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(struct lnet_nid *gw_nid,
124 struct list_head *route_list,
125 struct list_head *zombies);
128 rtr_sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
131 unsigned *sen = (unsigned *)kp->arg;
134 rc = kstrtoul(val, 0, &value);
136 CERROR("Invalid module parameter value for 'router_sensitivity_percentage'\n");
140 if (value < 0 || value > 100) {
141 CERROR("Invalid value: %lu for 'router_sensitivity_percentage'\n", value);
146 * The purpose of locking the api_mutex here is to ensure that
147 * the correct value ends up stored properly.
149 mutex_lock(&the_lnet.ln_api_mutex);
153 mutex_unlock(&the_lnet.ln_api_mutex);
159 lnet_move_route(struct lnet_route *route, struct lnet_peer *lp,
160 struct list_head *rt_list)
161 __must_hold(&the_lnet.ln_api_mutex)
163 struct lnet_remotenet *rnet;
164 struct list_head zombies;
167 INIT_LIST_HEAD(&zombies);
174 rnet = lnet_find_rnet_locked(route->lr_net);
177 CDEBUG(D_NET, "deleting route %s->%s\n",
178 libcfs_net2str(route->lr_net),
179 libcfs_nidstr(&route->lr_nid));
182 * use the gateway's lp_primary_nid to delete the route as the
183 * lr_nid can be a constituent NID of the peer
185 lnet_del_route_from_rnet(
186 &route->lr_gateway->lp_primary_nid,
187 &rnet->lrn_routes, l);
190 route = list_first_entry(l, struct lnet_route,
192 route->lr_gateway = lp;
193 lnet_add_route_to_rnet(rnet, route);
195 while (!list_empty(l) && !rt_list) {
196 route = list_first_entry(l, struct lnet_route,
198 list_del(&route->lr_list);
199 LIBCFS_FREE(route, sizeof(*route));
205 lnet_rtr_transfer_to_peer(struct lnet_peer *src, struct lnet_peer *target)
207 struct lnet_route *route;
208 struct lnet_route *tmp, *tmp2;
210 lnet_net_lock(LNET_LOCK_EX);
211 CDEBUG(D_NET, "transfering routes from %s -> %s\n",
212 libcfs_nidstr(&src->lp_primary_nid),
213 libcfs_nidstr(&target->lp_primary_nid));
214 list_for_each_entry(route, &src->lp_routes, lr_gwlist) {
215 CDEBUG(D_NET, "%s: %s->%s\n",
216 libcfs_nidstr(&src->lp_primary_nid),
217 libcfs_net2str(route->lr_net),
218 libcfs_nidstr(&route->lr_nid));
220 list_splice_init(&src->lp_rtrq, &target->lp_rtrq);
221 list_for_each_entry_safe(route, tmp, &src->lp_routes, lr_gwlist) {
222 struct lnet_route *r2;
223 bool present = false;
224 list_for_each_entry_safe(r2, tmp2, &target->lp_routes, lr_gwlist) {
225 if (route->lr_net == r2->lr_net) {
226 if (route->lr_priority >= r2->lr_priority)
228 else if (route->lr_hops >= r2->lr_hops)
231 lnet_move_route(r2, NULL, NULL);
235 lnet_move_route(route, NULL, NULL);
237 lnet_move_route(route, target, NULL);
240 if (list_empty(&target->lp_rtr_list)) {
241 lnet_peer_addref_locked(target);
242 list_add_tail(&target->lp_rtr_list, &the_lnet.ln_routers);
245 the_lnet.ln_routers_version++;
246 lnet_net_unlock(LNET_LOCK_EX);
250 lnet_peers_start_down(void)
252 return check_routers_before_use;
256 * The peer_net of a gateway is alive if at least one of the peer_ni's on
257 * that peer_net is alive.
260 lnet_is_gateway_net_alive(struct lnet_peer_net *lpn)
262 struct lnet_peer_ni *lpni;
264 list_for_each_entry(lpni, &lpn->lpn_peer_nis, lpni_peer_nis) {
265 if (lnet_is_peer_ni_alive(lpni))
273 * a gateway is alive only if all its nets are alive
274 * called with cpt lock held
276 bool lnet_is_gateway_alive(struct lnet_peer *gw)
278 struct lnet_peer_net *lpn;
283 list_for_each_entry(lpn, &gw->lp_peer_nets, lpn_peer_nets) {
284 if (!lnet_is_gateway_net_alive(lpn))
292 * lnet_is_route_alive() needs to be called with cpt lock held
293 * A route is alive if the gateway can route between the local network and
294 * the remote network of the route.
295 * This means at least one NI is alive on each of the local and remote
296 * networks of the gateway.
298 bool lnet_is_route_alive(struct lnet_route *route)
300 struct lnet_peer *gw = route->lr_gateway;
301 struct lnet_peer_net *llpn;
302 struct lnet_peer_net *rlpn;
304 /* If the gateway is down then all routes are considered down */
309 * if discovery is disabled then rely on the cached aliveness
310 * information. This is handicapped information which we log when
311 * we receive the discovery ping response. The most uptodate
312 * aliveness information can only be obtained when discovery is
315 if (lnet_is_discovery_disabled(gw))
316 return atomic_read(&route->lr_alive) == 1;
319 * check the gateway's interfaces on the local network
321 llpn = lnet_peer_get_net_locked(gw, route->lr_lnet);
325 if (!lnet_is_gateway_net_alive(llpn))
329 * For single hop routes avoid_asym_router_failure dictates
330 * that the remote net must exist on the gateway. For multi-hop
331 * routes the next-hop will not have the remote net.
