drivers/staging/lustre/lnet/lnet/router.c

   1 /*
   2  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
   3  *
   4  * Copyright (c) 2011, 2015, 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  */
  19
  20 #define DEBUG_SUBSYSTEM S_LNET
  21
  22 #include <linux/completion.h>
  23 #include <linux/lnet/lib-lnet.h>
  24
  25 #define LNET_NRB_TINY_MIN       512     /* min value for each CPT */
  26 #define LNET_NRB_TINY           (LNET_NRB_TINY_MIN * 4)
  27 #define LNET_NRB_SMALL_MIN      4096    /* min value for each CPT */
  28 #define LNET_NRB_SMALL          (LNET_NRB_SMALL_MIN * 4)
  29 #define LNET_NRB_SMALL_PAGES    1
  30 #define LNET_NRB_LARGE_MIN      256     /* min value for each CPT */
  31 #define LNET_NRB_LARGE          (LNET_NRB_LARGE_MIN * 4)
  32 #define LNET_NRB_LARGE_PAGES   ((LNET_MTU + PAGE_SIZE - 1) >> \
  33                                  PAGE_SHIFT)
  34
  35 static char *forwarding = "";
  36 module_param(forwarding, charp, 0444);
  37 MODULE_PARM_DESC(forwarding, "Explicitly enable/disable forwarding between networks");
  38
  39 static int tiny_router_buffers;
  40 module_param(tiny_router_buffers, int, 0444);
  41 MODULE_PARM_DESC(tiny_router_buffers, "# of 0 payload messages to buffer in the router");
  42 static int small_router_buffers;
  43 module_param(small_router_buffers, int, 0444);
  44 MODULE_PARM_DESC(small_router_buffers, "# of small (1 page) messages to buffer in the router");
  45 static int large_router_buffers;
  46 module_param(large_router_buffers, int, 0444);
  47 MODULE_PARM_DESC(large_router_buffers, "# of large messages to buffer in the router");
  48 static int peer_buffer_credits;
  49 module_param(peer_buffer_credits, int, 0444);
  50 MODULE_PARM_DESC(peer_buffer_credits, "# router buffer credits per peer");
  51
  52 static int auto_down = 1;
  53 module_param(auto_down, int, 0444);
  54 MODULE_PARM_DESC(auto_down, "Automatically mark peers down on comms error");
  55
  56 int
  57 lnet_peer_buffer_credits(struct lnet_ni *ni)
  58 {
  59         /* NI option overrides LNet default */
  60         if (ni->ni_peerrtrcredits > 0)
  61                 return ni->ni_peerrtrcredits;
  62         if (peer_buffer_credits > 0)
  63                 return peer_buffer_credits;
  64
  65         /*
  66          * As an approximation, allow this peer the same number of router
  67          * buffers as it is allowed outstanding sends
  68          */
  69         return ni->ni_peertxcredits;
  70 }
  71
  72 /* forward ref's */
  73 static int lnet_router_checker(void *);
  74
  75 static int check_routers_before_use;
  76 module_param(check_routers_before_use, int, 0444);
  77 MODULE_PARM_DESC(check_routers_before_use, "Assume routers are down and ping them before use");
  78
  79 int avoid_asym_router_failure = 1;
  80 module_param(avoid_asym_router_failure, int, 0644);
  81 MODULE_PARM_DESC(avoid_asym_router_failure, "Avoid asymmetrical router failures (0 to disable)");
  82
  83 static int dead_router_check_interval = 60;
  84 module_param(dead_router_check_interval, int, 0644);
  85 MODULE_PARM_DESC(dead_router_check_interval, "Seconds between dead router health checks (<= 0 to disable)");
  86
  87 static int live_router_check_interval = 60;
  88 module_param(live_router_check_interval, int, 0644);
  89 MODULE_PARM_DESC(live_router_check_interval, "Seconds between live router health checks (<= 0 to disable)");
  90
  91 static int router_ping_timeout = 50;
  92 module_param(router_ping_timeout, int, 0644);
  93 MODULE_PARM_DESC(router_ping_timeout, "Seconds to wait for the reply to a router health query");
  94
  95 int
  96 lnet_peers_start_down(void)
  97 {
  98         return check_routers_before_use;
  99 }
 100
 101 void
 102 lnet_notify_locked(struct lnet_peer *lp, int notifylnd, int alive,
 103                    unsigned long when)
 104 {
 105         if (time_before(when, lp->lp_timestamp)) { /* out of date information */
 106                 CDEBUG(D_NET, "Out of date\n");
 107                 return;
 108         }
 109
 110         lp->lp_timestamp = when;                /* update timestamp */
 111         lp->lp_ping_deadline = 0;              /* disable ping timeout */
 112
 113         if (lp->lp_alive_count &&         /* got old news */
 114             (!lp->lp_alive) == (!alive)) {      /* new date for old news */
 115                 CDEBUG(D_NET, "Old news\n");
 116                 return;
 117         }
 118
 119         /* Flag that notification is outstanding */
 120
 121         lp->lp_alive_count++;
 122         lp->lp_alive = !(!alive);              /* 1 bit! */
 123         lp->lp_notify = 1;
 124         lp->lp_notifylnd |= notifylnd;
 125         if (lp->lp_alive)
 126                 lp->lp_ping_feats = LNET_PING_FEAT_INVAL; /* reset */
 127
 128         CDEBUG(D_NET, "set %s %d\n", libcfs_nid2str(lp->lp_nid), alive);
 129 }
 130
 131 static void
 132 lnet_ni_notify_locked(struct lnet_ni *ni, struct lnet_peer *lp)
 133 {
 134         int alive;
 135         int notifylnd;
 136
 137         /*
 138          * Notify only in 1 thread at any time to ensure ordered notification.
 139          * NB individual events can be missed; the only guarantee is that you
 140          * always get the most recent news
 141          */
 142         if (lp->lp_notifying || !ni)
 143                 return;
 144
 145         lp->lp_notifying = 1;
 146
 147         while (lp->lp_notify) {
 148                 alive = lp->lp_alive;
 149                 notifylnd = lp->lp_notifylnd;
 150
 151                 lp->lp_notifylnd = 0;
 152                 lp->lp_notify    = 0;
 153
 154                 if (notifylnd && ni->ni_lnd->lnd_notify) {
 155                         lnet_net_unlock(lp->lp_cpt);
 156
 157                         /*
 158                          * A new notification could happen now; I'll handle it
 159                          * when control returns to me
 160                          */
 161                         ni->ni_lnd->lnd_notify(ni, lp->lp_nid, alive);
 162
 163                         lnet_net_lock(lp->lp_cpt);
 164                 }
 165         }
 166
 167         lp->lp_notifying = 0;
 168 }
 169
 170 static void
 171 lnet_rtr_addref_locked(struct lnet_peer *lp)
 172 {
 173         LASSERT(lp->lp_refcount > 0);
 174         LASSERT(lp->lp_rtr_refcount >= 0);
 175
 176         /* lnet_net_lock must be exclusively locked */
 177         lp->lp_rtr_refcount++;
 178         if (lp->lp_rtr_refcount == 1) {
 179                 struct list_head *pos;
 180
 181                 /* a simple insertion sort */
 182                 list_for_each_prev(pos, &the_lnet.ln_routers) {
 183                         struct lnet_peer *rtr;
 184
 185                         rtr = list_entry(pos, struct lnet_peer, lp_rtr_list);
 186                         if (rtr->lp_nid < lp->lp_nid)
 187                                 break;
 188                 }
 189
 190                 list_add(&lp->lp_rtr_list, pos);
 191                 /* addref for the_lnet.ln_routers */
 192                 lnet_peer_addref_locked(lp);
 193                 the_lnet.ln_routers_version++;
 194         }
 195 }
 196
 197 static void
 198 lnet_rtr_decref_locked(struct lnet_peer *lp)
 199 {
 200         LASSERT(lp->lp_refcount > 0);
 201         LASSERT(lp->lp_rtr_refcount > 0);
 202
 203         /* lnet_net_lock must be exclusively locked */
 204         lp->lp_rtr_refcount--;
 205         if (!lp->lp_rtr_refcount) {
 206                 LASSERT(list_empty(&lp->lp_routes));
 207
 208                 if (lp->lp_rcd) {
 209                         list_add(&lp->lp_rcd->rcd_list,
 210                                  &the_lnet.ln_rcd_deathrow);
 211                         lp->lp_rcd = NULL;
 212                 }
 213
 214                 list_del(&lp->lp_rtr_list);
 215                 /* decref for the_lnet.ln_routers */
