X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lnet%2Flnet%2Frouter.c;h=4aefdf149134bcf3f685d68c2fb3737ac8c29e59;hb=680da7444a061398ff30dd11fa514c241b64507e;hp=511847a0a587729e35c41080a32c1a092f3303fc;hpb=054ecbba403b2be103be7418289c87e1658cdbd6;p=fs%2Flustre-release.git diff --git a/lnet/lnet/router.c b/lnet/lnet/router.c index 511847a..4aefdf1 100644 --- a/lnet/lnet/router.c +++ b/lnet/lnet/router.c @@ -1,9 +1,9 @@ /* * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * - * Copyright (c) 2011, 2015, Intel Corporation. + * Copyright (c) 2011, 2017, Intel Corporation. * - * This file is part of Lustre, https://wiki.hpdd.intel.com/ + * This file is part of Lustre, https://wiki.whamcloud.com/ * * Portals is free software; you can redistribute it and/or * modify it under the terms of version 2 of the GNU General Public @@ -21,6 +21,8 @@ */ #define DEBUG_SUBSYSTEM S_LNET + +#include #include #define LNET_NRB_TINY_MIN 512 /* min value for each CPT */ @@ -55,22 +57,19 @@ module_param(auto_down, int, 0444); MODULE_PARM_DESC(auto_down, "Automatically mark peers down on comms error"); int -lnet_peer_buffer_credits(lnet_ni_t *ni) +lnet_peer_buffer_credits(struct lnet_net *net) { /* NI option overrides LNet default */ - if (ni->ni_peerrtrcredits > 0) - return ni->ni_peerrtrcredits; + if (net->net_tunables.lct_peer_rtr_credits > 0) + return net->net_tunables.lct_peer_rtr_credits; if (peer_buffer_credits > 0) return peer_buffer_credits; /* As an approximation, allow this peer the same number of router * buffers as it is allowed outstanding sends */ - return ni->ni_peertxcredits; + return net->net_tunables.lct_peer_tx_credits; } -/* forward ref's */ -static int lnet_router_checker(void *); - static int check_routers_before_use; module_param(check_routers_before_use, int, 0444); MODULE_PARM_DESC(check_routers_before_use, "Assume routers are down and ping them before use"); @@ -98,93 +97,106 @@ lnet_peers_start_down(void) } void -lnet_notify_locked(lnet_peer_t *lp, int notifylnd, int alive, cfs_time_t when) +lnet_notify_locked(struct lnet_peer_ni *lp, int notifylnd, int alive, + time64_t when) { - if (cfs_time_before(when, lp->lp_timestamp)) { /* out of date information */ + if (lp->lpni_timestamp > when) { /* out of date information */ CDEBUG(D_NET, "Out of date\n"); return; } - lp->lp_timestamp = when; /* update timestamp */ - lp->lp_ping_deadline = 0; /* disable ping timeout */ + /* + * This function can be called with different cpt locks being + * held. lpni_alive_count modification needs to be properly protected. + * Significant reads to lpni_alive_count are also protected with + * the same lock + */ + spin_lock(&lp->lpni_lock); + + lp->lpni_timestamp = when; /* update timestamp */ - if (lp->lp_alive_count != 0 && /* got old news */ - (!lp->lp_alive) == (!alive)) { /* new date for old news */ + if (lp->lpni_alive_count != 0 && /* got old news */ + (!lp->lpni_alive) == (!alive)) { /* new date for old news */ + spin_unlock(&lp->lpni_lock); CDEBUG(D_NET, "Old news\n"); return; } /* Flag that notification is outstanding */ - lp->lp_alive_count++; - lp->lp_alive = !(!alive); /* 1 bit! */ - lp->lp_notify = 1; - lp->lp_notifylnd |= notifylnd; - if (lp->lp_alive) - lp->lp_ping_feats = LNET_PING_FEAT_INVAL; /* reset */ + lp->lpni_alive_count++; + lp->lpni_alive = (alive) ? 1 : 0; + lp->lpni_notify = 1; + lp->lpni_notifylnd = notifylnd; + if (lp->lpni_alive) + lp->lpni_ping_feats = LNET_PING_FEAT_INVAL; /* reset */ - CDEBUG(D_NET, "set %s %d\n", libcfs_nid2str(lp->lp_nid), alive); + spin_unlock(&lp->lpni_lock); + + CDEBUG(D_NET, "set %s %d\n", libcfs_nid2str(lp->lpni_nid), alive); } +/* + * This function will always be called with lp->lpni_cpt lock held. + */ static void -lnet_ni_notify_locked(lnet_ni_t *ni, lnet_peer_t *lp) +lnet_ni_notify_locked(struct lnet_ni *ni, struct lnet_peer_ni *lp) { - int alive; - int notifylnd; + int alive; + int notifylnd; /* Notify only in 1 thread at any time to ensure ordered notification. * NB individual events can be missed; the only guarantee is that you * always get the most recent news */ - if (lp->lp_notifying || ni == NULL) + spin_lock(&lp->lpni_lock); + + if (lp->lpni_notifying || ni == NULL) { + spin_unlock(&lp->lpni_lock); return; + } - lp->lp_notifying = 1; + lp->lpni_notifying = 1; - while (lp->lp_notify) { - alive = lp->lp_alive; - notifylnd = lp->lp_notifylnd; + /* + * lp->lpni_notify needs to be protected because it can be set in + * lnet_notify_locked(). + */ + while (lp->lpni_notify) { + alive = lp->lpni_alive; + notifylnd = lp->lpni_notifylnd; - lp->lp_notifylnd = 0; - lp->lp_notify = 0; + lp->lpni_notifylnd = 0; + lp->lpni_notify = 0; - if (notifylnd && ni->ni_lnd->lnd_notify != NULL) { - lnet_net_unlock(lp->lp_cpt); + if (notifylnd && ni->ni_net->net_lnd->lnd_notify != NULL) { + spin_unlock(&lp->lpni_lock); + lnet_net_unlock(lp->lpni_cpt); /* A new notification could happen now; I'll handle it * when control returns to me */ - (ni->ni_lnd->lnd_notify)(ni, lp->lp_nid, alive); + (ni->ni_net->net_lnd->lnd_notify)(ni, lp->lpni_nid, + alive); - lnet_net_lock(lp->lp_cpt); + lnet_net_lock(lp->lpni_cpt); + spin_lock(&lp->lpni_lock); } } - lp->lp_notifying = 0; + lp->lpni_notifying = 0; + spin_unlock(&lp->lpni_lock); } - static void -lnet_rtr_addref_locked(lnet_peer_t *lp) +lnet_rtr_addref_locked(struct lnet_peer *lp) { - LASSERT(lp->lp_refcount > 0); LASSERT(lp->lp_rtr_refcount >= 0); /* lnet_net_lock must be exclusively locked */ lp->lp_rtr_refcount++; if (lp->lp_rtr_refcount == 1) { - struct list_head *pos; - - /* a simple insertion sort */ - list_for_each_prev(pos, &the_lnet.ln_routers) { - lnet_peer_t *rtr = list_entry(pos, lnet_peer_t, - lp_rtr_list); - - if (rtr->lp_nid < lp->lp_nid) - break; - } - - list_add(&lp->lp_rtr_list, pos); + list_add_tail(&lp->lp_rtr_list, &the_lnet.ln_routers); /* addref