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 extern unsigned int lnet_current_net_count;
40 static char *forwarding = "";
41 module_param(forwarding, charp, 0444);
42 MODULE_PARM_DESC(forwarding, "Explicitly enable/disable forwarding between networks");
44 static int tiny_router_buffers;
45 module_param(tiny_router_buffers, int, 0444);
46 MODULE_PARM_DESC(tiny_router_buffers, "# of 0 payload messages to buffer in the router");
47 static int small_router_buffers;
48 module_param(small_router_buffers, int, 0444);
49 MODULE_PARM_DESC(small_router_buffers, "# of small (1 page) messages to buffer in the router");
50 static int large_router_buffers;
51 module_param(large_router_buffers, int, 0444);
52 MODULE_PARM_DESC(large_router_buffers, "# of large messages to buffer in the router");
53 static int peer_buffer_credits;
54 module_param(peer_buffer_credits, int, 0444);
55 MODULE_PARM_DESC(peer_buffer_credits, "# router buffer credits per peer");
57 static int auto_down = 1;
58 module_param(auto_down, int, 0444);
59 MODULE_PARM_DESC(auto_down, "Automatically mark peers down on comms error");
62 lnet_peer_buffer_credits(struct lnet_net *net)
64 /* NI option overrides LNet default */
65 if (net->net_tunables.lct_peer_rtr_credits > 0)
66 return net->net_tunables.lct_peer_rtr_credits;
67 if (peer_buffer_credits > 0)
68 return peer_buffer_credits;
70 /* As an approximation, allow this peer the same number of router
71 * buffers as it is allowed outstanding sends */
72 return net->net_tunables.lct_peer_tx_credits;
75 static int check_routers_before_use;
76 module_param(check_routers_before_use, int, 0444);
77 MODULE_PARM_DESC(check_routers_before_use, "Assume routers are down and ping them before use");
79 int avoid_asym_router_failure = 1;
80 module_param(avoid_asym_router_failure, int, 0644);
81 MODULE_PARM_DESC(avoid_asym_router_failure, "Avoid asymmetrical router failures (0 to disable)");
83 int dead_router_check_interval = INT_MIN;
84 module_param(dead_router_check_interval, int, 0444);
85 MODULE_PARM_DESC(dead_router_check_interval, "(DEPRECATED - Use alive_router_check_interval)");
87 int live_router_check_interval = INT_MIN;
88 module_param(live_router_check_interval, int, 0444);
89 MODULE_PARM_DESC(live_router_check_interval, "(DEPRECATED - Use alive_router_check_interval)");
91 int alive_router_check_interval = 60;
92 module_param(alive_router_check_interval, int, 0644);
93 MODULE_PARM_DESC(alive_router_check_interval, "Seconds between live router health checks (<= 0 to disable)");
95 static int router_ping_timeout = 50;
96 module_param(router_ping_timeout, int, 0644);
97 MODULE_PARM_DESC(router_ping_timeout, "Seconds to wait for the reply to a router health query");
100 * A value between 0 and 100. 0 meaning that even if router's interfaces
101 * have the worse health still consider the gateway usable.
102 * 100 means that at least one interface on the route's remote net is 100%
103 * healthy to consider the route alive.
104 * The default is set to 100 to ensure we maintain the original behavior.
106 unsigned int router_sensitivity_percentage = 100;
107 static int rtr_sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp);
108 static struct kernel_param_ops param_ops_rtr_sensitivity = {
109 .set = rtr_sensitivity_set,
110 .get = param_get_int,
112 #define param_check_rtr_sensitivity(name, p) \
113 __param_check(name, p, int)
114 #ifdef HAVE_KERNEL_PARAM_OPS
115 module_param(router_sensitivity_percentage, rtr_sensitivity, S_IRUGO|S_IWUSR);
117 module_param_call(router_sensitivity_percentage, rtr_sensitivity_set, param_get_int,
118 &router_sensitivity_percentage, S_IRUGO|S_IWUSR);
120 MODULE_PARM_DESC(router_sensitivity_percentage,
121 "How healthy a gateway should be to be used in percent");
124 rtr_sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
127 unsigned *sen = (unsigned *)kp->arg;
130 rc = kstrtoul(val, 0, &value);
132 CERROR("Invalid module parameter value for 'router_sensitivity_percentage'\n");
136 if (value < 0 || value > 100) {
137 CERROR("Invalid value: %lu for 'router_sensitivity_percentage'\n", value);
142 * The purpose of locking the api_mutex here is to ensure that
143 * the correct value ends up stored properly.
145 mutex_lock(&the_lnet.ln_api_mutex);
149 mutex_unlock(&the_lnet.ln_api_mutex);
155 lnet_rtr_transfer_to_peer(struct lnet_peer *src, struct lnet_peer *target)
157 struct lnet_route *route;
159 lnet_net_lock(LNET_LOCK_EX);
160 target->lp_rtr_refcount += src->lp_rtr_refcount;
161 /* move the list of queued messages to the new peer */
162 list_splice_init(&src->lp_rtrq, &target->lp_rtrq);
163 /* move all the routes that reference the peer */
164 list_splice_init(&src->lp_routes, &target->lp_routes);
165 /* update all the routes to point to the new peer */
166 list_for_each_entry(route, &target->lp_routes, lr_gwlist)
167 route->lr_gateway = target;
168 /* remove the old peer from the ln_routers list */
169 list_del_init(&src->lp_rtr_list);
170 /* add the new peer to the ln_routers list */
171 if (list_empty(&target->lp_rtr_list)) {
172 lnet_peer_addref_locked(target);
173 list_add_tail(&target->lp_rtr_list, &the_lnet.ln_routers);
175 /* reset the ref count on the old peer and decrement its ref count */
176 src->lp_rtr_refcount = 0;
177 lnet_peer_decref_locked(src);
178 /* update the router version */
179 the_lnet.ln_routers_version++;
180 lnet_net_unlock(LNET_LOCK_EX);
184 lnet_peers_start_down(void)
186 return check_routers_before_use;
190 * A net is alive if at least one gateway NI on the network is alive.