 216                 lnet_peer_decref_locked(lp);
 217                 the_lnet.ln_routers_version++;
 218         }
 219 }
 220
 221 struct lnet_remotenet *
 222 lnet_find_net_locked(__u32 net)
 223 {
 224         struct lnet_remotenet *rnet;
 225         struct list_head *tmp;
 226         struct list_head *rn_list;
 227
 228         LASSERT(!the_lnet.ln_shutdown);
 229
 230         rn_list = lnet_net2rnethash(net);
 231         list_for_each(tmp, rn_list) {
 232                 rnet = list_entry(tmp, struct lnet_remotenet, lrn_list);
 233
 234                 if (rnet->lrn_net == net)
 235                         return rnet;
 236         }
 237         return NULL;
 238 }
 239
 240 static void lnet_shuffle_seed(void)
 241 {
 242         static int seeded;
 243         __u32 lnd_type, seed[2];
 244         struct timespec64 ts;
 245         struct lnet_ni *ni;
 246         struct list_head *tmp;
 247
 248         if (seeded)
 249                 return;
 250
 251         cfs_get_random_bytes(seed, sizeof(seed));
 252
 253         /*
 254          * Nodes with small feet have little entropy
 255          * the NID for this node gives the most entropy in the low bits
 256          */
 257         list_for_each(tmp, &the_lnet.ln_nis) {
 258                 ni = list_entry(tmp, struct lnet_ni, ni_list);
 259                 lnd_type = LNET_NETTYP(LNET_NIDNET(ni->ni_nid));
 260
 261                 if (lnd_type != LOLND)
 262                         seed[0] ^= (LNET_NIDADDR(ni->ni_nid) | lnd_type);
 263         }
 264
 265         ktime_get_ts64(&ts);
 266         cfs_srand(ts.tv_sec ^ seed[0], ts.tv_nsec ^ seed[1]);
 267         seeded = 1;
 268 }
 269
 270 /* NB expects LNET_LOCK held */
 271 static void
 272 lnet_add_route_to_rnet(struct lnet_remotenet *rnet, struct lnet_route *route)
 273 {
 274         unsigned int len = 0;
 275         unsigned int offset = 0;
 276         struct list_head *e;
 277
 278         lnet_shuffle_seed();
 279
 280         list_for_each(e, &rnet->lrn_routes) {
 281                 len++;
 282         }
 283
 284         /* len+1 positions to add a new entry, also prevents division by 0 */
 285         offset = cfs_rand() % (len + 1);
 286         list_for_each(e, &rnet->lrn_routes) {
 287                 if (!offset)
 288                         break;
 289                 offset--;
 290         }
 291         list_add(&route->lr_list, e);
 292         list_add(&route->lr_gwlist, &route->lr_gateway->lp_routes);
 293
 294         the_lnet.ln_remote_nets_version++;
 295         lnet_rtr_addref_locked(route->lr_gateway);
 296 }
 297
 298 int
 299 lnet_add_route(__u32 net, __u32 hops, lnet_nid_t gateway,
 300                unsigned int priority)
 301 {
 302         struct list_head *e;
 303         struct lnet_remotenet *rnet;
 304         struct lnet_remotenet *rnet2;
 305         struct lnet_route *route;
 306         struct lnet_ni *ni;
 307         int add_route;
 308         int rc;
 309
 310         CDEBUG(D_NET, "Add route: net %s hops %d priority %u gw %s\n",
 311                libcfs_net2str(net), hops, priority, libcfs_nid2str(gateway));
 312
 313         if (gateway == LNET_NID_ANY ||
 314             LNET_NETTYP(LNET_NIDNET(gateway)) == LOLND ||
 315             net == LNET_NIDNET(LNET_NID_ANY) ||
 316             LNET_NETTYP(net) == LOLND ||
 317             LNET_NIDNET(gateway) == net ||
 318             (hops != LNET_UNDEFINED_HOPS && (hops < 1 || hops > 255)))
 319                 return -EINVAL;
 320
 321         if (lnet_islocalnet(net))              /* it's a local network */
 322                 return -EEXIST;
 323
 324         /* Assume net, route, all new */
 325         LIBCFS_ALLOC(route, sizeof(*route));
 326         LIBCFS_ALLOC(rnet, sizeof(*rnet));
 327         if (!route || !rnet) {
 328                 CERROR("Out of memory creating route %s %d %s\n",
 329                        libcfs_net2str(net), hops, libcfs_nid2str(gateway));
 330                 if (route)
 331                         LIBCFS_FREE(route, sizeof(*route));
 332                 if (rnet)
 333                         LIBCFS_FREE(rnet, sizeof(*rnet));
 334                 return -ENOMEM;
 335         }
 336
 337         INIT_LIST_HEAD(&rnet->lrn_routes);
 338         rnet->lrn_net = net;
 339         route->lr_hops = hops;
 340         route->lr_net = net;
 341         route->lr_priority = priority;
 342
 343         lnet_net_lock(LNET_LOCK_EX);
 344
 345         rc = lnet_nid2peer_locked(&route->lr_gateway, gateway, LNET_LOCK_EX);
 346         if (rc) {
 347                 lnet_net_unlock(LNET_LOCK_EX);
 348
 349                 LIBCFS_FREE(route, sizeof(*route));
 350                 LIBCFS_FREE(rnet, sizeof(*rnet));
 351
 352                 if (rc == -EHOSTUNREACH) /* gateway is not on a local net */
 353                         return rc;      /* ignore the route entry */
 354                 CERROR("Error %d creating route %s %d %s\n", rc,
 355                        libcfs_net2str(net), hops,
 356                        libcfs_nid2str(gateway));
 357                 return rc;
 358         }
 359
 360         LASSERT(!the_lnet.ln_shutdown);
 361
 362         rnet2 = lnet_find_net_locked(net);
 363         if (!rnet2) {
 364                 /* new network */
 365                 list_add_tail(&rnet->lrn_list, lnet_net2rnethash(net));
 366                 rnet2 = rnet;
 367         }
 368
 369         /* Search for a duplicate route (it's a NOOP if it is) */
 370         add_route = 1;
 371         list_for_each(e, &rnet2->lrn_routes) {
 372                 struct lnet_route *route2;
 373
 374                 route2 = list_entry(e, struct lnet_route, lr_list);
 375                 if (route2->lr_gateway == route->lr_gateway) {
 376                         add_route = 0;
 377                         break;
 378                 }
 379
 380                 /* our lookups must be true */
 381                 LASSERT(route2->lr_gateway->lp_nid != gateway);
 382         }
 383
 384         if (add_route) {
 385                 lnet_peer_addref_locked(route->lr_gateway); /* +1 for notify */
 386                 lnet_add_route_to_rnet(rnet2, route);
 387
 388                 ni = route->lr_gateway->lp_ni;
 389                 lnet_net_unlock(LNET_LOCK_EX);
 390
 391                 /* XXX Assume alive */
 392                 if (ni->ni_lnd->lnd_notify)
 393                         ni->ni_lnd->lnd_notify(ni, gateway, 1);
 394
 395                 lnet_net_lock(LNET_LOCK_EX);
 396         }
 397
 398         /* -1 for notify or !add_route */
 399         lnet_peer_decref_locked(route->lr_gateway);
 400         lnet_net_unlock(LNET_LOCK_EX);
 401         rc = 0;
 402
 403         if (!add_route) {
 404                 rc = -EEXIST;
 405                 LIBCFS_FREE(route, sizeof(*route));
 406         }
 407
 408         if (rnet != rnet2)
 409                 LIBCFS_FREE(rnet, sizeof(*rnet));
 410
 411         /* indicate to startup the router checker if configured */
 412         wake_up(&the_lnet.ln_rc_waitq);
 413
 414         return rc;
 415 }
 416
 417 int
 418 lnet_check_routes(void)
 419 {
 420         struct lnet_remotenet *rnet;
 421         struct lnet_route *route;
 422         struct lnet_route *route2;
 423         struct list_head *e1;
 424         struct list_head *e2;
 425         int cpt;
 426         struct list_head *rn_list;
 427         int i;
 428
 429         cpt = lnet_net_lock_current();
 430
 431         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) {
 432                 rn_list = &the_lnet.ln_remote_nets_hash[i];
 433                 list_for_each(e1, rn_list) {
 434                         rnet = list_entry(e1, struct lnet_remotenet, lrn_list);
 435
 436                         route2 = NULL;
 437                         list_for_each(e2, &rnet->lrn_routes) {
 438                                 lnet_nid_t nid1;
