kp->arg;
133 rc = kstrtoul(val, 0, &value);
135 CERROR("Invalid module parameter value for 'router_sensitivity_percentage'\n");
139 if (value < 0 || value > 100) {
140 CERROR("Invalid value: %lu for 'router_sensitivity_percentage'\n", value);
145 * The purpose of locking the api_mutex here is to ensure that
146 * the correct value ends up stored properly.
148 mutex_lock(&the_lnet.ln_api_mutex);
152 mutex_unlock(&the_lnet.ln_api_mutex);
158 lnet_move_route(struct lnet_route *route, struct lnet_peer *lp,
159 struct list_head *rt_list)
161 struct lnet_remotenet *rnet;
162 struct list_head zombies;
165 INIT_LIST_HEAD(&zombies);
172 rnet = lnet_find_rnet_locked(route->lr_net);
175 CDEBUG(D_NET, "deleting route %s->%s\n",
176 libcfs_net2str(route->lr_net),
177 libcfs_nid2str(route->lr_nid));
180 * use the gateway's lp_primary_nid to delete the route as the
181 * lr_nid can be a constituent NID of the peer
183 lnet_del_route_from_rnet(route->lr_gateway->lp_primary_nid,
184 &rnet->lrn_routes, l);
187 route = list_first_entry(l, struct lnet_route,
189 route->lr_gateway = lp;
190 lnet_add_route_to_rnet(rnet, route);
192 while (!list_empty(l) && !rt_list) {
193 route = list_first_entry(l, struct lnet_route,
195 list_del(&route->lr_list);
196 LIBCFS_FREE(route, sizeof(*route));
202 lnet_rtr_transfer_to_peer(struct lnet_peer *src, struct lnet_peer *target)
204 struct lnet_route *route;
205 struct lnet_route *tmp, *tmp2;
207 lnet_net_lock(LNET_LOCK_EX);
208 CDEBUG(D_NET, "transfering routes from %s -> %s\n",
209 libcfs_nid2str(src->lp_primary_nid),
210 libcfs_nid2str(target->lp_primary_nid));
211 list_for_each_entry(route, &src->lp_routes, lr_gwlist) {
212 CDEBUG(D_NET, "%s: %s->%s\n", libcfs_nid2str(src->lp_primary_nid),
213 libcfs_net2str(route->lr_net),
214 libcfs_nid2str(route->lr_nid));
216 list_splice_init(&src->lp_rtrq, &target->lp_rtrq);
217 list_for_each_entry_safe(route, tmp, &src->lp_routes, lr_gwlist) {
218 struct lnet_route *r2;
219 bool present = false;
220 list_for_each_entry_safe(r2, tmp2, &target->lp_routes, lr_gwlist) {
221 if (route->lr_net == r2->lr_net) {
222 if (route->lr_priority >= r2->lr_priority)
224 else if (route->lr_hops >= r2->lr_hops)
227 lnet_move_route(r2, NULL, NULL);
231 lnet_move_route(route, NULL, NULL);
233 lnet_move_route(route, target, NULL);
236 if (list_empty(&target->lp_rtr_list)) {
237 lnet_peer_addref_locked(target);
238 list_add_tail(&target->lp_rtr_list, &the_lnet.ln_routers);
241 the_lnet.ln_routers_version++;
242 lnet_net_unlock(LNET_LOCK_EX);
246 lnet_peers_start_down(void)
248 return check_routers_before_use;
252 * The peer_net of a gateway is alive if at least one of the peer_ni's on
253 * that peer_net is alive.
256 lnet_is_gateway_net_alive(struct lnet_peer_net *lpn)
258 struct lnet_peer_ni *lpni;
260 list_for_each_entry(lpni, &lpn->lpn_peer_nis, lpni_peer_nis) {
261 if (lnet_is_peer_ni_alive(lpni))
269 * a gateway is alive only if all its nets are alive
270 * called with cpt lock held
272 bool lnet_is_gateway_alive(struct lnet_peer *gw)
274 struct lnet_peer_net *lpn;
279 list_for_each_entry(lpn, &gw->lp_peer_nets, lpn_peer_nets) {
280 if (!lnet_is_gateway_net_alive(lpn))
288 * lnet_is_route_alive() needs to be called with cpt lock held
289 * A route is alive if the gateway can route between the local network and
290 * the remote network of the route.
291 * This means at least one NI is alive on each of the local and remote
292 * networks of the gateway.
294 bool lnet_is_route_alive(struct lnet_route *route)
296 struct lnet_peer *gw = route->lr_gateway;
297 struct lnet_peer_net *llpn;
298 struct lnet_peer_net *rlpn;
300 /* If the gateway is down then all routes are considered down */
305 * if discovery is disabled then rely on the cached aliveness
306 * information. This is handicapped information which we log when
307 * we receive the discovery ping response. The most uptodate
308 * aliveness information can only be obtained when discovery is
311 if (lnet_is_discovery_disabled(gw))
312 return route->lr_alive;
315 * check the gateway's interfaces on the local network
317 llpn = lnet_peer_get_net_locked(gw, route->lr_lnet);
321 if (!lnet_is_gateway_net_alive(llpn))
325 * For single hop routes avoid_asym_router_failure dictates
326 * that the remote net must exist on the gateway. For multi-hop
327 * routes the next-hop will not have the remote net.
329 if (avoid_asym_router_failure && route->lr_single_hop) {
330 rlpn = lnet_peer_get_net_locked(gw, route->lr_net);
333 if (!lnet_is_gateway_net_alive(rlpn))
337 spin_lock(&gw->lp_lock);
338 if (!(gw->lp_state & LNET_PEER_ROUTER_ENABLED)) {
339 spin_unlock(&gw->lp_lock);
340 if (gw->lp_rtr_refcount > 0)
341 CERROR("peer %s is being used as a gateway but routing feature is not turned on\n",
342 libcfs_nid2str(gw->lp_primary_nid));
345 spin_unlock(&gw->lp_lock);
351 lnet_consolidate_routes_locked(struct lnet_peer *orig_lp,
352 struct lnet_peer *new_lp)
354 struct lnet_peer_ni *lpni;
355 struct lnet_route *route;
358 * Although a route is correlated with a peer, but when it's added
359 * a specific NID is used. That NID refers to a peer_ni within
360 * a peer. There could be other peer_nis on the same net, which
361 * can be used to send to that gateway. However when we are
362 * consolidating gateways because of discovery, the nid used to
363 * add the route might've moved between gateway peers. In this
364 * case we want to move the route to the new gateway as well. The
365 * intent here is not to confuse the user who added the route.
