lnet/lnet/router.c

   1 /*
   2  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
   3  *
   4  * Copyright (c) 2011, 2014, Intel Corporation.
   5  *
   6  *   This file is part of Portals
   7  *   http://sourceforge.net/projects/sandiaportals/
   8  *
   9  *   Portals is free software; you can redistribute it and/or
  10  *   modify it under the terms of version 2 of the GNU General Public
  11  *   License as published by the Free Software Foundation.
  12  *
  13  *   Portals is distributed in the hope that it will be useful,
  14  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  16  *   GNU General Public License for more details.
  17  *
  18  *   You should have received a copy of the GNU General Public License
  19  *   along with Portals; if not, write to the Free Software
  20  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  21  *
  22  */
  23
  24 #define DEBUG_SUBSYSTEM S_LNET
  25 #include <lnet/lib-lnet.h>
  26
  27 #if defined(__KERNEL__) && defined(LNET_ROUTER)
  28
  29 #define LNET_NRB_TINY_MIN       512     /* min value for each CPT */
  30 #define LNET_NRB_TINY           (LNET_NRB_TINY_MIN * 4)
  31 #define LNET_NRB_SMALL_MIN      4096    /* min value for each CPT */
  32 #define LNET_NRB_SMALL          (LNET_NRB_SMALL_MIN * 4)
  33 #define LNET_NRB_SMALL_PAGES    1
  34 #define LNET_NRB_LARGE_MIN      256     /* min value for each CPT */
  35 #define LNET_NRB_LARGE          (LNET_NRB_LARGE_MIN * 4)
  36 #define LNET_NRB_LARGE_PAGES    ((LNET_MTU + PAGE_CACHE_SIZE - 1) >> \
  37                                   PAGE_CACHE_SHIFT)
  38
  39 static char *forwarding = "";
  40 CFS_MODULE_PARM(forwarding, "s", charp, 0444,
  41                 "Explicitly enable/disable forwarding between networks");
  42
  43 static int tiny_router_buffers;
  44 CFS_MODULE_PARM(tiny_router_buffers, "i", int, 0444,
  45                 "# of 0 payload messages to buffer in the router");
  46 static int small_router_buffers;
  47 CFS_MODULE_PARM(small_router_buffers, "i", int, 0444,
  48                 "# of small (1 page) messages to buffer in the router");
  49 static int large_router_buffers;
  50 CFS_MODULE_PARM(large_router_buffers, "i", int, 0444,
  51                 "# of large messages to buffer in the router");
  52 static int peer_buffer_credits = 0;
  53 CFS_MODULE_PARM(peer_buffer_credits, "i", int, 0444,
  54                 "# router buffer credits per peer");
  55
  56 static int auto_down = 1;
  57 CFS_MODULE_PARM(auto_down, "i", int, 0444,
  58                 "Automatically mark peers down on comms error");
  59
  60 int
  61 lnet_peer_buffer_credits(lnet_ni_t *ni)
  62 {
  63         /* NI option overrides LNet default */
  64         if (ni->ni_peerrtrcredits > 0)
  65                 return ni->ni_peerrtrcredits;
  66         if (peer_buffer_credits > 0)
  67                 return peer_buffer_credits;
  68
  69         /* As an approximation, allow this peer the same number of router
  70          * buffers as it is allowed outstanding sends */
  71         return ni->ni_peertxcredits;
  72 }
  73
  74 /* forward ref's */
  75 static int lnet_router_checker(void *);
  76 #else
  77
  78 int
  79 lnet_peer_buffer_credits(lnet_ni_t *ni)
  80 {
  81         return 0;
  82 }
  83
  84 #endif
  85
  86 static int check_routers_before_use = 0;
  87 CFS_MODULE_PARM(check_routers_before_use, "i", int, 0444,
  88                 "Assume routers are down and ping them before use");
  89
  90 int avoid_asym_router_failure = 1;
  91 CFS_MODULE_PARM(avoid_asym_router_failure, "i", int, 0644,
  92                 "Avoid asymmetrical router failures (0 to disable)");
  93
  94 static int dead_router_check_interval = 60;
  95 CFS_MODULE_PARM(dead_router_check_interval, "i", int, 0644,
  96                 "Seconds between dead router health checks (<= 0 to disable)");
  97
  98 static int live_router_check_interval = 60;
  99 CFS_MODULE_PARM(live_router_check_interval, "i", int, 0644,
 100                 "Seconds between live router health checks (<= 0 to disable)");
 101
 102 static int router_ping_timeout = 50;
 103 CFS_MODULE_PARM(router_ping_timeout, "i", int, 0644,
 104                 "Seconds to wait for the reply to a router health query");
 105
 106 int
 107 lnet_peers_start_down(void)
 108 {
 109         return check_routers_before_use;
 110 }
 111
 112 void
 113 lnet_notify_locked(lnet_peer_t *lp, int notifylnd, int alive, cfs_time_t when)
 114 {
 115         if (cfs_time_before(when, lp->lp_timestamp)) { /* out of date information */
 116                 CDEBUG(D_NET, "Out of date\n");
 117                 return;
 118         }
 119
 120         lp->lp_timestamp = when;                /* update timestamp */
 121         lp->lp_ping_deadline = 0;               /* disable ping timeout */
 122
 123         if (lp->lp_alive_count != 0 &&          /* got old news */
 124             (!lp->lp_alive) == (!alive)) {      /* new date for old news */
 125                 CDEBUG(D_NET, "Old news\n");
 126                 return;
 127         }
 128
 129         /* Flag that notification is outstanding */
 130
 131         lp->lp_alive_count++;
 132         lp->lp_alive = !(!alive);               /* 1 bit! */
 133         lp->lp_notify = 1;
 134         lp->lp_notifylnd |= notifylnd;
 135         if (lp->lp_alive)
 136                 lp->lp_ping_feats = LNET_PING_FEAT_INVAL; /* reset */
 137
 138         CDEBUG(D_NET, "set %s %d\n", libcfs_nid2str(lp->lp_nid), alive);
 139 }
 140
 141 static void
 142 lnet_ni_notify_locked(lnet_ni_t *ni, lnet_peer_t *lp)
 143 {
 144         int        alive;
 145         int        notifylnd;
 146
 147         /* Notify only in 1 thread at any time to ensure ordered notification.
 148          * NB individual events can be missed; the only guarantee is that you
 149          * always get the most recent news */
 150
 151         if (lp->lp_notifying)
 152                 return;
 153
 154         lp->lp_notifying = 1;
 155
 156         while (lp->lp_notify) {
 157                 alive     = lp->lp_alive;
 158                 notifylnd = lp->lp_notifylnd;
 159
 160                 lp->lp_notifylnd = 0;
 161                 lp->lp_notify    = 0;
 162
 163                 if (notifylnd && ni->ni_lnd->lnd_notify != NULL) {
 164                         lnet_net_unlock(lp->lp_cpt);
 165
 166                         /* A new notification could happen now; I'll handle it
 167                          * when control returns to me */
 168
 169                         (ni->ni_lnd->lnd_notify)(ni, lp->lp_nid, alive);
 170
 171                         lnet_net_lock(lp->lp_cpt);
 172                 }
 173         }
 174
 175         lp->lp_notifying = 0;
 176 }
 177
 178
 179 static void
 180 lnet_rtr_addref_locked(lnet_peer_t *lp)
 181 {
 182         LASSERT(lp->lp_refcount > 0);
 183         LASSERT(lp->lp_rtr_refcount >= 0);
 184
 185         /* lnet_net_lock must be exclusively locked */
 186         lp->lp_rtr_refcount++;
 187         if (lp->lp_rtr_refcount == 1) {
 188                 struct list_head *pos;
 189
 190                 /* a simple insertion sort */
 191                 list_for_each_prev(pos, &the_lnet.ln_routers) {
 192                         lnet_peer_t *rtr = list_entry(pos, lnet_peer_t,
 193                                                       lp_rtr_list);
 194
 195                         if (rtr->lp_nid < lp->lp_nid)
 196                                 break;
 197                 }
 198
 199                 list_add(&lp->lp_rtr_list, pos);
 200                 /* addref for the_lnet.ln_routers */
 201                 lnet_peer_addref_locked(lp);
 202                 the_lnet.ln_routers_version++;
 203         }
 204 }
 205
 206 static void
 207 lnet_rtr_decref_locked(lnet_peer_t *lp)
 208 {
 209         LASSERT(lp->lp_refcount > 0);
 210         LASSERT(lp->lp_rtr_refcount > 0);
 211
 212         /* lnet_net_lock must be exclusively locked */
 213         lp->lp_rtr_refcount--;
 214         if (lp->lp_rtr_refcount == 0) {
 215                 LASSERT(list_empty(&lp->lp_routes));
 216
 217                 if (lp->lp_rcd != NULL) {
 218                         list_add(&lp->lp_rcd->rcd_list,
 219                                  &the_lnet.ln_rcd_deathrow);