 439                                 lnet_nid_t nid2;
 440                                 int net;
 441
 442                                 route = list_entry(e2, struct lnet_route, lr_list);
 443
 444                                 if (!route2) {
 445                                         route2 = route;
 446                                         continue;
 447                                 }
 448
 449                                 if (route->lr_gateway->lp_ni ==
 450                                     route2->lr_gateway->lp_ni)
 451                                         continue;
 452
 453                                 nid1 = route->lr_gateway->lp_nid;
 454                                 nid2 = route2->lr_gateway->lp_nid;
 455                                 net = rnet->lrn_net;
 456
 457                                 lnet_net_unlock(cpt);
 458
 459                                 CERROR("Routes to %s via %s and %s not supported\n",
 460                                        libcfs_net2str(net),
 461                                        libcfs_nid2str(nid1),
 462                                        libcfs_nid2str(nid2));
 463                                 return -EINVAL;
 464                         }
 465                 }
 466         }
 467
 468         lnet_net_unlock(cpt);
 469         return 0;
 470 }
 471
 472 int
 473 lnet_del_route(__u32 net, lnet_nid_t gw_nid)
 474 {
 475         struct lnet_peer *gateway;
 476         struct lnet_remotenet *rnet;
 477         struct lnet_route *route;
 478         struct list_head *e1;
 479         struct list_head *e2;
 480         int rc = -ENOENT;
 481         struct list_head *rn_list;
 482         int idx = 0;
 483
 484         CDEBUG(D_NET, "Del route: net %s : gw %s\n",
 485                libcfs_net2str(net), libcfs_nid2str(gw_nid));
 486
 487         /*
 488          * NB Caller may specify either all routes via the given gateway
 489          * or a specific route entry actual NIDs)
 490          */
 491         lnet_net_lock(LNET_LOCK_EX);
 492         if (net == LNET_NIDNET(LNET_NID_ANY))
 493                 rn_list = &the_lnet.ln_remote_nets_hash[0];
 494         else
 495                 rn_list = lnet_net2rnethash(net);
 496
 497  again:
 498         list_for_each(e1, rn_list) {
 499                 rnet = list_entry(e1, struct lnet_remotenet, lrn_list);
 500
 501                 if (!(net == LNET_NIDNET(LNET_NID_ANY) ||
 502                       net == rnet->lrn_net))
 503                         continue;
 504
 505                 list_for_each(e2, &rnet->lrn_routes) {
 506                         route = list_entry(e2, struct lnet_route, lr_list);
 507
 508                         gateway = route->lr_gateway;
 509                         if (!(gw_nid == LNET_NID_ANY ||
 510                               gw_nid == gateway->lp_nid))
 511                                 continue;
 512
 513                         list_del(&route->lr_list);
 514                         list_del(&route->lr_gwlist);
 515                         the_lnet.ln_remote_nets_version++;
 516
 517                         if (list_empty(&rnet->lrn_routes))
 518                                 list_del(&rnet->lrn_list);
 519                         else
 520                                 rnet = NULL;
 521
 522                         lnet_rtr_decref_locked(gateway);
 523                         lnet_peer_decref_locked(gateway);
 524
 525                         lnet_net_unlock(LNET_LOCK_EX);
 526
 527                         LIBCFS_FREE(route, sizeof(*route));
 528
 529                         if (rnet)
 530                                 LIBCFS_FREE(rnet, sizeof(*rnet));
 531
 532                         rc = 0;
 533                         lnet_net_lock(LNET_LOCK_EX);
 534                         goto again;
 535                 }
 536         }
 537
 538         if (net == LNET_NIDNET(LNET_NID_ANY) &&
 539             ++idx < LNET_REMOTE_NETS_HASH_SIZE) {
 540                 rn_list = &the_lnet.ln_remote_nets_hash[idx];
 541                 goto again;
 542         }
 543         lnet_net_unlock(LNET_LOCK_EX);
 544
 545         return rc;
 546 }
 547
 548 void
 549 lnet_destroy_routes(void)
 550 {
 551         lnet_del_route(LNET_NIDNET(LNET_NID_ANY), LNET_NID_ANY);
 552 }
 553
 554 int lnet_get_rtr_pool_cfg(int idx, struct lnet_ioctl_pool_cfg *pool_cfg)
 555 {
 556         int i, rc = -ENOENT, j;
 557
 558         if (!the_lnet.ln_rtrpools)
 559                 return rc;
 560
 561         for (i = 0; i < LNET_NRBPOOLS; i++) {
 562                 struct lnet_rtrbufpool *rbp;
 563
 564                 lnet_net_lock(LNET_LOCK_EX);
 565                 cfs_percpt_for_each(rbp, j, the_lnet.ln_rtrpools) {
 566                         if (i++ != idx)
 567                                 continue;
 568
 569                         pool_cfg->pl_pools[i].pl_npages = rbp[i].rbp_npages;
 570                         pool_cfg->pl_pools[i].pl_nbuffers = rbp[i].rbp_nbuffers;
 571                         pool_cfg->pl_pools[i].pl_credits = rbp[i].rbp_credits;
 572                         pool_cfg->pl_pools[i].pl_mincredits = rbp[i].rbp_mincredits;
 573                         rc = 0;
 574                         break;
 575                 }
 576                 lnet_net_unlock(LNET_LOCK_EX);
 577         }
 578
 579         lnet_net_lock(LNET_LOCK_EX);
 580         pool_cfg->pl_routing = the_lnet.ln_routing;
 581         lnet_net_unlock(LNET_LOCK_EX);
 582
 583         return rc;
 584 }
 585
 586 int
 587 lnet_get_route(int idx, __u32 *net, __u32 *hops,
 588                lnet_nid_t *gateway, __u32 *alive, __u32 *priority)
 589 {
 590         struct list_head *e1;
 591         struct list_head *e2;
 592         struct lnet_remotenet *rnet;
 593         struct lnet_route *route;
 594         int cpt;
 595         int i;
 596         struct list_head *rn_list;
 597
 598         cpt = lnet_net_lock_current();
 599
 600         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++) {
 601                 rn_list = &the_lnet.ln_remote_nets_hash[i];
 602                 list_for_each(e1, rn_list) {
 603                         rnet = list_entry(e1, struct lnet_remotenet, lrn_list);
 604
 605                         list_for_each(e2, &rnet->lrn_routes) {
 606                                 route = list_entry(e2, struct lnet_route,
 607                                                    lr_list);
 608
 609                                 if (!idx--) {
 610                                         *net      = rnet->lrn_net;
 611                                         *hops     = route->lr_hops;
 612                                         *priority = route->lr_priority;
 613                                         *gateway  = route->lr_gateway->lp_nid;
 614                                         *alive = lnet_is_route_alive(route);
 615                                         lnet_net_unlock(cpt);
 616                                         return 0;
 617                                 }
 618                         }
 619                 }
 620         }
 621
 622         lnet_net_unlock(cpt);
 623         return -ENOENT;
 624 }
 625
 626 void
 627 lnet_swap_pinginfo(struct lnet_ping_info *info)
 628 {
 629         int i;
 630         struct lnet_ni_status *stat;
 631
 632         __swab32s(&info->pi_magic);
 633         __swab32s(&info->pi_features);
 634         __swab32s(&info->pi_pid);
 635         __swab32s(&info->pi_nnis);
 636         for (i = 0; i < info->pi_nnis && i < LNET_MAX_RTR_NIS; i++) {
 637                 stat = &info->pi_ni[i];
 638                 __swab64s(&stat->ns_nid);
 639                 __swab32s(&stat->ns_status);
 640         }
 641 }
 642
 643 /**
 644  * parse router-checker pinginfo, record number of down NIs for remote
 645  * networks on that router.