333 if (avoid_asym_router_failure &&
334 (route->lr_hops == 1 || route->lr_single_hop)) {
335 rlpn = lnet_peer_get_net_locked(gw, route->lr_net);
338 if (!lnet_is_gateway_net_alive(rlpn))
342 spin_lock(&gw->lp_lock);
343 if (!(gw->lp_state & LNET_PEER_ROUTER_ENABLED)) {
344 spin_unlock(&gw->lp_lock);
345 if (gw->lp_rtr_refcount > 0)
346 CERROR("peer %s is being used as a gateway but routing feature is not turned on\n",
347 libcfs_nidstr(&gw->lp_primary_nid));
350 spin_unlock(&gw->lp_lock);
356 lnet_consolidate_routes_locked(struct lnet_peer *orig_lp,
357 struct lnet_peer *new_lp)
359 struct lnet_peer_ni *lpni;
360 struct lnet_route *route;
363 * Although a route is correlated with a peer, but when it's added
364 * a specific NID is used. That NID refers to a peer_ni within
365 * a peer. There could be other peer_nis on the same net, which
366 * can be used to send to that gateway. However when we are
367 * consolidating gateways because of discovery, the nid used to
368 * add the route might've moved between gateway peers. In this
369 * case we want to move the route to the new gateway as well. The
370 * intent here is not to confuse the user who added the route.
372 list_for_each_entry(route, &orig_lp->lp_routes, lr_gwlist) {
373 lpni = lnet_peer_ni_get_locked(orig_lp, &route->lr_nid);
375 lnet_net_lock(LNET_LOCK_EX);
376 list_move(&route->lr_gwlist, &new_lp->lp_routes);
377 lnet_net_unlock(LNET_LOCK_EX);
383 lnet_check_route_inconsistency(struct lnet_route *route)
385 if (!route->lr_single_hop &&
386 (route->lr_hops == 1 || route->lr_hops == LNET_UNDEFINED_HOPS) &&
387 avoid_asym_router_failure) {
388 CWARN("route %s->%s is detected to be multi-hop but hop count is set to %d\n",
389 libcfs_net2str(route->lr_net),
390 libcfs_nidstr(&route->lr_gateway->lp_primary_nid),
391 (int) route->lr_hops);
395 /* Routes are added and removed under both ln_api_mutex and net_lock/EX
396 * Since we are not modifying anything we simply require the ln_api_mutex be
397 * held so that things are not modified underneath us
400 lnet_router_discovery_ping_reply(struct lnet_peer *lp,
401 struct lnet_ping_buffer *pbuf)
402 __must_hold(&the_lnet.ln_api_mutex)
404 struct lnet_ping_iter piter;
405 struct lnet_peer_net *llpn;
406 struct lnet_route *route;
408 bool single_hop = false;
412 if (pbuf->pb_info.pi_features & LNET_PING_FEAT_RTE_DISABLED) {
413 CERROR("Peer %s is being used as a gateway but routing feature is not turned on\n",
414 libcfs_nidstr(&lp->lp_primary_nid));
415 list_for_each_entry(route, &lp->lp_routes, lr_gwlist)
416 lnet_set_route_aliveness(route, false);
420 CDEBUG(D_NET, "Processing reply for gw: %s nnis %d\n",
421 libcfs_nidstr(&lp->lp_primary_nid), pbuf->pb_info.pi_nnis);
423 /* examine the ping response to determine if the routes on that
424 * gateway should be declared alive.
425 * The route is alive if:
426 * 1. local network to reach the route is alive and
427 * 2. route is single hop, avoid_async_router_failure is set and
428 * there exists at least one NI on the route's remote net
430 list_for_each_entry(route, &lp->lp_routes, lr_gwlist) {
431 llpn = lnet_peer_get_net_locked(lp, route->lr_lnet);
433 lnet_set_route_aliveness(route, false);
437 if (!lnet_is_gateway_net_alive(llpn)) {
438 lnet_set_route_aliveness(route, false);
442 single_hop = net_up = false;
443 for (stp = ping_iter_first(&piter, pbuf, &nid);
445 stp = ping_iter_next(&piter, &nid)) {
446 if (route->lr_net == LNET_NID_NET(&nid)) {
448 if (*stp == LNET_NI_STATUS_UP) {
455 route->lr_single_hop = single_hop;
456 if (avoid_asym_router_failure &&
457 (route->lr_hops == 1 || route->lr_single_hop))
458 lnet_set_route_aliveness(route, net_up);
460 lnet_set_route_aliveness(route, true);
463 * warn that the route is configured as single-hop but it
464 * really is multi-hop as far as we can tell.
466 lnet_check_route_inconsistency(route);
471 lnet_router_discovery_complete(struct lnet_peer *lp)
473 struct lnet_peer_ni *lpni = NULL;
474 struct lnet_route *route;
476 spin_lock(&lp->lp_lock);
477 lp->lp_state &= ~LNET_PEER_RTR_DISCOVERY;
478 lp->lp_state |= LNET_PEER_RTR_DISCOVERED;
479 lp->lp_alive = lp->lp_dc_error == 0;
480 spin_unlock(&lp->lp_lock);
482 if (!lp->lp_dc_error)
486 * We do not send messages directly to the remote interfaces
487 * of an LNet router. As such, we rely on the PING response
488 * to determine the up/down status of these interfaces. If
489 * a PING response is not receieved, or some other problem with
490 * discovery occurs that prevents us from getting this status,
491 * we assume all interfaces are down until we're able to
492 * determine otherwise.