 220                         lp->lp_rcd = NULL;
 221                 }
 222
 223                 list_del(&lp->lp_rtr_list);
 224                 /* decref for the_lnet.ln_routers */
 225                 lnet_peer_decref_locked(lp);
 226                 the_lnet.ln_routers_version++;
 227         }
 228 }
 229
 230 lnet_remotenet_t *
 231 lnet_find_net_locked (__u32 net)
 232 {
 233         lnet_remotenet_t *rnet;
 234         struct list_head *tmp;
 235         struct list_head *rn_list;
 236
 237         LASSERT(!the_lnet.ln_shutdown);
 238
 239         rn_list = lnet_net2rnethash(net);
 240         list_for_each(tmp, rn_list) {
 241                 rnet = list_entry(tmp, lnet_remotenet_t, lrn_list);
 242
 243                 if (rnet->lrn_net == net)
 244                         return rnet;
 245         }
 246         return NULL;
 247 }
 248
 249 static void lnet_shuffle_seed(void)
 250 {
 251         static int seeded = 0;
 252         int lnd_type, seed[2];
 253         struct timeval tv;
 254         lnet_ni_t *ni;
 255         struct list_head *tmp;
 256
 257         if (seeded)
 258                 return;
 259
 260         cfs_get_random_bytes(seed, sizeof(seed));
 261
 262         /* Nodes with small feet have little entropy
 263          * the NID for this node gives the most entropy in the low bits */
 264         list_for_each(tmp, &the_lnet.ln_nis) {
 265                 ni = list_entry(tmp, lnet_ni_t, ni_list);
 266                 lnd_type = LNET_NETTYP(LNET_NIDNET(ni->ni_nid));
 267
 268                 if (lnd_type != LOLND)
 269                         seed[0] ^= (LNET_NIDADDR(ni->ni_nid) | lnd_type);
 270         }
 271
 272         do_gettimeofday(&tv);
 273         cfs_srand(tv.tv_sec ^ seed[0], tv.tv_usec ^ seed[1]);
 274         seeded = 1;
 275         return;
 276 }
 277
 278 /* NB expects LNET_LOCK held */
 279 static void
 280 lnet_add_route_to_rnet(lnet_remotenet_t *rnet, lnet_route_t *route)
 281 {
 282         unsigned int      len = 0;
 283         unsigned int      offset = 0;
 284         struct list_head *e;
 285
 286         lnet_shuffle_seed();
 287
 288         list_for_each(e, &rnet->lrn_routes) {
 289                 len++;
 290         }
 291
 292         /* len+1 positions to add a new entry, also prevents division by 0 */
 293         offset = cfs_rand() % (len + 1);
 294         list_for_each(e, &rnet->lrn_routes) {
 295                 if (offset == 0)
 296                         break;
 297                 offset--;
 298         }
 299         list_add(&route->lr_list, e);
 300         list_add(&route->lr_gwlist, &route->lr_gateway->lp_routes);
 301
 302         the_lnet.ln_remote_nets_version++;
 303         lnet_rtr_addref_locked(route->lr_gateway);
 304 }
 305
 306 int
 307 lnet_add_route(__u32 net, unsigned int hops, lnet_nid_t gateway,
 308                unsigned int priority)
 309 {
 310         struct list_head        *e;
 311         lnet_remotenet_t        *rnet;
 312         lnet_remotenet_t        *rnet2;
 313         lnet_route_t            *route;
 314         lnet_ni_t               *ni;
 315         int                     add_route;
 316         int                     rc;
 317
 318         CDEBUG(D_NET, "Add route: net %s hops %u priority %u gw %s\n",
 319                libcfs_net2str(net), hops, priority, libcfs_nid2str(gateway));
 320
 321         if (gateway == LNET_NID_ANY ||
 322             LNET_NETTYP(LNET_NIDNET(gateway)) == LOLND ||
 323             net == LNET_NIDNET(LNET_NID_ANY) ||
 324             LNET_NETTYP(net) == LOLND ||
 325             LNET_NIDNET(gateway) == net ||
 326             hops < 1 || hops > 255)
 327                 return -EINVAL;
 328
 329         if (lnet_islocalnet(net))       /* it's a local network */
 330                 return -EEXIST;
 331
 332         /* Assume net, route, all new */
 333         LIBCFS_ALLOC(route, sizeof(*route));
 334         LIBCFS_ALLOC(rnet, sizeof(*rnet));
 335         if (route == NULL || rnet == NULL) {
 336                 CERROR("Out of memory creating route %s %d %s\n",
 337                        libcfs_net2str(net), hops, libcfs_nid2str(gateway));
 338                 if (route != NULL)
 339                         LIBCFS_FREE(route, sizeof(*route));
 340                 if (rnet != NULL)
 341                         LIBCFS_FREE(rnet, sizeof(*rnet));
 342                 return -ENOMEM;
 343         }
 344
 345         INIT_LIST_HEAD(&rnet->lrn_routes);
 346         rnet->lrn_net = net;
 347         route->lr_hops = hops;
 348         route->lr_net = net;
 349         route->lr_priority = priority;
 350
 351         lnet_net_lock(LNET_LOCK_EX);
 352
 353         rc = lnet_nid2peer_locked(&route->lr_gateway, gateway, LNET_LOCK_EX);
 354         if (rc != 0) {
 355                 lnet_net_unlock(LNET_LOCK_EX);
 356
 357                 LIBCFS_FREE(route, sizeof(*route));
 358                 LIBCFS_FREE(rnet, sizeof(*rnet));
 359
 360                 if (rc == -EHOSTUNREACH) /* gateway is not on a local net. */
 361                         return rc;       /* ignore the route entry */
 362                 CERROR("Error %d creating route %s %d %s\n", rc,
 363                         libcfs_net2str(net), hops,
 364                         libcfs_nid2str(gateway));
 365                 return rc;
 366         }
 367
 368         LASSERT(!the_lnet.ln_shutdown);
 369
 370         rnet2 = lnet_find_net_locked(net);
 371         if (rnet2 == NULL) {
 372                 /* new network */
 373                 list_add_tail(&rnet->lrn_list, lnet_net2rnethash(net));
 374                 rnet2 = rnet;
 375         }
 376
 377         /* Search for a duplicate route (it's a NOOP if it is) */
 378         add_route = 1;
 379         list_for_each(e, &rnet2->lrn_routes) {
 380                 lnet_route_t *route2 = list_entry(e, lnet_route_t, lr_list);
 381
 382                 if (route2->lr_gateway == route->lr_gateway) {
 383                         add_route = 0;
 384                         break;
 385                 }
 386
 387                 /* our lookups must be true */
 388                 LASSERT(route2->lr_gateway->lp_nid != gateway);
 389         }
 390
 391         if (add_route) {
 392                 lnet_peer_addref_locked(route->lr_gateway); /* +1 for notify */
 393                 lnet_add_route_to_rnet(rnet2, route);
 394
 395                 ni = route->lr_gateway->lp_ni;
 396                 lnet_net_unlock(LNET_LOCK_EX);
 397
 398                 /* XXX Assume alive */
 399                 if (ni->ni_lnd->lnd_notify != NULL)
 400                         (ni->ni_lnd->lnd_notify)(ni, gateway, 1);
 401
 402                 lnet_net_lock(LNET_LOCK_EX);
 403         }
 404
 405         /* -1 for notify or !add_route */
 406         lnet_peer_decref_locked(route->lr_gateway);
 407         lnet_net_unlock(LNET_LOCK_EX);
 408
 409         rc = 0;
 410
 411         if (!add_route) {
 412                 rc = -EEXIST;
 413                 LIBCFS_FREE(route, sizeof(*route));
 414         }
 415
 416         if (rnet != rnet2)
 417                 LIBCFS_FREE(rnet, sizeof(*rnet));
 418
 419         return rc;
 420 }
 421
 422 int
 423 lnet_check_routes(void)
 424 {
 425         lnet_remotenet_t *rnet;
 426         lnet_route_t     *route;
 427         lnet_route_t     *route2;
 428         struct list_head *e1;
 429         struct list_head *e2;
 430         int               cpt;
 431         struct list_head *rn_list;
 432         int               i;
 433
 434         cpt = lnet_net_lock_current();
 435
 436         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) {
 437                 rn_list = &the_lnet.ln_remote_nets_hash[i];
 438                 list_for_each(e1, rn_list) {
 439                         rnet = list_entry(e1, lnet_remotenet_t, lrn_list);
 440
 441                         route2 = NULL;
 442                         list_for_each(e2, &rnet->lrn_routes) {
 443                                 lnet_nid_t      nid1;
 444                                 lnet_nid_t      nid2;
 445                                 int             net;
 446
 447                                 route = list_entry(e2, lnet_route_t,