367 list_for_each_entry(route, &orig_lp->lp_routes, lr_gwlist) {
368 lpni = lnet_peer_get_ni_locked(orig_lp, route->lr_nid);
370 lnet_net_lock(LNET_LOCK_EX);
371 list_move(&route->lr_gwlist, &new_lp->lp_routes);
372 lnet_net_unlock(LNET_LOCK_EX);
378 lnet_check_route_inconsistency(struct lnet_route *route)
380 if (!route->lr_single_hop && (int)route->lr_hops <= 1) {
381 CWARN("route %s->%s is detected to be multi-hop but hop count is set to %d\n",
382 libcfs_net2str(route->lr_net),
383 libcfs_nid2str(route->lr_gateway->lp_primary_nid),
384 (int) route->lr_hops);
389 lnet_set_route_hop_type(struct lnet_peer *gw, struct lnet_route *route)
391 struct lnet_peer_net *lpn;
392 bool single_hop = false;
394 list_for_each_entry(lpn, &gw->lp_peer_nets, lpn_peer_nets) {
395 if (route->lr_net == lpn->lpn_net_id) {
400 route->lr_single_hop = single_hop;
401 lnet_check_route_inconsistency(route);
405 lnet_set_route_aliveness(struct lnet_route *route, bool alive)
407 /* Log when there's a state change */
408 if (route->lr_alive != alive) {
409 CERROR("route to %s through %s has gone from %s to %s\n",
410 libcfs_net2str(route->lr_net),
411 libcfs_nid2str(route->lr_gateway->lp_primary_nid),
412 (route->lr_alive) ? "up" : "down",
413 alive ? "up" : "down");
414 route->lr_alive = alive;
419 lnet_router_discovery_ping_reply(struct lnet_peer *lp)
421 struct lnet_ping_buffer *pbuf = lp->lp_data;
422 struct lnet_peer_net *llpn;
423 struct lnet_route *route;
424 bool single_hop = false;
431 spin_lock(&lp->lp_lock);
432 lp_state = lp->lp_state;
434 /* only handle replies if discovery is disabled. */
435 if (!lnet_is_discovery_disabled_locked(lp)) {
436 spin_unlock(&lp->lp_lock);
440 spin_unlock(&lp->lp_lock);
442 if (lp_state & LNET_PEER_PING_FAILED ||
443 pbuf->pb_info.pi_features & LNET_PING_FEAT_RTE_DISABLED) {
444 CDEBUG(D_NET, "Set routes down for gw %s because %s %d\n",
445 libcfs_nid2str(lp->lp_primary_nid),
446 lp_state & LNET_PEER_PING_FAILED ? "ping failed" :
447 "route feature is disabled", lp->lp_ping_error);
448 /* If the ping failed or the peer has routing disabled then
449 * mark the routes served by this peer down
451 list_for_each_entry(route, &lp->lp_routes, lr_gwlist)
452 lnet_set_route_aliveness(route, false);
456 CDEBUG(D_NET, "Discovery is disabled. Processing reply for gw: %s:%d\n",
457 libcfs_nid2str(lp->lp_primary_nid), pbuf->pb_info.pi_nnis);
460 * examine the ping response to determine if the routes on that
461 * gateway should be declared alive.
462 * The route is alive if:
463 * 1. local network to reach the route is alive and
464 * 2. route is single hop, avoid_async_router_failure is set and
465 * there exists at least one NI on the route's remote net
467 list_for_each_entry(route, &lp->lp_routes, lr_gwlist) {
468 llpn = lnet_peer_get_net_locked(lp, route->lr_lnet);
470 lnet_set_route_aliveness(route, false);
474 if (!lnet_is_gateway_net_alive(llpn)) {
475 lnet_set_route_aliveness(route, false);
479 single_hop = net_up = false;
480 for (i = 1; i < pbuf->pb_info.pi_nnis; i++) {
481 net = LNET_NIDNET(pbuf->pb_info.pi_ni[i].ns_nid);
483 if (route->lr_net == net) {
485 if (pbuf->pb_info.pi_ni[i].ns_status ==
493 route->lr_single_hop = single_hop;
494 if (avoid_asym_router_failure && single_hop)
495 lnet_set_route_aliveness(route, net_up);
497 lnet_set_route_aliveness(route, true);
500 * warn that the route is configured as single-hop but it
501 * really is multi-hop as far as we can tell.
503 lnet_check_route_inconsistency(route);
508 lnet_router_discovery_complete(struct lnet_peer *lp)
510 struct lnet_peer_ni *lpni = NULL;
511 struct lnet_route *route;
513 spin_lock(&lp->lp_lock);
514 lp->lp_state &= ~LNET_PEER_RTR_DISCOVERY;
515 lp->lp_state |= LNET_PEER_RTR_DISCOVERED;
516 lp->lp_alive = lp->lp_dc_error == 0;
517 spin_unlock(&lp->lp_lock);
519 /* ping replies are being handled when discovery is disabled */
520 if (lnet_is_discovery_disabled_locked(lp))
523 if (!lp->lp_dc_error) {
525 * mark single-hop routes. If the remote net is not configured on
526 * the gateway we assume this is intentional and we mark the
527 * gateway as multi-hop
529 list_for_each_entry(route, &lp->lp_routes, lr_gwlist) {
530 lnet_set_route_aliveness(route, true);
531 lnet_set_route_hop_type(lp, route);
538 * We do not send messages directly to the remote interfaces
539 * of an LNet router. As such, we rely on the PING response
540 * to determine the up/down status of these interfaces. If
541 * a PING response is not receieved, or some other problem with
542 * discovery occurs that prevents us from getting this status,
543 * we assume all interfaces are down until we're able to
544 * determine otherwise.