193 lnet_is_gateway_net_alive(struct lnet_peer_net *lpn)
195 struct lnet_peer_ni *lpni;
197 list_for_each_entry(lpni, &lpn->lpn_peer_nis, lpni_peer_nis) {
198 if (lnet_is_peer_ni_alive(lpni))
206 * a gateway is alive only if all its nets are alive
207 * called with cpt lock held
209 bool lnet_is_gateway_alive(struct lnet_peer *gw)
211 struct lnet_peer_net *lpn;
213 list_for_each_entry(lpn, &gw->lp_peer_nets, lpn_peer_nets) {
214 if (!lnet_is_gateway_net_alive(lpn))
222 * lnet_is_route_alive() needs to be called with cpt lock held
223 * A route is alive if the gateway can route between the local network and
224 * the remote network of the route.
225 * This means at least one NI is alive on each of the local and remote
226 * networks of the gateway.
228 bool lnet_is_route_alive(struct lnet_route *route)
230 struct lnet_peer *gw = route->lr_gateway;
231 struct lnet_peer_net *llpn;
232 struct lnet_peer_net *rlpn;
236 * if discovery is disabled then rely on the cached aliveness
237 * information. This is handicapped information which we log when
238 * we receive the discovery ping response. The most uptodate
239 * aliveness information can only be obtained when discovery is
242 if (lnet_is_discovery_disabled(gw))
243 return route->lr_alive;
246 * check the gateway's interfaces on the route rnet to make sure
247 * that the gateway is viable.
249 llpn = lnet_peer_get_net_locked(gw, route->lr_lnet);
253 route_alive = lnet_is_gateway_net_alive(llpn);
255 if (avoid_asym_router_failure) {
256 rlpn = lnet_peer_get_net_locked(gw, route->lr_net);
259 route_alive = route_alive &&
260 lnet_is_gateway_net_alive(rlpn);
266 spin_lock(&gw->lp_lock);
267 if (!(gw->lp_state & LNET_PEER_ROUTER_ENABLED)) {
268 if (gw->lp_rtr_refcount > 0)
269 CERROR("peer %s is being used as a gateway but routing feature is not turned on\n",
270 libcfs_nid2str(gw->lp_primary_nid));
273 spin_unlock(&gw->lp_lock);
279 lnet_consolidate_routes_locked(struct lnet_peer *orig_lp,
280 struct lnet_peer *new_lp)
282 struct lnet_peer_ni *lpni;
283 struct lnet_route *route;
286 * Although a route is correlated with a peer, but when it's added
287 * a specific NID is used. That NID refers to a peer_ni within
288 * a peer. There could be other peer_nis on the same net, which
289 * can be used to send to that gateway. However when we are
290 * consolidating gateways because of discovery, the nid used to
291 * add the route might've moved between gateway peers. In this
292 * case we want to move the route to the new gateway as well. The
293 * intent here is not to confuse the user who added the route.
295 list_for_each_entry(route, &orig_lp->lp_routes, lr_gwlist) {
296 lpni = lnet_peer_get_ni_locked(orig_lp, route->lr_nid);
298 lnet_net_lock(LNET_LOCK_EX);
299 list_move(&route->lr_gwlist, &new_lp->lp_routes);
300 lnet_net_unlock(LNET_LOCK_EX);
307 lnet_set_route_aliveness(struct lnet_route *route, bool alive)
309 /* Log when there's a state change */
310 if (route->lr_alive != alive) {
311 CERROR("route to %s through %s has gone from %s to %s\n",
312 libcfs_net2str(route->lr_net),
313 libcfs_nid2str(route->lr_gateway->lp_primary_nid),
314 (route->lr_alive) ? "up" : "down",
315 alive ? "up" : "down");
316 route->lr_alive = alive;
321 lnet_router_discovery_ping_reply(struct lnet_peer *lp)
323 struct lnet_ping_buffer *pbuf = lp->lp_data;
324 struct lnet_remotenet *rnet;
325 struct lnet_peer_net *llpn;
326 struct lnet_route *route;
333 spin_lock(&lp->lp_lock);
334 lp_state = lp->lp_state;
336 /* only handle replies if discovery is disabled. */
337 if (!lnet_is_discovery_disabled_locked(lp)) {
338 spin_unlock(&lp->lp_lock);
342 spin_unlock(&lp->lp_lock);
344 if (lp_state & LNET_PEER_PING_FAILED ||
345 pbuf->pb_info.pi_features & LNET_PING_FEAT_RTE_DISABLED) {
346 CDEBUG(D_NET, "Set routes down for gw %s because %s %d\n",
347 libcfs_nid2str(lp->lp_primary_nid),
348 lp_state & LNET_PEER_PING_FAILED ? "ping failed" :
349 "route feature is disabled", lp->lp_ping_error);
350 /* If the ping failed or the peer has routing disabled then
351 * mark the routes served by this peer down
353 list_for_each_entry(route, &lp->lp_routes, lr_gwlist)
354 lnet_set_route_aliveness(route, false);
358 CDEBUG(D_NET, "Discovery is disabled. Processing reply for gw: %s\n",
359 libcfs_nid2str(lp->lp_primary_nid));
362 * examine the ping response:
363 * For each NID in the ping response, extract the net
364 * if the net exists on our remote net list then
365 * iterate over the routes on the rnet and if:
366 * The route's local net is healthy and
367 * The remote net status is UP, then mark the route up
368 * otherwise mark the route down
370 for (i = 1; i < pbuf->pb_info.pi_nnis; i++) {
371 net = LNET_NIDNET(pbuf->pb_info.pi_ni[i].ns_nid);
372 rnet = lnet_find_rnet_locked(net);
375 list_for_each_entry(route, &rnet->lrn_routes, lr_list) {
376 /* check if this is the route's gateway */
377 if (lp->lp_primary_nid !=
378 route->lr_gateway->lp_primary_nid)
381 llpn = lnet_peer_get_net_locked(lp, route->lr_lnet);
383 lnet_set_route_aliveness(route, false);
387 if (!lnet_is_gateway_net_alive(llpn)) {
388 lnet_set_route_aliveness(route, false);
392 if (avoid_asym_router_failure &&
393 pbuf->pb_info.pi_ni[i].ns_status !=
398 * revisit all previous NIDs and check if
399 * any on the network we're examining is
400 * up. If at least one is up then we consider
401 * the route to be alive.