 646  */
 647 static void
 648 lnet_parse_rc_info(struct lnet_rc_data *rcd)
 649 {
 650         struct lnet_ping_info *info = rcd->rcd_pinginfo;
 651         struct lnet_peer *gw = rcd->rcd_gateway;
 652         struct lnet_route *rte;
 653
 654         if (!gw->lp_alive)
 655                 return;
 656
 657         if (info->pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
 658                 lnet_swap_pinginfo(info);
 659
 660         /* NB always racing with network! */
 661         if (info->pi_magic != LNET_PROTO_PING_MAGIC) {
 662                 CDEBUG(D_NET, "%s: Unexpected magic %08x\n",
 663                        libcfs_nid2str(gw->lp_nid), info->pi_magic);
 664                 gw->lp_ping_feats = LNET_PING_FEAT_INVAL;
 665                 return;
 666         }
 667
 668         gw->lp_ping_feats = info->pi_features;
 669         if (!(gw->lp_ping_feats & LNET_PING_FEAT_MASK)) {
 670                 CDEBUG(D_NET, "%s: Unexpected features 0x%x\n",
 671                        libcfs_nid2str(gw->lp_nid), gw->lp_ping_feats);
 672                 return; /* nothing I can understand */
 673         }
 674
 675         if (!(gw->lp_ping_feats & LNET_PING_FEAT_NI_STATUS))
 676                 return; /* can't carry NI status info */
 677
 678         list_for_each_entry(rte, &gw->lp_routes, lr_gwlist) {
 679                 int down = 0;
 680                 int up = 0;
 681                 int i;
 682
 683                 if (gw->lp_ping_feats & LNET_PING_FEAT_RTE_DISABLED) {
 684                         rte->lr_downis = 1;
 685                         continue;
 686                 }
 687
 688                 for (i = 0; i < info->pi_nnis && i < LNET_MAX_RTR_NIS; i++) {
 689                         struct lnet_ni_status *stat = &info->pi_ni[i];
 690                         lnet_nid_t nid = stat->ns_nid;
 691
 692                         if (nid == LNET_NID_ANY) {
 693                                 CDEBUG(D_NET, "%s: unexpected LNET_NID_ANY\n",
 694                                        libcfs_nid2str(gw->lp_nid));
 695                                 gw->lp_ping_feats = LNET_PING_FEAT_INVAL;
 696                                 return;
 697                         }
 698
 699                         if (LNET_NETTYP(LNET_NIDNET(nid)) == LOLND)
 700                                 continue;
 701
 702                         if (stat->ns_status == LNET_NI_STATUS_DOWN) {
 703                                 down++;
 704                                 continue;
 705                         }
 706
 707                         if (stat->ns_status == LNET_NI_STATUS_UP) {
 708                                 if (LNET_NIDNET(nid) == rte->lr_net) {
 709                                         up = 1;
 710                                         break;
 711                                 }
 712                                 continue;
 713                         }
 714
 715                         CDEBUG(D_NET, "%s: Unexpected status 0x%x\n",
 716                                libcfs_nid2str(gw->lp_nid), stat->ns_status);
 717                         gw->lp_ping_feats = LNET_PING_FEAT_INVAL;
 718                         return;
 719                 }
 720
 721                 if (up) { /* ignore downed NIs if NI for dest network is up */
 722                         rte->lr_downis = 0;
 723                         continue;
 724                 }
 725                 /**
 726                  * if @down is zero and this route is single-hop, it means
 727                  * we can't find NI for target network
 728                  */
 729                 if (!down && rte->lr_hops == 1)
 730                         down = 1;
 731
 732                 rte->lr_downis = down;
 733         }
 734 }
 735
 736 static void
 737 lnet_router_checker_event(struct lnet_event *event)
 738 {
 739         struct lnet_rc_data *rcd = event->md.user_ptr;
 740         struct lnet_peer *lp;
 741
 742         LASSERT(rcd);
 743
 744         if (event->unlinked) {
 745                 LNetInvalidateMDHandle(&rcd->rcd_mdh);
 746                 return;
 747         }
 748
 749         LASSERT(event->type == LNET_EVENT_SEND ||
 750                 event->type == LNET_EVENT_REPLY);
 751
 752         lp = rcd->rcd_gateway;
 753         LASSERT(lp);
 754
 755         /*
 756          * NB: it's called with holding lnet_res_lock, we have a few
 757          * places need to hold both locks at the same time, please take
 758          * care of lock ordering
 759          */
 760         lnet_net_lock(lp->lp_cpt);
 761         if (!lnet_isrouter(lp) || lp->lp_rcd != rcd) {
 762                 /* ignore if no longer a router or rcd is replaced */
 763                 goto out;
 764         }
 765
 766         if (event->type == LNET_EVENT_SEND) {
 767                 lp->lp_ping_notsent = 0;
 768                 if (!event->status)
 769                         goto out;
 770         }
 771
 772         /* LNET_EVENT_REPLY */
 773         /*
 774          * A successful REPLY means the router is up.  If _any_ comms
 775          * to the router fail I assume it's down (this will happen if
 776          * we ping alive routers to try to detect router death before
 777          * apps get burned).
 778          */
 779         lnet_notify_locked(lp, 1, !event->status, cfs_time_current());
 780
 781         /*
 782          * The router checker will wake up very shortly and do the
 783          * actual notification.
 784          * XXX If 'lp' stops being a router before then, it will still
 785          * have the notification pending!!!
 786          */
 787         if (avoid_asym_router_failure && !event->status)
 788                 lnet_parse_rc_info(rcd);
 789
 790  out:
 791         lnet_net_unlock(lp->lp_cpt);
 792 }
 793
 794 static void
 795 lnet_wait_known_routerstate(void)
 796 {
 797         struct lnet_peer *rtr;
 798         struct list_head *entry;
 799         int all_known;
 800
 801         LASSERT(the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING);
 802
 803         for (;;) {
 804                 int cpt = lnet_net_lock_current();
 805
 806                 all_known = 1;
 807                 list_for_each(entry, &the_lnet.ln_routers) {
 808                         rtr = list_entry(entry, struct lnet_peer, lp_rtr_list);
 809
 810                         if (!rtr->lp_alive_count) {
 811                                 all_known = 0;
 812                                 break;
 813                         }
 814                 }
 815
 816                 lnet_net_unlock(cpt);
 817
 818                 if (all_known)
 819                         return;
 820
 821                 set_current_state(TASK_UNINTERRUPTIBLE);
 822                 schedule_timeout(cfs_time_seconds(1));
 823         }
 824 }
 825
 826 void
 827 lnet_router_ni_update_locked(struct lnet_peer *gw, __u32 net)
 828 {
 829         struct lnet_route *rte;
 830
 831         if ((gw->lp_ping_feats & LNET_PING_FEAT_NI_STATUS)) {
 832                 list_for_each_entry(rte, &gw->lp_routes, lr_gwlist) {
 833                         if (rte->lr_net == net) {
 834                                 rte->lr_downis = 0;
 835                                 break;
 836                         }
 837                 }
 838         }
 839 }
 840
 841 static void
 842 lnet_update_ni_status_locked(void)
 843 {
 844         struct lnet_ni *ni;