546 CDEBUG(D_NET, "%s: Router discovery failed %d\n",
547 libcfs_nid2str(lp->lp_primary_nid), lp->lp_dc_error);
548 while ((lpni = lnet_get_next_peer_ni_locked(lp, NULL, lpni)) != NULL)
549 lpni->lpni_ns_status = LNET_NI_STATUS_DOWN;
551 list_for_each_entry(route, &lp->lp_routes, lr_gwlist)
552 lnet_set_route_aliveness(route, false);
556 lnet_rtr_addref_locked(struct lnet_peer *lp)
558 LASSERT(lp->lp_rtr_refcount >= 0);
560 /* lnet_net_lock must be exclusively locked */
561 lp->lp_rtr_refcount++;
562 if (lp->lp_rtr_refcount == 1) {
563 list_add_tail(&lp->lp_rtr_list, &the_lnet.ln_routers);
564 /* addref for the_lnet.ln_routers */
565 lnet_peer_addref_locked(lp);
566 the_lnet.ln_routers_version++;
571 lnet_rtr_decref_locked(struct lnet_peer *lp)
573 LASSERT(atomic_read(&lp->lp_refcount) > 0);
574 LASSERT(lp->lp_rtr_refcount > 0);
576 /* lnet_net_lock must be exclusively locked */
577 lp->lp_rtr_refcount--;
578 if (lp->lp_rtr_refcount == 0) {
579 LASSERT(list_empty(&lp->lp_routes));
581 list_del(&lp->lp_rtr_list);
582 /* decref for the_lnet.ln_routers */
583 lnet_peer_decref_locked(lp);
584 the_lnet.ln_routers_version++;
588 struct lnet_remotenet *
589 lnet_find_rnet_locked(__u32 net)
591 struct lnet_remotenet *rnet;
592 struct list_head *tmp;
593 struct list_head *rn_list;
595 LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
597 rn_list = lnet_net2rnethash(net);
598 list_for_each(tmp, rn_list) {
599 rnet = list_entry(tmp, struct lnet_remotenet, lrn_list);
601 if (rnet->lrn_net == net)
607 static void lnet_shuffle_seed(void)
610 struct lnet_ni *ni = NULL;
615 /* Nodes with small feet have little entropy
616 * the NID for this node gives the most entropy in the low bits */
617 while ((ni = lnet_get_next_ni_locked(NULL, ni)))
618 add_device_randomness(&ni->ni_nid, sizeof(ni->ni_nid));
623 /* NB expects LNET_LOCK held */
625 lnet_add_route_to_rnet(struct lnet_remotenet *rnet, struct lnet_route *route)
627 struct lnet_peer_net *lpn;
628 unsigned int offset = 0;
629 unsigned int len = 0;
634 list_for_each(e, &rnet->lrn_routes)
638 * Randomly adding routes to the list is done to ensure that when
639 * different nodes are using the same list of routers, they end up
640 * preferring different routers.
642 offset = prandom_u32_max(len + 1);
643 list_for_each(e, &rnet->lrn_routes) {
648 list_add(&route->lr_list, e);
650 * force a router check on the gateway to make sure the route is
653 list_for_each_entry(lpn, &route->lr_gateway->lp_peer_nets,
655 lpn->lpn_rtrcheck_timestamp = 0;
658 the_lnet.ln_remote_nets_version++;
660 /* add the route on the gateway list */
661 list_add(&route->lr_gwlist, &route->lr_gateway->lp_routes);
663 /* take a router reference count on the gateway */
664 lnet_rtr_addref_locked(route->lr_gateway);
668 lnet_add_route(__u32 net, __u32 hops, lnet_nid_t gateway,
669 __u32 priority, __u32 sensitivity)
671 struct list_head *route_entry;
672 struct lnet_remotenet *rnet;
673 struct lnet_remotenet *rnet2;
674 struct lnet_route *route;
675 struct lnet_peer_ni *lpni;
676 struct lnet_peer *gw;
680 CDEBUG(D_NET, "Add route: remote net %s hops %d priority %u gw %s\n",
681 libcfs_net2str(net), hops, priority, libcfs_nid2str(gateway));
683 if (gateway == LNET_NID_ANY ||
684 gateway == LNET_NID_LO_0 ||
685 net == LNET_NET_ANY ||
686 LNET_NETTYP(net) == LOLND ||
687 LNET_NIDNET(gateway) == net ||
688 (hops != LNET_UNDEFINED_HOPS && (hops < 1 || hops > 255)))
691 /* it's a local network */
692 if (lnet_islocalnet(net))
695 if (!lnet_islocalnet(LNET_NIDNET(gateway))) {
696 CERROR("Cannot add route with gateway %s. There is no local interface configured on LNet %s\n",
697 libcfs_nid2str(gateway),
698 libcfs_net2str(LNET_NIDNET(gateway)));
699 return -EHOSTUNREACH;
702 /* Assume net, route, all new */
703 LIBCFS_ALLOC(route, sizeof(*route));
704 LIBCFS_ALLOC(rnet, sizeof(*rnet));
705 if (route == NULL || rnet == NULL) {
706 CERROR("Out of memory creating route %s %d %s\n",
707 libcfs_net2str(net), hops, libcfs_nid2str(gateway));
709 LIBCFS_FREE(route, sizeof(*route));
711 LIBCFS_FREE(rnet, sizeof(*rnet));
715 INIT_LIST_HEAD(&rnet->lrn_routes);
717 /* store the local and remote net that the route represents */
718 route->lr_lnet = LNET_NIDNET(gateway);
720 route->lr_nid = gateway;
721 route->lr_priority = priority;
722 route->lr_hops = hops;
724 lnet_net_lock(LNET_LOCK_EX);
727 * lnet_nid2peerni_ex() grabs a ref on the lpni. We will need to
728 * lose that once we're done
730 lpni = lnet_nid2peerni_ex(gateway, LNET_LOCK_EX);
732 lnet_net_unlock(LNET_LOCK_EX);
734 LIBCFS_FREE(route, sizeof(*route));
735 LIBCFS_FREE(rnet, sizeof(*rnet));
738 CERROR("Error %d creating route %s %d %s\n", rc,
