lnet/lnet/api-ni.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #define DEBUG_SUBSYSTEM S_LNET
  34
  35 #include <linux/ctype.h>
  36 #include <linux/log2.h>
  37 #include <linux/ktime.h>
  38 #include <linux/moduleparam.h>
  39 #include <linux/uaccess.h>
  40
  41 #include <lnet/lib-lnet.h>
  42
  43 #define D_LNI D_CONSOLE
  44
  45 /*
  46  * initialize ln_api_mutex statically, since it needs to be used in
  47  * discovery_set callback. That module parameter callback can be called
  48  * before module init completes. The mutex needs to be ready for use then.
  49  */
  50 struct lnet the_lnet = {
  51         .ln_api_mutex = __MUTEX_INITIALIZER(the_lnet.ln_api_mutex),
  52 };              /* THE state of the network */
  53 EXPORT_SYMBOL(the_lnet);
  54
  55 static char *ip2nets = "";
  56 module_param(ip2nets, charp, 0444);
  57 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
  58
  59 static char *networks = "";
  60 module_param(networks, charp, 0444);
  61 MODULE_PARM_DESC(networks, "local networks");
  62
  63 static char *routes = "";
  64 module_param(routes, charp, 0444);
  65 MODULE_PARM_DESC(routes, "routes to non-local networks");
  66
  67 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
  68 module_param(rnet_htable_size, int, 0444);
  69 MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
  70
  71 static int use_tcp_bonding = false;
  72 module_param(use_tcp_bonding, int, 0444);
  73 MODULE_PARM_DESC(use_tcp_bonding,
  74                  "Set to 1 to use socklnd bonding. 0 to use Multi-Rail");
  75
  76 unsigned int lnet_numa_range = 0;
  77 module_param(lnet_numa_range, uint, 0444);
  78 MODULE_PARM_DESC(lnet_numa_range,
  79                 "NUMA range to consider during Multi-Rail selection");
  80
  81 /*
  82  * lnet_health_sensitivity determines by how much we decrement the health
  83  * value on sending error. The value defaults to 100, which means health
  84  * interface health is decremented by 100 points every failure.
  85  */
  86 unsigned int lnet_health_sensitivity = 100;
  87 static int sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp);
  88 #ifdef HAVE_KERNEL_PARAM_OPS
  89 static struct kernel_param_ops param_ops_health_sensitivity = {
  90         .set = sensitivity_set,
  91         .get = param_get_int,
  92 };
  93 #define param_check_health_sensitivity(name, p) \
  94                 __param_check(name, p, int)
  95 module_param(lnet_health_sensitivity, health_sensitivity, S_IRUGO|S_IWUSR);
  96 #else
  97 module_param_call(lnet_health_sensitivity, sensitivity_set, param_get_int,
  98                   &lnet_health_sensitivity, S_IRUGO|S_IWUSR);
  99 #endif
 100 MODULE_PARM_DESC(lnet_health_sensitivity,
 101                 "Value to decrement the health value by on error");
 102
 103 /*
 104  * lnet_recovery_interval determines how often we should perform recovery
 105  * on unhealthy interfaces.