 448                                                    lr_list);
 449
 450                                 if (route2 == NULL) {
 451                                         route2 = route;
 452                                         continue;
 453                                 }
 454
 455                                 if (route->lr_gateway->lp_ni ==
 456                                     route2->lr_gateway->lp_ni)
 457                                         continue;
 458
 459                                 nid1 = route->lr_gateway->lp_nid;
 460                                 nid2 = route2->lr_gateway->lp_nid;
 461                                 net = rnet->lrn_net;
 462
 463                                 lnet_net_unlock(cpt);
 464
 465                                 CERROR("Routes to %s via %s and %s not "
 466                                        "supported\n",
 467                                        libcfs_net2str(net),
 468                                        libcfs_nid2str(nid1),
 469                                        libcfs_nid2str(nid2));
 470                                 return -EINVAL;
 471                         }
 472                 }
 473         }
 474
 475         lnet_net_unlock(cpt);
 476         return 0;
 477 }
 478
 479 int
 480 lnet_del_route(__u32 net, lnet_nid_t gw_nid)
 481 {
 482         struct lnet_peer        *gateway;
 483         lnet_remotenet_t        *rnet;
 484         lnet_route_t            *route;
 485         struct list_head        *e1;
 486         struct list_head        *e2;
 487         int                     rc = -ENOENT;
 488         struct list_head        *rn_list;
 489         int                     idx = 0;
 490
 491         CDEBUG(D_NET, "Del route: net %s : gw %s\n",
 492                libcfs_net2str(net), libcfs_nid2str(gw_nid));
 493
 494         /* NB Caller may specify either all routes via the given gateway
 495          * or a specific route entry actual NIDs) */
 496
 497         lnet_net_lock(LNET_LOCK_EX);
 498         if (net == LNET_NIDNET(LNET_NID_ANY))
 499                 rn_list = &the_lnet.ln_remote_nets_hash[0];
 500         else
 501                 rn_list = lnet_net2rnethash(net);
 502
 503 again:
 504         list_for_each(e1, rn_list) {
 505                 rnet = list_entry(e1, lnet_remotenet_t, lrn_list);
 506
 507                 if (!(net == LNET_NIDNET(LNET_NID_ANY) ||
 508                         net == rnet->lrn_net))
 509                         continue;
 510
 511                 list_for_each(e2, &rnet->lrn_routes) {
 512                         route = list_entry(e2, lnet_route_t, lr_list);
 513
 514                         gateway = route->lr_gateway;
 515                         if (!(gw_nid == LNET_NID_ANY ||
 516                               gw_nid == gateway->lp_nid))
 517                                 continue;
 518
 519                         list_del(&route->lr_list);
 520                         list_del(&route->lr_gwlist);
 521                         the_lnet.ln_remote_nets_version++;
 522
 523                         if (list_empty(&rnet->lrn_routes))
 524                                 list_del(&rnet->lrn_list);
 525                         else
 526                                 rnet = NULL;
 527
 528                         lnet_rtr_decref_locked(gateway);
 529                         lnet_peer_decref_locked(gateway);
 530
 531                         lnet_net_unlock(LNET_LOCK_EX);
 532
 533                         LIBCFS_FREE(route, sizeof(*route));
 534
 535                         if (rnet != NULL)
 536                                 LIBCFS_FREE(rnet, sizeof(*rnet));
 537
 538                         rc = 0;
 539                         lnet_net_lock(LNET_LOCK_EX);
 540                         goto again;
 541                 }
 542         }
 543
 544         if (net == LNET_NIDNET(LNET_NID_ANY) &&
 545             ++idx < LNET_REMOTE_NETS_HASH_SIZE) {
 546                 rn_list = &the_lnet.ln_remote_nets_hash[idx];
 547                 goto again;
 548         }
 549         lnet_net_unlock(LNET_LOCK_EX);
 550
 551         return rc;
 552 }
 553
 554 void
 555 lnet_destroy_routes (void)
 556 {
 557         lnet_del_route(LNET_NIDNET(LNET_NID_ANY), LNET_NID_ANY);
 558 }
 559
 560 int lnet_get_rtr_pool_cfg(int idx, struct lnet_ioctl_pool_cfg *pool_cfg)
 561 {
 562         int i, rc = -ENOENT, lidx, j;
 563
 564         if (the_lnet.ln_rtrpools == NULL)
 565                 return rc;
 566
 567         for (i = 0; i < LNET_NRBPOOLS; i++) {
 568                 lnet_rtrbufpool_t *rbp;
 569
 570                 lnet_net_lock(LNET_LOCK_EX);
 571                 lidx = idx;
 572                 cfs_percpt_for_each(rbp, j, the_lnet.ln_rtrpools) {
 573                         if (lidx-- == 0) {
 574                                 rc = 0;
 575                                 pool_cfg->pl_pools[i].pl_npages =
 576                                         rbp[i].rbp_npages;
 577                                 pool_cfg->pl_pools[i].pl_nbuffers =
 578                                         rbp[i].rbp_nbuffers;
 579                                 pool_cfg->pl_pools[i].pl_credits =
 580                                         rbp[i].rbp_credits;
 581                                 pool_cfg->pl_pools[i].pl_mincredits =
 582                                         rbp[i].rbp_mincredits;
 583                                 break;
 584                         }
 585                 }
 586                 lnet_net_unlock(LNET_LOCK_EX);
 587         }
 588
 589         lnet_net_lock(LNET_LOCK_EX);
 590         pool_cfg->pl_routing = the_lnet.ln_routing;
 591         lnet_net_unlock(LNET_LOCK_EX);
 592
 593         return rc;
 594 }
 595
 596 int
 597 lnet_get_route(int idx, __u32 *net, __u32 *hops,
 598                lnet_nid_t *gateway, __u32 *alive, __u32 *priority)
 599 {
 600         struct list_head *e1;
 601         struct list_head *e2;
 602         lnet_remotenet_t *rnet;
 603         lnet_route_t     *route;
 604         int               cpt;
 605         int               i;
 606         struct list_head *rn_list;
 607
 608         cpt = lnet_net_lock_current();
 609
 610         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) {
 611                 rn_list = &the_lnet.ln_remote_nets_hash[i];
 612                 list_for_each(e1, rn_list) {
 613                         rnet = list_entry(e1, lnet_remotenet_t, lrn_list);
 614
 615                         list_for_each(e2, &rnet->lrn_routes) {
 616                                 route = list_entry(e2, lnet_route_t,
 617                                                    lr_list);
 618
 619                                 if (idx-- == 0) {
 620                                         *net      = rnet->lrn_net;
 621                                         *hops     = route->lr_hops;
 622                                         *priority = route->lr_priority;
 623                                         *gateway  = route->lr_gateway->lp_nid;
 624                                         *alive    =
 625                                                 route->lr_gateway->lp_alive &&
 626                                                         !route->lr_downis;
 627                                         lnet_net_unlock(cpt);
 628                                         return 0;
 629                                 }
 630                         }
 631                 }
 632         }
 633
 634         lnet_net_unlock(cpt);
 635         return -ENOENT;
 636 }
 637
 638 void
 639 lnet_swap_pinginfo(lnet_ping_info_t *info)
 640 {
 641         int               i;
 642         lnet_ni_status_t *stat;
 643
 644         __swab32s(&info->pi_magic);
 645         __swab32s(&info->pi_features);
 646         __swab32s(&info->pi_pid);
 647         __swab32s(&info->pi_nnis);
 648         for (i = 0; i < info->pi_nnis && i < LNET_MAX_RTR_NIS; i++) {
 649                 stat = &info->pi_ni[i];
 650                 __swab64s(&stat->ns_nid);
 651                 __swab32s(&stat->ns_status);
 652         }
 653         return;
 654 }
 655
 656 /**
 657  * parse router-checker pinginfo, record number of down NIs for remote
 658  * networks on that router.