 845         time64_t now;
 846         int timeout;
 847
 848         LASSERT(the_lnet.ln_routing);
 849
 850         timeout = router_ping_timeout +
 851                   max(live_router_check_interval, dead_router_check_interval);
 852
 853         now = ktime_get_real_seconds();
 854         list_for_each_entry(ni, &the_lnet.ln_nis, ni_list) {
 855                 if (ni->ni_lnd->lnd_type == LOLND)
 856                         continue;
 857
 858                 if (now < ni->ni_last_alive + timeout)
 859                         continue;
 860
 861                 lnet_ni_lock(ni);
 862                 /* re-check with lock */
 863                 if (now < ni->ni_last_alive + timeout) {
 864                         lnet_ni_unlock(ni);
 865                         continue;
 866                 }
 867
 868                 LASSERT(ni->ni_status);
 869
 870                 if (ni->ni_status->ns_status != LNET_NI_STATUS_DOWN) {
 871                         CDEBUG(D_NET, "NI(%s:%d) status changed to down\n",
 872                                libcfs_nid2str(ni->ni_nid), timeout);
 873                         /*
 874                          * NB: so far, this is the only place to set
 875                          * NI status to "down"
 876                          */
 877                         ni->ni_status->ns_status = LNET_NI_STATUS_DOWN;
 878                 }
 879                 lnet_ni_unlock(ni);
 880         }
 881 }
 882
 883 static void
 884 lnet_destroy_rc_data(struct lnet_rc_data *rcd)
 885 {
 886         LASSERT(list_empty(&rcd->rcd_list));
 887         /* detached from network */
 888         LASSERT(LNetMDHandleIsInvalid(rcd->rcd_mdh));
 889
 890         if (rcd->rcd_gateway) {
 891                 int cpt = rcd->rcd_gateway->lp_cpt;
 892
 893                 lnet_net_lock(cpt);
 894                 lnet_peer_decref_locked(rcd->rcd_gateway);
 895                 lnet_net_unlock(cpt);
 896         }
 897
 898         if (rcd->rcd_pinginfo)
 899                 LIBCFS_FREE(rcd->rcd_pinginfo, LNET_PINGINFO_SIZE);
 900
 901         LIBCFS_FREE(rcd, sizeof(*rcd));
 902 }
 903
 904 static struct lnet_rc_data *
 905 lnet_create_rc_data_locked(struct lnet_peer *gateway)
 906 {
 907         struct lnet_rc_data *rcd = NULL;
 908         struct lnet_ping_info *pi;
 909         struct lnet_md md;
 910         int rc;
 911         int i;
 912
 913         lnet_net_unlock(gateway->lp_cpt);
 914
 915         LIBCFS_ALLOC(rcd, sizeof(*rcd));
 916         if (!rcd)
 917                 goto out;
 918
 919         LNetInvalidateMDHandle(&rcd->rcd_mdh);
 920         INIT_LIST_HEAD(&rcd->rcd_list);
 921
 922         LIBCFS_ALLOC(pi, LNET_PINGINFO_SIZE);
 923         if (!pi)
 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         md.start = pi;
 933         md.user_ptr = rcd;
 934         md.length = LNET_PINGINFO_SIZE;
 935         md.threshold = LNET_MD_THRESH_INF;
 936         md.options = LNET_MD_TRUNCATE;
 937         md.eq_handle = the_lnet.ln_rc_eqh;
 938
 939         LASSERT(!LNetEQHandleIsInvalid(the_lnet.ln_rc_eqh));
 940         rc = LNetMDBind(md, LNET_UNLINK, &rcd->rcd_mdh);
 941         if (rc < 0) {
 942                 CERROR("Can't bind MD: %d\n", rc);
 943                 goto out;
 944         }
 945         LASSERT(!rc);
 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) {
 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) {
 963                 if (!LNetMDHandleIsInvalid(rcd->rcd_mdh)) {
 964                         rc = LNetMDUnlink(rcd->rcd_mdh);
 965                         LASSERT(!rc);
 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(struct lnet_peer *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(struct lnet_peer *rtr)
 989 {
 990         struct lnet_rc_data *rcd = NULL;
 991         unsigned long now = cfs_time_current();
 992         int secs;
 993
 994         lnet_peer_addref_locked(rtr);
 995
 996         if (rtr->lp_ping_deadline && /* 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 ?
1011               rtr->lp_rcd : lnet_create_rc_data_locked(rtr);
1012
1013         if (!rcd)
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 alive_count %d lp_ping_timestamp %lu\n",
1020                libcfs_nid2str(rtr->lp_nid), secs,
1021                rtr->lp_ping_deadline, rtr->lp_ping_notsent,
1022                rtr->lp_alive, rtr->lp_alive_count, rtr->lp_ping_timestamp);
1023
1024         if (secs && !rtr->lp_ping_notsent &&
1025             cfs_time_after(now, cfs_time_add(rtr->lp_ping_timestamp,
1026                                              cfs_time_seconds(secs)))) {
1027                 int rc;
1028                 struct lnet_process_id id;
1029                 struct lnet_handle_md mdh;
1030
1031                 id.nid = rtr->lp_nid;
1032                 id.pid = LNET_PID_LUSTRE;
1033                 CDEBUG(D_NET, "Check: %s\n", libcfs_id2str(id));
1034
1035                 rtr->lp_ping_notsent   = 1;
1036                 rtr->lp_ping_timestamp = now;
1037
1038                 mdh = rcd->rcd_mdh;
1039
1040                 if (!rtr->lp_ping_deadline) {
1041                         rtr->lp_ping_deadline =
1042                                 cfs_time_shift(router_ping_timeout);
1043                 }
1044
1045                 lnet_net_unlock(rtr->lp_cpt);
1046
1047                 rc = LNetGet(LNET_NID_ANY, mdh, id, LNET_RESERVED_PORTAL,
1048                              LNET_PROTO_PING_MATCHBITS, 0);
1049
1050                 lnet_net_lock(rtr->lp_cpt);
1051                 if (rc)
1052                         rtr->lp_ping_notsent = 0; /* no event pending */
1053         }
1054
1055         lnet_peer_decref_locked(rtr);
1056 }
1057
1058 int
1059 lnet_router_checker_start(void)
1060 {
1061         struct task_struct *task;
1062         int rc;
1063         int eqsz = 0;
1064
1065         LASSERT(the_lnet.ln_rc_state == LNET_RC_STATE_SHUTDOWN);
1066
1067         if (check_routers_before_use &&
1068             dead_router_check_interval <= 0) {
1069                 LCONSOLE_ERROR_MSG(0x10a, "'dead_router_check_interval' must be set if 'check_routers_before_use' is set\n");
1070                 return -EINVAL;
1071         }
1072
1073         init_completion(&the_lnet.ln_rc_signal);
1074
1075         rc = LNetEQAlloc(0, lnet_router_checker_event, &the_lnet.ln_rc_eqh);
1076         if (rc) {
1077                 CERROR("Can't allocate EQ(%d): %d\n", eqsz, rc);
1078                 return -ENOMEM;
1079         }
1080
1081         the_lnet.ln_rc_state = LNET_RC_STATE_RUNNING;
1082         task = kthread_run(lnet_router_checker, NULL, "router_checker");
1083         if (IS_ERR(task)) {
1084                 rc = PTR_ERR(task);
1085                 CERROR("Can't start router checker thread: %d\n", rc);
1086                 /* block until event callback signals exit */
1087                 wait_for_completion(&the_lnet.ln_rc_signal);
1088                 rc = LNetEQFree(the_lnet.ln_rc_eqh);
1089                 LASSERT(!rc);
1090                 the_lnet.ln_rc_state = LNET_RC_STATE_SHUTDOWN;
1091                 return -ENOMEM;
1092         }
1093
1094         if (check_routers_before_use) {
1095                 /*
1096                  * Note that a helpful side-effect of pinging all known routers
1097                  * at startup is that it makes them drop stale connections they
1098                  * may have to a previous instance of me.