403 for (j = 1; j < i; j++) {
404 net2 = LNET_NIDNET(pbuf->pb_info.
407 pbuf->pb_info.pi_ni[j].ns_status ==
412 lnet_set_route_aliveness(route, false);
417 lnet_set_route_aliveness(route, true);
423 lnet_router_discovery_complete(struct lnet_peer *lp)
425 struct lnet_peer_ni *lpni = NULL;
426 struct lnet_route *route;
428 spin_lock(&lp->lp_lock);
429 lp->lp_state &= ~LNET_PEER_RTR_DISCOVERY;
430 lp->lp_state |= LNET_PEER_RTR_DISCOVERED;
431 spin_unlock(&lp->lp_lock);
434 * Router discovery successful? All peer information would've been
435 * updated already. No need to do any more processing
437 if (!lp->lp_dc_error)
440 * discovery failed? then we need to set the status of each lpni
441 * to DOWN. It will be updated the next time we discover the
442 * router. For router peer NIs not on local networks, we never send
443 * messages directly to them, so their health will always remain
444 * at maximum. We can only tell if they are up or down from the
445 * status returned in the PING response. If we fail to get that
446 * status in our scheduled router discovery, then we'll assume
447 * it's down until we're told otherwise.
449 CDEBUG(D_NET, "%s: Router discovery failed %d\n",
450 libcfs_nid2str(lp->lp_primary_nid), lp->lp_dc_error);
451 while ((lpni = lnet_get_next_peer_ni_locked(lp, NULL, lpni)) != NULL)
452 lpni->lpni_ns_status = LNET_NI_STATUS_DOWN;
454 list_for_each_entry(route, &lp->lp_routes, lr_gwlist)
455 lnet_set_route_aliveness(route, false);
459 lnet_rtr_addref_locked(struct lnet_peer *lp)
461 LASSERT(lp->lp_rtr_refcount >= 0);
463 /* lnet_net_lock must be exclusively locked */
464 lp->lp_rtr_refcount++;
465 if (lp->lp_rtr_refcount == 1) {
466 list_add_tail(&lp->lp_rtr_list, &the_lnet.ln_routers);
467 /* addref for the_lnet.ln_routers */
468 lnet_peer_addref_locked(lp);
469 the_lnet.ln_routers_version++;
474 lnet_rtr_decref_locked(struct lnet_peer *lp)
476 LASSERT(atomic_read(&lp->lp_refcount) > 0);
477 LASSERT(lp->lp_rtr_refcount > 0);
479 /* lnet_net_lock must be exclusively locked */
480 lp->lp_rtr_refcount--;
481 if (lp->lp_rtr_refcount == 0) {
482 LASSERT(list_empty(&lp->lp_routes));
484 list_del(&lp->lp_rtr_list);
485 /* decref for the_lnet.ln_routers */
486 lnet_peer_decref_locked(lp);
487 the_lnet.ln_routers_version++;
491 struct lnet_remotenet *
492 lnet_find_rnet_locked(__u32 net)
494 struct lnet_remotenet *rnet;
495 struct list_head *tmp;
496 struct list_head *rn_list;
498 LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
500 rn_list = lnet_net2rnethash(net);
501 list_for_each(tmp, rn_list) {
502 rnet = list_entry(tmp, struct lnet_remotenet, lrn_list);
504 if (rnet->lrn_net == net)
510 static void lnet_shuffle_seed(void)
513 struct lnet_ni *ni = NULL;
518 /* Nodes with small feet have little entropy
519 * the NID for this node gives the most entropy in the low bits */
520 while ((ni = lnet_get_next_ni_locked(NULL, ni)))
521 add_device_randomness(&ni->ni_nid, sizeof(ni->ni_nid));
526 /* NB expects LNET_LOCK held */
528 lnet_add_route_to_rnet(struct lnet_remotenet *rnet, struct lnet_route *route)
530 struct lnet_peer_net *lpn;
531 unsigned int offset = 0;
532 unsigned int len = 0;
537 list_for_each(e, &rnet->lrn_routes)
541 * Randomly adding routes to the list is done to ensure that when
542 * different nodes are using the same list of routers, they end up
543 * preferring different routers.