 659  */
 660 static void
 661 lnet_parse_rc_info(lnet_rc_data_t *rcd)
 662 {
 663         lnet_ping_info_t        *info = rcd->rcd_pinginfo;
 664         struct lnet_peer        *gw   = rcd->rcd_gateway;
 665         lnet_route_t            *rte;
 666
 667         if (!gw->lp_alive)
 668                 return;
 669
 670         if (info->pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
 671                 lnet_swap_pinginfo(info);
 672
 673         /* NB always racing with network! */
 674         if (info->pi_magic != LNET_PROTO_PING_MAGIC) {
 675                 CDEBUG(D_NET, "%s: Unexpected magic %08x\n",
 676                        libcfs_nid2str(gw->lp_nid), info->pi_magic);
 677                 gw->lp_ping_feats = LNET_PING_FEAT_INVAL;
 678                 return;
 679         }
 680
 681         gw->lp_ping_feats = info->pi_features;
 682         if ((gw->lp_ping_feats & LNET_PING_FEAT_MASK) == 0) {
 683                 CDEBUG(D_NET, "%s: Unexpected features 0x%x\n",
 684                        libcfs_nid2str(gw->lp_nid), gw->lp_ping_feats);
 685                 return; /* nothing I can understand */
 686         }
 687
 688         if ((gw->lp_ping_feats & LNET_PING_FEAT_NI_STATUS) == 0)
 689                 return; /* can't carry NI status info */
 690
 691         list_for_each_entry(rte, &gw->lp_routes, lr_gwlist) {
 692                 int     down = 0;
 693                 int     up = 0;
 694                 int     i;
 695
 696                 if ((gw->lp_ping_feats & LNET_PING_FEAT_RTE_DISABLED) != 0) {
 697                         rte->lr_downis = 1;
 698                         continue;
 699                 }
 700
 701                 for (i = 0; i < info->pi_nnis && i < LNET_MAX_RTR_NIS; i++) {
 702                         lnet_ni_status_t *stat = &info->pi_ni[i];
 703                         lnet_nid_t       nid = stat->ns_nid;
 704
 705                         if (nid == LNET_NID_ANY) {
 706                                 CDEBUG(D_NET, "%s: unexpected LNET_NID_ANY\n",
 707                                        libcfs_nid2str(gw->lp_nid));
 708                                 gw->lp_ping_feats = LNET_PING_FEAT_INVAL;
 709                                 return;
 710                         }
 711
 712                         if (LNET_NETTYP(LNET_NIDNET(nid)) == LOLND)
 713                                 continue;
 714
 715                         if (stat->ns_status == LNET_NI_STATUS_DOWN) {
 716                                 down++;
 717                                 continue;
 718                         }
 719
 720                         if (stat->ns_status == LNET_NI_STATUS_UP) {
 721                                 if (LNET_NIDNET(nid) == rte->lr_net) {
 722                                         up = 1;
 723                                         break;
 724                                 }
 725                                 continue;
 726                         }
 727
 728                         CDEBUG(D_NET, "%s: Unexpected status 0x%x\n",
 729                                libcfs_nid2str(gw->lp_nid), stat->ns_status);
 730                         gw->lp_ping_feats = LNET_PING_FEAT_INVAL;
 731                         return;
 732                 }
 733
 734                 if (up) { /* ignore downed NIs if NI for dest network is up */
 735                         rte->lr_downis = 0;
 736                         continue;
 737                 }
 738                 rte->lr_downis = down;
 739         }
 740 }
 741
 742 static void
 743 lnet_router_checker_event(lnet_event_t *event)
 744 {
 745         lnet_rc_data_t          *rcd = event->md.user_ptr;
 746         struct lnet_peer        *lp;
 747
 748         LASSERT(rcd != NULL);
 749
 750         if (event->unlinked) {
 751                 LNetInvalidateHandle(&rcd->rcd_mdh);
 752                 return;
 753         }
 754
 755         LASSERT(event->type == LNET_EVENT_SEND ||
 756                 event->type == LNET_EVENT_REPLY);
 757
 758         lp = rcd->rcd_gateway;
 759         LASSERT(lp != NULL);
 760
 761          /* NB: it's called with holding lnet_res_lock, we have a few
 762           * places need to hold both locks at the same time, please take
 763           * care of lock ordering */
 764         lnet_net_lock(lp->lp_cpt);
 765         if (!lnet_isrouter(lp) || lp->lp_rcd != rcd) {
 766                 /* ignore if no longer a router or rcd is replaced */
 767                 goto out;
 768         }
 769
 770         if (event->type == LNET_EVENT_SEND) {
 771                 lp->lp_ping_notsent = 0;
 772                 if (event->status == 0)
 773                         goto out;
 774         }
 775
 776         /* LNET_EVENT_REPLY */
 777         /* A successful REPLY means the router is up.  If _any_ comms
 778          * to the router fail I assume it's down (this will happen if
 779          * we ping alive routers to try to detect router death before
 780          * apps get burned). */
 781
 782         lnet_notify_locked(lp, 1, (event->status == 0), cfs_time_current());
 783         /* The router checker will wake up very shortly and do the
 784          * actual notification.
 785          * XXX If 'lp' stops being a router before then, it will still
 786          * have the notification pending!!! */
 787
 788         if (avoid_asym_router_failure && event->status == 0)
 789                 lnet_parse_rc_info(rcd);
 790
 791  out:
 792         lnet_net_unlock(lp->lp_cpt);
 793 }
 794
 795 static void
 796 lnet_wait_known_routerstate(void)
 797 {
 798         lnet_peer_t      *rtr;
 799         struct list_head *entry;
 800         int               all_known;
 801
 802         LASSERT(the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING);
 803
 804         for (;;) {
 805                 int cpt = lnet_net_lock_current();
 806
 807                 all_known = 1;
 808                 list_for_each(entry, &the_lnet.ln_routers) {
 809                         rtr = list_entry(entry, lnet_peer_t, lp_rtr_list);
 810
 811                         if (rtr->lp_alive_count == 0) {
 812                                 all_known = 0;
 813                                 break;
 814                         }
 815                 }
 816
 817                 lnet_net_unlock(cpt);
 818
 819                 if (all_known)
 820                         return;
 821
 822 #ifndef __KERNEL__
 823                 lnet_router_checker();
 824 #endif
 825                 cfs_pause(cfs_time_seconds(1));
 826         }
 827 }
 828
 829 void
 830 lnet_router_ni_update_locked(lnet_peer_t *gw, __u32 net)
 831 {
 832         lnet_route_t *rte;
 833
 834         if ((gw->lp_ping_feats & LNET_PING_FEAT_NI_STATUS) != 0) {
 835                 list_for_each_entry(rte, &gw->lp_routes, lr_gwlist) {
 836                         if (rte->lr_net == net) {
 837                                 rte->lr_downis = 0;
 838                                 break;
 839                         }
 840                 }
 841         }
 842 }
 843
 844 static void
 845 lnet_update_ni_status_locked(void)
 846 {
 847         lnet_ni_t       *ni;
 848         long            now;
 849         int             timeout;
 850
 851         LASSERT(the_lnet.ln_routing);
 852
 853         timeout = router_ping_timeout +
 854                   MAX(live_router_check_interval, dead_router_check_interval);
 855
 856         now = cfs_time_current_sec();
 857         list_for_each_entry(ni, &the_lnet.ln_nis, ni_list) {
 858                 if (ni->ni_lnd->lnd_type == LOLND)
 859                         continue;
 860
 861                 if (now < ni->ni_last_alive + timeout)
 862                         continue;
 863
 864                 lnet_ni_lock(ni);
 865                 /* re-check with lock */
 866                 if (now < ni->ni_last_alive + timeout) {
 867                         lnet_ni_unlock(ni);
 868                         continue;
 869                 }
 870
 871                 LASSERT(ni->ni_status != NULL);
 872
 873                 if (ni->ni_status->ns_status != LNET_NI_STATUS_DOWN) {
 874                         CDEBUG(D_NET, "NI(%s:%d) status changed to down\n",
 875                                libcfs_nid2str(ni->ni_nid), timeout);
 876                         /* NB: so far, this is the only place to set
 877                          * NI status to "down" */
 878                         ni->ni_status->ns_status = LNET_NI_STATUS_DOWN;
 879                 }
 880                 lnet_ni_unlock(ni);
 881         }
 882 }
 883
 884 static void
 885 lnet_destroy_rc_data(lnet_rc_data_t *rcd)
 886 {
 887         LASSERT(list_empty(&rcd->rcd_list));
 888         /* detached from network */
 889         LASSERT(LNetHandleIsInvalid(rcd->rcd_mdh));
 890
 891         if (rcd->rcd_gateway != NULL) {
 892                 int cpt = rcd->rcd_gateway->lp_cpt;
 893
 894                 lnet_net_lock(cpt);
 895                 lnet_peer_decref_locked(rcd->rcd_gateway);
 896                 lnet_net_unlock(cpt);
 897         }
 898
 899         if (rcd->rcd_pinginfo != NULL)
 900                 LIBCFS_FREE(rcd->rcd_pinginfo, LNET_PINGINFO_SIZE);
 901
 902         LIBCFS_FREE(rcd, sizeof(*rcd));
 903 }
 904
 905 static lnet_rc_data_t *
 906 lnet_create_rc_data_locked(lnet_peer_t *gateway)
 907 {
 908         lnet_rc_data_t          *rcd = NULL;
 909         lnet_ping_info_t        *pi;
 910         int                     rc;
 911         int                     i;
 912
 913         lnet_net_unlock(gateway->lp_cpt);
 914
 915         LIBCFS_ALLOC(rcd, sizeof(*rcd));
 916         if (rcd == NULL)
 917                 goto out;
 918
 919         LNetInvalidateHandle(&rcd->rcd_mdh);
 920         INIT_LIST_HEAD(&rcd->rcd_list);
 921
 922         LIBCFS_ALLOC(pi, LNET_PINGINFO_SIZE);
 923         if (pi == NULL)
 924                 goto out;
 925
 926         for (i = 0; i < LNET_MAX_RTR_NIS; i++) {
 927                 pi->pi_ni[i].ns_nid = LNET_NID_ANY;
 928                 pi->pi_ni[i].ns_status = LNET_NI_STATUS_INVALID;
 929         }
 930         rcd->rcd_pinginfo = pi;
 931
 932         LASSERT (!LNetHandleIsInvalid(the_lnet.ln_rc_eqh));
 933         rc = LNetMDBind((lnet_md_t){.start     = pi,
 934                                     .user_ptr  = rcd,
 935                                     .length    = LNET_PINGINFO_SIZE,
 936                                     .threshold = LNET_MD_THRESH_INF,
 937                                     .options   = LNET_MD_TRUNCATE,
 938                                     .eq_handle = the_lnet.ln_rc_eqh},
 939                         LNET_UNLINK,
 940                         &rcd->rcd_mdh);
 941         if (rc < 0) {
 942                 CERROR("Can't bind MD: %d\n", rc);
 943                 goto out;
 944         }
 945         LASSERT(rc == 0);
 946
 947         lnet_net_lock(gateway->lp_cpt);
 948         /* router table changed or someone has created rcd for this gateway */
 949         if (!lnet_isrouter(gateway) || gateway->lp_rcd != NULL) {
 950                 lnet_net_unlock(gateway->lp_cpt);
 951                 goto out;
 952         }
 953
 954         lnet_peer_addref_locked(gateway);
 955         rcd->rcd_gateway = gateway;
 956         gateway->lp_rcd = rcd;
 957         gateway->lp_ping_notsent = 0;
 958
 959         return rcd;
 960
 961  out:
 962         if (rcd != NULL) {
 963                 if (!LNetHandleIsInvalid(rcd->rcd_mdh)) {
 964                         rc = LNetMDUnlink(rcd->rcd_mdh);
 965                         LASSERT(rc == 0);
 966                 }
 967                 lnet_destroy_rc_data(rcd);
 968         }
 969
 970         lnet_net_lock(gateway->lp_cpt);
 971         return gateway->lp_rcd;
 972 }
 973
 974 static int
 975 lnet_router_check_interval (lnet_peer_t *rtr)
 976 {
 977         int secs;
 978
 979         secs = rtr->lp_alive ? live_router_check_interval :
 980                                dead_router_check_interval;
 981         if (secs < 0)
 982                 secs = 0;
 983
 984         return secs;
 985 }
 986
 987 static void
 988 lnet_ping_router_locked (lnet_peer_t *rtr)
 989 {
 990         lnet_rc_data_t *rcd = NULL;
 991         cfs_time_t      now = cfs_time_current();
 992         int             secs;
 993
 994         lnet_peer_addref_locked(rtr);
 995
 996         if (rtr->lp_ping_deadline != 0 && /* ping timed out? */
 997             cfs_time_after(now, rtr->lp_ping_deadline))
 998                 lnet_notify_locked(rtr, 1, 0, now);
 999
1000         /* Run any outstanding notifications */
1001         lnet_ni_notify_locked(rtr->lp_ni, rtr);
1002
1003         if (!lnet_isrouter(rtr) ||
1004             the_lnet.ln_rc_state != LNET_RC_STATE_RUNNING) {
1005                 /* router table changed or router checker is shutting down */
1006                 lnet_peer_decref_locked(rtr);
1007                 return;
1008         }
1009
1010         rcd = rtr->lp_rcd != NULL ?