 106  */
 107 unsigned int lnet_recovery_interval = 1;
 108 static int recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp);
 109 #ifdef HAVE_KERNEL_PARAM_OPS
 110 static struct kernel_param_ops param_ops_recovery_interval = {
 111         .set = recovery_interval_set,
 112         .get = param_get_int,
 113 };
 114 #define param_check_recovery_interval(name, p) \
 115                 __param_check(name, p, int)
 116 module_param(lnet_recovery_interval, recovery_interval, S_IRUGO|S_IWUSR);
 117 #else
 118 module_param_call(lnet_recovery_interval, recovery_interval_set, param_get_int,
 119                   &lnet_recovery_interval, S_IRUGO|S_IWUSR);
 120 #endif
 121 MODULE_PARM_DESC(lnet_recovery_interval,
 122                 "Interval to recover unhealthy interfaces in seconds");
 123
 124 static int lnet_interfaces_max = LNET_INTERFACES_MAX_DEFAULT;
 125 static int intf_max_set(const char *val, cfs_kernel_param_arg_t *kp);
 126
 127 static struct kernel_param_ops param_ops_interfaces_max = {
 128         .set = intf_max_set,
 129         .get = param_get_int,
 130 };
 131
 132 #define param_check_interfaces_max(name, p) \
 133                 __param_check(name, p, int)
 134
 135 #ifdef HAVE_KERNEL_PARAM_OPS
 136 module_param(lnet_interfaces_max, interfaces_max, 0644);
 137 #else
 138 module_param_call(lnet_interfaces_max, intf_max_set, param_get_int,
 139                   &param_ops_interfaces_max, 0644);
 140 #endif
 141 MODULE_PARM_DESC(lnet_interfaces_max,
 142                 "Maximum number of interfaces in a node.");
 143
 144 unsigned lnet_peer_discovery_disabled = 0;
 145 static int discovery_set(const char *val, cfs_kernel_param_arg_t *kp);
 146
 147 static struct kernel_param_ops param_ops_discovery_disabled = {
 148         .set = discovery_set,
 149         .get = param_get_int,
 150 };
 151
 152 #define param_check_discovery_disabled(name, p) \
 153                 __param_check(name, p, int)
 154 #ifdef HAVE_KERNEL_PARAM_OPS
 155 module_param(lnet_peer_discovery_disabled, discovery_disabled, 0644);
 156 #else
 157 module_param_call(lnet_peer_discovery_disabled, discovery_set, param_get_int,
 158                   &param_ops_discovery_disabled, 0644);
 159 #endif
 160 MODULE_PARM_DESC(lnet_peer_discovery_disabled,
 161                 "Set to 1 to disable peer discovery on this node.");
 162
 163 unsigned int lnet_drop_asym_route;
 164 static int drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp);
 165
 166 static struct kernel_param_ops param_ops_drop_asym_route = {
 167         .set = drop_asym_route_set,
 168         .get = param_get_int,
 169 };
 170
 171 #define param_check_drop_asym_route(name, p)    \
 172         __param_check(name, p, int)
 173 #ifdef HAVE_KERNEL_PARAM_OPS
 174 module_param(lnet_drop_asym_route, drop_asym_route, 0644);
 175 #else
 176 module_param_call(lnet_drop_asym_route, drop_asym_route_set, param_get_int,
 177                   &param_ops_drop_asym_route, 0644);
 178 #endif
 179 MODULE_PARM_DESC(lnet_drop_asym_route,
 180                  "Set to 1 to drop asymmetrical route messages.");
 181
 182 #define LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT 50
 183 #define LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT 50
 184
 185 unsigned lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
 186 static int transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp);
 187 #ifdef HAVE_KERNEL_PARAM_OPS
 188 static struct kernel_param_ops param_ops_transaction_timeout = {
 189         .set = transaction_to_set,
 190         .get = param_get_int,
 191 };
 192
 193 #define param_check_transaction_timeout(name, p) \
 194                 __param_check(name, p, int)
 195 module_param(lnet_transaction_timeout, transaction_timeout, S_IRUGO|S_IWUSR);
 196 #else
 197 module_param_call(lnet_transaction_timeout, transaction_to_set, param_get_int,
 198                   &lnet_transaction_timeout, S_IRUGO|S_IWUSR);
 199 #endif
 200 MODULE_PARM_DESC(lnet_transaction_timeout,
 201                 "Maximum number of seconds to wait for a peer response.");
 202
 203 #define LNET_RETRY_COUNT_HEALTH_DEFAULT 2
 204 unsigned lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
 205 static int retry_count_set(const char *val, cfs_kernel_param_arg_t *kp);
 206 #ifdef HAVE_KERNEL_PARAM_OPS
 207 static struct kernel_param_ops param_ops_retry_count = {
 208         .set = retry_count_set,
 209         .get = param_get_int,
 210 };
 211
 212 #define param_check_retry_count(name, p) \
 213                 __param_check(name, p, int)
 214 module_param(lnet_retry_count, retry_count, S_IRUGO|S_IWUSR);
 215 #else
 216 module_param_call(lnet_retry_count, retry_count_set, param_get_int,
 217                   &lnet_retry_count, S_IRUGO|S_IWUSR);
 218 #endif
 219 MODULE_PARM_DESC(lnet_retry_count,
 220                  "Maximum number of times to retry transmitting a message");
 221
 222
 223 unsigned lnet_lnd_timeout = LNET_LND_DEFAULT_TIMEOUT;
 224 unsigned int lnet_current_net_count;
 225
 226 /*
 227  * This sequence number keeps track of how many times DLC was used to
 228  * update the local NIs. It is incremented when a NI is added or
 229  * removed and checked when sending a message to determine if there is
 230  * a need to re-run the selection algorithm. See lnet_select_pathway()
 231  * for more details on its usage.