739 libcfs_net2str(net), hops,
740 libcfs_nid2str(gateway));
744 LASSERT(lpni->lpni_peer_net && lpni->lpni_peer_net->lpn_peer);
745 gw = lpni->lpni_peer_net->lpn_peer;
747 route->lr_gateway = gw;
749 rnet2 = lnet_find_rnet_locked(net);
752 list_add_tail(&rnet->lrn_list, lnet_net2rnethash(net));
756 /* Search for a duplicate route (it's a NOOP if it is) */
758 list_for_each(route_entry, &rnet2->lrn_routes) {
759 struct lnet_route *route2;
761 route2 = list_entry(route_entry, struct lnet_route, lr_list);
762 if (route2->lr_gateway == route->lr_gateway) {
767 /* our lookups must be true */
768 LASSERT(route2->lr_gateway->lp_primary_nid != gateway);
772 * It is possible to add multiple routes through the same peer,
773 * but it'll be using a different NID of that peer. When the
774 * gateway is discovered, discovery will consolidate the different
775 * peers into one peer. In this case the discovery code will have
776 * to move the routes from the peer that's being deleted to the
777 * consolidated peer lp_routes list
780 gw->lp_health_sensitivity = sensitivity;
781 lnet_add_route_to_rnet(rnet2, route);
782 if (lnet_peer_discovery_disabled)
783 CWARN("Consider turning discovery on to enable full "
784 "Multi-Rail routing functionality\n");
788 * get rid of the reference on the lpni.
790 lnet_peer_ni_decref_locked(lpni);
791 lnet_net_unlock(LNET_LOCK_EX);
797 LIBCFS_FREE(route, sizeof(*route));
801 LIBCFS_FREE(rnet, sizeof(*rnet));
803 /* kick start the monitor thread to handle the added route */
804 complete(&the_lnet.ln_mt_wait_complete);
810 lnet_del_route_from_rnet(lnet_nid_t gw_nid, struct list_head *route_list,
811 struct list_head *zombies)
813 struct lnet_peer *gateway;
814 struct lnet_route *route;
815 struct lnet_route *tmp;
817 list_for_each_entry_safe(route, tmp, route_list, lr_list) {
818 gateway = route->lr_gateway;
819 if (gw_nid != LNET_NID_ANY &&
820 gw_nid != gateway->lp_primary_nid)
824 * move to zombie to delete outside the lock
825 * Note that this function is called with the
826 * ln_api_mutex held as well as the exclusive net
827 * lock. Adding to the remote net list happens
828 * under the same conditions. Same goes for the
829 * gateway router list
831 list_move(&route->lr_list, zombies);
832 the_lnet.ln_remote_nets_version++;
834 list_del(&route->lr_gwlist);
835 lnet_rtr_decref_locked(gateway);
840 lnet_del_route(__u32 net, lnet_nid_t gw_nid)
842 LIST_HEAD(rnet_zombies);
843 struct lnet_remotenet *rnet;
844 struct lnet_remotenet *tmp;
845 struct list_head *rn_list;
846 struct lnet_peer_ni *lpni;
847 struct lnet_route *route;
849 struct lnet_peer *lp = NULL;
852 CDEBUG(D_NET, "Del route: net %s : gw %s\n",
853 libcfs_net2str(net), libcfs_nid2str(gw_nid));
855 /* NB Caller may specify either all routes via the given gateway
856 * or a specific route entry actual NIDs) */
858 lnet_net_lock(LNET_LOCK_EX);
860 lpni = lnet_find_peer_ni_locked(gw_nid);
862 lp = lpni->lpni_peer_net->lpn_peer;
864 gw_nid = lp->lp_primary_nid;
865 lnet_peer_ni_decref_locked(lpni);
868 if (net != LNET_NET_ANY) {
869 rnet = lnet_find_rnet_locked(net);
871 lnet_net_unlock(LNET_LOCK_EX);
874 lnet_del_route_from_rnet(gw_nid, &rnet->lrn_routes,
876 if (list_empty(&rnet->lrn_routes))
877 list_move(&rnet->lrn_list, &rnet_zombies);
881 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) {
882 rn_list = &the_lnet.ln_remote_nets_hash[i];
884 list_for_each_entry_safe(rnet, tmp, rn_list, lrn_list) {
885 lnet_del_route_from_rnet(gw_nid, &rnet->lrn_routes,
887 if (list_empty(&rnet->lrn_routes))
888 list_move(&rnet->lrn_list, &rnet_zombies);
894 * check if there are any routes remaining on the gateway
895 * If there are no more routes make sure to set the peer's
896 * lp_disc_net_id to 0 (invalid), in case we add more routes in
897 * the future on that gateway, then we start our discovery process
901 if (list_empty(&lp->lp_routes))
902 lp->lp_disc_net_id = 0;
905 lnet_net_unlock(LNET_LOCK_EX);
907 while (!list_empty(&zombies)) {
908 route = list_first_entry(&zombies, struct lnet_route, lr_list);
909 list_del(&route->lr_list);
910 LIBCFS_FREE(route, sizeof(*route));
913 while (!list_empty(&rnet_zombies)) {
914 rnet = list_first_entry(&rnet_zombies, struct lnet_remotenet,
916 list_del(&rnet->lrn_list);
917 LIBCFS_FREE(rnet, sizeof(*rnet));
924 lnet_destroy_routes (void)
926 lnet_del_route(LNET_NET_ANY, LNET_NID_ANY);
929 int lnet_get_rtr_pool_cfg(int cpt, struct lnet_ioctl_pool_cfg *pool_cfg)
931 struct lnet_rtrbufpool *rbp;
932 int i, rc = -ENOENT, j;
934 if (the_lnet.ln_rtrpools == NULL)
938 cfs_percpt_for_each(rbp, i, the_lnet.ln_rtrpools) {
943 for (j = 0; j < LNET_NRBPOOLS; j++) {
944 pool_cfg->pl_pools[j].pl_npages = rbp[j].rbp_npages;