1011               rtr->lp_rcd : lnet_create_rc_data_locked(rtr);
1012
1013         if (rcd == NULL)
1014                 return;
1015
1016         secs = lnet_router_check_interval(rtr);
1017
1018         CDEBUG(D_NET,
1019                "rtr %s %d: deadline %lu ping_notsent %d alive %d "
1020                "alive_count %d lp_ping_timestamp %lu\n",
1021                libcfs_nid2str(rtr->lp_nid), secs,
1022                rtr->lp_ping_deadline, rtr->lp_ping_notsent,
1023                rtr->lp_alive, rtr->lp_alive_count, rtr->lp_ping_timestamp);
1024
1025         if (secs != 0 && !rtr->lp_ping_notsent &&
1026             cfs_time_after(now, cfs_time_add(rtr->lp_ping_timestamp,
1027                                              cfs_time_seconds(secs)))) {
1028                 int               rc;
1029                 lnet_process_id_t id;
1030                 lnet_handle_md_t  mdh;
1031
1032                 id.nid = rtr->lp_nid;
1033                 id.pid = LNET_PID_LUSTRE;
1034                 CDEBUG(D_NET, "Check: %s\n", libcfs_id2str(id));
1035
1036                 rtr->lp_ping_notsent   = 1;
1037                 rtr->lp_ping_timestamp = now;
1038
1039                 mdh = rcd->rcd_mdh;
1040
1041                 if (rtr->lp_ping_deadline == 0) {
1042                         rtr->lp_ping_deadline =
1043                                 cfs_time_shift(router_ping_timeout);
1044                 }
1045
1046                 lnet_net_unlock(rtr->lp_cpt);
1047
1048                 rc = LNetGet(LNET_NID_ANY, mdh, id, LNET_RESERVED_PORTAL,
1049                              LNET_PROTO_PING_MATCHBITS, 0);
1050
1051                 lnet_net_lock(rtr->lp_cpt);
1052                 if (rc != 0)
1053                         rtr->lp_ping_notsent = 0; /* no event pending */
1054         }
1055
1056         lnet_peer_decref_locked(rtr);
1057         return;
1058 }
1059
1060 int
1061 lnet_router_checker_start(void)
1062 {
1063         int                     rc;
1064         int                     eqsz;
1065 #ifdef __KERNEL__
1066         struct task_struct     *task;
1067 #else /* __KERNEL__ */
1068         lnet_peer_t            *rtr;
1069         __u64                   version;
1070         int                     nrtr = 0;
1071         int                     router_checker_max_eqsize = 10240;
1072
1073         LASSERT (check_routers_before_use);
1074         LASSERT (dead_router_check_interval > 0);
1075
1076         lnet_net_lock(0);
1077
1078         /* As an approximation, allow each router the same number of
1079          * outstanding events as it is allowed outstanding sends */
1080         eqsz = 0;
1081         version = the_lnet.ln_routers_version;
1082         list_for_each_entry(rtr, &the_lnet.ln_routers, lp_rtr_list) {
1083                 lnet_ni_t         *ni = rtr->lp_ni;
1084                 lnet_process_id_t  id;
1085
1086                 nrtr++;
1087                 eqsz += ni->ni_peertxcredits;
1088
1089                 /* one async ping reply per router */
1090                 id.nid = rtr->lp_nid;
1091                 id.pid = LNET_PID_LUSTRE;
1092
1093                 lnet_net_unlock(0);
1094
1095                 rc = LNetSetAsync(id, 1);
1096                 if (rc != 0) {
1097                         CWARN("LNetSetAsync %s failed: %d\n",
1098                               libcfs_id2str(id), rc);
1099                         return rc;
1100                 }
1101
1102                 lnet_net_lock(0);
1103                 /* NB router list doesn't change in userspace */
1104                 LASSERT(version == the_lnet.ln_routers_version);
1105         }
1106
1107         lnet_net_unlock(0);
1108
1109         if (nrtr == 0) {
1110                 CDEBUG(D_NET,
1111                        "No router found, not starting router checker\n");
1112                 return 0;
1113         }
1114
1115         /* at least allow a SENT and a REPLY per router */
1116         if (router_checker_max_eqsize < 2 * nrtr)
1117                 router_checker_max_eqsize = 2 * nrtr;
1118
1119         LASSERT (eqsz > 0);
1120         if (eqsz > router_checker_max_eqsize)
1121                 eqsz = router_checker_max_eqsize;
1122 #endif
1123
1124         LASSERT (the_lnet.ln_rc_state == LNET_RC_STATE_SHUTDOWN);
1125
1126         if (check_routers_before_use &&
1127             dead_router_check_interval <= 0) {
1128                 LCONSOLE_ERROR_MSG(0x10a, "'dead_router_check_interval' must be"
1129                                    " set if 'check_routers_before_use' is set"
1130                                    "\n");
1131                 return -EINVAL;
1132         }
1133
1134         if (!the_lnet.ln_routing &&
1135             live_router_check_interval <= 0 &&
1136             dead_router_check_interval <= 0)
1137                 return 0;
1138
1139 #ifdef __KERNEL__
1140         sema_init(&the_lnet.ln_rc_signal, 0);
1141         /* EQ size doesn't matter; the callback is guaranteed to get every
1142          * event */
1143         eqsz = 0;
1144         rc = LNetEQAlloc(eqsz, lnet_router_checker_event,
1145                          &the_lnet.ln_rc_eqh);
1146 #else
1147         rc = LNetEQAlloc(eqsz, LNET_EQ_HANDLER_NONE,
1148                          &the_lnet.ln_rc_eqh);
1149 #endif
1150         if (rc != 0) {
1151                 CERROR("Can't allocate EQ(%d): %d\n", eqsz, rc);
1152                 return -ENOMEM;
1153         }
1154
1155         the_lnet.ln_rc_state = LNET_RC_STATE_RUNNING;
1156 #ifdef __KERNEL__
1157         task = kthread_run(lnet_router_checker, NULL, "router_checker");
1158         if (IS_ERR(task)) {
1159                 rc = PTR_ERR(task);
1160                 CERROR("Can't start router checker thread: %d\n", rc);
1161                 /* block until event callback signals exit */
1162                 down(&the_lnet.ln_rc_signal);
1163                 rc = LNetEQFree(the_lnet.ln_rc_eqh);
1164                 LASSERT(rc == 0);
1165                 the_lnet.ln_rc_state = LNET_RC_STATE_SHUTDOWN;
1166                 return -ENOMEM;
1167         }
1168 #endif
1169
1170         if (check_routers_before_use) {
1171                 /* Note that a helpful side-effect of pinging all known routers
1172                  * at startup is that it makes them drop stale connections they
1173                  * may have to a previous instance of me. */
1174                 lnet_wait_known_routerstate();
1175         }
1176
1177         return 0;
1178 }
1179
1180 void
1181 lnet_router_checker_stop (void)
1182 {
1183         int rc;
1184
1185         if (the_lnet.ln_rc_state == LNET_RC_STATE_SHUTDOWN)
1186                 return;
1187
1188         LASSERT (the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING);
1189         the_lnet.ln_rc_state = LNET_RC_STATE_STOPPING;
1190
1191 #ifdef __KERNEL__
1192         /* block until event callback signals exit */
1193         down(&the_lnet.ln_rc_signal);
1194 #else
1195         lnet_router_checker();
1196 #endif
1197         LASSERT(the_lnet.ln_rc_state == LNET_RC_STATE_SHUTDOWN);
1198
1199         rc = LNetEQFree(the_lnet.ln_rc_eqh);
1200         LASSERT (rc == 0);
1201         return;
1202 }
1203
1204 static void
1205 lnet_prune_rc_data(int wait_unlink)
1206 {
1207         lnet_rc_data_t          *rcd;
1208         lnet_rc_data_t          *tmp;
1209         lnet_peer_t             *lp;
1210         struct list_head         head;
1211         int                      i = 2;
1212
1213         if (likely(the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING &&
1214                    list_empty(&the_lnet.ln_rcd_deathrow) &&
1215                    list_empty(&the_lnet.ln_rcd_zombie)))
1216                 return;
1217
1218         INIT_LIST_HEAD(&head);
1219
1220         lnet_net_lock(LNET_LOCK_EX);
1221
1222         if (the_lnet.ln_rc_state != LNET_RC_STATE_RUNNING) {
1223                 /* router checker is stopping, prune all */
1224                 list_for_each_entry(lp, &the_lnet.ln_routers,
1225                                     lp_rtr_list) {
1226                         if (lp->lp_rcd == NULL)
1227                                 continue;
1228
1229                         LASSERT(list_empty(&lp->lp_rcd->rcd_list));
1230                         list_add(&lp->lp_rcd->rcd_list,
1231                                  &the_lnet.ln_rcd_deathrow);
1232                         lp->lp_rcd = NULL;
1233                 }
1234         }
1235
1236         /* unlink all RCDs on deathrow list */
1237         list_splice_init(&the_lnet.ln_rcd_deathrow, &head);
1238
1239         if (!list_empty(&head)) {