494 CDEBUG(D_NET, "%s: Router discovery failed %d\n",
495 libcfs_nidstr(&lp->lp_primary_nid), lp->lp_dc_error);
496 while ((lpni = lnet_get_next_peer_ni_locked(lp, NULL, lpni)) != NULL)
497 lpni->lpni_ns_status = LNET_NI_STATUS_DOWN;
499 list_for_each_entry(route, &lp->lp_routes, lr_gwlist)
500 lnet_set_route_aliveness(route, false);
504 lnet_rtr_addref_locked(struct lnet_peer *lp)
506 LASSERT(lp->lp_rtr_refcount >= 0);
508 /* lnet_net_lock must be exclusively locked */
509 lp->lp_rtr_refcount++;
510 if (lp->lp_rtr_refcount == 1) {
511 list_add_tail(&lp->lp_rtr_list, &the_lnet.ln_routers);
512 /* addref for the_lnet.ln_routers */
513 lnet_peer_addref_locked(lp);
514 the_lnet.ln_routers_version++;
519 lnet_rtr_decref_locked(struct lnet_peer *lp)
521 LASSERT(atomic_read(&lp->lp_refcount) > 0);
522 LASSERT(lp->lp_rtr_refcount > 0);
524 /* lnet_net_lock must be exclusively locked */
525 lp->lp_rtr_refcount--;
526 if (lp->lp_rtr_refcount == 0) {
527 LASSERT(list_empty(&lp->lp_routes));
529 list_del(&lp->lp_rtr_list);
530 /* decref for the_lnet.ln_routers */
531 lnet_peer_decref_locked(lp);
532 the_lnet.ln_routers_version++;
536 struct lnet_remotenet *
537 lnet_find_rnet_locked(__u32 net)
539 struct lnet_remotenet *rnet;
540 struct list_head *rn_list;
542 LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
544 rn_list = lnet_net2rnethash(net);
545 list_for_each_entry(rnet, rn_list, lrn_list) {
546 if (rnet->lrn_net == net)
552 static void lnet_shuffle_seed(void)
555 struct lnet_ni *ni = NULL;
560 /* Nodes with small feet have little entropy
561 * the NID for this node gives the most entropy in the low bits */
562 while ((ni = lnet_get_next_ni_locked(NULL, ni)))
563 add_device_randomness(&ni->ni_nid, sizeof(ni->ni_nid));
568 /* NB expects LNET_LOCK held */
570 lnet_add_route_to_rnet(struct lnet_remotenet *rnet, struct lnet_route *route)
572 struct lnet_peer_net *lpn;
573 unsigned int offset = 0;
574 unsigned int len = 0;
580 list_for_each(e, &rnet->lrn_routes)
584 * Randomly adding routes to the list is done to ensure that when
585 * different nodes are using the same list of routers, they end up
586 * preferring different routers.
588 offset = prandom_u32_max(len + 1);
589 list_for_each(e, &rnet->lrn_routes) {
594 list_add(&route->lr_list, e);
596 * force a router check on the gateway to make sure the route is
599 now = ktime_get_real_seconds();
600 list_for_each_entry(lpn, &route->lr_gateway->lp_peer_nets,
602 lpn->lpn_next_ping = now;
605 the_lnet.ln_remote_nets_version++;
607 /* add the route on the gateway list */
608 list_add(&route->lr_gwlist, &route->lr_gateway->lp_routes);
610 /* take a router reference count on the gateway */
611 lnet_rtr_addref_locked(route->lr_gateway);
615 lnet_add_route(__u32 net, __u32 hops, struct lnet_nid *gateway,
616 __u32 priority, __u32 sensitivity)
617 __must_hold(&the_lnet.ln_api_mutex)
619 struct list_head *route_entry;
620 struct lnet_remotenet *rnet;
621 struct lnet_remotenet *rnet2;
622 struct lnet_route *route;
623 struct lnet_peer_ni *lpni;
624 struct lnet_peer *gw;
628 CDEBUG(D_NET, "Add route: remote net %s hops %d priority %u gw %s\n",
629 libcfs_net2str(net), hops, priority, libcfs_nidstr(gateway));
631 if (LNET_NID_IS_ANY(gateway) ||
632 nid_is_lo0(gateway) ||
633 net == LNET_NET_ANY ||
634 LNET_NETTYP(net) == LOLND ||
635 LNET_NID_NET(gateway) == net ||
636 (hops != LNET_UNDEFINED_HOPS && (hops < 1 || hops > 255)))
639 /* it's a local network */
640 if (lnet_islocalnet(net))
643 if (!lnet_islocalnet(LNET_NID_NET(gateway))) {
644 CERROR("Cannot add route with gateway %s. There is no local interface configured on LNet %s\n",
645 libcfs_nidstr(gateway),
646 libcfs_net2str(LNET_NID_NET(gateway)));
647 return -EHOSTUNREACH;
650 /* Assume net, route, all new */
651 LIBCFS_ALLOC(route, sizeof(*route));
652 LIBCFS_ALLOC(rnet, sizeof(*rnet));
653 if (route == NULL || rnet == NULL) {
654 CERROR("Out of memory creating route %s %d %s\n",
655 libcfs_net2str(net), hops, libcfs_nidstr(gateway));
657 LIBCFS_FREE(route, sizeof(*route));
659 LIBCFS_FREE(rnet, sizeof(*rnet));
663 INIT_LIST_HEAD(&rnet->lrn_routes);
665 /* store the local and remote net that the route represents */
666 route->lr_lnet = LNET_NID_NET(gateway);
668 route->lr_nid = *gateway;
669 route->lr_priority = priority;
670 route->lr_hops = hops;
671 if (lnet_peers_start_down())
672 atomic_set(&route->lr_alive, 0);
674 atomic_set(&route->lr_alive, 1);
676 lnet_net_lock(LNET_LOCK_EX);
679 * lnet_nid2peerni_ex() grabs a ref on the lpni. We will need to
680 * lose that once we're done
682 lpni = lnet_nid2peerni_ex(gateway);
684 lnet_net_unlock(LNET_LOCK_EX);
686 LIBCFS_FREE(route, sizeof(*route));
687 LIBCFS_FREE(rnet, sizeof(*rnet));
690 CERROR("Error %d creating route %s %d %s\n", rc,
691 libcfs_net2str(net), hops,
692 libcfs_nidstr(gateway));
697 LASSERT(lpni->lpni_peer_net);
698 LASSERT(lpni->lpni_peer_net->lpn_peer);
699 gw = lpni->lpni_peer_net->lpn_peer;
701 route->lr_gateway = gw;
703 rnet2 = lnet_find_rnet_locked(net);
706 list_add_tail(&rnet->lrn_list, lnet_net2rnethash(net));
710 /* Search for a duplicate route (it's a NOOP if it is) */
712 list_for_each(route_entry, &rnet2->lrn_routes) {
713 struct lnet_route *route2;
715 route2 = list_entry(route_entry, struct lnet_route, lr_list);
716 if (route2->lr_gateway == route->lr_gateway) {
721 /* our lookups must be true */
722 LASSERT(!nid_same(&route2->lr_gateway->lp_primary_nid,
727 * It is possible to add multiple routes through the same peer,
728 * but it'll be using a different NID of that peer. When the
729 * gateway is discovered, discovery will consolidate the different
730 * peers into one peer. In this case the discovery code will have
731 * to move the routes from the peer that's being deleted to the
732 * consolidated peer lp_routes list
735 gw->lp_health_sensitivity = sensitivity;
736 lnet_add_route_to_rnet(rnet2, route);
737 if (lnet_peer_discovery_disabled)
738 CWARN("Consider turning discovery on to enable full Multi-Rail routing functionality\n");
742 * get rid of the reference on the lpni.