545 offset = prandom_u32_max(len + 1);
546 list_for_each(e, &rnet->lrn_routes) {
551 list_add(&route->lr_list, e);
553 * force a router check on the gateway to make sure the route is
556 list_for_each_entry(lpn, &route->lr_gateway->lp_peer_nets,
558 lpn->lpn_rtrcheck_timestamp = 0;
561 the_lnet.ln_remote_nets_version++;
563 /* add the route on the gateway list */
564 list_add(&route->lr_gwlist, &route->lr_gateway->lp_routes);
566 /* take a router reference count on the gateway */
567 lnet_rtr_addref_locked(route->lr_gateway);
571 lnet_add_route(__u32 net, __u32 hops, lnet_nid_t gateway,
572 __u32 priority, __u32 sensitivity)
574 struct list_head *route_entry;
575 struct lnet_remotenet *rnet;
576 struct lnet_remotenet *rnet2;
577 struct lnet_route *route;
578 struct lnet_peer_ni *lpni;
579 struct lnet_peer *gw;
583 CDEBUG(D_NET, "Add route: remote net %s hops %d priority %u gw %s\n",
584 libcfs_net2str(net), hops, priority, libcfs_nid2str(gateway));
586 if (gateway == LNET_NID_ANY ||
587 LNET_NETTYP(LNET_NIDNET(gateway)) == LOLND ||
588 net == LNET_NIDNET(LNET_NID_ANY) ||
589 LNET_NETTYP(net) == LOLND ||
590 LNET_NIDNET(gateway) == net ||
591 (hops != LNET_UNDEFINED_HOPS && (hops < 1 || hops > 255)))
594 /* it's a local network */
595 if (lnet_islocalnet(net))
598 if (!lnet_islocalnet(LNET_NIDNET(gateway))) {
599 CERROR("Cannot add route with gateway %s. There is no local interface configured on LNet %s\n",
600 libcfs_nid2str(gateway),
601 libcfs_net2str(LNET_NIDNET(gateway)));
602 return -EHOSTUNREACH;
605 /* Assume net, route, all new */
606 LIBCFS_ALLOC(route, sizeof(*route));
607 LIBCFS_ALLOC(rnet, sizeof(*rnet));
608 if (route == NULL || rnet == NULL) {
609 CERROR("Out of memory creating route %s %d %s\n",
610 libcfs_net2str(net), hops, libcfs_nid2str(gateway));
612 LIBCFS_FREE(route, sizeof(*route));
614 LIBCFS_FREE(rnet, sizeof(*rnet));
618 INIT_LIST_HEAD(&rnet->lrn_routes);
620 /* store the local and remote net that the route represents */
621 route->lr_lnet = LNET_NIDNET(gateway);
623 route->lr_nid = gateway;
624 route->lr_priority = priority;
625 route->lr_hops = hops;
627 lnet_net_lock(LNET_LOCK_EX);
630 * lnet_nid2peerni_ex() grabs a ref on the lpni. We will need to
631 * lose that once we're done
633 lpni = lnet_nid2peerni_ex(gateway, LNET_LOCK_EX);
635 lnet_net_unlock(LNET_LOCK_EX);
637 LIBCFS_FREE(route, sizeof(*route));
638 LIBCFS_FREE(rnet, sizeof(*rnet));
641 CERROR("Error %d creating route %s %d %s\n", rc,
642 libcfs_net2str(net), hops,
643 libcfs_nid2str(gateway));
647 LASSERT(lpni->lpni_peer_net && lpni->lpni_peer_net->lpn_peer);
648 gw = lpni->lpni_peer_net->lpn_peer;
650 route->lr_gateway = gw;
652 rnet2 = lnet_find_rnet_locked(net);
655 list_add_tail(&rnet->lrn_list, lnet_net2rnethash(net));
659 /* Search for a duplicate route (it's a NOOP if it is) */
661 list_for_each(route_entry, &rnet2->lrn_routes) {
662 struct lnet_route *route2;
664 route2 = list_entry(route_entry, struct lnet_route, lr_list);
665 if (route2->lr_gateway == route->lr_gateway) {
670 /* our lookups must be true */
671 LASSERT(route2->lr_gateway->lp_primary_nid != gateway);
675 * It is possible to add multiple routes through the same peer,
676 * but it'll be using a different NID of that peer. When the
677 * gateway is discovered, discovery will consolidate the different
678 * peers into one peer. In this case the discovery code will have
679 * to move the routes from the peer that's being deleted to the
680 * consolidated peer lp_routes list
683 gw->lp_health_sensitivity = sensitivity;
684 lnet_add_route_to_rnet(rnet2, route);
685 if (lnet_peer_discovery_disabled)
686 CWARN("Consider turning discovery on to enable full "
687 "Multi-Rail routing functionality\n");
691 * get rid of the reference on the lpni.
693 lnet_peer_ni_decref_locked(lpni);
694 lnet_net_unlock(LNET_LOCK_EX);
700 LIBCFS_FREE(route, sizeof(*route));
704 LIBCFS_FREE(rnet, sizeof(*rnet));
706 /* kick start the monitor thread to handle the added route */
707 complete(&the_lnet.ln_mt_wait_complete);
713 lnet_del_route_from_rnet(lnet_nid_t gw_nid, struct list_head *route_list,
714 struct list_head *zombies)
716 struct lnet_peer *gateway;
717 struct lnet_route *route;
718 struct lnet_route *tmp;
720 list_for_each_entry_safe(route, tmp, route_list, lr_list) {
721 gateway = route->lr_gateway;
722 if (gw_nid != LNET_NID_ANY &&
723 gw_nid != gateway->lp_primary_nid)
727 * move to zombie to delete outside the lock
728 * Note that this function is called with the
729 * ln_api_mutex held as well as the exclusive net
730 * lock. Adding to the remote net list happens
731 * under the same conditions. Same goes for the
732 * gateway router list
734 list_move(&route->lr_list, zombies);
735 the_lnet.ln_remote_nets_version++;
737 list_del(&route->lr_gwlist);
738 lnet_rtr_decref_locked(gateway);
743 lnet_del_route(__u32 net, lnet_nid_t gw_nid)
745 struct list_head rnet_zombies;
746 struct lnet_remotenet *rnet;
747 struct lnet_remotenet *tmp;
748 struct list_head *rn_list;
749 struct lnet_peer_ni *lpni;
750 struct lnet_route *route;
751 struct list_head zombies;
752 struct lnet_peer *lp = NULL;
755 INIT_LIST_HEAD(&rnet_zombies);
756 INIT_LIST_HEAD(&zombies);
758 CDEBUG(D_NET, "Del route: net %s : gw %s\n",
759 libcfs_net2str(net), libcfs_nid2str(gw_nid));
761 /* NB Caller may specify either all routes via the given gateway
762 * or a specific route entry actual NIDs) */
764 lnet_net_lock(LNET_LOCK_EX);
766 lpni = lnet_find_peer_ni_locked(gw_nid);
768 lp = lpni->lpni_peer_net->lpn_peer;
770 gw_nid = lp->lp_primary_nid;
771 lnet_peer_ni_decref_locked(lpni);
774 if (net != LNET_NIDNET(LNET_NID_ANY)) {