1240                 lnet_net_unlock(LNET_LOCK_EX);
1241
1242                 list_for_each_entry(rcd, &head, rcd_list)
1243                         LNetMDUnlink(rcd->rcd_mdh);
1244
1245                 lnet_net_lock(LNET_LOCK_EX);
1246         }
1247
1248         list_splice_init(&head, &the_lnet.ln_rcd_zombie);
1249
1250         /* release all zombie RCDs */
1251         while (!list_empty(&the_lnet.ln_rcd_zombie)) {
1252                 list_for_each_entry_safe(rcd, tmp, &the_lnet.ln_rcd_zombie,
1253                                          rcd_list) {
1254                         if (LNetHandleIsInvalid(rcd->rcd_mdh))
1255                                 list_move(&rcd->rcd_list, &head);
1256                 }
1257
1258                 wait_unlink = wait_unlink &&
1259                               !list_empty(&the_lnet.ln_rcd_zombie);
1260
1261                 lnet_net_unlock(LNET_LOCK_EX);
1262
1263                 while (!list_empty(&head)) {
1264                         rcd = list_entry(head.next,
1265                                          lnet_rc_data_t, rcd_list);
1266                         list_del_init(&rcd->rcd_list);
1267                         lnet_destroy_rc_data(rcd);
1268                 }
1269
1270                 if (!wait_unlink)
1271                         return;
1272
1273                 i++;
1274                 CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET,
1275                        "Waiting for rc buffers to unlink\n");
1276                 cfs_pause(cfs_time_seconds(1) / 4);
1277
1278                 lnet_net_lock(LNET_LOCK_EX);
1279         }
1280
1281         lnet_net_unlock(LNET_LOCK_EX);
1282 }
1283
1284
1285 #if defined(__KERNEL__) && defined(LNET_ROUTER)
1286
1287 static int
1288 lnet_router_checker(void *arg)
1289 {
1290         lnet_peer_t       *rtr;
1291         struct list_head  *entry;
1292
1293         cfs_block_allsigs();
1294
1295         LASSERT (the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING);
1296
1297         while (the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING) {
1298                 __u64   version;
1299                 int     cpt;
1300                 int     cpt2;
1301
1302                 cpt = lnet_net_lock_current();
1303 rescan:
1304                 version = the_lnet.ln_routers_version;
1305
1306                 list_for_each(entry, &the_lnet.ln_routers) {
1307                         rtr = list_entry(entry, lnet_peer_t, lp_rtr_list);
1308
1309                         cpt2 = lnet_cpt_of_nid_locked(rtr->lp_nid);
1310                         if (cpt != cpt2) {
1311                                 lnet_net_unlock(cpt);
1312                                 cpt = cpt2;
1313                                 lnet_net_lock(cpt);
1314                                 /* the routers list has changed */
1315                                 if (version != the_lnet.ln_routers_version)
1316                                         goto rescan;
1317                         }
1318
1319                         lnet_ping_router_locked(rtr);
1320
1321                         /* NB dropped lock */
1322                         if (version != the_lnet.ln_routers_version) {
1323                                 /* the routers list has changed */
1324                                 goto rescan;
1325                         }
1326                 }
1327
1328                 if (the_lnet.ln_routing)
1329                         lnet_update_ni_status_locked();
1330
1331                 lnet_net_unlock(cpt);
1332
1333                 lnet_prune_rc_data(0); /* don't wait for UNLINK */
1334
1335                 /* Call cfs_pause() here always adds 1 to load average
1336                  * because kernel counts # active tasks as nr_running
1337                  * + nr_uninterruptible. */
1338                 schedule_timeout_and_set_state(TASK_INTERRUPTIBLE,
1339                                                    cfs_time_seconds(1));
1340         }
1341
1342         LASSERT(the_lnet.ln_rc_state == LNET_RC_STATE_STOPPING);
1343
1344         lnet_prune_rc_data(1); /* wait for UNLINK */
1345
1346         the_lnet.ln_rc_state = LNET_RC_STATE_SHUTDOWN;
1347         up(&the_lnet.ln_rc_signal);
1348         /* The unlink event callback will signal final completion */
1349         return 0;
1350 }
1351
1352 void
1353 lnet_destroy_rtrbuf(lnet_rtrbuf_t *rb, int npages)
1354 {
1355         int sz = offsetof(lnet_rtrbuf_t, rb_kiov[npages]);
1356
1357         while (--npages >= 0)
1358                 __free_page(rb->rb_kiov[npages].kiov_page);
1359
1360         LIBCFS_FREE(rb, sz);
1361 }
1362
1363 static lnet_rtrbuf_t *
1364 lnet_new_rtrbuf(lnet_rtrbufpool_t *rbp, int cpt)
1365 {
1366         int            npages = rbp->rbp_npages;
1367         int            sz = offsetof(lnet_rtrbuf_t, rb_kiov[npages]);
1368         struct page   *page;
1369         lnet_rtrbuf_t *rb;
1370         int            i;
1371
1372         LIBCFS_CPT_ALLOC(rb, lnet_cpt_table(), cpt, sz);
1373         if (rb == NULL)
1374                 return NULL;
1375
1376         rb->rb_pool = rbp;
1377
1378         for (i = 0; i < npages; i++) {
1379                 page = cfs_page_cpt_alloc(lnet_cpt_table(), cpt,
1380                                           __GFP_ZERO | GFP_IOFS);
1381                 if (page == NULL) {
1382                         while (--i >= 0)
1383                                 __free_page(rb->rb_kiov[i].kiov_page);
1384
1385                         LIBCFS_FREE(rb, sz);
1386                         return NULL;
1387                 }
1388
1389                 rb->rb_kiov[i].kiov_len = PAGE_CACHE_SIZE;
1390                 rb->rb_kiov[i].kiov_offset = 0;
1391                 rb->rb_kiov[i].kiov_page = page;
1392         }
1393
1394         return rb;
1395 }
1396
1397 static void
1398 lnet_rtrpool_free_bufs(lnet_rtrbufpool_t *rbp, int cpt)
1399 {
1400         int              npages = rbp->rbp_npages;
1401         lnet_rtrbuf_t    *rb;
1402         struct list_head tmp;
1403
1404         if (rbp->rbp_nbuffers == 0) /* not initialized or already freed */
1405                 return;
1406
1407         INIT_LIST_HEAD(&tmp);
1408
1409         lnet_net_lock(cpt);
1410         lnet_drop_routed_msgs_locked(&rbp->rbp_msgs, cpt);
1411         list_splice_init(&rbp->rbp_bufs, &tmp);
1412         rbp->rbp_nbuffers = rbp->rbp_credits = 0;
1413         rbp->rbp_mincredits = 0;
1414         lnet_net_unlock(cpt);
1415
1416         /* Free buffers on the free list. */
1417         while (!list_empty(&tmp)) {
1418                 rb = list_entry(tmp.next, lnet_rtrbuf_t, rb_list);
1419                 list_del(&rb->rb_list);
1420                 lnet_destroy_rtrbuf(rb, npages);
1421         }
1422 }
1423
1424 static int
1425 lnet_rtrpool_adjust_bufs(lnet_rtrbufpool_t *rbp, int nbufs, int cpt)
1426 {
1427         struct list_head rb_list;
1428         lnet_rtrbuf_t   *rb;
1429         int             num_rb;
1430         int             num_buffers = 0;
1431         int             npages = rbp->rbp_npages;
1432
1433         /* If we are called for less buffers than already in the pool, we
1434          * just lower the nbuffers number and excess buffers will be
1435          * thrown away as they are returned to the free list.  Credits
1436          * then get adjusted as well. */
1437         if (nbufs <= rbp->rbp_nbuffers) {
1438                 lnet_net_lock(cpt);
1439                 rbp->rbp_nbuffers = nbufs;
1440                 lnet_net_unlock(cpt);
1441                 return 0;
1442         }
1443
1444         INIT_LIST_HEAD(&rb_list);
1445
1446         /* allocate the buffers on a local list first.  If all buffers are
1447          * allocated successfully then join this list to the rbp buffer
1448          * list.  If not then free all allocated buffers. */
1449         num_rb = rbp->rbp_nbuffers;
1450
1451         while (num_rb < nbufs) {
1452                 rb = lnet_new_rtrbuf(rbp, cpt);
1453                 if (rb == NULL) {
1454                         CERROR("Failed to allocate %d route bufs of %d pages\n",
1455                                nbufs, npages);
1456                         goto failed;
1457                 }
1458
1459                 list_add(&rb->rb_list, &rb_list);
1460                 num_buffers++;
1461                 num_rb++;
1462         }
1463
1464         lnet_net_lock(cpt);
1465
1466         list_splice_tail(&rb_list, &rbp->rbp_bufs);
1467         rbp->rbp_nbuffers += num_buffers;
1468         rbp->rbp_credits += num_buffers;
1469         rbp->rbp_mincredits = rbp->rbp_credits;
1470         /* We need to schedule blocked msg using the newly