744 lnet_peer_ni_decref_locked(lpni);
745 lnet_net_unlock(LNET_LOCK_EX);
747 /* If avoid_asym_router_failure is enabled and hop count is not
748 * set to 1 for a route that is actually single-hop, then the
749 * feature will fail to prevent the router from being selected
750 * if it is missing a NI on the remote network due to misconfiguration.
752 if (avoid_asym_router_failure && hops == LNET_UNDEFINED_HOPS)
753 CWARN("Use hops = 1 for a single-hop route when avoid_asym_router_failure feature is enabled\n");
759 LIBCFS_FREE(route, sizeof(*route));
763 LIBCFS_FREE(rnet, sizeof(*rnet));
765 /* kick start the monitor thread to handle the added route */
766 complete(&the_lnet.ln_mt_wait_complete);
772 lnet_del_route_from_rnet(struct lnet_nid *gw_nid,
773 struct list_head *route_list,
774 struct list_head *zombies)
776 struct lnet_peer *gateway;
777 struct lnet_route *route;
778 struct lnet_route *tmp;
780 list_for_each_entry_safe(route, tmp, route_list, lr_list) {
781 gateway = route->lr_gateway;
782 if (gw_nid && !nid_same(gw_nid, &gateway->lp_primary_nid))
786 * move to zombie to delete outside the lock
787 * Note that this function is called with the
788 * ln_api_mutex held as well as the exclusive net
789 * lock. Adding to the remote net list happens
790 * under the same conditions. Same goes for the
791 * gateway router list
793 list_move(&route->lr_list, zombies);
794 the_lnet.ln_remote_nets_version++;
796 list_del(&route->lr_gwlist);
797 lnet_rtr_decref_locked(gateway);
802 lnet_del_route(__u32 net, struct lnet_nid *gw)
803 __must_hold(&the_lnet.ln_api_mutex)
805 LIST_HEAD(rnet_zombies);
806 struct lnet_remotenet *rnet;
807 struct lnet_remotenet *tmp;
808 struct list_head *rn_list;
809 struct lnet_peer_ni *lpni;
810 struct lnet_route *route;
811 struct lnet_nid gw_nid;
813 struct lnet_peer *lp = NULL;
816 CDEBUG(D_NET, "Del route: net %s : gw %s\n",
817 libcfs_net2str(net), libcfs_nidstr(gw));
819 /* NB Caller may specify either all routes via the given gateway
820 * or a specific route entry actual NIDs) */
822 lnet_net_lock(LNET_LOCK_EX);
825 lpni = lnet_peer_ni_find_locked(gw);
829 lp = lpni->lpni_peer_net->lpn_peer;
831 gw_nid = lp->lp_primary_nid;
833 lnet_peer_ni_decref_locked(lpni);
836 if (net != LNET_NET_ANY) {
837 rnet = lnet_find_rnet_locked(net);
839 lnet_net_unlock(LNET_LOCK_EX);
842 lnet_del_route_from_rnet(gw, &rnet->lrn_routes,
844 if (list_empty(&rnet->lrn_routes))
845 list_move(&rnet->lrn_list, &rnet_zombies);
849 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) {
850 rn_list = &the_lnet.ln_remote_nets_hash[i];
852 list_for_each_entry_safe(rnet, tmp, rn_list, lrn_list) {
853 lnet_del_route_from_rnet(gw, &rnet->lrn_routes,
855 if (list_empty(&rnet->lrn_routes))
856 list_move(&rnet->lrn_list, &rnet_zombies);
862 * check if there are any routes remaining on the gateway
863 * If there are no more routes make sure to set the peer's
864 * lp_disc_net_id to 0 (invalid), in case we add more routes in
865 * the future on that gateway, then we start our discovery process
869 if (list_empty(&lp->lp_routes))
870 lp->lp_disc_net_id = 0;
873 lnet_net_unlock(LNET_LOCK_EX);
875 while (!list_empty(&zombies)) {
876 route = list_first_entry(&zombies, struct lnet_route, lr_list);
877 list_del(&route->lr_list);
878 LIBCFS_FREE(route, sizeof(*route));
881 while (!list_empty(&rnet_zombies)) {
882 rnet = list_first_entry(&rnet_zombies, struct lnet_remotenet,
884 list_del(&rnet->lrn_list);
885 LIBCFS_FREE(rnet, sizeof(*rnet));
892 lnet_destroy_routes(void)
894 lnet_del_route(LNET_NET_ANY, NULL);
897 int lnet_get_rtr_pool_cfg(int cpt, struct lnet_ioctl_pool_cfg *pool_cfg)
899 struct lnet_rtrbufpool *rbp;
900 int i, rc = -ENOENT, j;
902 if (the_lnet.ln_rtrpools == NULL)
906 cfs_percpt_for_each(rbp, i, the_lnet.ln_rtrpools) {
911 for (j = 0; j < LNET_NRBPOOLS; j++) {
912 pool_cfg->pl_pools[j].pl_npages = rbp[j].rbp_npages;
913 pool_cfg->pl_pools[j].pl_nbuffers = rbp[j].rbp_nbuffers;
914 pool_cfg->pl_pools[j].pl_credits = rbp[j].rbp_credits;
915 pool_cfg->pl_pools[j].pl_mincredits = rbp[j].rbp_mincredits;