945 pool_cfg->pl_pools[j].pl_nbuffers = rbp[j].rbp_nbuffers;
946 pool_cfg->pl_pools[j].pl_credits = rbp[j].rbp_credits;
947 pool_cfg->pl_pools[j].pl_mincredits = rbp[j].rbp_mincredits;
954 lnet_net_lock(LNET_LOCK_EX);
955 pool_cfg->pl_routing = the_lnet.ln_routing;
956 lnet_net_unlock(LNET_LOCK_EX);
962 lnet_get_route(int idx, __u32 *net, __u32 *hops,
963 lnet_nid_t *gateway, __u32 *flags, __u32 *priority, __u32 *sensitivity)
965 struct lnet_remotenet *rnet;
966 struct list_head *rn_list;
967 struct lnet_route *route;
968 struct list_head *e1;
969 struct list_head *e2;
973 cpt = lnet_net_lock_current();
975 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) {
976 rn_list = &the_lnet.ln_remote_nets_hash[i];
977 list_for_each(e1, rn_list) {
978 rnet = list_entry(e1, struct lnet_remotenet, lrn_list);
980 list_for_each(e2, &rnet->lrn_routes) {
981 route = list_entry(e2, struct lnet_route,
985 *net = rnet->lrn_net;
986 *gateway = route->lr_nid;
987 *hops = route->lr_hops;
988 *priority = route->lr_priority;
989 *sensitivity = route->lr_gateway->
990 lp_health_sensitivity;
991 if (lnet_is_route_alive(route))
992 *flags |= LNET_RT_ALIVE;
994 *flags &= ~LNET_RT_ALIVE;
995 if (route->lr_single_hop)
996 *flags &= ~LNET_RT_MULTI_HOP;
998 *flags |= LNET_RT_MULTI_HOP;
999 lnet_net_unlock(cpt);
1006 lnet_net_unlock(cpt);
1011 lnet_wait_known_routerstate(void)
1013 struct lnet_peer *rtr;
1014 struct list_head *entry;
1017 LASSERT(the_lnet.ln_mt_state == LNET_MT_STATE_RUNNING);
1020 int cpt = lnet_net_lock_current();
1023 list_for_each(entry, &the_lnet.ln_routers) {
1024 rtr = list_entry(entry, struct lnet_peer,
1027 spin_lock(&rtr->lp_lock);
1029 if ((rtr->lp_state & LNET_PEER_RTR_DISCOVERED) == 0) {
1031 spin_unlock(&rtr->lp_lock);
1034 spin_unlock(&rtr->lp_lock);
1037 lnet_net_unlock(cpt);
1042 schedule_timeout_uninterruptible(cfs_time_seconds(1));
1047 lnet_net_set_status_locked(struct lnet_net *net, __u32 status)
1050 bool update = false;
1052 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1053 if (lnet_ni_set_status(ni, status))
1060 lnet_update_ni_status_locked(void)
1062 struct lnet_net *net;
1068 LASSERT(the_lnet.ln_routing);
1070 timeout = router_ping_timeout + alive_router_check_interval;
1072 now = ktime_get_real_seconds();
1073 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1074 if (net->net_lnd->lnd_type == LOLND)
1077 if (now < net->net_last_alive + timeout)
1078 goto check_ni_fatal;
1080 spin_lock(&net->net_lock);
1081 /* re-check with lock */
1082 if (now < net->net_last_alive + timeout) {
1083 spin_unlock(&net->net_lock);
1084 goto check_ni_fatal;
1086 spin_unlock(&net->net_lock);
1089 * if the net didn't receive any traffic for past the
1090 * timeout on any of its constituent NIs, then mark all
1093 if (lnet_net_set_status_locked(net, LNET_NI_STATUS_DOWN)) {
1099 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1100 /* lnet_ni_set_status() will perform the same check of
1101 * ni_status while holding the ni lock. We can safely
1102 * check ni_status without that lock because it is only
1103 * written to under net_lock/EX and our caller is
1104 * holding a net lock.
1106 if (atomic_read(&ni->ni_fatal_error_on) &&
1108 ni->ni_status->ns_status != LNET_NI_STATUS_DOWN &&
1109 lnet_ni_set_status(ni, LNET_NI_STATUS_DOWN))
1117 void lnet_wait_router_start(void)
1119 if (check_routers_before_use) {
1120 /* Note that a helpful side-effect of pinging all known routers
1121 * at startup is that it makes them drop stale connections they
1122 * may have to a previous instance of me. */
1123 lnet_wait_known_routerstate();
1128 * This function is called from the monitor thread to check if there are
1129 * any active routers that need to be checked.
1132 lnet_router_checker_active(void)
1134 /* Router Checker thread needs to run when routing is enabled in
1135 * order to call lnet_update_ni_status_locked() */
1136 if (the_lnet.ln_routing)
1139 return !list_empty(&the_lnet.ln_routers) &&
1140 alive_router_check_interval > 0;
1144 lnet_check_routers(void)
1146 struct lnet_peer_net *first_lpn = NULL;
1147 struct lnet_peer_net *lpn;
1148 struct lnet_peer_ni *lpni;
1149 struct list_head *entry;
1150 struct lnet_peer *rtr;
1159 cpt = lnet_net_lock_current();
1161 version = the_lnet.ln_routers_version;
1163 list_for_each(entry, &the_lnet.ln_routers) {
1164 rtr = list_entry(entry, struct lnet_peer,
1167 now = ktime_get_real_seconds();
1170 * only discover the router if we've passed
1171 * alive_router_check_interval seconds. Some of the router
1172 * interfaces could be down and in that case they would be
1173 * undergoing recovery separately from this discovery.