1471          * added buffers. */
1472         while (!list_empty(&rbp->rbp_bufs) &&
1473                !list_empty(&rbp->rbp_msgs))
1474                 lnet_schedule_blocked_locked(rbp);
1475
1476         lnet_net_unlock(cpt);
1477
1478         return 0;
1479
1480 failed:
1481         while (!list_empty(&rb_list)) {
1482                 rb = list_entry(rb_list.next, lnet_rtrbuf_t, rb_list);
1483                 list_del(&rb->rb_list);
1484                 lnet_destroy_rtrbuf(rb, npages);
1485         }
1486
1487         return -ENOMEM;
1488 }
1489
1490 static void
1491 lnet_rtrpool_init(lnet_rtrbufpool_t *rbp, int npages)
1492 {
1493         INIT_LIST_HEAD(&rbp->rbp_msgs);
1494         INIT_LIST_HEAD(&rbp->rbp_bufs);
1495
1496         rbp->rbp_npages = npages;
1497         rbp->rbp_credits = 0;
1498         rbp->rbp_mincredits = 0;
1499 }
1500
1501 void
1502 lnet_rtrpools_free(int keep_pools)
1503 {
1504         lnet_rtrbufpool_t *rtrp;
1505         int               i;
1506
1507         if (the_lnet.ln_rtrpools == NULL) /* uninitialized or freed */
1508                 return;
1509
1510         cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1511                 lnet_rtrpool_free_bufs(&rtrp[LNET_TINY_BUF_IDX], i);
1512                 lnet_rtrpool_free_bufs(&rtrp[LNET_SMALL_BUF_IDX], i);
1513                 lnet_rtrpool_free_bufs(&rtrp[LNET_LARGE_BUF_IDX], i);
1514         }
1515
1516         if (!keep_pools) {
1517                 cfs_percpt_free(the_lnet.ln_rtrpools);
1518                 the_lnet.ln_rtrpools = NULL;
1519         }
1520 }
1521
1522 static int
1523 lnet_nrb_tiny_calculate(void)
1524 {
1525         int     nrbs = LNET_NRB_TINY;
1526
1527         if (tiny_router_buffers < 0) {
1528                 LCONSOLE_ERROR_MSG(0x10c,
1529                                    "tiny_router_buffers=%d invalid when "
1530                                    "routing enabled\n", tiny_router_buffers);
1531                 return -1;
1532         }
1533
1534         if (tiny_router_buffers > 0)
1535                 nrbs = tiny_router_buffers;
1536
1537         nrbs /= LNET_CPT_NUMBER;
1538         return max(nrbs, LNET_NRB_TINY_MIN);
1539 }
1540
1541 static int
1542 lnet_nrb_small_calculate(void)
1543 {
1544         int     nrbs = LNET_NRB_SMALL;
1545
1546         if (small_router_buffers < 0) {
1547                 LCONSOLE_ERROR_MSG(0x10c,
1548                                    "small_router_buffers=%d invalid when "
1549                                    "routing enabled\n", small_router_buffers);
1550                 return -1;
1551         }
1552
1553         if (small_router_buffers > 0)
1554                 nrbs = small_router_buffers;
1555
1556         nrbs /= LNET_CPT_NUMBER;
1557         return max(nrbs, LNET_NRB_SMALL_MIN);
1558 }
1559
1560 static int
1561 lnet_nrb_large_calculate(void)
1562 {
1563         int     nrbs = LNET_NRB_LARGE;
1564
1565         if (large_router_buffers < 0) {
1566                 LCONSOLE_ERROR_MSG(0x10c,
1567                                    "large_router_buffers=%d invalid when "
1568                                    "routing enabled\n", large_router_buffers);
1569                 return -1;
1570         }
1571
1572         if (large_router_buffers > 0)
1573                 nrbs = large_router_buffers;
1574
1575         nrbs /= LNET_CPT_NUMBER;
1576         return max(nrbs, LNET_NRB_LARGE_MIN);
1577 }
1578
1579 int
1580 lnet_rtrpools_alloc(int im_a_router)
1581 {
1582         lnet_rtrbufpool_t *rtrp;
1583         int     nrb_tiny;
1584         int     nrb_small;
1585         int     nrb_large;
1586         int     rc;
1587         int     i;
1588
1589         if (!strcmp(forwarding, "")) {
1590                 /* not set either way */
1591                 if (!im_a_router)
1592                         return 0;
1593         } else if (!strcmp(forwarding, "disabled")) {
1594                 /* explicitly disabled */
1595                 return 0;
1596         } else if (!strcmp(forwarding, "enabled")) {
1597                 /* explicitly enabled */
1598         } else {
1599                 LCONSOLE_ERROR_MSG(0x10b, "'forwarding' not set to either "
1600                                    "'enabled' or 'disabled'\n");
1601                 return -EINVAL;
1602         }
1603
1604         nrb_tiny = lnet_nrb_tiny_calculate();
1605         if (nrb_tiny < 0)
1606                 return -EINVAL;
1607
1608         nrb_small = lnet_nrb_small_calculate();
1609         if (nrb_small < 0)
1610                 return -EINVAL;
1611
1612         nrb_large = lnet_nrb_large_calculate();
1613         if (nrb_large < 0)
1614                 return -EINVAL;
1615
1616         the_lnet.ln_rtrpools = cfs_percpt_alloc(lnet_cpt_table(),
1617                                                 LNET_NRBPOOLS *
1618                                                 sizeof(lnet_rtrbufpool_t));
1619         if (the_lnet.ln_rtrpools == NULL) {
1620                 LCONSOLE_ERROR_MSG(0x10c,
1621                                    "Failed to initialize router buffe pool\n");
1622                 return -ENOMEM;
1623         }
1624
1625         cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1626                 lnet_rtrpool_init(&rtrp[LNET_TINY_BUF_IDX], 0);
1627                 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_TINY_BUF_IDX],
1628                                               nrb_tiny, i);
1629                 if (rc != 0)
1630                         goto failed;
1631
1632                 lnet_rtrpool_init(&rtrp[LNET_SMALL_BUF_IDX],
1633                                   LNET_NRB_SMALL_PAGES);
1634                 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_SMALL_BUF_IDX],
1635                                               nrb_small, i);
1636                 if (rc != 0)
1637                         goto failed;
1638
1639                 lnet_rtrpool_init(&rtrp[LNET_LARGE_BUF_IDX],
1640                                   LNET_NRB_LARGE_PAGES);
1641                 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_LARGE_BUF_IDX],
1642                                               nrb_large, i);
1643                 if (rc != 0)
1644                         goto failed;
1645         }
1646
1647         lnet_net_lock(LNET_LOCK_EX);
1648         the_lnet.ln_routing = 1;
1649         lnet_net_unlock(LNET_LOCK_EX);
1650         return 0;
1651
1652  failed:
1653         lnet_rtrpools_free(0);
1654         return rc;
1655 }
1656
1657 static int
1658 lnet_rtrpools_adjust_helper(int tiny, int small, int large)
1659 {
1660         int nrb = 0;
1661         int rc = 0;
1662         int i;
1663         lnet_rtrbufpool_t *rtrp;
1664
1665         /* If the provided values for each buffer pool are different than the
1666          * configured values, we need to take action. */
1667         if (tiny >= 0) {
1668                 tiny_router_buffers = tiny;
1669                 nrb = lnet_nrb_tiny_calculate();
1670                 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1671                         rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_TINY_BUF_IDX],
1672                                                       nrb, i);
1673                         if (rc != 0)
1674                                 return rc;
1675                 }
1676         }
1677         if (small >= 0) {
1678                 small_router_buffers = small;
1679                 nrb = lnet_nrb_small_calculate();
1680                 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1681                         rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_SMALL_BUF_IDX],
1682                                                       nrb, i);
1683                         if (rc != 0)
1684                                 return rc;
1685                 }
1686         }
1687         if (large >= 0) {
1688                 large_router_buffers = large;
1689                 nrb = lnet_nrb_large_calculate();
1690                 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1691                         rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_LARGE_BUF_IDX],
1692                                                       nrb, i);
1693                         if (rc != 0)
1694                                 return rc;
1695                 }
1696         }
1697
1698         return 0;
1699 }
1700
1701 int
1702 lnet_rtrpools_adjust(int tiny, int small, int large)
1703 {
1704         /* this function doesn't revert the changes if adding new buffers
1705          * failed.  It's up to the user space caller to revert the
1706          * changes. */