922 lnet_net_lock(LNET_LOCK_EX);
923 pool_cfg->pl_routing = the_lnet.ln_routing;
924 lnet_net_unlock(LNET_LOCK_EX);
930 lnet_get_route(int idx, __u32 *net, __u32 *hops, lnet_nid_t *gateway,
931 __u32 *flags, __u32 *priority, __u32 *sensitivity)
933 struct lnet_remotenet *rnet;
934 struct list_head *rn_list;
935 struct lnet_route *route;
939 cpt = lnet_net_lock_current();
941 for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) {
942 rn_list = &the_lnet.ln_remote_nets_hash[i];
943 list_for_each_entry(rnet, rn_list, lrn_list) {
944 list_for_each_entry(route, &rnet->lrn_routes, lr_list) {
946 *net = rnet->lrn_net;
947 *gateway = lnet_nid_to_nid4(&route->lr_nid);
948 *hops = route->lr_hops;
949 *priority = route->lr_priority;
950 *sensitivity = route->lr_gateway->
951 lp_health_sensitivity;
952 if (lnet_is_route_alive(route))
953 *flags |= LNET_RT_ALIVE;
955 *flags &= ~LNET_RT_ALIVE;
956 if (route->lr_single_hop)
957 *flags &= ~LNET_RT_MULTI_HOP;
959 *flags |= LNET_RT_MULTI_HOP;
960 lnet_net_unlock(cpt);
967 lnet_net_unlock(cpt);
972 lnet_wait_known_routerstate(void)
974 struct lnet_peer *rtr;
977 LASSERT(the_lnet.ln_mt_state == LNET_MT_STATE_RUNNING);
980 int cpt = lnet_net_lock_current();
983 list_for_each_entry(rtr, &the_lnet.ln_routers, lp_rtr_list) {
984 spin_lock(&rtr->lp_lock);
986 if ((rtr->lp_state & LNET_PEER_RTR_DISCOVERED) == 0) {
988 spin_unlock(&rtr->lp_lock);
991 spin_unlock(&rtr->lp_lock);
994 lnet_net_unlock(cpt);
999 schedule_timeout_uninterruptible(cfs_time_seconds(1));
1004 lnet_net_set_status_locked(struct lnet_net *net, __u32 status)
1007 bool update = false;
1009 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1010 if (lnet_ni_set_status(ni, status))
1017 lnet_update_ni_status_locked(void)
1019 struct lnet_net *net;
1025 LASSERT(the_lnet.ln_routing);
1027 timeout = router_ping_timeout + alive_router_check_interval;
1029 now = ktime_get_seconds();
1030 list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1031 if (net->net_lnd->lnd_type == LOLND)
1034 if (now < net->net_last_alive + timeout)
1035 goto check_ni_fatal;
1037 spin_lock(&net->net_lock);
1038 /* re-check with lock */
1039 if (now < net->net_last_alive + timeout) {
1040 spin_unlock(&net->net_lock);
1041 goto check_ni_fatal;
1043 spin_unlock(&net->net_lock);
1046 * if the net didn't receive any traffic for past the
1047 * timeout on any of its constituent NIs, then mark all
1050 if (lnet_net_set_status_locked(net, LNET_NI_STATUS_DOWN)) {
1056 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1057 /* lnet_ni_set_status() will perform the same check of
1058 * ni_status while holding the ni lock. We can safely
1059 * check ni_status without that lock because it is only
1060 * written to under net_lock/EX and our caller is
1061 * holding a net lock.
1063 if (atomic_read(&ni->ni_fatal_error_on) &&
1065 *ni->ni_status != LNET_NI_STATUS_DOWN &&
1066 lnet_ni_set_status(ni, LNET_NI_STATUS_DOWN))
1074 void lnet_wait_router_start(void)
1076 if (check_routers_before_use) {
1077 /* Note that a helpful side-effect of pinging all known routers
1078 * at startup is that it makes them drop stale connections they
1079 * may have to a previous instance of me. */
1080 lnet_wait_known_routerstate();
1085 * This function is called from the monitor thread to check if there are
1086 * any active routers that need to be checked.
1088 bool lnet_router_checker_active(void)
1090 /* Router Checker thread needs to run when routing is enabled in
1091 * order to call lnet_update_ni_status_locked() */
1092 if (the_lnet.ln_routing)
1095 return !list_empty(&the_lnet.ln_routers) &&
1096 alive_router_check_interval > 0;
1100 lnet_check_routers(void)
1102 struct lnet_peer_net *first_lpn;
1103 struct lnet_peer_net *lpn;
1104 struct lnet_peer_ni *lpni;
1105 struct lnet_peer *rtr;
1115 cpt = lnet_net_lock_current();
1117 version = the_lnet.ln_routers_version;
1119 list_for_each_entry(rtr, &the_lnet.ln_routers, lp_rtr_list) {
1120 /* If we're currently discovering the peer then don't
1121 * issue another discovery
1123 if (rtr->lp_state & LNET_PEER_RTR_DISCOVERY)