775 rnet = lnet_find_rnet_locked(net);
777 lnet_net_unlock(LNET_LOCK_EX);
780 lnet_del_route_from_rnet(gw_nid, &rnet->lrn_routes,
782 if (list_empty(&rnet->lrn_routes))
783 list_move(&rnet->lrn_list, &rnet_zombies);
787 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) {
788 rn_list = &the_lnet.ln_remote_nets_hash[i];
790 list_for_each_entry_safe(rnet, tmp, rn_list, lrn_list) {
791 lnet_del_route_from_rnet(gw_nid, &rnet->lrn_routes,
793 if (list_empty(&rnet->lrn_routes))
794 list_move(&rnet->lrn_list, &rnet_zombies);
800 * check if there are any routes remaining on the gateway
801 * If there are no more routes make sure to set the peer's
802 * lp_disc_net_id to 0 (invalid), in case we add more routes in
803 * the future on that gateway, then we start our discovery process
807 if (list_empty(&lp->lp_routes))
808 lp->lp_disc_net_id = 0;
811 lnet_net_unlock(LNET_LOCK_EX);
813 while (!list_empty(&zombies)) {
814 route = list_first_entry(&zombies, struct lnet_route, lr_list);
815 list_del(&route->lr_list);
816 LIBCFS_FREE(route, sizeof(*route));
819 while (!list_empty(&rnet_zombies)) {
820 rnet = list_first_entry(&rnet_zombies, struct lnet_remotenet,
822 list_del(&rnet->lrn_list);
823 LIBCFS_FREE(rnet, sizeof(*rnet));
830 lnet_destroy_routes (void)
832 lnet_del_route(LNET_NIDNET(LNET_NID_ANY), LNET_NID_ANY);
835 int lnet_get_rtr_pool_cfg(int cpt, struct lnet_ioctl_pool_cfg *pool_cfg)
837 struct lnet_rtrbufpool *rbp;
838 int i, rc = -ENOENT, j;
840 if (the_lnet.ln_rtrpools == NULL)
844 cfs_percpt_for_each(rbp, i, the_lnet.ln_rtrpools) {
849 for (j = 0; j < LNET_NRBPOOLS; j++) {
850 pool_cfg->pl_pools[j].pl_npages = rbp[j].rbp_npages;
851 pool_cfg->pl_pools[j].pl_nbuffers = rbp[j].rbp_nbuffers;
852 pool_cfg->pl_pools[j].pl_credits = rbp[j].rbp_credits;
853 pool_cfg->pl_pools[j].pl_mincredits = rbp[j].rbp_mincredits;
860 lnet_net_lock(LNET_LOCK_EX);
861 pool_cfg->pl_routing = the_lnet.ln_routing;
862 lnet_net_unlock(LNET_LOCK_EX);
868 lnet_get_route(int idx, __u32 *net, __u32 *hops,
869 lnet_nid_t *gateway, __u32 *alive, __u32 *priority, __u32 *sensitivity)
871 struct lnet_remotenet *rnet;
872 struct list_head *rn_list;
873 struct lnet_route *route;
874 struct list_head *e1;
875 struct list_head *e2;
879 cpt = lnet_net_lock_current();
881 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) {
882 rn_list = &the_lnet.ln_remote_nets_hash[i];
883 list_for_each(e1, rn_list) {
884 rnet = list_entry(e1, struct lnet_remotenet, lrn_list);
886 list_for_each(e2, &rnet->lrn_routes) {
887 route = list_entry(e2, struct lnet_route,
891 *net = rnet->lrn_net;
892 *gateway = route->lr_nid;
893 *hops = route->lr_hops;
894 *priority = route->lr_priority;
895 *sensitivity = route->lr_gateway->
896 lp_health_sensitivity;
897 *alive = lnet_is_route_alive(route);
898 lnet_net_unlock(cpt);
905 lnet_net_unlock(cpt);
910 lnet_wait_known_routerstate(void)
912 struct lnet_peer *rtr;
913 struct list_head *entry;
916 LASSERT(the_lnet.ln_mt_state == LNET_MT_STATE_RUNNING);
919 int cpt = lnet_net_lock_current();
922 list_for_each(entry, &the_lnet.ln_routers) {
923 rtr = list_entry(entry, struct lnet_peer,
926 spin_lock(&rtr->lp_lock);
928 if ((rtr->lp_state & LNET_PEER_RTR_DISCOVERED) == 0) {
930 spin_unlock(&rtr->lp_lock);
933 spin_unlock(&rtr->lp_lock);
936 lnet_net_unlock(cpt);
941 set_current_state(TASK_UNINTERRUPTIBLE);
942 schedule_timeout(cfs_time_seconds(1));
947 lnet_net_set_status_locked(struct lnet_net *net, __u32 status)
952 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
955 ni->ni_status->ns_status != status) {
956 ni->ni_status->ns_status = status;
966 lnet_update_ni_status_locked(void)
968 struct lnet_net *net;
973 LASSERT(the_lnet.ln_routing);
975 timeout = router_ping_timeout + alive_router_check_interval;
977 now = ktime_get_real_seconds();
978 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
979 if (net->net_lnd->lnd_type == LOLND)
982 if (now < net->net_last_alive + timeout)
985 spin_lock(&net->net_lock);
986 /* re-check with lock */
987 if (now < net->net_last_alive + timeout) {
988 spin_unlock(&net->net_lock);
991 spin_unlock(&net->net_lock);
994 * if the net didn't receive any traffic for past the
995 * timeout on any of its constituent NIs, then mark all
998 push = lnet_net_set_status_locked(net, LNET_NI_STATUS_DOWN);
1004 void lnet_wait_router_start(void)
1006 if (check_routers_before_use) {
1007 /* Note that a helpful side-effect of pinging all known routers
1008 * at startup is that it makes them drop stale connections they
1009 * may have to a previous instance of me. */
1010 lnet_wait_known_routerstate();
1015 * This function is called from the monitor thread to check if there are
1016 * any active routers that need to be checked.
1019 lnet_router_checker_active(void)
1021 /* Router Checker thread needs to run when routing is enabled in
1022 * order to call lnet_update_ni_status_locked() */
1023 if (the_lnet.ln_routing)
1026 return !list_empty(&the_lnet.ln_routers) &&
1027 alive_router_check_interval > 0;
1031 lnet_check_routers(void)
1033 struct lnet_peer_net *first_lpn = NULL;
1034 struct lnet_peer_net *lpn;
1035 struct lnet_peer_ni *lpni;
1036 struct list_head *entry;
1037 struct lnet_peer *rtr;
1046 cpt = lnet_net_lock_current();
1048 version = the_lnet.ln_routers_version;
1050 list_for_each(entry, &the_lnet.ln_routers) {
1051 rtr = list_entry(entry, struct lnet_peer,
1054 now = ktime_get_real_seconds();
1057 * only discover the router if we've passed
1058 * alive_router_check_interval seconds. Some of the router
1059 * interfaces could be down and in that case they would be
1060 * undergoing recovery separately from this discovery.