1099                  */
1100                 lnet_wait_known_routerstate();
1101         }
1102
1103         return 0;
1104 }
1105
1106 void
1107 lnet_router_checker_stop(void)
1108 {
1109         int rc;
1110
1111         if (the_lnet.ln_rc_state == LNET_RC_STATE_SHUTDOWN)
1112                 return;
1113
1114         LASSERT(the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING);
1115         the_lnet.ln_rc_state = LNET_RC_STATE_STOPPING;
1116         /* wakeup the RC thread if it's sleeping */
1117         wake_up(&the_lnet.ln_rc_waitq);
1118
1119         /* block until event callback signals exit */
1120         wait_for_completion(&the_lnet.ln_rc_signal);
1121         LASSERT(the_lnet.ln_rc_state == LNET_RC_STATE_SHUTDOWN);
1122
1123         rc = LNetEQFree(the_lnet.ln_rc_eqh);
1124         LASSERT(!rc);
1125 }
1126
1127 static void
1128 lnet_prune_rc_data(int wait_unlink)
1129 {
1130         struct lnet_rc_data *rcd;
1131         struct lnet_rc_data *tmp;
1132         struct lnet_peer *lp;
1133         struct list_head head;
1134         int i = 2;
1135
1136         if (likely(the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING &&
1137                    list_empty(&the_lnet.ln_rcd_deathrow) &&
1138                    list_empty(&the_lnet.ln_rcd_zombie)))
1139                 return;
1140
1141         INIT_LIST_HEAD(&head);
1142
1143         lnet_net_lock(LNET_LOCK_EX);
1144
1145         if (the_lnet.ln_rc_state != LNET_RC_STATE_RUNNING) {
1146                 /* router checker is stopping, prune all */
1147                 list_for_each_entry(lp, &the_lnet.ln_routers,
1148                                     lp_rtr_list) {
1149                         if (!lp->lp_rcd)
1150                                 continue;
1151
1152                         LASSERT(list_empty(&lp->lp_rcd->rcd_list));
1153                         list_add(&lp->lp_rcd->rcd_list,
1154                                  &the_lnet.ln_rcd_deathrow);
1155                         lp->lp_rcd = NULL;
1156                 }
1157         }
1158
1159         /* unlink all RCDs on deathrow list */
1160         list_splice_init(&the_lnet.ln_rcd_deathrow, &head);
1161
1162         if (!list_empty(&head)) {
1163                 lnet_net_unlock(LNET_LOCK_EX);
1164
1165                 list_for_each_entry(rcd, &head, rcd_list)
1166                         LNetMDUnlink(rcd->rcd_mdh);
1167
1168                 lnet_net_lock(LNET_LOCK_EX);
1169         }
1170
1171         list_splice_init(&head, &the_lnet.ln_rcd_zombie);
1172
1173         /* release all zombie RCDs */
1174         while (!list_empty(&the_lnet.ln_rcd_zombie)) {
1175                 list_for_each_entry_safe(rcd, tmp, &the_lnet.ln_rcd_zombie,
1176                                          rcd_list) {
1177                         if (LNetMDHandleIsInvalid(rcd->rcd_mdh))
1178                                 list_move(&rcd->rcd_list, &head);
1179                 }
1180
1181                 wait_unlink = wait_unlink &&
1182                               !list_empty(&the_lnet.ln_rcd_zombie);
1183
1184                 lnet_net_unlock(LNET_LOCK_EX);
1185
1186                 while (!list_empty(&head)) {
1187                         rcd = list_entry(head.next,
1188                                          struct lnet_rc_data, rcd_list);
1189                         list_del_init(&rcd->rcd_list);
1190                         lnet_destroy_rc_data(rcd);
1191                 }
1192
1193                 if (!wait_unlink)
1194                         return;
1195
1196                 i++;
1197                 CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET,
1198                        "Waiting for rc buffers to unlink\n");
1199                 set_current_state(TASK_UNINTERRUPTIBLE);
1200                 schedule_timeout(cfs_time_seconds(1) / 4);
1201
1202                 lnet_net_lock(LNET_LOCK_EX);
1203         }
1204
1205         lnet_net_unlock(LNET_LOCK_EX);
1206 }
1207
1208 /*
1209  * This function is called to check if the RC should block indefinitely.
1210  * It's called from lnet_router_checker() as well as being passed to
1211  * wait_event_interruptible() to avoid the lost wake_up problem.
1212  *
1213  * When it's called from wait_event_interruptible() it is necessary to
1214  * also not sleep if the rc state is not running to avoid a deadlock
1215  * when the system is shutting down
1216  */
1217 static inline bool
1218 lnet_router_checker_active(void)
1219 {
1220         if (the_lnet.ln_rc_state != LNET_RC_STATE_RUNNING)
1221                 return true;
1222
1223         /*
1224          * Router Checker thread needs to run when routing is enabled in
1225          * order to call lnet_update_ni_status_locked()
1226          */
1227         if (the_lnet.ln_routing)
1228                 return true;
1229
1230         return !list_empty(&the_lnet.ln_routers) &&
1231                 (live_router_check_interval > 0 ||
1232                  dead_router_check_interval > 0);
1233 }
1234
1235 static int
1236 lnet_router_checker(void *arg)
1237 {
1238         struct lnet_peer *rtr;
1239         struct list_head *entry;
1240
1241         cfs_block_allsigs();
1242
1243         while (the_lnet.ln_rc_state == LNET_RC_STATE_RUNNING) {
1244                 __u64 version;
1245                 int cpt;
1246                 int cpt2;
1247
1248                 cpt = lnet_net_lock_current();
1249 rescan:
1250                 version = the_lnet.ln_routers_version;
1251
1252                 list_for_each(entry, &the_lnet.ln_routers) {
1253                         rtr = list_entry(entry, struct lnet_peer, lp_rtr_list);
1254
1255                         cpt2 = lnet_cpt_of_nid_locked(rtr->lp_nid);
1256                         if (cpt != cpt2) {
1257                                 lnet_net_unlock(cpt);
1258                                 cpt = cpt2;
1259                                 lnet_net_lock(cpt);
1260                                 /* the routers list has changed */
1261                                 if (version != the_lnet.ln_routers_version)
1262                                         goto rescan;
1263                         }
1264
1265                         lnet_ping_router_locked(rtr);
1266
1267                         /* NB dropped lock */
1268                         if (version != the_lnet.ln_routers_version) {
1269                                 /* the routers list has changed */
1270                                 goto rescan;
1271                         }
1272                 }
1273
1274                 if (the_lnet.ln_routing)
1275                         lnet_update_ni_status_locked();
1276
1277                 lnet_net_unlock(cpt);
1278
1279                 lnet_prune_rc_data(0); /* don't wait for UNLINK */
1280
1281                 /*
1282                  * Call schedule_timeout() here always adds 1 to load average
1283                  * because kernel counts # active tasks as nr_running
1284                  * + nr_uninterruptible.
1285                  */
1286                 /*
1287                  * if there are any routes then wakeup every second.  If
1288                  * there are no routes then sleep indefinitely until woken
1289                  * up by a user adding a route
1290                  */
1291                 if (!lnet_router_checker_active())
1292                         wait_event_interruptible(the_lnet.ln_rc_waitq,
1293                                                  lnet_router_checker_active());
1294                 else
1295                         wait_event_interruptible_timeout(the_lnet.ln_rc_waitq,
1296                                                          false,
1297                                                          cfs_time_seconds(1));
1298         }
1299
1300         lnet_prune_rc_data(1); /* wait for UNLINK */
1301
1302         the_lnet.ln_rc_state = LNET_RC_STATE_SHUTDOWN;
1303         complete(&the_lnet.ln_rc_signal);
1304         /* The unlink event callback will signal final completion */
1305         return 0;
1306 }
1307
1308 void
1309 lnet_destroy_rtrbuf(struct lnet_rtrbuf *rb, int npages)
1310 {
1311         int sz = offsetof(struct lnet_rtrbuf, rb_kiov[npages]);
1312
1313         while (--npages >= 0)
1314                 __free_page(rb->rb_kiov[npages].bv_page);
1315
1316         LIBCFS_FREE(rb, sz);
1317 }
1318
1319 static struct lnet_rtrbuf *
1320 lnet_new_rtrbuf(struct lnet_rtrbufpool *rbp, int cpt)
1321 {
1322         int npages = rbp->rbp_npages;
1323         int sz = offsetof(struct lnet_rtrbuf, rb_kiov[npages]);
1324         struct page *page;
1325         struct lnet_rtrbuf *rb;
1326         int i;
1327
1328         LIBCFS_CPT_ALLOC(rb, lnet_cpt_table(), cpt, sz);
1329         if (!rb)
1330                 return NULL;
1331
1332         rb->rb_pool = rbp;
1333
1334         for (i = 0; i < npages; i++) {
1335                 page = alloc_pages_node(
1336                                 cfs_cpt_spread_node(lnet_cpt_table(), cpt),
1337                                 GFP_KERNEL | __GFP_ZERO, 0);
1338                 if (!page) {
1339                         while (--i >= 0)
1340                                 __free_page(rb->rb_kiov[i].bv_page);
1341
1342                         LIBCFS_FREE(rb, sz);
1343                         return NULL;
1344                 }
1345
1346                 rb->rb_kiov[i].bv_len = PAGE_SIZE;
1347                 rb->rb_kiov[i].bv_offset = 0;
1348                 rb->rb_kiov[i].bv_page = page;
1349         }
1350
1351         return rb;
1352 }
1353
1354 static void
1355 lnet_rtrpool_free_bufs(struct lnet_rtrbufpool *rbp, int cpt)
1356 {
1357         int npages = rbp->rbp_npages;
1358         struct list_head tmp;
1359         struct lnet_rtrbuf *rb;
1360         struct lnet_rtrbuf *temp;
1361
1362         if (!rbp->rbp_nbuffers) /* not initialized or already freed */
1363                 return;
1364
1365         INIT_LIST_HEAD(&tmp);
1366
1367         lnet_net_lock(cpt);
1368         lnet_drop_routed_msgs_locked(&rbp->rbp_msgs, cpt);
1369         list_splice_init(&rbp->rbp_bufs, &tmp);
1370         rbp->rbp_req_nbuffers = 0;
1371         rbp->rbp_nbuffers = 0;
1372         rbp->rbp_credits = 0;
1373         rbp->rbp_mincredits = 0;
1374         lnet_net_unlock(cpt);
1375
1376         /* Free buffers on the free list. */
1377         list_for_each_entry_safe(rb, temp, &tmp, rb_list) {
1378                 list_del(&rb->rb_list);
1379                 lnet_destroy_rtrbuf(rb, npages);
1380         }
1381 }
1382
1383 static int
1384 lnet_rtrpool_adjust_bufs(struct lnet_rtrbufpool *rbp, int nbufs, int cpt)
1385 {
1386         struct list_head rb_list;
1387         struct lnet_rtrbuf *rb;
1388         int num_rb;
1389         int num_buffers = 0;
1390         int old_req_nbufs;
1391         int npages = rbp->rbp_npages;
1392
1393         lnet_net_lock(cpt);
1394         /*
1395          * If we are called for less buffers than already in the pool, we
1396          * just lower the req_nbuffers number and excess buffers will be
1397          * thrown away as they are returned to the free list.  Credits
1398          * then get adjusted as well.