1126 now = ktime_get_real_seconds();
1128 /* find the next local peer net which needs to be ping'd */
1132 net_id = rtr->lp_disc_net_id;
1134 lpn = lnet_get_next_peer_net_locked(rtr, net_id);
1136 CERROR("gateway %s has no networks\n",
1137 libcfs_nidstr(&rtr->lp_primary_nid));
1141 /* We looped back to the first peer net */
1142 if (first_lpn == lpn)
1147 net_id = lpn->lpn_net_id;
1148 if (!lnet_islocalnet_locked(net_id))
1153 CDEBUG(D_NET, "rtr %s(%p) %s(%p) next ping %lld\n",
1154 libcfs_nidstr(&rtr->lp_primary_nid), rtr,
1155 libcfs_net2str(net_id), lpn,
1156 lpn->lpn_next_ping);
1158 needs_ping = now >= lpn->lpn_next_ping;
1160 } while (!needs_ping);
1162 if (!found_lpn || !lpn) {
1163 CERROR("no local network found for gateway %s\n",
1164 libcfs_nidstr(&rtr->lp_primary_nid));
1171 spin_lock(&rtr->lp_lock);
1172 /* make sure we fully discover the router */
1173 rtr->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
1174 rtr->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH |
1175 LNET_PEER_RTR_DISCOVERY;
1176 spin_unlock(&rtr->lp_lock);
1178 /* find the peer_ni associated with the primary NID */
1179 lpni = lnet_peer_ni_get_locked(rtr, &rtr->lp_primary_nid);
1181 CDEBUG(D_NET, "Expected to find an lpni for %s, but non found\n",
1182 libcfs_nidstr(&rtr->lp_primary_nid));
1185 lnet_peer_ni_addref_locked(lpni);
1187 /* specify the net to use */
1188 rtr->lp_disc_net_id = lpn->lpn_net_id;
1190 /* discover the router */
1191 CDEBUG(D_NET, "discover %s, cpt = %d\n",
1192 libcfs_nidstr(&lpni->lpni_nid), cpt);
1193 rc = lnet_discover_peer_locked(lpni, cpt, false);
1195 /* drop ref taken above */
1196 lnet_peer_ni_decref_locked(lpni);
1199 lpn->lpn_next_ping = now + alive_router_check_interval;
1201 CERROR("Failed to discover router %s\n",
1202 libcfs_nidstr(&rtr->lp_primary_nid));
1204 /* NB cpt lock was dropped in lnet_discover_peer_locked() */
1205 if (version != the_lnet.ln_routers_version) {
1206 /* the routers list has changed */
1211 if (the_lnet.ln_routing)
1212 push = lnet_update_ni_status_locked();
1214 lnet_net_unlock(cpt);
1216 /* if the status of the ni changed update the peers */
1218 lnet_push_update_to_peers(1);
1222 lnet_destroy_rtrbuf(struct lnet_rtrbuf *rb, int npages)
1224 int sz = offsetof(struct lnet_rtrbuf, rb_kiov[npages]);
1226 while (--npages >= 0)
1227 __free_page(rb->rb_kiov[npages].bv_page);
1229 LIBCFS_FREE(rb, sz);
1232 static struct lnet_rtrbuf *
1233 lnet_new_rtrbuf(struct lnet_rtrbufpool *rbp, int cpt)
1235 int npages = rbp->rbp_npages;
1236 int sz = offsetof(struct lnet_rtrbuf, rb_kiov[npages]);
1238 struct lnet_rtrbuf *rb;
1241 LIBCFS_CPT_ALLOC(rb, lnet_cpt_table(), cpt, sz);
1247 for (i = 0; i < npages; i++) {
1248 page = cfs_page_cpt_alloc(lnet_cpt_table(), cpt, GFP_KERNEL |
1249 __GFP_ZERO | __GFP_NORETRY);
1252 __free_page(rb->rb_kiov[i].bv_page);
1254 LIBCFS_FREE(rb, sz);
1258 rb->rb_kiov[i].bv_len = PAGE_SIZE;
1259 rb->rb_kiov[i].bv_offset = 0;
1260 rb->rb_kiov[i].bv_page = page;
1267 lnet_rtrpool_free_bufs(struct lnet_rtrbufpool *rbp, int cpt)
1269 int npages = rbp->rbp_npages;
1270 struct lnet_rtrbuf *rb;
1273 if (rbp->rbp_nbuffers == 0) /* not initialized or already freed */
1277 list_splice_init(&rbp->rbp_msgs, &tmp);
1278 lnet_drop_routed_msgs_locked(&tmp, cpt);
1279 list_splice_init(&rbp->rbp_bufs, &tmp);
1280 rbp->rbp_req_nbuffers = 0;
1281 rbp->rbp_nbuffers = rbp->rbp_credits = 0;
1282 rbp->rbp_mincredits = 0;
1283 lnet_net_unlock(cpt);
1285 /* Free buffers on the free list. */
1286 while (!list_empty(&tmp)) {
1287 rb = list_first_entry(&tmp, struct lnet_rtrbuf, rb_list);
1288 list_del(&rb->rb_list);
1289 lnet_destroy_rtrbuf(rb, npages);
1294 lnet_rtrpool_adjust_bufs(struct lnet_rtrbufpool *rbp, int nbufs, int cpt)
1297 struct lnet_rtrbuf *rb;
1299 int num_buffers = 0;
1301 int npages = rbp->rbp_npages;
1304 /* If we are called for less buffers than already in the pool, we
1305 * just lower the req_nbuffers number and excess buffers will be
1306 * thrown away as they are returned to the free list. Credits
1307 * then get adjusted as well.