1062 /* find next peer net which is also local */
1063 net_id = rtr->lp_disc_net_id;
1065 lpn = lnet_get_next_peer_net_locked(rtr, net_id);
1067 CERROR("gateway %s has no networks\n",
1068 libcfs_nid2str(rtr->lp_primary_nid));
1071 if (first_lpn == lpn)
1075 found_lpn = lnet_islocalnet_locked(lpn->lpn_net_id);
1076 net_id = lpn->lpn_net_id;
1077 } while (!found_lpn);
1079 if (!found_lpn || !lpn) {
1080 CERROR("no local network found for gateway %s\n",
1081 libcfs_nid2str(rtr->lp_primary_nid));
1085 if (now - lpn->lpn_rtrcheck_timestamp <
1086 alive_router_check_interval / lnet_current_net_count)
1090 * If we're currently discovering the peer then don't
1091 * issue another discovery
1093 spin_lock(&rtr->lp_lock);
1094 if (rtr->lp_state & LNET_PEER_RTR_DISCOVERY) {
1095 spin_unlock(&rtr->lp_lock);
1098 /* make sure we actively discover the router */
1099 rtr->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
1100 rtr->lp_state |= LNET_PEER_RTR_DISCOVERY;
1101 spin_unlock(&rtr->lp_lock);
1103 /* find the peer_ni associated with the primary NID */
1104 lpni = lnet_peer_get_ni_locked(rtr, rtr->lp_primary_nid);
1106 CDEBUG(D_NET, "Expected to find an lpni for %s, but non found\n",
1107 libcfs_nid2str(rtr->lp_primary_nid));
1110 lnet_peer_ni_addref_locked(lpni);
1112 /* specify the net to use */
1113 rtr->lp_disc_net_id = lpn->lpn_net_id;
1115 /* discover the router */
1116 CDEBUG(D_NET, "discover %s, cpt = %d\n",
1117 libcfs_nid2str(lpni->lpni_nid), cpt);
1118 rc = lnet_discover_peer_locked(lpni, cpt, false);
1120 /* decrement ref count acquired by find_peer_ni_locked() */
1121 lnet_peer_ni_decref_locked(lpni);
1124 lpn->lpn_rtrcheck_timestamp = now;
1126 CERROR("Failed to discover router %s\n",
1127 libcfs_nid2str(rtr->lp_primary_nid));
1129 /* NB dropped lock */
1130 if (version != the_lnet.ln_routers_version) {
1131 /* the routers list has changed */
1136 if (the_lnet.ln_routing)
1137 push = lnet_update_ni_status_locked();
1139 lnet_net_unlock(cpt);
1141 /* if the status of the ni changed update the peers */
1143 lnet_push_update_to_peers(1);
1147 lnet_destroy_rtrbuf(struct lnet_rtrbuf *rb, int npages)
1149 int sz = offsetof(struct lnet_rtrbuf, rb_kiov[npages]);
1151 while (--npages >= 0)
1152 __free_page(rb->rb_kiov[npages].kiov_page);
1154 LIBCFS_FREE(rb, sz);
1157 static struct lnet_rtrbuf *
1158 lnet_new_rtrbuf(struct lnet_rtrbufpool *rbp, int cpt)
1160 int npages = rbp->rbp_npages;
1161 int sz = offsetof(struct lnet_rtrbuf, rb_kiov[npages]);
1163 struct lnet_rtrbuf *rb;
1166 LIBCFS_CPT_ALLOC(rb, lnet_cpt_table(), cpt, sz);
1172 for (i = 0; i < npages; i++) {
1173 page = cfs_page_cpt_alloc(lnet_cpt_table(), cpt,
1174 GFP_KERNEL | __GFP_ZERO);
1177 __free_page(rb->rb_kiov[i].kiov_page);
1179 LIBCFS_FREE(rb, sz);
1183 rb->rb_kiov[i].kiov_len = PAGE_SIZE;
1184 rb->rb_kiov[i].kiov_offset = 0;
1185 rb->rb_kiov[i].kiov_page = page;
1192 lnet_rtrpool_free_bufs(struct lnet_rtrbufpool *rbp, int cpt)
1194 int npages = rbp->rbp_npages;
1195 struct lnet_rtrbuf *rb;
1196 struct list_head tmp;
1198 if (rbp->rbp_nbuffers == 0) /* not initialized or already freed */
1201 INIT_LIST_HEAD(&tmp);
1204 list_splice_init(&rbp->rbp_msgs, &tmp);
1205 lnet_drop_routed_msgs_locked(&tmp, cpt);
1206 list_splice_init(&rbp->rbp_bufs, &tmp);
1207 rbp->rbp_req_nbuffers = 0;
1208 rbp->rbp_nbuffers = rbp->rbp_credits = 0;
1209 rbp->rbp_mincredits = 0;
1210 lnet_net_unlock(cpt);
1212 /* Free buffers on the free list. */
1213 while (!list_empty(&tmp)) {
1214 rb = list_entry(tmp.next, struct lnet_rtrbuf, rb_list);
1215 list_del(&rb->rb_list);
1216 lnet_destroy_rtrbuf(rb, npages);
1221 lnet_rtrpool_adjust_bufs(struct lnet_rtrbufpool *rbp, int nbufs, int cpt)
1223 struct list_head rb_list;
1224 struct lnet_rtrbuf *rb;
1226 int num_buffers = 0;
1228 int npages = rbp->rbp_npages;
1231 /* If we are called for less buffers than already in the pool, we
1232 * just lower the req_nbuffers number and excess buffers will be
1233 * thrown away as they are returned to the free list. Credits
1234 * then get adjusted as well.