1399          * If we already have enough buffers allocated to serve the
1400          * increase requested, then we can treat that the same way as we
1401          * do the decrease.
1402          */
1403         num_rb = nbufs - rbp->rbp_nbuffers;
1404         if (nbufs <= rbp->rbp_req_nbuffers || num_rb <= 0) {
1405                 rbp->rbp_req_nbuffers = nbufs;
1406                 lnet_net_unlock(cpt);
1407                 return 0;
1408         }
1409         /*
1410          * store the older value of rbp_req_nbuffers and then set it to
1411          * the new request to prevent lnet_return_rx_credits_locked() from
1412          * freeing buffers that we need to keep around
1413          */
1414         old_req_nbufs = rbp->rbp_req_nbuffers;
1415         rbp->rbp_req_nbuffers = nbufs;
1416         lnet_net_unlock(cpt);
1417
1418         INIT_LIST_HEAD(&rb_list);
1419
1420         /*
1421          * allocate the buffers on a local list first.  If all buffers are
1422          * allocated successfully then join this list to the rbp buffer
1423          * list. If not then free all allocated buffers.
1424          */
1425         while (num_rb-- > 0) {
1426                 rb = lnet_new_rtrbuf(rbp, cpt);
1427                 if (!rb) {
1428                         CERROR("Failed to allocate %d route bufs of %d pages\n",
1429                                nbufs, npages);
1430
1431                         lnet_net_lock(cpt);
1432                         rbp->rbp_req_nbuffers = old_req_nbufs;
1433                         lnet_net_unlock(cpt);
1434
1435                         goto failed;
1436                 }
1437
1438                 list_add(&rb->rb_list, &rb_list);
1439                 num_buffers++;
1440         }
1441
1442         lnet_net_lock(cpt);
1443
1444         list_splice_tail(&rb_list, &rbp->rbp_bufs);
1445         rbp->rbp_nbuffers += num_buffers;
1446         rbp->rbp_credits += num_buffers;
1447         rbp->rbp_mincredits = rbp->rbp_credits;
1448         /*
1449          * We need to schedule blocked msg using the newly
1450          * added buffers.
1451          */
1452         while (!list_empty(&rbp->rbp_bufs) &&
1453                !list_empty(&rbp->rbp_msgs))
1454                 lnet_schedule_blocked_locked(rbp);
1455
1456         lnet_net_unlock(cpt);
1457
1458         return 0;
1459
1460 failed:
1461         while (!list_empty(&rb_list)) {
1462                 rb = list_entry(rb_list.next, struct lnet_rtrbuf, rb_list);
1463                 list_del(&rb->rb_list);
1464                 lnet_destroy_rtrbuf(rb, npages);
1465         }
1466
1467         return -ENOMEM;
1468 }
1469
1470 static void
1471 lnet_rtrpool_init(struct lnet_rtrbufpool *rbp, int npages)
1472 {
1473         INIT_LIST_HEAD(&rbp->rbp_msgs);
1474         INIT_LIST_HEAD(&rbp->rbp_bufs);
1475
1476         rbp->rbp_npages = npages;
1477         rbp->rbp_credits = 0;
1478         rbp->rbp_mincredits = 0;
1479 }
1480
1481 void
1482 lnet_rtrpools_free(int keep_pools)
1483 {
1484         struct lnet_rtrbufpool *rtrp;
1485         int i;
1486
1487         if (!the_lnet.ln_rtrpools) /* uninitialized or freed */
1488                 return;
1489
1490         cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1491                 lnet_rtrpool_free_bufs(&rtrp[LNET_TINY_BUF_IDX], i);
1492                 lnet_rtrpool_free_bufs(&rtrp[LNET_SMALL_BUF_IDX], i);
1493                 lnet_rtrpool_free_bufs(&rtrp[LNET_LARGE_BUF_IDX], i);
1494         }
1495
1496         if (!keep_pools) {
1497                 cfs_percpt_free(the_lnet.ln_rtrpools);
1498                 the_lnet.ln_rtrpools = NULL;
1499         }
1500 }
1501
1502 static int
1503 lnet_nrb_tiny_calculate(void)
1504 {
1505         int nrbs = LNET_NRB_TINY;
1506
1507         if (tiny_router_buffers < 0) {
1508                 LCONSOLE_ERROR_MSG(0x10c,
1509                                    "tiny_router_buffers=%d invalid when routing enabled\n",
1510                                    tiny_router_buffers);
1511                 return -EINVAL;
1512         }
1513
1514         if (tiny_router_buffers > 0)
1515                 nrbs = tiny_router_buffers;
1516
1517         nrbs /= LNET_CPT_NUMBER;
1518         return max(nrbs, LNET_NRB_TINY_MIN);
1519 }
1520
1521 static int
1522 lnet_nrb_small_calculate(void)
1523 {
1524         int nrbs = LNET_NRB_SMALL;
1525
1526         if (small_router_buffers < 0) {
1527                 LCONSOLE_ERROR_MSG(0x10c,
1528                                    "small_router_buffers=%d invalid when routing enabled\n",
1529                                    small_router_buffers);
1530                 return -EINVAL;
1531         }
1532
1533         if (small_router_buffers > 0)
1534                 nrbs = small_router_buffers;
1535
1536         nrbs /= LNET_CPT_NUMBER;
1537         return max(nrbs, LNET_NRB_SMALL_MIN);
1538 }
1539
1540 static int
1541 lnet_nrb_large_calculate(void)
1542 {
1543         int nrbs = LNET_NRB_LARGE;
1544
1545         if (large_router_buffers < 0) {
1546                 LCONSOLE_ERROR_MSG(0x10c,
1547                                    "large_router_buffers=%d invalid when routing enabled\n",
1548                                    large_router_buffers);
1549                 return -EINVAL;
1550         }
1551
1552         if (large_router_buffers > 0)
1553                 nrbs = large_router_buffers;
1554
1555         nrbs /= LNET_CPT_NUMBER;
1556         return max(nrbs, LNET_NRB_LARGE_MIN);
1557 }
1558
1559 int
1560 lnet_rtrpools_alloc(int im_a_router)
1561 {
1562         struct lnet_rtrbufpool *rtrp;
1563         int nrb_tiny;
1564         int nrb_small;
1565         int nrb_large;
1566         int rc;
1567         int i;
1568
1569         if (!strcmp(forwarding, "")) {
1570                 /* not set either way */
1571                 if (!im_a_router)
1572                         return 0;
1573         } else if (!strcmp(forwarding, "disabled")) {
1574                 /* explicitly disabled */
1575                 return 0;
1576         } else if (!strcmp(forwarding, "enabled")) {
1577                 /* explicitly enabled */
1578         } else {
1579                 LCONSOLE_ERROR_MSG(0x10b, "'forwarding' not set to either 'enabled' or 'disabled'\n");
1580                 return -EINVAL;
1581         }
1582
1583         nrb_tiny = lnet_nrb_tiny_calculate();
1584         if (nrb_tiny < 0)
1585                 return -EINVAL;
1586
1587         nrb_small = lnet_nrb_small_calculate();
1588         if (nrb_small < 0)
1589                 return -EINVAL;
1590
1591         nrb_large = lnet_nrb_large_calculate();
1592         if (nrb_large < 0)
1593                 return -EINVAL;
1594
1595         the_lnet.ln_rtrpools = cfs_percpt_alloc(lnet_cpt_table(),
1596                                                 LNET_NRBPOOLS *
1597                                                 sizeof(struct lnet_rtrbufpool));
1598         if (!the_lnet.ln_rtrpools) {
1599                 LCONSOLE_ERROR_MSG(0x10c,
1600                                    "Failed to initialize router buffe pool\n");
1601                 return -ENOMEM;
1602         }
1603
1604         cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1605                 lnet_rtrpool_init(&rtrp[LNET_TINY_BUF_IDX], 0);
1606                 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_TINY_BUF_IDX],
1607                                               nrb_tiny, i);
1608                 if (rc)
1609                         goto failed;
1610
1611                 lnet_rtrpool_init(&rtrp[LNET_SMALL_BUF_IDX],
1612                                   LNET_NRB_SMALL_PAGES);
1613                 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_SMALL_BUF_IDX],
1614                                               nrb_small, i);
1615                 if (rc)
1616                         goto failed;
1617
1618                 lnet_rtrpool_init(&rtrp[LNET_LARGE_BUF_IDX],
1619                                   LNET_NRB_LARGE_PAGES);
1620                 rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_LARGE_BUF_IDX],
1621                                               nrb_large, i);
1622                 if (rc)
1623                         goto failed;
1624         }
1625
1626         lnet_net_lock(LNET_LOCK_EX);
1627         the_lnet.ln_routing = 1;
1628         lnet_net_unlock(LNET_LOCK_EX);
1629
1630         return 0;
1631
1632  failed:
1633         lnet_rtrpools_free(0);
1634         return rc;
1635 }
1636
1637 static int
1638 lnet_rtrpools_adjust_helper(int tiny, int small, int large)
1639 {
1640         int nrb = 0;
1641         int rc = 0;
1642         int i;
1643         struct lnet_rtrbufpool *rtrp;
1644
1645         /*
1646          * If the provided values for each buffer pool are different than the
1647          * configured values, we need to take action.