1707
1708         if (!the_lnet.ln_routing)
1709                 return 0;
1710
1711         return lnet_rtrpools_adjust_helper(tiny, small, large);
1712 }
1713
1714 int
1715 lnet_rtrpools_enable(void)
1716 {
1717         int rc;
1718
1719         if (the_lnet.ln_routing)
1720                 return 0;
1721
1722         if (the_lnet.ln_rtrpools == NULL)
1723                 /* If routing is turned off, and we have never
1724                  * initialized the pools before, just call the
1725                  * standard buffer pool allocation routine as
1726                  * if we are just configuring this for the first
1727                  * time. */
1728                 return lnet_rtrpools_alloc(1);
1729
1730         rc = lnet_rtrpools_adjust_helper(0, 0, 0);
1731         if (rc != 0)
1732                 return rc;
1733
1734         lnet_net_lock(LNET_LOCK_EX);
1735         the_lnet.ln_routing = 1;
1736
1737         the_lnet.ln_ping_info->pi_features &= ~LNET_PING_FEAT_RTE_DISABLED;
1738         lnet_net_unlock(LNET_LOCK_EX);
1739
1740         return 0;
1741 }
1742
1743 void
1744 lnet_rtrpools_disable(void)
1745 {
1746         if (!the_lnet.ln_routing)
1747                 return;
1748
1749         lnet_net_lock(LNET_LOCK_EX);
1750         the_lnet.ln_routing = 0;
1751         the_lnet.ln_ping_info->pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1752
1753         tiny_router_buffers = 0;
1754         small_router_buffers = 0;
1755         large_router_buffers = 0;
1756         lnet_net_unlock(LNET_LOCK_EX);
1757         lnet_rtrpools_free(1);
1758 }
1759
1760 int
1761 lnet_notify(lnet_ni_t *ni, lnet_nid_t nid, int alive, cfs_time_t when)
1762 {
1763         struct lnet_peer        *lp = NULL;
1764         cfs_time_t              now = cfs_time_current();
1765         int                     cpt = lnet_cpt_of_nid(nid);
1766
1767         LASSERT (!in_interrupt ());
1768
1769         CDEBUG (D_NET, "%s notifying %s: %s\n",
1770                 (ni == NULL) ? "userspace" : libcfs_nid2str(ni->ni_nid),
1771                 libcfs_nid2str(nid),
1772                 alive ? "up" : "down");
1773
1774         if (ni != NULL &&
1775             LNET_NIDNET(ni->ni_nid) != LNET_NIDNET(nid)) {
1776                 CWARN ("Ignoring notification of %s %s by %s (different net)\n",
1777                         libcfs_nid2str(nid), alive ? "birth" : "death",
1778                         libcfs_nid2str(ni->ni_nid));
1779                 return -EINVAL;
1780         }
1781
1782         /* can't do predictions... */
1783         if (cfs_time_after(when, now)) {
1784                 CWARN ("Ignoring prediction from %s of %s %s "
1785                        "%ld seconds in the future\n",
1786                        (ni == NULL) ? "userspace" : libcfs_nid2str(ni->ni_nid),
1787                        libcfs_nid2str(nid), alive ? "up" : "down",
1788                        cfs_duration_sec(cfs_time_sub(when, now)));
1789                 return -EINVAL;
1790         }
1791
1792         if (ni != NULL && !alive &&             /* LND telling me she's down */
1793             !auto_down) {                       /* auto-down disabled */
1794                 CDEBUG(D_NET, "Auto-down disabled\n");
1795                 return 0;
1796         }
1797
1798         lnet_net_lock(cpt);
1799
1800         if (the_lnet.ln_shutdown) {
1801                 lnet_net_unlock(cpt);
1802                 return -ESHUTDOWN;
1803         }
1804
1805         lp = lnet_find_peer_locked(the_lnet.ln_peer_tables[cpt], nid);
1806         if (lp == NULL) {
1807                 /* nid not found */
1808                 lnet_net_unlock(cpt);
1809                 CDEBUG(D_NET, "%s not found\n", libcfs_nid2str(nid));
1810                 return 0;
1811         }
1812
1813         /* We can't fully trust LND on reporting exact peer last_alive
1814          * if he notifies us about dead peer. For example ksocklnd can
1815          * call us with when == _time_when_the_node_was_booted_ if
1816          * no connections were successfully established */
1817         if (ni != NULL && !alive && when < lp->lp_last_alive)
1818                 when = lp->lp_last_alive;
1819
1820         lnet_notify_locked(lp, ni == NULL, alive, when);
1821
1822         if (ni != NULL)
1823                 lnet_ni_notify_locked(ni, lp);
1824
1825         lnet_peer_decref_locked(lp);
1826
1827         lnet_net_unlock(cpt);
1828         return 0;
1829 }
1830 EXPORT_SYMBOL(lnet_notify);
1831
1832 void
1833 lnet_get_tunables (void)
1834 {
1835         return;
1836 }
1837
1838 #else
1839
1840 int
1841 lnet_notify (lnet_ni_t *ni, lnet_nid_t nid, int alive, cfs_time_t when)
1842 {
1843         return -EOPNOTSUPP;
1844 }
1845
1846 void
1847 lnet_router_checker (void)
1848 {
1849         static time_t last = 0;
1850         static int    running = 0;
1851
1852         time_t            now = cfs_time_current_sec();
1853         int               interval = now - last;
1854         int               rc;
1855         __u64             version;
1856         lnet_peer_t      *rtr;
1857
1858         /* It's no use to call me again within a sec - all intervals and
1859          * timeouts are measured in seconds */
1860         if (last != 0 && interval < 2)
1861                 return;
1862
1863         if (last != 0 &&
1864             interval > MAX(live_router_check_interval,
1865                            dead_router_check_interval))
1866                 CNETERR("Checker(%d/%d) not called for %d seconds\n",
1867                         live_router_check_interval, dead_router_check_interval,
1868                         interval);
1869
1870         LASSERT(LNET_CPT_NUMBER == 1);
1871
1872         lnet_net_lock(0);
1873         LASSERT(!running); /* recursion check */
1874         running = 1;
1875         lnet_net_unlock(0);
1876
1877         last = now;
1878
1879         if (the_lnet.ln_rc_state == LNET_RC_STATE_STOPPING)
1880                 lnet_prune_rc_data(0); /* unlink all rcd and nowait */
1881
1882         /* consume all pending events */
1883         while (1) {
1884                 int          i;
1885                 lnet_event_t ev;
1886
1887                 /* NB ln_rc_eqh must be the 1st in 'eventqs' otherwise the
1888                  * recursion breaker in LNetEQPoll would fail */
1889                 rc = LNetEQPoll(&the_lnet.ln_rc_eqh, 1, 0, &ev, &i);
1890                 if (rc == 0)   /* no event pending */
1891                         break;
1892
1893                 /* NB a lost SENT prevents me from pinging a router again */
1894                 if (rc == -EOVERFLOW) {
1895                         CERROR("Dropped an event!!!\n");
1896                         abort();
1897                 }
1898
1899                 LASSERT (rc == 1);
1900
1901                 lnet_router_checker_event(&ev);
1902         }
1903
1904         if (the_lnet.ln_rc_state == LNET_RC_STATE_STOPPING) {
1905                 lnet_prune_rc_data(1); /* release rcd */
1906                 the_lnet.ln_rc_state = LNET_RC_STATE_SHUTDOWN;
1907                 running = 0;
1908                 return;
1909         }
1910
1911         LASSERT (the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING);
1912
1913         lnet_net_lock(0);
1914
1915         version = the_lnet.ln_routers_version;
1916         list_for_each_entry(rtr, &the_lnet.ln_routers, lp_rtr_list) {
1917                 lnet_ping_router_locked(rtr);
1918                 LASSERT(version == the_lnet.ln_routers_version);
1919         }
1920
1921         lnet_net_unlock(0);
1922
1923         running = 0; /* lock only needed for the recursion check */
1924         return;
1925 }
1926
1927 /* NB lnet_peers_start_down depends on me,
1928  * so must be called before any peer creation */
1929 void
1930 lnet_get_tunables (void)
1931 {
1932         char *s;
1933
1934         s = getenv("LNET_ROUTER_PING_TIMEOUT");
1935         if (s != NULL) router_ping_timeout = atoi(s);
1936
1937         s = getenv("LNET_LIVE_ROUTER_CHECK_INTERVAL");
1938         if (s != NULL) live_router_check_interval = atoi(s);
1939
1940         s = getenv("LNET_DEAD_ROUTER_CHECK_INTERVAL");
1941         if (s != NULL) dead_router_check_interval = atoi(s);
1942
1943         /* This replaces old lnd_notify mechanism */
1944         check_routers_before_use = 1;
1945         if (dead_router_check_interval <= 0)
1946                 dead_router_check_interval = 30;
1947 }
1948
1949 void
1950 lnet_rtrpools_free(int keep_pools)
1951 {
1952 }
1953
1954 int
1955 lnet_rtrpools_alloc(int im_a_arouter)
1956 {
1957         return 0;
1958 }
1959
1960 #endif