1175 /* find next peer net which is also local */
1176 net_id = rtr->lp_disc_net_id;
1178 lpn = lnet_get_next_peer_net_locked(rtr, net_id);
1180 CERROR("gateway %s has no networks\n",
1181 libcfs_nid2str(rtr->lp_primary_nid));
1184 if (first_lpn == lpn)
1188 found_lpn = lnet_islocalnet_locked(lpn->lpn_net_id);
1189 net_id = lpn->lpn_net_id;
1190 } while (!found_lpn);
1192 if (!found_lpn || !lpn) {
1193 CERROR("no local network found for gateway %s\n",
1194 libcfs_nid2str(rtr->lp_primary_nid));
1198 if (now - lpn->lpn_rtrcheck_timestamp <
1199 alive_router_check_interval / lnet_current_net_count)
1203 * If we're currently discovering the peer then don't
1204 * issue another discovery
1206 spin_lock(&rtr->lp_lock);
1207 if (rtr->lp_state & LNET_PEER_RTR_DISCOVERY) {
1208 spin_unlock(&rtr->lp_lock);
1211 /* make sure we fully discover the router */
1212 rtr->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
1213 rtr->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH |
1214 LNET_PEER_RTR_DISCOVERY;
1215 spin_unlock(&rtr->lp_lock);
1217 /* find the peer_ni associated with the primary NID */
1218 lpni = lnet_peer_get_ni_locked(rtr, rtr->lp_primary_nid);
1220 CDEBUG(D_NET, "Expected to find an lpni for %s, but non found\n",
1221 libcfs_nid2str(rtr->lp_primary_nid));
1224 lnet_peer_ni_addref_locked(lpni);
1226 /* specify the net to use */
1227 rtr->lp_disc_net_id = lpn->lpn_net_id;
1229 /* discover the router */
1230 CDEBUG(D_NET, "discover %s, cpt = %d\n",
1231 libcfs_nid2str(lpni->lpni_nid), cpt);
1232 rc = lnet_discover_peer_locked(lpni, cpt, false);
1234 /* decrement ref count acquired by find_peer_ni_locked() */
1235 lnet_peer_ni_decref_locked(lpni);
1238 lpn->lpn_rtrcheck_timestamp = now;
1240 CERROR("Failed to discover router %s\n",
1241 libcfs_nid2str(rtr->lp_primary_nid));
1243 /* NB dropped lock */
1244 if (version != the_lnet.ln_routers_version) {
1245 /* the routers list has changed */
1250 if (the_lnet.ln_routing)
1251 push = lnet_update_ni_status_locked();
1253 lnet_net_unlock(cpt);
1255 /* if the status of the ni changed update the peers */
1257 lnet_push_update_to_peers(1);
1261 lnet_destroy_rtrbuf(struct lnet_rtrbuf *rb, int npages)
1263 int sz = offsetof(struct lnet_rtrbuf, rb_kiov[npages]);
1265 while (--npages >= 0)
1266 __free_page(rb->rb_kiov[npages].bv_page);
1268 LIBCFS_FREE(rb, sz);
1271 static struct lnet_rtrbuf *
1272 lnet_new_rtrbuf(struct lnet_rtrbufpool *rbp, int cpt)
1274 int npages = rbp->rbp_npages;
1275 int sz = offsetof(struct lnet_rtrbuf, rb_kiov[npages]);
1277 struct lnet_rtrbuf *rb;
1280 LIBCFS_CPT_ALLOC(rb, lnet_cpt_table(), cpt, sz);
1286 for (i = 0; i < npages; i++) {
1287 page = cfs_page_cpt_alloc(lnet_cpt_table(), cpt,
1288 GFP_KERNEL | __GFP_ZERO);
1291 __free_page(rb->rb_kiov[i].bv_page);
1293 LIBCFS_FREE(rb, sz);
1297 rb->rb_kiov[i].bv_len = PAGE_SIZE;
1298 rb->rb_kiov[i].bv_offset = 0;
1299 rb->rb_kiov[i].bv_page = page;
1306 lnet_rtrpool_free_bufs(struct lnet_rtrbufpool *rbp, int cpt)
1308 int npages = rbp->rbp_npages;
1309 struct lnet_rtrbuf *rb;
1312 if (rbp->rbp_nbuffers == 0) /* not initialized or already freed */
1316 list_splice_init(&rbp->rbp_msgs, &tmp);
1317 lnet_drop_routed_msgs_locked(&tmp, cpt);
1318 list_splice_init(&rbp->rbp_bufs, &tmp);
1319 rbp->rbp_req_nbuffers = 0;
1320 rbp->rbp_nbuffers = rbp->rbp_credits = 0;
1321 rbp->rbp_mincredits = 0;
1322 lnet_net_unlock(cpt);
1324 /* Free buffers on the free list. */
1325 while (!list_empty(&tmp)) {
1326 rb = list_entry(tmp.next, struct lnet_rtrbuf, rb_list);
1327 list_del(&rb->rb_list);
1328 lnet_destroy_rtrbuf(rb, npages);
1333 lnet_rtrpool_adjust_bufs(struct lnet_rtrbufpool *rbp, int nbufs, int cpt)
1336 struct lnet_rtrbuf *rb;
1338 int num_buffers = 0;
1340 int npages = rbp->rbp_npages;
1343 /* If we are called for less buffers than already in the pool, we
1344 * just lower the req_nbuffers number and excess buffers will be
1345 * thrown away as they are returned to the free list. Credits
1346 * then get adjusted as well.