1235 * If we already have enough buffers allocated to serve the
1236 * increase requested, then we can treat that the same way as we
1237 * do the decrease. */
1238 num_rb = nbufs - rbp->rbp_nbuffers;
1239 if (nbufs <= rbp->rbp_req_nbuffers || num_rb <= 0) {
1240 rbp->rbp_req_nbuffers = nbufs;
1241 lnet_net_unlock(cpt);
1244 /* store the older value of rbp_req_nbuffers and then set it to
1245 * the new request to prevent lnet_return_rx_credits_locked() from
1246 * freeing buffers that we need to keep around */
1247 old_req_nbufs = rbp->rbp_req_nbuffers;
1248 rbp->rbp_req_nbuffers = nbufs;
1249 lnet_net_unlock(cpt);
1251 INIT_LIST_HEAD(&rb_list);
1253 /* allocate the buffers on a local list first. If all buffers are
1254 * allocated successfully then join this list to the rbp buffer
1255 * list. If not then free all allocated buffers. */
1256 while (num_rb-- > 0) {
1257 rb = lnet_new_rtrbuf(rbp, cpt);
1259 CERROR("Failed to allocate %d route bufs of %d pages\n",
1263 rbp->rbp_req_nbuffers = old_req_nbufs;
1264 lnet_net_unlock(cpt);
1269 list_add(&rb->rb_list, &rb_list);
1275 list_splice_tail(&rb_list, &rbp->rbp_bufs);
1276 rbp->rbp_nbuffers += num_buffers;
1277 rbp->rbp_credits += num_buffers;
1278 rbp->rbp_mincredits = rbp->rbp_credits;
1279 /* We need to schedule blocked msg using the newly
1281 while (!list_empty(&rbp->rbp_bufs) &&
1282 !list_empty(&rbp->rbp_msgs))
1283 lnet_schedule_blocked_locked(rbp);
1285 lnet_net_unlock(cpt);
1290 while (!list_empty(&rb_list)) {
1291 rb = list_entry(rb_list.next, struct lnet_rtrbuf, rb_list);
1292 list_del(&rb->rb_list);
1293 lnet_destroy_rtrbuf(rb, npages);
1300 lnet_rtrpool_init(struct lnet_rtrbufpool *rbp, int npages)
1302 INIT_LIST_HEAD(&rbp->rbp_msgs);
1303 INIT_LIST_HEAD(&rbp->rbp_bufs);
1305 rbp->rbp_npages = npages;
1306 rbp->rbp_credits = 0;
1307 rbp->rbp_mincredits = 0;
1311 lnet_rtrpools_free(int keep_pools)
1313 struct lnet_rtrbufpool *rtrp;
1316 if (the_lnet.ln_rtrpools == NULL) /* uninitialized or freed */
1319 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1320 lnet_rtrpool_free_bufs(&rtrp[LNET_TINY_BUF_IDX], i);
1321 lnet_rtrpool_free_bufs(&rtrp[LNET_SMALL_BUF_IDX], i);
1322 lnet_rtrpool_free_bufs(&rtrp[LNET_LARGE_BUF_IDX], i);
1326 cfs_percpt_free(the_lnet.ln_rtrpools);
1327 the_lnet.ln_rtrpools = NULL;
1332 lnet_nrb_tiny_calculate(void)
1334 int nrbs = LNET_NRB_TINY;
1336 if (tiny_router_buffers < 0) {
1337 LCONSOLE_ERROR_MSG(0x10c,
1338 "tiny_router_buffers=%d invalid when "
1339 "routing enabled\n", tiny_router_buffers);
1343 if (tiny_router_buffers > 0)
1344 nrbs = tiny_router_buffers;
1346 nrbs /= LNET_CPT_NUMBER;
1347 return max(nrbs, LNET_NRB_TINY_MIN);
1351 lnet_nrb_small_calculate(void)
1353 int nrbs = LNET_NRB_SMALL;
1355 if (small_router_buffers < 0) {
1356 LCONSOLE_ERROR_MSG(0x10c,
1357 "small_router_buffers=%d invalid when "
1358 "routing enabled\n", small_router_buffers);
1362 if (small_router_buffers > 0)
1363 nrbs = small_router_buffers;
1365 nrbs /= LNET_CPT_NUMBER;
1366 return max(nrbs, LNET_NRB_SMALL_MIN);
1370 lnet_nrb_large_calculate(void)
1372 int nrbs = LNET_NRB_LARGE;
1374 if (large_router_buffers < 0) {
1375 LCONSOLE_ERROR_MSG(0x10c,
1376 "large_router_buffers=%d invalid when "
1377 "routing enabled\n", large_router_buffers);
1381 if (large_router_buffers > 0)
1382 nrbs = large_router_buffers;
1384 nrbs /= LNET_CPT_NUMBER;
1385 return max(nrbs, LNET_NRB_LARGE_MIN);
1389 lnet_rtrpools_alloc(int im_a_router)
1391 struct lnet_rtrbufpool *rtrp;
1398 if (!strcmp(forwarding, "")) {
1399 /* not set either way */
1402 } else if (!strcmp(forwarding, "disabled")) {
1403 /* explicitly disabled */
1405 } else if (!strcmp(forwarding, "enabled")) {
1406 /* explicitly enabled */
1408 LCONSOLE_ERROR_MSG(0x10b, "'forwarding' not set to either "
1409 "'enabled' or 'disabled'\n");
1413 nrb_tiny = lnet_nrb_tiny_calculate();
1417 nrb_small = lnet_nrb_small_calculate();
1421 nrb_large = lnet_nrb_large_calculate();
1425 the_lnet.ln_rtrpools = cfs_percpt_alloc(lnet_cpt_table(),
1427 sizeof(struct lnet_rtrbufpool));
1428 if (the_lnet.ln_rtrpools == NULL) {
1429 LCONSOLE_ERROR_MSG(0x10c,
1430 "Failed to initialize router buffe pool\n");
1434 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1435 lnet_rtrpool_init(&rtrp[LNET_TINY_BUF_IDX], 0);
1436 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_TINY_BUF_IDX],
1441 lnet_rtrpool_init(&rtrp[LNET_SMALL_BUF_IDX],
1442 LNET_NRB_SMALL_PAGES);
1443 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_SMALL_BUF_IDX],
1448 lnet_rtrpool_init(&rtrp[LNET_LARGE_BUF_IDX],
1449 LNET_NRB_LARGE_PAGES);
1450 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_LARGE_BUF_IDX],