for the_lnet.ln_routers */ lnet_peer_addref_locked(lp); the_lnet.ln_routers_version++; @@ -192,9 +204,9 @@ lnet_rtr_addref_locked(lnet_peer_t *lp) } static void -lnet_rtr_decref_locked(lnet_peer_t *lp) +lnet_rtr_decref_locked(struct lnet_peer *lp) { - LASSERT(lp->lp_refcount > 0); + LASSERT(atomic_read(&lp->lp_refcount) > 0); LASSERT(lp->lp_rtr_refcount > 0); /* lnet_net_lock must be exclusively locked */ @@ -202,12 +214,6 @@ lnet_rtr_decref_locked(lnet_peer_t *lp) if (lp->lp_rtr_refcount == 0) { LASSERT(list_empty(&lp->lp_routes)); - if (lp->lp_rcd != NULL) { - list_add(&lp->lp_rcd->rcd_list, - &the_lnet.ln_rcd_deathrow); - lp->lp_rcd = NULL; - } - list_del(&lp->lp_rtr_list); /* decref for the_lnet.ln_routers */ lnet_peer_decref_locked(lp); @@ -215,18 +221,18 @@ lnet_rtr_decref_locked(lnet_peer_t *lp) } } -lnet_remotenet_t * -lnet_find_net_locked (__u32 net) +struct lnet_remotenet * +lnet_find_rnet_locked(__u32 net) { - lnet_remotenet_t *rnet; + struct lnet_remotenet *rnet; struct list_head *tmp; struct list_head *rn_list; - LASSERT(!the_lnet.ln_shutdown); + LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING); rn_list = lnet_net2rnethash(net); list_for_each(tmp, rn_list) { - rnet = list_entry(tmp, lnet_remotenet_t, lrn_list); + rnet = list_entry(tmp, struct lnet_remotenet, lrn_list); if (rnet->lrn_net == net) return rnet; @@ -237,48 +243,38 @@ lnet_find_net_locked (__u32 net) static void lnet_shuffle_seed(void) { static int seeded; - __u32 lnd_type; - __u32 seed[2]; - struct timeval tv; - lnet_ni_t *ni; - struct list_head *tmp; + struct lnet_ni *ni = NULL; if (seeded) return; - cfs_get_random_bytes(seed, sizeof(seed)); - /* Nodes with small feet have little entropy * the NID for this node gives the most entropy in the low bits */ - list_for_each(tmp, &the_lnet.ln_nis) { - ni = list_entry(tmp, lnet_ni_t, ni_list); - lnd_type = LNET_NETTYP(LNET_NIDNET(ni->ni_nid)); + while ((ni = lnet_get_next_ni_locked(NULL, ni))) + add_device_randomness(&ni->ni_nid, sizeof(ni->ni_nid)); - if (lnd_type != LOLND) - seed[0] ^= (LNET_NIDADDR(ni->ni_nid) | lnd_type); - } - - do_gettimeofday(&tv); - cfs_srand(tv.tv_sec ^ seed[0], tv.tv_usec ^ seed[1]); seeded = 1; return; } /* NB expects LNET_LOCK held */ static void -lnet_add_route_to_rnet(lnet_remotenet_t *rnet, lnet_route_t *route) +lnet_add_route_to_rnet(struct lnet_remotenet *rnet, struct lnet_route *route) { - unsigned int len = 0; - unsigned int offset = 0; + unsigned int len = 0; + unsigned int offset = 0; struct list_head *e; lnet_shuffle_seed(); - list_for_each(e, &rnet->lrn_routes) { + list_for_each(e, &rnet->lrn_routes) len++; - } - /* len+1 positions to add a new entry, also prevents division by 0 */ + /* + * Randomly adding routes to the list is done to ensure that when + * different nodes are using the same list of routers, they end up + * preferring different routers. + */ offset = cfs_rand() % (len + 1); list_for_each(e, &rnet->lrn_routes) { if (offset == 0) @@ -286,9 +282,18 @@ lnet_add_route_to_rnet(lnet_remotenet_t *rnet, lnet_route_t *route) offset--; } list_add(&route->lr_list, e); - list_add(&route->lr_gwlist, &route->lr_gateway->lp_routes); + /* + * force a router check on the gateway to make sure the route is + * alive + */ + route->lr_gateway->lp_rtrcheck_timestamp = 0; the_lnet.ln_remote_nets_version++; + + /* add the route on the gateway list */ + list_add(&route->lr_gwlist, &route->lr_gateway->lp_routes); + + /* take a router reference count on the gateway */ lnet_rtr_addref_locked(route->lr_gateway); } @@ -296,15 +301,16 @@ int lnet_add_route(__u32 net, __u32 hops, lnet_nid_t gateway, unsigned int priority) { - struct list_head *e; - lnet_remotenet_t *rnet; - lnet_remotenet_t *rnet2; - lnet_route_t *route; - lnet_ni_t *ni; - int add_route; - int rc; - - CDEBUG(D_NET, "Add route: net %s hops %d priority %u gw %s\n", + struct list_head *route_entry; + struct lnet_remotenet *rnet; + struct lnet_remotenet *rnet2; + struct lnet_route *route; + struct lnet_peer_ni *lpni; + struct lnet_peer *gw; + int add_route; + int rc; + + CDEBUG(D_NET, "Add route: remote net %s hops %d priority %u gw %s\n", libcfs_net2str(net), hops, priority, libcfs_nid2str(gateway)); if (gateway == LNET_NID_ANY || @@ -315,7 +321,8 @@ lnet_add_route(__u32 net, __u32 hops, lnet_nid_t gateway, (hops != LNET_UNDEFINED_HOPS && (hops < 1 || hops > 255))) return -EINVAL; - if (lnet_islocalnet(net)) /* it's a local network */ + /* it's a local network */ + if (lnet_islocalnet(net)) return -EEXIST; /* Assume net, route, all new */ @@ -333,30 +340,38 @@ lnet_add_route(__u32 net, __u32 hops, lnet_nid_t gateway, INIT_LIST_HEAD(&rnet->lrn_routes); rnet->lrn_net = net; - route->lr_hops = hops; + /* store the local and remote net that the route represents */ + route->lr_lnet = LNET_NIDNET(gateway); route->lr_net = net; route->lr_priority = priority; + route->lr_hops = hops; lnet_net_lock(LNET_LOCK_EX); - rc = lnet_nid2peer_locked(&route->lr_gateway, gateway, LNET_LOCK_EX); - if (rc != 0) { + /* + * lnet_nid2peerni_ex() grabs a ref on the lpni. We will need to + * lose that once we're done + */ + lpni = lnet_nid2peerni_ex(gateway, LNET_LOCK_EX); + if (IS_ERR(lpni)) { lnet_net_unlock(LNET_LOCK_EX); LIBCFS_FREE(route, sizeof(*route)); LIBCFS_FREE(rnet, sizeof(*rnet)); - if (rc == -EHOSTUNREACH) /* gateway is not on a local net. */ - return rc; /* ignore the route entry */ + rc = PTR_ERR(lpni); CERROR("Error %d creating route %s %d %s\n", rc, libcfs_net2str(net), hops, libcfs_nid2str(gateway)); return rc; } - LASSERT(!the_lnet.ln_shutdown); + LASSERT(lpni->lpni_peer_net && lpni->lpni_peer_net->lpn_peer); + gw = lpni->lpni_peer_net->lpn_peer; - rnet2 = lnet_find_net_locked(net); + route->lr_gateway = gw; + + rnet2 = lnet_find_rnet_locked(net); if (rnet2 == NULL) { /* new network */ list_add_tail(&rnet->lrn_list, lnet_net2rnethash(net)); @@ -365,34 +380,34 @@ lnet_add_route(__u32 net, __u32 hops, lnet_nid_t gateway, /* Search for a duplicate route (it's a NOOP if it is) */ add_route = 1; - list_for_each(e, &rnet2->lrn_routes) { - lnet_route_t *route2 = list_entry(e, lnet_route_t, lr_list); + list_for_each(route_entry, &rnet2->lrn_routes) { + struct lnet_route *route2; + route2 = list_entry(route_entry, struct lnet_route, lr_list); if (route2->lr_gateway == route->lr_gateway) { add_route = 0; break; } /* our lookups must be true */ - LASSERT(route2->lr_gateway->lp_nid != gateway); - } - - if (add_route) { - lnet_peer_addref_locked(route->lr_gateway); /* +1 for notify */ + LASSERT(route2->lr_gateway->lp_primary_nid != gateway); + } + + /* + * It is possible to add multiple routes through the same peer, + * but it'll be using a different NID of that peer. When the + * gateway is discovered, discovery will consolidate the different + * peers into one peer. In this case the discovery code will have + * to move the routes from the peer that's being deleted to the + * consolidated peer lp_routes list + */ + if (add_route) lnet_add_route_to_rnet(rnet2, route); - ni = route->lr_gateway->lp_ni; - lnet_net_unlock(LNET_LOCK_EX); - - /* XXX Assume alive */ - if (ni->ni_lnd->lnd_notify != NULL) - (ni->ni_lnd->lnd_notify)(ni, gateway, 1); - - lnet_net_lock(LNET_LOCK_EX); - } - - /* -1 for notify or !add_route */ - lnet_peer_decref_locked(route->lr_gateway); + /* + * get rid of the reference on the lpni. + */ + lnet_peer_ni_decref_locked(lpni); lnet_net_unlock(LNET_LOCK_EX); rc = 0; @@ -405,80 +420,57 @@ lnet_add_route(__u32 net, __u32 hops, lnet_nid_t gateway, if (rnet != rnet2) LIBCFS_FREE(rnet, sizeof(*rnet)); - /* indicate to startup the router checker if configured */ - wake_up(&the_lnet.ln_rc_waitq); + /* kick start the monitor thread to handle the added route */ + wake_up(&the_lnet.ln_mt_waitq); return rc; } -int -lnet_check_routes(void) +static void +lnet_del_route_from_rnet(lnet_nid_t gw_nid, struct list_head *route_list, + struct list_head *zombies) { - lnet_remotenet_t *rnet; - lnet_route_t *route; - lnet_route_t *route2; - struct list_head *e1; - struct list_head *e2; - int cpt; - struct list_head *rn_list; - int i; - - cpt = lnet_net_lock_current(); - - for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) { - rn_list = &the_lnet.ln_remote_nets_hash[i]; - list_for_each(e1, rn_list) { - rnet = list_entry(e1, lnet_remotenet_t, lrn_list); - - route2 = NULL; - list_for_each(e2, &rnet->lrn_routes) { - lnet_nid_t nid1; - lnet_nid_t nid2; - int net; - - route = list_entry(e2, lnet_route_t, - lr_list); - - if (route2 == NULL) { - route2 = route; - continue; - } - - if (route->lr_gateway->lp_ni == - route2->lr_gateway->lp_ni) - continue; - - nid1 = route->lr_gateway->lp_nid; - nid2 = route2->lr_gateway->lp_nid; - net = rnet->lrn_net; + struct lnet_peer *gateway; + struct lnet_route *route; + struct lnet_route *tmp; + + list_for_each_entry_safe(route, tmp, route_list, lr_list) { + gateway = route->lr_gateway; + if (gw_nid != LNET_NID_ANY && + gw_nid != gateway->lp_primary_nid) + continue; - lnet_net_unlock(cpt); + /* + * move to zombie to delete outside the lock + * Note that this function is called with the + * ln_api_mutex held as well as the exclusive net + * lock. Adding to the remote net list happens + * under the same conditions. Same goes for the + * gateway router list + */ + list_move(&route->lr_list, zombies); + the_lnet.ln_remote_nets_version++; - CERROR("Routes to %s via %s and %s not " - "supported\n", - libcfs_net2str(net), - libcfs_nid2str(nid1), - libcfs_nid2str(nid2)); - return -EINVAL; - } - } + list_del(&route->lr_gwlist); + lnet_rtr_decref_locked(gateway); } - - lnet_net_unlock(cpt); - return 0; } int lnet_del_route(__u32 net, lnet_nid_t gw_nid) { - struct lnet_peer *gateway; - lnet_remotenet_t *rnet; - lnet_route_t *route; - struct list_head *e1; - struct list_head *e2; - int rc = -ENOENT; - struct list_head *rn_list; - int idx = 0; + struct list_head rnet_zombies; + struct lnet_remotenet *rnet; + struct lnet_remotenet *tmp; + struct list_head *rn_list; + struct lnet_peer_ni *lpni; + struct lnet_route *route; + struct list_head zombies; + struct lnet_peer *lp; + int i = 0; + + INIT_LIST_HEAD(&rnet_zombies); + INIT_LIST_HEAD(&zombies); CDEBUG(D_NET, "Del route: net %s : gw %s\n", libcfs_net2str(net), libcfs_nid2str(gw_nid)); @@ -487,60 +479,56 @@ lnet_del_route(__u32 net, lnet_nid_t gw_nid) * or a specific route entry actual NIDs) */ lnet_net_lock(LNET_LOCK_EX); - if (net == LNET_NIDNET(LNET_NID_ANY)) - rn_list = &the_lnet.ln_remote_nets_hash[0]; - else - rn_list = lnet_net2rnethash(net); - -again: - list_for_each(e1, rn_list) { - rnet = list_entry(e1, lnet_remotenet_t, lrn_list); - if (!(net == LNET_NIDNET(LNET_NID_ANY) || - net == rnet->lrn_net)) - continue; - - list_for_each(e2, &rnet->lrn_routes) { - route = list_entry(e2, lnet_route_t, lr_list); + lpni = lnet_find_peer_ni_locked(gw_nid); + if (lpni) { + lp = lpni->lpni_peer_net->lpn_peer; + LASSERT(lp); + gw_nid = lp->lp_primary_nid; + lnet_peer_ni_decref_locked(lpni); + } - gateway = route->lr_gateway; - if (!