1347 * If we already have enough buffers allocated to serve the
1348 * increase requested, then we can treat that the same way as we
1349 * do the decrease. */
1350 num_rb = nbufs - rbp->rbp_nbuffers;
1351 if (nbufs <= rbp->rbp_req_nbuffers || num_rb <= 0) {
1352 rbp->rbp_req_nbuffers = nbufs;
1353 lnet_net_unlock(cpt);
1356 /* store the older value of rbp_req_nbuffers and then set it to
1357 * the new request to prevent lnet_return_rx_credits_locked() from
1358 * freeing buffers that we need to keep around */
1359 old_req_nbufs = rbp->rbp_req_nbuffers;
1360 rbp->rbp_req_nbuffers = nbufs;
1361 lnet_net_unlock(cpt);
1363 /* allocate the buffers on a local list first. If all buffers are
1364 * allocated successfully then join this list to the rbp buffer
1365 * list. If not then free all allocated buffers. */
1366 while (num_rb-- > 0) {
1367 rb = lnet_new_rtrbuf(rbp, cpt);
1369 CERROR("Failed to allocate %d route bufs of %d pages\n",
1373 rbp->rbp_req_nbuffers = old_req_nbufs;
1374 lnet_net_unlock(cpt);
1379 list_add(&rb->rb_list, &rb_list);
1385 list_splice_tail(&rb_list, &rbp->rbp_bufs);
1386 rbp->rbp_nbuffers += num_buffers;
1387 rbp->rbp_credits += num_buffers;
1388 rbp->rbp_mincredits = rbp->rbp_credits;
1389 /* We need to schedule blocked msg using the newly
1391 while (!list_empty(&rbp->rbp_bufs) &&
1392 !list_empty(&rbp->rbp_msgs))
1393 lnet_schedule_blocked_locked(rbp);
1395 lnet_net_unlock(cpt);
1400 while (!list_empty(&rb_list)) {
1401 rb = list_entry(rb_list.next, struct lnet_rtrbuf, rb_list);
1402 list_del(&rb->rb_list);
1403 lnet_destroy_rtrbuf(rb, npages);
1410 lnet_rtrpool_init(struct lnet_rtrbufpool *rbp, int npages)
1412 INIT_LIST_HEAD(&rbp->rbp_msgs);
1413 INIT_LIST_HEAD(&rbp->rbp_bufs);
1415 rbp->rbp_npages = npages;
1416 rbp->rbp_credits = 0;
1417 rbp->rbp_mincredits = 0;
1421 lnet_rtrpools_free(int keep_pools)
1423 struct lnet_rtrbufpool *rtrp;
1426 if (the_lnet.ln_rtrpools == NULL) /* uninitialized or freed */
1429 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1430 lnet_rtrpool_free_bufs(&rtrp[LNET_TINY_BUF_IDX], i);
1431 lnet_rtrpool_free_bufs(&rtrp[LNET_SMALL_BUF_IDX], i);
1432 lnet_rtrpool_free_bufs(&rtrp[LNET_LARGE_BUF_IDX], i);
1436 cfs_percpt_free(the_lnet.ln_rtrpools);
1437 the_lnet.ln_rtrpools = NULL;
1442 lnet_nrb_tiny_calculate(void)
1444 int nrbs = LNET_NRB_TINY;
1446 if (tiny_router_buffers < 0) {
1447 LCONSOLE_ERROR_MSG(0x10c,
1448 "tiny_router_buffers=%d invalid when "
1449 "routing enabled\n", tiny_router_buffers);
1453 if (tiny_router_buffers > 0)
1454 nrbs = tiny_router_buffers;
1456 nrbs /= LNET_CPT_NUMBER;
1457 return max(nrbs, LNET_NRB_TINY_MIN);
1461 lnet_nrb_small_calculate(void)
1463 int nrbs = LNET_NRB_SMALL;
1465 if (small_router_buffers < 0) {
1466 LCONSOLE_ERROR_MSG(0x10c,
1467 "small_router_buffers=%d invalid when "
1468 "routing enabled\n", small_router_buffers);
1472 if (small_router_buffers > 0)
1473 nrbs = small_router_buffers;
1475 nrbs /= LNET_CPT_NUMBER;
1476 return max(nrbs, LNET_NRB_SMALL_MIN);
1480 lnet_nrb_large_calculate(void)
1482 int nrbs = LNET_NRB_LARGE;
1484 if (large_router_buffers < 0) {
1485 LCONSOLE_ERROR_MSG(0x10c,
1486 "large_router_buffers=%d invalid when "
1487 "routing enabled\n", large_router_buffers);
1491 if (large_router_buffers > 0)
1492 nrbs = large_router_buffers;
1494 nrbs /= LNET_CPT_NUMBER;
1495 return max(nrbs, LNET_NRB_LARGE_MIN);
1499 lnet_rtrpools_alloc(int im_a_router)
1501 struct lnet_rtrbufpool *rtrp;
1508 if (!strcmp(forwarding, "")) {
1509 /* not set either way */
1512 } else if (!strcmp(forwarding, "disabled")) {
1513 /* explicitly disabled */
1515 } else if (!strcmp(forwarding, "enabled")) {
1516 /* explicitly enabled */
1518 LCONSOLE_ERROR_MSG(0x10b, "'forwarding' not set to either "
1519 "'enabled' or 'disabled'\n");
1523 nrb_tiny = lnet_nrb_tiny_calculate();
1527 nrb_small = lnet_nrb_small_calculate();
1531 nrb_large = lnet_nrb_large_calculate();
1535 the_lnet.ln_rtrpools = cfs_percpt_alloc(lnet_cpt_table(),
1537 sizeof(struct lnet_rtrbufpool));
1538 if (the_lnet.ln_rtrpools == NULL) {
1539 LCONSOLE_ERROR_MSG(0x10c,
1540 "Failed to initialize router buffe pool\n");
1544 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1545 lnet_rtrpool_init(&rtrp[LNET_TINY_BUF_IDX], 0);
1546 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_TINY_BUF_IDX],
1551 lnet_rtrpool_init(&rtrp[LNET_SMALL_BUF_IDX],
1552 LNET_NRB_SMALL_PAGES);
1553 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_SMALL_BUF_IDX],
1558 lnet_rtrpool_init(&rtrp[LNET_LARGE_BUF_IDX],
1559 LNET_NRB_LARGE_PAGES);
1560 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_LARGE_BUF_IDX],
1566 lnet_net_lock(LNET_LOCK_EX);
1567 the_lnet.ln_routing = 1;
1568 lnet_net_unlock(LNET_LOCK_EX);
1569 complete(&the_lnet.ln_mt_wait_complete);