1308 * If we already have enough buffers allocated to serve the
1309 * increase requested, then we can treat that the same way as we
1310 * do the decrease. */
1311 num_rb = nbufs - rbp->rbp_nbuffers;
1312 if (nbufs <= rbp->rbp_req_nbuffers || num_rb <= 0) {
1313 rbp->rbp_req_nbuffers = nbufs;
1314 lnet_net_unlock(cpt);
1317 /* store the older value of rbp_req_nbuffers and then set it to
1318 * the new request to prevent lnet_return_rx_credits_locked() from
1319 * freeing buffers that we need to keep around */
1320 old_req_nbufs = rbp->rbp_req_nbuffers;
1321 rbp->rbp_req_nbuffers = nbufs;
1322 lnet_net_unlock(cpt);
1324 /* allocate the buffers on a local list first. If all buffers are
1325 * allocated successfully then join this list to the rbp buffer
1326 * list. If not then free all allocated buffers. */
1327 while (num_rb-- > 0) {
1328 rb = lnet_new_rtrbuf(rbp, cpt);
1330 CERROR("lnet: error allocating %ux%u page router buffers on CPT %u: rc = %d\n",
1331 nbufs, npages, cpt, -ENOMEM);
1334 rbp->rbp_req_nbuffers = old_req_nbufs;
1335 lnet_net_unlock(cpt);
1340 list_add(&rb->rb_list, &rb_list);
1346 list_splice_tail(&rb_list, &rbp->rbp_bufs);
1347 rbp->rbp_nbuffers += num_buffers;
1348 rbp->rbp_credits += num_buffers;
1349 rbp->rbp_mincredits = rbp->rbp_credits;
1350 /* We need to schedule blocked msg using the newly
1352 while (!list_empty(&rbp->rbp_bufs) &&
1353 !list_empty(&rbp->rbp_msgs))
1354 lnet_schedule_blocked_locked(rbp);
1356 lnet_net_unlock(cpt);
1361 while ((rb = list_first_entry_or_null(&rb_list,
1363 rb_list)) != NULL) {
1364 list_del(&rb->rb_list);
1365 lnet_destroy_rtrbuf(rb, npages);
1372 lnet_rtrpool_init(struct lnet_rtrbufpool *rbp, int npages)
1374 INIT_LIST_HEAD(&rbp->rbp_msgs);
1375 INIT_LIST_HEAD(&rbp->rbp_bufs);
1377 rbp->rbp_npages = npages;
1378 rbp->rbp_credits = 0;
1379 rbp->rbp_mincredits = 0;
1383 lnet_rtrpools_free(int keep_pools)
1385 struct lnet_rtrbufpool *rtrp;
1388 if (the_lnet.ln_rtrpools == NULL) /* uninitialized or freed */
1391 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1392 lnet_rtrpool_free_bufs(&rtrp[LNET_TINY_BUF_IDX], i);
1393 lnet_rtrpool_free_bufs(&rtrp[LNET_SMALL_BUF_IDX], i);
1394 lnet_rtrpool_free_bufs(&rtrp[LNET_LARGE_BUF_IDX], i);
1398 cfs_percpt_free(the_lnet.ln_rtrpools);
1399 the_lnet.ln_rtrpools = NULL;
1404 lnet_nrb_tiny_calculate(void)
1406 int nrbs = LNET_NRB_TINY;
1408 if (tiny_router_buffers < 0) {
1409 LCONSOLE_ERROR_MSG(0x10c,
1410 "tiny_router_buffers=%d invalid when "
1411 "routing enabled\n", tiny_router_buffers);
1415 if (tiny_router_buffers > 0) {
1416 if (tiny_router_buffers < LNET_NRB_TINY_MIN)
1417 CWARN("tiny_router_buffers=%d less than recommended minimum %d\n",
1418 tiny_router_buffers, LNET_NRB_TINY_MIN);
1419 nrbs = tiny_router_buffers;
1422 nrbs /= LNET_CPT_NUMBER;
1423 return max(nrbs, 1);
1427 lnet_nrb_small_calculate(void)
1429 int nrbs = LNET_NRB_SMALL;
1431 if (small_router_buffers < 0) {
1432 LCONSOLE_ERROR_MSG(0x10c,
1433 "small_router_buffers=%d invalid when "
1434 "routing enabled\n", small_router_buffers);
1438 if (small_router_buffers > 0) {
1439 if (small_router_buffers < LNET_NRB_SMALL_MIN)
1440 CWARN("small_router_buffers=%d less than recommended minimum %d\n",
1441 small_router_buffers, LNET_NRB_SMALL_MIN);
1442 nrbs = small_router_buffers;
1445 nrbs /= LNET_CPT_NUMBER;
1446 return max(nrbs, 1);
1450 lnet_nrb_large_calculate(void)
1452 int nrbs = LNET_NRB_LARGE;
1454 if (large_router_buffers < 0) {
1455 LCONSOLE_ERROR_MSG(0x10c,
1456 "large_router_buffers=%d invalid when "
1457 "routing enabled\n", large_router_buffers);
1461 if (large_router_buffers > 0) {
1462 if (large_router_buffers < LNET_NRB_LARGE_MIN)
1463 CWARN("large_router_buffers=%d less than recommended minimum %d\n",
1464 large_router_buffers, LNET_NRB_LARGE_MIN);
1465 nrbs = large_router_buffers;
1468 nrbs /= LNET_CPT_NUMBER;
1469 return max(nrbs, 1);
1473 lnet_rtrpools_alloc(int im_a_router)
1475 struct lnet_rtrbufpool *rtrp;
1482 if (!strcmp(forwarding, "")) {
1483 /* not set either way */
1486 } else if (!strcmp(forwarding, "disabled")) {
1487 /* explicitly disabled */
1489 } else if (!strcmp(forwarding, "enabled")) {
1490 /* explicitly enabled */
1493 LCONSOLE_ERROR_MSG(0x10b,
1494 "lnet: forwarding='%s' not set to either 'enabled' or 'disabled': rc = %d\n",
1499 nrb_tiny = lnet_nrb_tiny_calculate();
1503 nrb_small = lnet_nrb_small_calculate();
1507 nrb_large = lnet_nrb_large_calculate();
1511 the_lnet.ln_rtrpools = cfs_percpt_alloc(lnet_cpt_table(),
1513 sizeof(struct lnet_rtrbufpool));
1514 if (the_lnet.ln_rtrpools == NULL) {
1516 LCONSOLE_ERROR_MSG(0x10c,
1517 "lnet: error allocating router buffer pool: rc = %d\n",
1522 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1523 lnet_rtrpool_init(&rtrp[LNET_TINY_BUF_IDX], 0);
1524 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_TINY_BUF_IDX],
1529 lnet_rtrpool_init(&rtrp[LNET_SMALL_BUF_IDX],
1530 LNET_NRB_SMALL_PAGES);
1531 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_SMALL_BUF_IDX],
1536 lnet_rtrpool_init(&rtrp[LNET_LARGE_BUF_IDX],
1537 LNET_NRB_LARGE_PAGES);
1538 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_LARGE_BUF_IDX],
1544 lnet_net_lock(LNET_LOCK_EX);
1545 the_lnet.ln_routing = 1;