1456 lnet_net_lock(LNET_LOCK_EX);
1457 the_lnet.ln_routing = 1;
1458 lnet_net_unlock(LNET_LOCK_EX);
1459 complete(&the_lnet.ln_mt_wait_complete);
1463 lnet_rtrpools_free(0);
1468 lnet_rtrpools_adjust_helper(int tiny, int small, int large)
1473 struct lnet_rtrbufpool *rtrp;
1475 /* If the provided values for each buffer pool are different than the
1476 * configured values, we need to take action. */
1478 tiny_router_buffers = tiny;
1479 nrb = lnet_nrb_tiny_calculate();
1480 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1481 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_TINY_BUF_IDX],
1488 small_router_buffers = small;
1489 nrb = lnet_nrb_small_calculate();
1490 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1491 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_SMALL_BUF_IDX],
1498 large_router_buffers = large;
1499 nrb = lnet_nrb_large_calculate();
1500 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1501 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_LARGE_BUF_IDX],
1512 lnet_rtrpools_adjust(int tiny, int small, int large)
1514 /* this function doesn't revert the changes if adding new buffers
1515 * failed. It's up to the user space caller to revert the
1518 if (!the_lnet.ln_routing)
1521 return lnet_rtrpools_adjust_helper(tiny, small, large);
1525 lnet_rtrpools_enable(void)
1529 if (the_lnet.ln_routing)
1532 if (the_lnet.ln_rtrpools == NULL)
1533 /* If routing is turned off, and we have never
1534 * initialized the pools before, just call the
1535 * standard buffer pool allocation routine as
1536 * if we are just configuring this for the first
1538 rc = lnet_rtrpools_alloc(1);
1540 rc = lnet_rtrpools_adjust_helper(0, 0, 0);
1544 lnet_net_lock(LNET_LOCK_EX);
1545 the_lnet.ln_routing = 1;
1547 the_lnet.ln_ping_target->pb_info.pi_features &=
1548 ~LNET_PING_FEAT_RTE_DISABLED;
1549 lnet_net_unlock(LNET_LOCK_EX);
1551 if (lnet_peer_discovery_disabled)
1552 CWARN("Consider turning discovery on to enable full "
1553 "Multi-Rail routing functionality\n");
1559 lnet_rtrpools_disable(void)
1561 if (!the_lnet.ln_routing)
1564 lnet_net_lock(LNET_LOCK_EX);
1565 the_lnet.ln_routing = 0;
1566 the_lnet.ln_ping_target->pb_info.pi_features |=
1567 LNET_PING_FEAT_RTE_DISABLED;
1569 tiny_router_buffers = 0;
1570 small_router_buffers = 0;
1571 large_router_buffers = 0;
1572 lnet_net_unlock(LNET_LOCK_EX);
1573 lnet_rtrpools_free(1);
1577 lnet_notify_peer_down(struct lnet_ni *ni, lnet_nid_t nid)
1579 if (ni->ni_net->net_lnd->lnd_notify_peer_down != NULL)
1580 (ni->ni_net->net_lnd->lnd_notify_peer_down)(nid);
1584 * ni: local NI used to communicate with the peer
1586 * alive: true if peer is alive, false otherwise
1587 * reset: reset health value. This is requested by the LND.
1588 * when: notificaiton time.
1591 lnet_notify(struct lnet_ni *ni, lnet_nid_t nid, bool alive, bool reset,
1594 struct lnet_peer_ni *lpni = NULL;
1595 struct lnet_route *route;
1596 struct lnet_peer *lp;
1597 time64_t now = ktime_get_seconds();
1600 LASSERT (!in_interrupt ());
1602 CDEBUG (D_NET, "%s notifying %s: %s\n",
1603 (ni == NULL) ? "userspace" : libcfs_nid2str(ni->ni_nid),
1604 libcfs_nid2str(nid),
1605 alive ? "up" : "down");
1608 LNET_NIDNET(ni->ni_nid) != LNET_NIDNET(nid)) {
1609 CWARN("Ignoring notification of %s %s by %s (different net)\n",
1610 libcfs_nid2str(nid), alive ? "birth" : "death",
1611 libcfs_nid2str(ni->ni_nid));
1615 /* can't do predictions... */
1617 CWARN("Ignoring prediction from %s of %s %s "
1618 "%lld seconds in the future\n",
1619 (ni == NULL) ? "userspace" : libcfs_nid2str(ni->ni_nid),
1620 libcfs_nid2str(nid), alive ? "up" : "down", when - now);
1624 if (ni != NULL && !alive && /* LND telling me she's down */
1625 !auto_down) { /* auto-down disabled */
1626 CDEBUG(D_NET, "Auto-down disabled\n");
1630 /* must lock 0 since this is used for synchronization */
1633 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
1638 lpni = lnet_find_peer_ni_locked(nid);
1642 CDEBUG(D_NET, "%s not found\n", libcfs_nid2str(nid));
1648 lnet_set_healthv(&lpni->lpni_healthv,
1649 LNET_MAX_HEALTH_VALUE);
1651 lnet_inc_healthv(&lpni->lpni_healthv);
1653 lnet_handle_remote_failure_locked(lpni);
1656 /* recalculate aliveness */
1657 alive = lnet_is_peer_ni_alive(lpni);
1660 if (ni != NULL && !alive)
1661 lnet_notify_peer_down(ni, lpni->lpni_nid);
1663 cpt = lpni->lpni_cpt;
1665 lnet_peer_ni_decref_locked(lpni);
1666 if (lpni && lpni->lpni_peer_net && lpni->lpni_peer_net->lpn_peer) {
1667 lp = lpni->lpni_peer_net->lpn_peer;
1668 list_for_each_entry(route, &lp->lp_routes, lr_gwlist)
1669 lnet_set_route_aliveness(route, alive);
1671 lnet_net_unlock(cpt);
1675 EXPORT_SYMBOL(lnet_notify);