(gw_nid == LNET_NID_ANY || - gw_nid == gateway->lp_nid)) - continue; + if (net != LNET_NIDNET(LNET_NID_ANY)) { + rnet = lnet_find_rnet_locked(net); + if (!rnet) { + lnet_net_unlock(LNET_LOCK_EX); + return -ENOENT; + } + lnet_del_route_from_rnet(gw_nid, &rnet->lrn_routes, + &zombies); + if (list_empty(&rnet->lrn_routes)) + list_move(&rnet->lrn_list, &rnet_zombies); + goto delete_zombies; + } - list_del(&route->lr_list); - list_del(&route->lr_gwlist); - the_lnet.ln_remote_nets_version++; + for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) { + rn_list = &the_lnet.ln_remote_nets_hash[i]; + list_for_each_entry_safe(rnet, tmp, rn_list, lrn_list) { + lnet_del_route_from_rnet(gw_nid, &rnet->lrn_routes, + &zombies); if (list_empty(&rnet->lrn_routes)) - list_del(&rnet->lrn_list); - else - rnet = NULL; - - lnet_rtr_decref_locked(gateway); - lnet_peer_decref_locked(gateway); - - lnet_net_unlock(LNET_LOCK_EX); - - LIBCFS_FREE(route, sizeof(*route)); + list_move(&rnet->lrn_list, &rnet_zombies); + } + } - if (rnet != NULL) - LIBCFS_FREE(rnet, sizeof(*rnet)); +delete_zombies: + lnet_net_unlock(LNET_LOCK_EX); - rc = 0; - lnet_net_lock(LNET_LOCK_EX); - goto again; - } + while (!list_empty(&zombies)) { + route = list_first_entry(&zombies, struct lnet_route, lr_list); + list_del(&route->lr_list); + LIBCFS_FREE(route, sizeof(*route)); } - if (net == LNET_NIDNET(LNET_NID_ANY) && - ++idx < LNET_REMOTE_NETS_HASH_SIZE) { - rn_list = &the_lnet.ln_remote_nets_hash[idx]; - goto again; + while (!list_empty(&rnet_zombies)) { + rnet = list_first_entry(&rnet_zombies, struct lnet_remotenet, + lrn_list); + list_del(&rnet->lrn_list); + LIBCFS_FREE(rnet, sizeof(*rnet)); } - lnet_net_unlock(LNET_LOCK_EX); - return rc; + return 0; } void @@ -549,33 +537,29 @@ lnet_destroy_routes (void) lnet_del_route(LNET_NIDNET(LNET_NID_ANY), LNET_NID_ANY); } -int lnet_get_rtr_pool_cfg(int idx, struct lnet_ioctl_pool_cfg *pool_cfg) +int lnet_get_rtr_pool_cfg(int cpt, struct lnet_ioctl_pool_cfg *pool_cfg) { - int i, rc = -ENOENT, lidx, j; + struct lnet_rtrbufpool *rbp; + int i, rc = -ENOENT, j; if (the_lnet.ln_rtrpools == NULL) return rc; - for (i = 0; i < LNET_NRBPOOLS; i++) { - lnet_rtrbufpool_t *rbp; - - lnet_net_lock(LNET_LOCK_EX); - lidx = idx; - cfs_percpt_for_each(rbp, j, the_lnet.ln_rtrpools) { - if (lidx-- == 0) { - rc = 0; - pool_cfg->pl_pools[i].pl_npages = - rbp[i].rbp_npages; - pool_cfg->pl_pools[i].pl_nbuffers = - rbp[i].rbp_nbuffers; - pool_cfg->pl_pools[i].pl_credits = - rbp[i].rbp_credits; - pool_cfg->pl_pools[i].pl_mincredits = - rbp[i].rbp_mincredits; - break; - } + + cfs_percpt_for_each(rbp, i, the_lnet.ln_rtrpools) { + if (i != cpt) + continue; + + lnet_net_lock(i); + for (j = 0; j < LNET_NRBPOOLS; j++) { + pool_cfg->pl_pools[j].pl_npages = rbp[j].rbp_npages; + pool_cfg->pl_pools[j].pl_nbuffers = rbp[j].rbp_nbuffers; + pool_cfg->pl_pools[j].pl_credits = rbp[j].rbp_credits; + pool_cfg->pl_pools[j].pl_mincredits = rbp[j].rbp_mincredits; } - lnet_net_unlock(LNET_LOCK_EX); + lnet_net_unlock(i); + rc = 0; + break; } lnet_net_lock(LNET_LOCK_EX); @@ -591,8 +575,8 @@ lnet_get_route(int idx, __u32 *net, __u32 *hops, { struct list_head *e1; struct list_head *e2; - lnet_remotenet_t *rnet; - lnet_route_t *route; + struct lnet_remotenet *rnet; + struct lnet_route *route; int cpt; int i; struct list_head *rn_list; @@ -602,17 +586,17 @@ lnet_get_route(int idx, __u32 *net, __u32 *hops, for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) { rn_list = &the_lnet.ln_remote_nets_hash[i]; list_for_each(e1, rn_list) { - rnet = list_entry(e1, lnet_remotenet_t, lrn_list); + rnet = list_entry(e1, struct lnet_remotenet, lrn_list); list_for_each(e2, &rnet->lrn_routes) { - route = list_entry(e2, lnet_route_t, + route = list_entry(e2, struct lnet_route, lr_list); if (idx-- == 0) { *net = rnet->lrn_net; *hops = route->lr_hops; *priority = route->lr_priority; - *gateway = route->lr_gateway->lp_nid; + *gateway = route->lr_gateway->lp_primary_nid; *alive = lnet_is_route_alive(route); lnet_net_unlock(cpt); return 0; @@ -626,17 +610,21 @@ lnet_get_route(int idx, __u32 *net, __u32 *hops, } void -lnet_swap_pinginfo(lnet_ping_info_t *info) +lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf) { - int i; - lnet_ni_status_t *stat; - - __swab32s(&info->pi_magic); - __swab32s(&info->pi_features); - __swab32s(&info->pi_pid); - __swab32s(&info->pi_nnis); - for (i = 0; i < info->pi_nnis && i < LNET_MAX_RTR_NIS; i++) { - stat = &info->pi_ni[i]; + struct lnet_ni_status *stat; + int nnis; + int i; + + __swab32s(&pbuf->pb_info.pi_magic); + __swab32s(&pbuf->pb_info.pi_features); + __swab32s(&pbuf->pb_info.pi_pid); + __swab32s(&pbuf->pb_info.pi_nnis); + nnis = pbuf->pb_info.pi_nnis; + if (nnis > pbuf->pb_nnis) + nnis = pbuf->pb_nnis; + for (i = 0; i < nnis; i++) { + stat = &pbuf->pb_info.pi_ni[i]; __swab64s(&stat->ns_nid); __swab32s(&stat->ns_status); } @@ -644,106 +632,24 @@ lnet_swap_pinginfo(lnet_ping_info_t *info) } /** - * parse router-checker pinginfo, record number of down NIs for remote - * networks on that router. + * TODO: re-implement */ static void -lnet_parse_rc_info(lnet_rc_data_t *rcd) +lnet_parse_rc_info(struct lnet_rc_data *rcd) { - lnet_ping_info_t *info = rcd->rcd_pinginfo; - struct lnet_peer *gw = rcd->rcd_gateway; - lnet_route_t *rte; - - if (!gw->lp_alive) - return; - - if (info->pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) - lnet_swap_pinginfo(info); - - /* NB always racing with network! */ - if (info->pi_magic != LNET_PROTO_PING_MAGIC) { - CDEBUG(D_NET, "%s: Unexpected magic %08x\n", - libcfs_nid2str(gw->lp_nid), info->pi_magic); - gw->lp_ping_feats = LNET_PING_FEAT_INVAL; - return; - } - - gw->lp_ping_feats = info->pi_features; - if ((gw->lp_ping_feats & LNET_PING_FEAT_MASK) == 0) { - CDEBUG(D_NET, "%s: Unexpected features 0x%x\n", - libcfs_nid2str(gw->lp_nid), gw->lp_ping_feats); - return; /* nothing I can understand */ - } - - if ((gw->lp_ping_feats & LNET_PING_FEAT_NI_STATUS) == 0) - return; /* can't carry NI status info */ - - list_for_each_entry(rte, &gw->lp_routes, lr_gwlist) { - int down = 0; - int up = 0; - int i; - - if ((gw->lp_ping_feats & LNET_PING_FEAT_RTE_DISABLED) != 0) { - rte->lr_downis = 1; - continue; - } - - for (i = 0; i < info->pi_nnis && i < LNET_MAX_RTR_NIS; i++) { - lnet_ni_status_t *stat = &info->pi_ni[i]; - lnet_nid_t