1648          */
1649         if (tiny >= 0) {
1650                 tiny_router_buffers = tiny;
1651                 nrb = lnet_nrb_tiny_calculate();
1652                 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1653                         rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_TINY_BUF_IDX],
1654                                                       nrb, i);
1655                         if (rc)
1656                                 return rc;
1657                 }
1658         }
1659         if (small >= 0) {
1660                 small_router_buffers = small;
1661                 nrb = lnet_nrb_small_calculate();
1662                 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1663                         rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_SMALL_BUF_IDX],
1664                                                       nrb, i);
1665                         if (rc)
1666                                 return rc;
1667                 }
1668         }
1669         if (large >= 0) {
1670                 large_router_buffers = large;
1671                 nrb = lnet_nrb_large_calculate();
1672                 cfs_percpt_for_each(rtrp, i, the_lnet.ln_rtrpools) {
1673                         rc = lnet_rtrpool_adjust_bufs(&rtrp[LNET_LARGE_BUF_IDX],
1674                                                       nrb, i);
1675                         if (rc)
1676                                 return rc;
1677                 }
1678         }
1679
1680         return 0;
1681 }
1682
1683 int
1684 lnet_rtrpools_adjust(int tiny, int small, int large)
1685 {
1686         /*
1687          * this function doesn't revert the changes if adding new buffers
1688          * failed.  It's up to the user space caller to revert the
1689          * changes.
1690          */
1691         if (!the_lnet.ln_routing)
1692                 return 0;
1693
1694         return lnet_rtrpools_adjust_helper(tiny, small, large);
1695 }
1696
1697 int
1698 lnet_rtrpools_enable(void)
1699 {
1700         int rc = 0;
1701
1702         if (the_lnet.ln_routing)
1703                 return 0;
1704
1705         if (!the_lnet.ln_rtrpools)
1706                 /*
1707                  * If routing is turned off, and we have never
1708                  * initialized the pools before, just call the
1709                  * standard buffer pool allocation routine as
1710                  * if we are just configuring this for the first
1711                  * time.
1712                  */
1713                 rc = lnet_rtrpools_alloc(1);
1714         else
1715                 rc = lnet_rtrpools_adjust_helper(0, 0, 0);
1716         if (rc)
1717                 return rc;
1718
1719         lnet_net_lock(LNET_LOCK_EX);
1720         the_lnet.ln_routing = 1;
1721
1722         the_lnet.ln_ping_info->pi_features &= ~LNET_PING_FEAT_RTE_DISABLED;
1723         lnet_net_unlock(LNET_LOCK_EX);
1724
1725         return rc;
1726 }
1727
1728 void
1729 lnet_rtrpools_disable(void)
1730 {
1731         if (!the_lnet.ln_routing)
1732                 return;
1733
1734         lnet_net_lock(LNET_LOCK_EX);
1735         the_lnet.ln_routing = 0;
1736         the_lnet.ln_ping_info->pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1737
1738         tiny_router_buffers = 0;
1739         small_router_buffers = 0;
1740         large_router_buffers = 0;
1741         lnet_net_unlock(LNET_LOCK_EX);
1742         lnet_rtrpools_free(1);
1743 }
1744
1745 int
1746 lnet_notify(struct lnet_ni *ni, lnet_nid_t nid, int alive, unsigned long when)
1747 {
1748         struct lnet_peer *lp = NULL;
1749         unsigned long now = cfs_time_current();
1750         int cpt = lnet_cpt_of_nid(nid);
1751
1752         LASSERT(!in_interrupt());
1753
1754         CDEBUG(D_NET, "%s notifying %s: %s\n",
1755                !ni ? "userspace" : libcfs_nid2str(ni->ni_nid),
1756                libcfs_nid2str(nid),
1757                alive ? "up" : "down");
1758
1759         if (ni &&
1760             LNET_NIDNET(ni->ni_nid) != LNET_NIDNET(nid)) {
1761                 CWARN("Ignoring notification of %s %s by %s (different net)\n",
1762                       libcfs_nid2str(nid), alive ? "birth" : "death",
1763                       libcfs_nid2str(ni->ni_nid));
1764                 return -EINVAL;
1765         }
1766
1767         /* can't do predictions... */
1768         if (cfs_time_after(when, now)) {
1769                 CWARN("Ignoring prediction from %s of %s %s %ld seconds in the future\n",
1770                       !ni ? "userspace" : libcfs_nid2str(ni->ni_nid),
1771                       libcfs_nid2str(nid), alive ? "up" : "down",
1772                       cfs_duration_sec(cfs_time_sub(when, now)));
1773                 return -EINVAL;
1774         }
1775
1776         if (ni && !alive &&          /* LND telling me she's down */
1777             !auto_down) {                      /* auto-down disabled */
1778                 CDEBUG(D_NET, "Auto-down disabled\n");
1779                 return 0;
1780         }
1781
1782         lnet_net_lock(cpt);
1783
1784         if (the_lnet.ln_shutdown) {
1785                 lnet_net_unlock(cpt);
1786                 return -ESHUTDOWN;
1787         }
1788
1789         lp = lnet_find_peer_locked(the_lnet.ln_peer_tables[cpt], nid);
1790         if (!lp) {
1791                 /* nid not found */
1792                 lnet_net_unlock(cpt);
1793                 CDEBUG(D_NET, "%s not found\n", libcfs_nid2str(nid));
1794                 return 0;
1795         }
1796
1797         /*
1798          * We can't fully trust LND on reporting exact peer last_alive
1799          * if he notifies us about dead peer. For example ksocklnd can
1800          * call us with when == _time_when_the_node_was_booted_ if
1801          * no connections were successfully established
1802          */
1803         if (ni && !alive && when < lp->lp_last_alive)
1804                 when = lp->lp_last_alive;
1805
1806         lnet_notify_locked(lp, !ni, alive, when);
1807
1808         if (ni)
1809                 lnet_ni_notify_locked(ni, lp);
1810
1811         lnet_peer_decref_locked(lp);
1812
1813         lnet_net_unlock(cpt);
1814         return 0;
1815 }
1816 EXPORT_SYMBOL(lnet_notify);