1573 lnet_rtrpools_free(0);
1578 lnet_rtrpools_adjust_helper(int tiny, int small, int large)
1583 struct lnet_rtrbufpool *rtrp;
1585 /* If the provided values for each buffer pool are different than the
1586 * configured values, we need to take action. */
1588 tiny_router_buffers = tiny;
1589 nrb = lnet_nrb_tiny_calculate();
1590 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1591 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_TINY_BUF_IDX],
1598 small_router_buffers = small;
1599 nrb = lnet_nrb_small_calculate();
1600 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1601 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_SMALL_BUF_IDX],
1608 large_router_buffers = large;
1609 nrb = lnet_nrb_large_calculate();
1610 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1611 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_LARGE_BUF_IDX],
1622 lnet_rtrpools_adjust(int tiny, int small, int large)
1624 /* this function doesn't revert the changes if adding new buffers
1625 * failed. It's up to the user space caller to revert the
1628 if (!the_lnet.ln_routing)
1631 return lnet_rtrpools_adjust_helper(tiny, small, large);
1635 lnet_rtrpools_enable(void)
1639 if (the_lnet.ln_routing)
1642 if (the_lnet.ln_rtrpools == NULL)
1643 /* If routing is turned off, and we have never
1644 * initialized the pools before, just call the
1645 * standard buffer pool allocation routine as
1646 * if we are just configuring this for the first
1648 rc = lnet_rtrpools_alloc(1);
1650 rc = lnet_rtrpools_adjust_helper(0, 0, 0);
1654 lnet_net_lock(LNET_LOCK_EX);
1655 the_lnet.ln_routing = 1;
1657 the_lnet.ln_ping_target->pb_info.pi_features &=
1658 ~LNET_PING_FEAT_RTE_DISABLED;
1659 lnet_net_unlock(LNET_LOCK_EX);
1661 if (lnet_peer_discovery_disabled)
1662 CWARN("Consider turning discovery on to enable full "
1663 "Multi-Rail routing functionality\n");
1669 lnet_rtrpools_disable(void)
1671 if (!the_lnet.ln_routing)
1674 lnet_net_lock(LNET_LOCK_EX);
1675 the_lnet.ln_routing = 0;
1676 the_lnet.ln_ping_target->pb_info.pi_features |=
1677 LNET_PING_FEAT_RTE_DISABLED;
1679 tiny_router_buffers = 0;
1680 small_router_buffers = 0;
1681 large_router_buffers = 0;
1682 lnet_net_unlock(LNET_LOCK_EX);
1683 lnet_rtrpools_free(1);
1687 lnet_notify_peer_down(struct lnet_ni *ni, lnet_nid_t nid)
1689 if (ni->ni_net->net_lnd->lnd_notify_peer_down != NULL)
1690 (ni->ni_net->net_lnd->lnd_notify_peer_down)(nid);
1694 * ni: local NI used to communicate with the peer
1696 * alive: true if peer is alive, false otherwise
1697 * reset: reset health value. This is requested by the LND.
1698 * when: notificaiton time.
1701 lnet_notify(struct lnet_ni *ni, lnet_nid_t nid, bool alive, bool reset,
1704 struct lnet_peer_ni *lpni = NULL;
1705 struct lnet_route *route;
1706 struct lnet_peer *lp;
1707 time64_t now = ktime_get_seconds();
1710 LASSERT(!in_interrupt());
1712 CDEBUG(D_NET, "%s notifying %s: %s\n",
1713 (ni == NULL) ? "userspace" : libcfs_nid2str(ni->ni_nid),
1714 libcfs_nid2str(nid), alive ? "up" : "down");
1717 LNET_NIDNET(ni->ni_nid) != LNET_NIDNET(nid)) {
1718 CWARN("Ignoring notification of %s %s by %s (different net)\n",
1719 libcfs_nid2str(nid), alive ? "birth" : "death",
1720 libcfs_nid2str(ni->ni_nid));
1724 /* can't do predictions... */
1726 CWARN("Ignoring prediction from %s of %s %s "
1727 "%lld seconds in the future\n",
1728 (ni == NULL) ? "userspace" : libcfs_nid2str(ni->ni_nid),
1729 libcfs_nid2str(nid), alive ? "up" : "down", when - now);
1733 if (ni != NULL && !alive && /* LND telling me she's down */
1734 !auto_down) { /* auto-down disabled */
1735 CDEBUG(D_NET, "Auto-down disabled\n");
1739 /* must lock 0 since this is used for synchronization */
1742 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
1747 lpni = lnet_find_peer_ni_locked(nid);
1751 CDEBUG(D_NET, "%s not found\n", libcfs_nid2str(nid));
1757 lpni->lpni_ns_status = LNET_NI_STATUS_UP;
1758 lnet_set_lpni_healthv_locked(lpni,
1759 LNET_MAX_HEALTH_VALUE);
1761 __u32 sensitivity = lpni->lpni_peer_net->
1762 lpn_peer->lp_health_sensitivity;
1764 lnet_inc_lpni_healthv_locked(lpni,
1765 (sensitivity) ? sensitivity :
1766 lnet_health_sensitivity);
1769 lpni->lpni_ns_status = LNET_NI_STATUS_DOWN;
1772 /* recalculate aliveness */
1773 alive = lnet_is_peer_ni_alive(lpni);
1776 if (ni != NULL && !alive)
1777 lnet_notify_peer_down(ni, lpni->lpni_nid);
1779 cpt = lpni->lpni_cpt;
1781 lnet_peer_ni_decref_locked(lpni);
1782 if (lpni && lpni->lpni_peer_net && lpni->lpni_peer_net->lpn_peer) {
1783 lp = lpni->lpni_peer_net->lpn_peer;
1784 lp->lp_alive = alive;
1785 list_for_each_entry(route, &lp->lp_routes, lr_gwlist)
1786 lnet_set_route_aliveness(route, alive);
1788 lnet_net_unlock(cpt);
1792 EXPORT_SYMBOL(lnet_notify);