nid = stat->ns_nid; - - if (nid == LNET_NID_ANY) { - CDEBUG(D_NET, "%s: unexpected LNET_NID_ANY\n", - libcfs_nid2str(gw->lp_nid)); - gw->lp_ping_feats = LNET_PING_FEAT_INVAL; - return; - } - - if (LNET_NETTYP(LNET_NIDNET(nid)) == LOLND) - continue; - - if (stat->ns_status == LNET_NI_STATUS_DOWN) { - down++; - continue; - } - - if (stat->ns_status == LNET_NI_STATUS_UP) { - if (LNET_NIDNET(nid) == rte->lr_net) { - up = 1; - break; - } - continue; - } - - CDEBUG(D_NET, "%s: Unexpected status 0x%x\n", - libcfs_nid2str(gw->lp_nid), stat->ns_status); - gw->lp_ping_feats = LNET_PING_FEAT_INVAL; - return; - } - - if (up) { /* ignore downed NIs if NI for dest network is up */ - rte->lr_downis = 0; - continue; - } - /* if @down is zero and this route is single-hop, it means - * we can't find NI for target network */ - if (down == 0 && rte->lr_hops == 1) - down = 1; - - rte->lr_downis = down; - } + rcd = rcd; } static void -lnet_router_checker_event(lnet_event_t *event) +lnet_router_checker_event(struct lnet_event *event) { - lnet_rc_data_t *rcd = event->md.user_ptr; - struct lnet_peer *lp; + struct lnet_rc_data *rcd = event->md.user_ptr; + struct lnet_peer_ni *lp; LASSERT(rcd != NULL); if (event->unlinked) { - LNetInvalidateHandle(&rcd->rcd_mdh); + LNetInvalidateMDHandle(&rcd->rcd_mdh); return; } @@ -756,14 +662,13 @@ lnet_router_checker_event(lnet_event_t *event) /* NB: it's called with holding lnet_res_lock, we have a few * places need to hold both locks at the same time, please take * care of lock ordering */ - lnet_net_lock(lp->lp_cpt); - if (!lnet_isrouter(lp) || lp->lp_rcd != rcd) { + lnet_net_lock(lp->lpni_cpt); + if (!lnet_isrouter(lp) || lp->lpni_rcd != rcd) { /* ignore if no longer a router or rcd is replaced */ goto out; } if (event->type == LNET_EVENT_SEND) { - lp->lp_ping_notsent = 0; if (event->status == 0) goto out; } @@ -774,7 +679,7 @@ lnet_router_checker_event(lnet_event_t *event) * we ping alive routers to try to detect router death before * apps get burned). */ - lnet_notify_locked(lp, 1, (event->status == 0), cfs_time_current()); + lnet_notify_locked(lp, 1, !event->status, ktime_get_seconds()); /* The router checker will wake up very shortly and do the * actual notification. * XXX If 'lp' stops being a router before then, it will still @@ -784,29 +689,34 @@ lnet_router_checker_event(lnet_event_t *event) lnet_parse_rc_info(rcd); out: - lnet_net_unlock(lp->lp_cpt); + lnet_net_unlock(lp->lpni_cpt); } static void lnet_wait_known_routerstate(void) { - lnet_peer_t *rtr; + struct lnet_peer *rtr; struct list_head *entry; - int all_known; + int all_known; - LASSERT(the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING); + LASSERT(the_lnet.ln_mt_state == LNET_MT_STATE_RUNNING); for (;;) { int cpt = lnet_net_lock_current(); all_known = 1; list_for_each(entry, &the_lnet.ln_routers) { - rtr = list_entry(entry, lnet_peer_t, lp_rtr_list); + rtr = list_entry(entry, struct lnet_peer, + lp_rtr_list); - if (rtr->lp_alive_count == 0) { + spin_lock(&rtr->lp_lock); + + if ((rtr->lp_state & LNET_PEER_DISCOVERED) == 0) { all_known = 0; + spin_unlock(&rtr->lp_lock); break; } + spin_unlock(&rtr->lp_lock); } lnet_net_unlock(cpt); @@ -819,17 +729,22 @@ lnet_wait_known_routerstate(void) } } +/* TODO: reimplement */ void -lnet_router_ni_update_locked(lnet_peer_t *gw, __u32 net) +lnet_router_ni_update_locked(struct lnet_peer_ni *gw, __u32 net) { - lnet_route_t *rte; + struct lnet_route *rte; + struct lnet_peer *lp; - if ((gw->lp_ping_feats & LNET_PING_FEAT_NI_STATUS) != 0) { - list_for_each_entry(rte, &gw->lp_routes, lr_gwlist) { - if (rte->lr_net == net) { - rte->lr_downis = 0; - break; - } + if ((gw->lpni_ping_feats & LNET_PING_FEAT_NI_STATUS) != 0) + lp = gw->lpni_peer_net->lpn_peer; + else + return; + + list_for_each_entry(rte, &lp->lp_routes, lr_gwlist) { + if (rte->lr_net == net) { + rte->lr_downis = 0; + break; } } } @@ -837,18 +752,18 @@ lnet_router_ni_update_locked(lnet_peer_t *gw, __u32 net) static void lnet_update_ni_status_locked(void) { - lnet_ni_t *ni; - long now; - int timeout; + struct lnet_ni *ni = NULL; + time64_t now; + time64_t timeout; LASSERT(the_lnet.ln_routing); timeout = router_ping_timeout + MAX(live_router_check_interval, dead_router_check_interval); - now = cfs_time_current_sec(); - list_for_each_entry(ni, &the_lnet.ln_nis, ni_list) { - if (ni->ni_lnd->lnd_type == LOLND) + now = ktime_get_real_seconds(); + while ((ni = lnet_get_next_ni_locked(NULL, ni))) { + if (ni->ni_net->net_lnd->lnd_type == LOLND) continue; if (now < ni->ni_last_alive + timeout) @@ -864,7 +779,7 @@ lnet_update_ni_status_locked(void) LASSERT(ni->ni_status != NULL); if (ni->ni_status->ns_status != LNET_NI_STATUS_DOWN) { - CDEBUG(D_NET, "NI(%s:%d) status changed to down\n", + CDEBUG(D_NET, "NI(%s:%lld) status changed to down\n", libcfs_nid2str(ni->ni_nid), timeout); /* NB: so far, this is the only place to set * NI status to "down" */ @@ -874,190 +789,9 @@ lnet_update_ni_status_locked(void) } } -static void -lnet_destroy_rc_data(lnet_rc_data_t *rcd) -{ - LASSERT(list_empty(&rcd->rcd_list)); - /* detached from network */ - LASSERT(LNetHandleIsInvalid(rcd->rcd_mdh)); - - if (rcd->rcd_gateway != NULL) { - int cpt = rcd->rcd_gateway->lp_cpt; - - lnet_net_lock(cpt); - lnet_peer_decref_locked(rcd->rcd_gateway); - lnet_net_unlock(cpt); - } - - if (rcd->rcd_pinginfo != NULL) - LIBCFS_FREE(rcd->rcd_pinginfo, LNET_PINGINFO_SIZE); - - LIBCFS_FREE(rcd, sizeof(*rcd)); -} - -static lnet_rc_data_t * -lnet_create_rc_data_locked(lnet_peer_t *gateway) -{ - lnet_rc_data_t *rcd = NULL; - lnet_ping_info_t *pi; - int rc; - int i; - - lnet_net_unlock(gateway->lp_cpt); - - LIBCFS_ALLOC(rcd, sizeof(*rcd)); - if (rcd == NULL) - goto out; - - LNetInvalidateHandle(&rcd->rcd_mdh); - INIT_LIST_HEAD(&rcd->rcd_list); - - LIBCFS_ALLOC(pi, LNET_PINGINFO_SIZE); - if (pi == NULL) - goto out; - - for (i = 0; i < LNET_MAX_RTR_NIS; i++) { - pi->pi_ni[i].ns_nid = LNET_NID_ANY; - pi->pi_ni[i].ns_status = LNET_NI_STATUS_INVALID; - } - rcd->rcd_pinginfo = pi; - - LASSERT(!LNetHandleIsInvalid(the_lnet.ln_rc_eqh)); - rc = LNetMDBind((lnet_md_t){.start = pi, - .user_ptr = rcd, - .length = LNET_PINGINFO_SIZE, - .threshold = LNET_MD_THRESH_INF, - .options = LNET_MD_TRUNCATE, - .eq_handle = the_lnet.ln_rc_eqh}, - LNET_UNLINK, - &rcd->rcd_mdh); - if (rc < 0) { - CERROR("Can't bind MD: %d\n", rc); - goto out; - } - LASSERT(rc == 0); - - lnet_net_lock(gateway->lp_cpt); - /* router table changed or someone has created rcd for this gateway */ - if (!lnet_isrouter(gateway) || gateway->lp_rcd != NULL) { - lnet_net_unlock(gateway->lp_cpt); - goto out; - } - - lnet_peer_addref_locked(gateway); - rcd->rcd_gateway = gateway; - gateway->lp_rcd = rcd; - gateway->lp_ping_notsent = 0; - - return rcd; - - out: - if (rcd != NULL) { - if (!LNetHandleIsInvalid(rcd->rcd_mdh)) { - rc = LNetMDUnlink(rcd->rcd_mdh); - LASSERT(rc == 0); - } - lnet_destroy_rc_data(rcd); - } - - lnet_net_lock(gateway->lp_cpt); - return gateway->lp_rcd; -} - -static int -lnet_router_check_interval (lnet_peer_t *rtr) -{ - int secs; - - secs = rtr->lp_alive ? live_router_check_interval : - dead_router_check_interval; - if (secs < 0) - secs = 0; - - return secs; -} - -static void -lnet_ping_router_locked (lnet_peer_t *rtr) -{ - lnet_rc_data_t *rcd = NULL; - cfs_time_t now = cfs_time_current(); - int secs; - - lnet_peer_addref_locked(rtr); - - if (rtr->lp_ping_deadline != 0 && /* ping timed out? */ - cfs_time_after(now, rtr->lp_ping_deadline)) - lnet_notify_locked(rtr, 1, 0, now); - - /* Run any outstanding notifications */ - lnet_ni_notify_locked(rtr->lp_ni, rtr); - - if (!lnet_isrouter(rtr) || - the_lnet.ln_rc_state != LNET_RC_STATE_RUNNING) { - /* router table changed or router checker is shutting down */ - lnet_peer_decref_locked(rtr); - return; - } - - rcd = rtr->lp_rcd != NULL ? - rtr->lp_rcd : lnet_create_rc_data_locked(rtr); - - if (rcd == NULL) - return; - - secs = lnet_router_check_interval(rtr); - - CDEBUG(D_NET, - "rtr %s %d: deadline %lu ping_notsent %d alive %d " - "alive_count %d lp_ping_timestamp %lu\n", - libcfs_nid2str(rtr->lp_nid), secs, - rtr->lp_ping_deadline, rtr->lp_ping_notsent, - rtr->lp_alive, rtr->lp_alive_count, rtr->lp_ping_timestamp); - - if (secs != 0 && !rtr->lp_ping_notsent && - cfs_time_after(now, cfs_time_add(rtr->lp_ping_timestamp, - cfs_time_seconds(secs)))) { - int rc; - lnet_process_id_t id; - lnet_handle_md_t mdh; - - id.nid = rtr->lp_nid; - id.pid = LNET_PID_LUSTRE; - CDEBUG(D_NET, "Check: %s\n", libcfs_id2str(id)); - - rtr->lp_ping_notsent = 1; - rtr->lp_ping_timestamp = now; - - mdh = rcd->rcd_mdh; - - if (rtr->lp_ping_deadline == 0) { - rtr->lp_ping_deadline = - cfs_time_shift(router_ping_timeout); - } - - lnet_net_unlock(rtr->lp_cpt); - - rc = LNetGet(LNET_NID_ANY, mdh, id, LNET_RESERVED_PORTAL, - LNET_PROTO_PING_MATCHBITS, 0); - - lnet_net_lock(rtr->lp_cpt); - if (rc != 0) - rtr->lp_ping_notsent = 0; /* no event pending */ - } - - lnet_peer_decref_locked(rtr); - return; -} - -int -lnet_router_checker_start(void) +int lnet_router_pre_mt_start(void) { - int rc; - int eqsz = 0; - struct task_struct *task; - - LASSERT(the_lnet.ln_rc_state == LNET_RC_STATE_SHUTDOWN); + int rc; if (check_routers_before_use && dead_router_check_interval <= 0) { @@ -1067,153 +801,49 @@ lnet_router_checker_start(void) return -EINVAL; } - sema_init(&the_lnet.ln_rc_signal, 0); - rc = LNetEQAlloc(0, lnet_router_checker_event, &the_lnet.ln_rc_eqh); if (rc != 0) { - CERROR("Can't allocate EQ(%d): %d\n", eqsz, rc); + CERROR("Can't allocate EQ(0): %d\n", rc); return -ENOMEM; } - the_lnet.ln_rc_state = LNET_RC_STATE_RUNNING; - task = kthread_run(lnet_router_checker, NULL, "router_checker"); - if (IS_ERR(task)) { - rc = PTR_ERR(task); - CERROR("Can't start router checker thread: %d\n", rc); - /* block until event callback signals exit */ - down(&the_lnet.ln_rc_signal); - rc = LNetEQFree(the_lnet.ln_rc_eqh); - LASSERT(rc == 0); - the_lnet.ln_rc_state = LNET_RC_STATE_SHUTDOWN; - return -ENOMEM; - } + return 0; +} +void lnet_router_post_mt_start(void) +{ if (check_routers_before_use) { /* Note that a helpful side-effect of pinging all known routers * at startup is that it makes them drop stale connections they * may have to a previous instance of me. */ lnet_wait_known_routerstate(); } - - return 0; } void -lnet_router_checker_stop (void) +lnet_router_cleanup(void) { int rc; - if (the_lnet.ln_rc_state == LNET_RC_STATE_SHUTDOWN) - return; - - LASSERT (the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING); - the_lnet.ln_rc_state = LNET_RC_STATE_STOPPING; - /* wakeup the RC thread if it's sleeping */ - wake_up(&the_lnet.ln_rc_waitq); - - /* block until event callback signals exit */ - down(&the_lnet.ln_rc_signal); - LASSERT(the_lnet.ln_rc_state == LNET_RC_STATE_SHUTDOWN); - rc = LNetEQFree(the_lnet.ln_rc_eqh); LASSERT(rc == 0); return; } -static void +void lnet_prune_rc_data(int wait_unlink) { - lnet_rc_data_t *rcd; - lnet_rc_data_t *tmp; - lnet_peer_t *lp; - struct list_head head; - int i = 2; - - if (likely(the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING && - list_empty(&the_lnet.ln_rcd_deathrow) && - list_empty(&the_lnet.ln_rcd_zombie))) - return; - - INIT_LIST_HEAD(&head); - - lnet_net_lock(LNET_LOCK_EX); - - if (the_lnet.ln_rc_state != LNET_RC_STATE_RUNNING) { - /* router checker is stopping, prune all */ - list_for_each_entry(lp, &the_lnet.ln_routers, - lp_rtr_list) { - if (lp->lp_rcd == NULL) - continue; - - LASSERT(list_empty(&lp->lp_rcd->rcd_list)); - list_add(&lp->lp_rcd->rcd_list, - &the_lnet.ln_rcd_deathrow); - lp->lp_rcd = NULL; - } - } - - /* unlink all RCDs on deathrow list */ - list_splice_init(&the_lnet.ln_rcd_deathrow, &head); - - if (!list_empty(&head)) { - lnet_net_unlock(LNET_LOCK_EX); - - list_for_each_entry(rcd, &head, rcd_list) - LNetMDUnlink(rcd->rcd_mdh); - - lnet_net_lock(LNET_LOCK_EX); - } - - list_splice_init(&head, &the_lnet.ln_rcd_zombie); - - /* release all zombie RCDs */ - while (!list_empty(&the_lnet.ln_rcd_zombie)) { - list_for_each_entry_safe(rcd, tmp, &the_lnet.ln_rcd_zombie, - rcd_list) { - if (LNetHandleIsInvalid(rcd->rcd_mdh)) - list_move(&rcd->rcd_list, &head); - } - - wait_unlink = wait_unlink && - !list_empty(&the_lnet.ln_rcd_zombie); - - lnet_net_unlock(LNET_LOCK_EX); - - while (!list_empty(&head)) { - rcd = list_entry(head.next, - lnet_rc_data_t, rcd_list); - list_del_init(&rcd->rcd_list); - lnet_destroy_rc_data(rcd); - } - - if (!wait_unlink) - return; - - i++; - CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET, - "Waiting for rc buffers to unlink\n"); - set_current_state(TASK_UNINTERRUPTIBLE); - schedule_timeout(cfs_time_seconds(1) / 4); - - lnet_net_lock(LNET_LOCK_EX); - } - - lnet_net_unlock(LNET_LOCK_EX); + wait_unlink = wait_unlink; } /* - * This function is called to check if the RC should block indefinitely. - * It's called from lnet_router_checker() as well as being passed to - * wait_event_interruptible() to avoid the lost wake_up problem. - * - * When it's called from wait_event_interruptible() it is necessary to - * also not sleep if the rc state is not running to avoid a deadlock - * when the system is shutting down + * This function is called from the monitor thread to check if there are + * any active routers that need to be checked. */ -static inline bool +inline bool lnet_router_checker_active(void) { - if (the_lnet.ln_rc_state != LNET_RC_STATE_RUNNING) + if (the_lnet.ln_mt_state != LNET_MT_STATE_RUNNING) return true; /* Router Checker thread needs to run when routing is enabled in @@ -1221,84 +851,53 @@ lnet_router_checker_active(void) if (the_lnet.ln_routing) return true; + /* if there are routers that need to be cleaned up then do so */ + if (!list_empty(&the_lnet.ln_rcd_deathrow) || + !list_empty(&the_lnet.ln_rcd_zombie)) + return true; + return !list_empty(&the_lnet.ln_routers) && (live_router_check_interval > 0 || dead_router_check_interval > 0); } -static int -lnet_router_checker(void *arg) +void +lnet_check_routers(void) { - lnet_peer_t *rtr; - struct list_head *entry; - - cfs_block_allsigs(); - - while (the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING) { - __u64 version; - int cpt; - int cpt2; + struct lnet_peer *rtr; + struct list_head *entry; + __u64 version; + int cpt; - cpt = lnet_net_lock_current(); + cpt = lnet_net_lock_current(); rescan: - version = the_lnet.ln_routers_version; + version = the_lnet.ln_routers_version; - list_for_each(entry, &the_lnet.ln_routers) { - rtr = list_entry(entry, lnet_peer_t, lp_rtr_list); - - cpt2 = lnet_cpt_of_nid_locked(rtr->lp_nid); - if (cpt != cpt2) { - lnet_net_unlock(cpt); - cpt = cpt2; - lnet_net_lock(cpt); - /* the routers list has changed */ - if (version != the_lnet.ln_routers_version) - goto rescan; - } + list_for_each(entry, &the_lnet.ln_routers) { + rtr = list_entry(entry, struct lnet_peer, + lp_rtr_list); - lnet_ping_router_locked(rtr); + /* TODO use discovery to determine if router is alive */ - /* NB dropped lock */ - if (version != the_lnet.ln_routers_version) { - /* the routers list has changed */ - goto rescan; - } + /* NB dropped lock */ + if (version != the_lnet.ln_routers_version) { + /* the routers list has changed */ + goto rescan; } - - if (the_lnet.ln_routing) - lnet_update_ni_status_locked(); - - lnet_net_unlock(cpt); - - lnet_prune_rc_data(0); /* don't wait for UNLINK */ - - /* Call schedule_timeout() here always adds 1 to load average - * because kernel counts # active tasks as nr_running - * + nr_uninterruptible. */ - /* if there are any routes then wakeup every second. If - * there are no routes then sleep indefinitely until woken - * up by a user adding a route */ - if (!lnet_router_checker_active()) - wait_event_interruptible(the_lnet.ln_rc_waitq, - lnet_router_checker_active()); - else - wait_event_interruptible_timeout(the_lnet.ln_rc_waitq, - false, - cfs_time_seconds(1)); } - lnet_prune_rc_data(1); /* wait for UNLINK */ + if (the_lnet.ln_routing) + lnet_update_ni_status_locked(); - the_lnet.ln_rc_state = LNET_RC_STATE_SHUTDOWN; - up(&the_lnet.ln_rc_signal); - /* The unlink event callback will signal final completion */ - return 0; + lnet_net_unlock(cpt); + + lnet_prune_rc_data(0); /* don't wait for UNLINK */ } void -lnet_destroy_rtrbuf(lnet_rtrbuf_t *rb, int npages) +lnet_destroy_rtrbuf(struct lnet_rtrbuf *rb, int npages) { - int sz = offsetof(lnet_rtrbuf_t, rb_kiov[npages]); + int sz = offsetof(struct lnet_rtrbuf, rb_kiov[npages]); while (--npages >= 0) __free_page(rb->rb_kiov[npages].kiov_page); @@ -1306,13 +905,13 @@ lnet_destroy_rtrbuf(lnet_rtrbuf_t *rb, int npages) LIBCFS_FREE(rb, sz); } -static lnet_rtrbuf_t * -lnet_new_rtrbuf(lnet_rtrbufpool_t *rbp, int cpt) +static struct lnet_rtrbuf * +lnet_new_rtrbuf(struct lnet_rtrbufpool *rbp, int cpt) { int npages = rbp->rbp_npages; - int sz = offsetof(lnet_rtrbuf_t, rb_kiov[npages]); + int sz = offsetof(struct lnet_rtrbuf, rb_kiov[npages]); struct page *page; - lnet_rtrbuf_t *rb; + struct lnet_rtrbuf *rb; int i; LIBCFS_CPT_ALLOC(rb, lnet_cpt_table(), cpt, sz); @@ -1341,10 +940,10 @@ lnet_new_rtrbuf(lnet_rtrbufpool_t *rbp, int cpt) } static void -lnet_rtrpool_free_bufs(lnet_rtrbufpool_t *rbp, int cpt) +lnet_rtrpool_free_bufs(struct lnet_rtrbufpool *rbp, int cpt) { - int npages = rbp->rbp_npages; - lnet_rtrbuf_t *rb; + int npages = rbp->rbp_npages; + struct lnet_rtrbuf *rb; struct list_head tmp; if (rbp->rbp_nbuffers == 