1546 lnet_net_unlock(LNET_LOCK_EX);
1547 complete(&the_lnet.ln_mt_wait_complete);
1551 lnet_rtrpools_free(0);
1556 lnet_rtrpools_adjust_helper(int tiny, int small, int large)
1561 struct lnet_rtrbufpool *rtrp;
1563 /* If the provided values for each buffer pool are different than the
1564 * configured values, we need to take action. */
1566 tiny_router_buffers = tiny;
1567 nrb = lnet_nrb_tiny_calculate();
1568 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1569 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_TINY_BUF_IDX],
1576 small_router_buffers = small;
1577 nrb = lnet_nrb_small_calculate();
1578 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1579 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_SMALL_BUF_IDX],
1586 large_router_buffers = large;
1587 nrb = lnet_nrb_large_calculate();
1588 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1589 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_LARGE_BUF_IDX],
1600 lnet_rtrpools_adjust(int tiny, int small, int large)
1602 /* this function doesn't revert the changes if adding new buffers
1603 * failed. It's up to the user space caller to revert the
1606 if (!the_lnet.ln_routing)
1609 return lnet_rtrpools_adjust_helper(tiny, small, large);
1613 lnet_rtrpools_enable(void)
1617 if (the_lnet.ln_routing)
1620 if (the_lnet.ln_rtrpools == NULL)
1621 /* If routing is turned off, and we have never
1622 * initialized the pools before, just call the
1623 * standard buffer pool allocation routine as
1624 * if we are just configuring this for the first
1626 rc = lnet_rtrpools_alloc(1);
1628 rc = lnet_rtrpools_adjust_helper(0, 0, 0);
1632 lnet_net_lock(LNET_LOCK_EX);
1633 the_lnet.ln_routing = 1;
1635 the_lnet.ln_ping_target->pb_info.pi_features &=
1636 ~LNET_PING_FEAT_RTE_DISABLED;
1637 lnet_net_unlock(LNET_LOCK_EX);
1639 if (lnet_peer_discovery_disabled)
1640 CWARN("Consider turning discovery on to enable full "
1641 "Multi-Rail routing functionality\n");
1647 lnet_rtrpools_disable(void)
1649 if (!the_lnet.ln_routing)
1652 lnet_net_lock(LNET_LOCK_EX);
1653 the_lnet.ln_routing = 0;
1654 the_lnet.ln_ping_target->pb_info.pi_features |=
1655 LNET_PING_FEAT_RTE_DISABLED;
1657 tiny_router_buffers = 0;
1658 small_router_buffers = 0;
1659 large_router_buffers = 0;
1660 lnet_net_unlock(LNET_LOCK_EX);
1661 lnet_rtrpools_free(1);
1665 lnet_notify_peer_down(struct lnet_ni *ni, struct lnet_nid *nid)
1667 if (ni->ni_net->net_lnd->lnd_notify_peer_down != NULL)
1668 (ni->ni_net->net_lnd->lnd_notify_peer_down)(nid);
1672 * ni: local NI used to communicate with the peer
1674 * alive: true if peer is alive, false otherwise
1675 * reset: reset health value. This is requested by the LND.
1676 * when: notificaiton time.
1679 lnet_notify(struct lnet_ni *ni, struct lnet_nid *nid, bool alive, bool reset,
1682 struct lnet_peer_ni *lpni = NULL;
1683 struct lnet_route *route;
1684 struct lnet_peer *lp;
1685 time64_t now = ktime_get_seconds();
1688 LASSERT(!in_interrupt());
1690 CDEBUG(D_NET, "%s notifying %s: %s\n",
1691 (ni == NULL) ? "userspace" : libcfs_nidstr(&ni->ni_nid),
1692 libcfs_nidstr(nid), alive ? "up" : "down");
1695 LNET_NID_NET(&ni->ni_nid) != LNET_NID_NET(nid)) {
1696 CWARN("Ignoring notification of %s %s by %s (different net)\n",
1697 libcfs_nidstr(nid), alive ? "birth" : "death",
1698 libcfs_nidstr(&ni->ni_nid));
1702 /* can't do predictions... */
1704 CWARN("Ignoring prediction from %s of %s %s %lld seconds in the future\n",
1705 ni ? libcfs_nidstr(&ni->ni_nid) : "userspace",
1706 libcfs_nidstr(nid), alive ? "up" : "down", when - now);
1710 if (ni != NULL && !alive && /* LND telling me she's down */
1711 !auto_down) { /* auto-down disabled */
1712 CDEBUG(D_NET, "Auto-down disabled\n");
1716 /* must lock 0 since this is used for synchronization */
1719 if (the_lnet.ln_state != LNET_STATE_RUNNING) {
1724 lpni = lnet_peer_ni_find_locked(nid);
1728 CDEBUG(D_NET, "%s not found\n", libcfs_nidstr(nid));
1734 lpni->lpni_ns_status = LNET_NI_STATUS_UP;
1735 lnet_set_lpni_healthv_locked(lpni,
1736 LNET_MAX_HEALTH_VALUE);
1738 lnet_inc_lpni_healthv_locked(lpni);
1741 lpni->lpni_ns_status = LNET_NI_STATUS_DOWN;
1744 /* recalculate aliveness */
1745 alive = lnet_is_peer_ni_alive(lpni);
1747 lp = lpni->lpni_peer_net->lpn_peer;
1748 /* If this is an LNet router then update route aliveness */
1749 if (lp->lp_rtr_refcount) {
1751 /* reset flag indicates gateway peer went up or down */
1752 lp->lp_alive = alive;
1754 /* If discovery is disabled, locally or on the gateway, then
1755 * any routes using lpni as next-hop need to be updated
1757 * NB: We can get many notifications while a route is down, so
1758 * we try and avoid the expensive net_lock/EX here for the
1759 * common case of receiving duplicate lnet_notify() calls (i.e.
1760 * only grab EX lock when we actually need to update the route
1763 if (lnet_is_discovery_disabled(lp)) {
1764 list_for_each_entry(route, &lp->lp_routes, lr_gwlist) {
1765 if (nid_same(&route->lr_nid, &lpni->lpni_nid))
1766 lnet_set_route_aliveness(route, alive);
1773 if (ni != NULL && !alive)
1774 lnet_notify_peer_down(ni, &lpni->lpni_nid);
1776 cpt = lpni->lpni_cpt;
1778 lnet_peer_ni_decref_locked(lpni);
1779 lnet_net_unlock(cpt);
1783 EXPORT_SYMBOL(lnet_notify);