0) /* not initialized or already freed */ @@ -1353,7 +952,8 @@ lnet_rtrpool_free_bufs(lnet_rtrbufpool_t *rbp, int cpt) INIT_LIST_HEAD(&tmp); lnet_net_lock(cpt); - lnet_drop_routed_msgs_locked(&rbp->rbp_msgs, cpt); + list_splice_init(&rbp->rbp_msgs, &tmp); + lnet_drop_routed_msgs_locked(&tmp, cpt); list_splice_init(&rbp->rbp_bufs, &tmp); rbp->rbp_req_nbuffers = 0; rbp->rbp_nbuffers = rbp->rbp_credits = 0; @@ -1362,17 +962,17 @@ lnet_rtrpool_free_bufs(lnet_rtrbufpool_t *rbp, int cpt) /* Free buffers on the free list. */ while (!list_empty(&tmp)) { - rb = list_entry(tmp.next, lnet_rtrbuf_t, rb_list); + rb = list_entry(tmp.next, struct lnet_rtrbuf, rb_list); list_del(&rb->rb_list); lnet_destroy_rtrbuf(rb, npages); } } static int -lnet_rtrpool_adjust_bufs(lnet_rtrbufpool_t *rbp, int nbufs, int cpt) +lnet_rtrpool_adjust_bufs(struct lnet_rtrbufpool *rbp, int nbufs, int cpt) { struct list_head rb_list; - lnet_rtrbuf_t *rb; + struct lnet_rtrbuf *rb; int num_rb; int num_buffers = 0; int old_req_nbufs; @@ -1439,7 +1039,7 @@ lnet_rtrpool_adjust_bufs(lnet_rtrbufpool_t *rbp, int nbufs, int cpt) failed: while (!list_empty(&rb_list)) { - rb = list_entry(rb_list.next, lnet_rtrbuf_t, rb_list); + rb = list_entry(rb_list.next, struct lnet_rtrbuf, rb_list); list_del(&rb->rb_list); lnet_destroy_rtrbuf(rb, npages); } @@ -1448,7 +1048,7 @@ failed: } static void -lnet_rtrpool_init(lnet_rtrbufpool_t *rbp, int npages) +lnet_rtrpool_init(struct lnet_rtrbufpool *rbp, int npages) { INIT_LIST_HEAD(&rbp->rbp_msgs); INIT_LIST_HEAD(&rbp->rbp_bufs); @@ -1461,7 +1061,7 @@ lnet_rtrpool_init(lnet_rtrbufpool_t *rbp, int npages) void lnet_rtrpools_free(int keep_pools) { - lnet_rtrbufpool_t *rtrp; + struct lnet_rtrbufpool *rtrp; int i; if (the_lnet.ln_rtrpools == NULL) /* uninitialized or freed */ @@ -1539,7 +1139,7 @@ lnet_nrb_large_calculate(void) int lnet_rtrpools_alloc(int im_a_router) { - lnet_rtrbufpool_t *rtrp; + struct lnet_rtrbufpool *rtrp; int nrb_tiny; int nrb_small; int nrb_large; @@ -1575,7 +1175,7 @@ lnet_rtrpools_alloc(int im_a_router) the_lnet.ln_rtrpools = cfs_percpt_alloc(lnet_cpt_table(), LNET_NRBPOOLS * - sizeof(lnet_rtrbufpool_t)); + sizeof(struct lnet_rtrbufpool)); if (the_lnet.ln_rtrpools == NULL) { LCONSOLE_ERROR_MSG(0x10c, "Failed to initialize router buffe pool\n"); @@ -1607,6 +1207,7 @@ lnet_rtrpools_alloc(int im_a_router) lnet_net_lock(LNET_LOCK_EX); the_lnet.ln_routing = 1; lnet_net_unlock(LNET_LOCK_EX); + wake_up(&the_lnet.ln_mt_waitq); return 0; failed: @@ -1620,7 +1221,7 @@ lnet_rtrpools_adjust_helper(int tiny, int small, int large) int nrb = 0; int rc = 0; int i; - lnet_rtrbufpool_t *rtrp; + struct lnet_rtrbufpool *rtrp; /* If the provided values for each buffer pool are different than the * configured values, we need to take action. */ @@ -1694,7 +1295,8 @@ lnet_rtrpools_enable(void) lnet_net_lock(LNET_LOCK_EX); the_lnet.ln_routing = 1; - the_lnet.ln_ping_info->pi_features &= ~LNET_PING_FEAT_RTE_DISABLED; + the_lnet.ln_ping_target->pb_info.pi_features &= + ~LNET_PING_FEAT_RTE_DISABLED; lnet_net_unlock(LNET_LOCK_EX); return rc; @@ -1708,7 +1310,8 @@ lnet_rtrpools_disable(void) lnet_net_lock(LNET_LOCK_EX); the_lnet.ln_routing = 0; - the_lnet.ln_ping_info->pi_features |= LNET_PING_FEAT_RTE_DISABLED; + the_lnet.ln_ping_target->pb_info.pi_features |= + LNET_PING_FEAT_RTE_DISABLED; tiny_router_buffers = 0; small_router_buffers = 0; @@ -1718,11 +1321,11 @@ lnet_rtrpools_disable(void) } int -lnet_notify(lnet_ni_t *ni, lnet_nid_t nid, int alive, cfs_time_t when) +lnet_notify(struct lnet_ni *ni, lnet_nid_t nid, int alive, time64_t when) { - struct lnet_peer *lp = NULL; - cfs_time_t now = cfs_time_current(); - int cpt = lnet_cpt_of_nid(nid); + struct lnet_peer_ni *lp = NULL; + time64_t now = ktime_get_seconds(); + int cpt = lnet_cpt_of_nid(nid, ni); LASSERT (!in_interrupt ()); @@ -1740,12 +1343,11 @@ lnet_notify(lnet_ni_t *ni, lnet_nid_t nid, int alive, cfs_time_t when) } /* can't do predictions... */ - if (cfs_time_after(when, now)) { + if (when > now) { CWARN("Ignoring prediction from %s of %s %s " - "%ld seconds in the future\n", + "%lld seconds in the future\n", (ni == NULL) ? "userspace" : libcfs_nid2str(ni->ni_nid), - libcfs_nid2str(nid), alive ? "up" : "down", - cfs_duration_sec(cfs_time_sub(when, now))); + libcfs_nid2str(nid), alive ? "up" : "down", when - now); return -EINVAL; } @@ -1757,12 +1359,12 @@ lnet_notify(lnet_ni_t *ni, lnet_nid_t nid, int alive, cfs_time_t when) lnet_net_lock(cpt); - if (the_lnet.ln_shutdown) { + if (the_lnet.ln_state != LNET_STATE_RUNNING) { lnet_net_unlock(cpt); return -ESHUTDOWN; } - lp = lnet_find_peer_locked(the_lnet.ln_peer_tables[cpt], nid); + lp = lnet_find_peer_ni_locked(nid); if (lp == NULL) { /* nid not found */ lnet_net_unlock(cpt); @@ -1770,19 +1372,31 @@ lnet_notify(lnet_ni_t *ni, lnet_nid_t nid, int alive, cfs_time_t when) return 0; } + /* + * It is possible for this function to be called for the same peer + * but with different NIs. We want to synchronize the notification + * between the different calls. So we will use the lpni_cpt to + * grab the net lock. + */ + if (lp->lpni_cpt != cpt) { + lnet_net_unlock(cpt); + cpt = lp->lpni_cpt; + lnet_net_lock(cpt); + } + /* We can't fully trust LND on reporting exact peer last_alive * if he notifies us about dead peer. For example ksocklnd can * call us with when == _time_when_the_node_was_booted_ if * no connections were successfully established */ - if (ni != NULL && !alive && when < lp->lp_last_alive) - when = lp->lp_last_alive; + if (ni != NULL && !alive && when < lp->lpni_last_alive) + when = lp->lpni_last_alive; lnet_notify_locked(lp, ni == NULL, alive, when); if (ni != NULL) lnet_ni_notify_locked(ni, lp); - lnet_peer_decref_locked(lp); + lnet_peer_ni_decref_locked(lp); lnet_net_unlock(cpt); return 0;