 232  */
 233 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
 234
 235 static int lnet_ping(struct lnet_process_id id, signed long timeout,
 236                      struct lnet_process_id __user *ids, int n_ids);
 237
 238 static int lnet_discover(struct lnet_process_id id, __u32 force,
 239                          struct lnet_process_id __user *ids, int n_ids);
 240
 241 static int
 242 sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
 243 {
 244         int rc;
 245         unsigned *sensitivity = (unsigned *)kp->arg;
 246         unsigned long value;
 247
 248         rc = kstrtoul(val, 0, &value);
 249         if (rc) {
 250                 CERROR("Invalid module parameter value for 'lnet_health_sensitivity'\n");
 251                 return rc;
 252         }
 253
 254         /*
 255          * The purpose of locking the api_mutex here is to ensure that
 256          * the correct value ends up stored properly.
 257          */
 258         mutex_lock(&the_lnet.ln_api_mutex);
 259
 260         if (value > LNET_MAX_HEALTH_VALUE) {
 261                 mutex_unlock(&the_lnet.ln_api_mutex);
 262                 CERROR("Invalid health value. Maximum: %d value = %lu\n",
 263                        LNET_MAX_HEALTH_VALUE, value);
 264                 return -EINVAL;
 265         }
 266
 267         /*
 268          * if we're turning on health then use the health timeout
 269          * defaults.
 270          */
 271         if (*sensitivity == 0 && value != 0) {
 272                 lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
 273                 lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
 274         /*
 275          * if we're turning off health then use the no health timeout
 276          * default.
 277          */
 278         } else if (*sensitivity != 0 && value == 0) {
 279                 lnet_transaction_timeout =
 280                         LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT;
 281                 lnet_retry_count = 0;
 282         }
 283
 284         *sensitivity = value;
 285
 286         mutex_unlock(&the_lnet.ln_api_mutex);
 287
 288         return 0;
 289 }
 290
 291 static int
 292 recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp)
 293 {
 294         int rc;
 295         unsigned *interval = (unsigned *)kp->arg;
 296         unsigned long value;
 297
 298         rc = kstrtoul(val, 0, &value);
 299         if (rc) {
 300                 CERROR("Invalid module parameter value for 'lnet_recovery_interval'\n");
 301                 return rc;
 302         }
 303
 304         if (value < 1) {
 305                 CERROR("lnet_recovery_interval must be at least 1 second\n");
 306                 return -EINVAL;
 307         }
 308
 309         /*
 310          * The purpose of locking the api_mutex here is to ensure that
 311          * the correct value ends up stored properly.
 312          */
 313         mutex_lock(&the_lnet.ln_api_mutex);
 314
 315         *interval = value;
 316
 317         mutex_unlock(&the_lnet.ln_api_mutex);
 318
 319         return 0;
 320 }
 321
 322 static int
 323 discovery_set(const char *val, cfs_kernel_param_arg_t *kp)
 324 {
 325         int rc;
 326         unsigned *discovery = (unsigned *)kp->arg;
 327         unsigned long value;
 328         struct lnet_ping_buffer *pbuf;
 329
 330         rc = kstrtoul(val, 0, &value);
 331         if (rc) {
 332                 CERROR("Invalid module parameter value for 'lnet_peer_discovery_disabled'\n");
 333                 return rc;
 334         }
 335
 336         value = (value) ? 1 : 0;
 337
 338         /*
 339          * The purpose of locking the api_mutex here is to ensure that
 340          * the correct value ends up stored properly.
 341          */
 342         mutex_lock(&the_lnet.ln_api_mutex);
 343
 344         if (value == *discovery) {
 345                 mutex_unlock(&the_lnet.ln_api_mutex);
 346                 return 0;
 347         }
 348
 349         *discovery = value;
 350
 351         if (the_lnet.ln_state != LNET_STATE_RUNNING) {
 352                 mutex_unlock(&the_lnet.ln_api_mutex);
 353                 return 0;
 354         }
 355
 356         /* tell peers that discovery setting has changed */
 357         lnet_net_lock(LNET_LOCK_EX);
 358         pbuf = the_lnet.ln_ping_target;
 359         if (value)
 360                 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
 361         else
 362                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
 363         lnet_net_unlock(LNET_LOCK_EX);
 364
 365         lnet_push_update_to_peers(1);
 366
 367         mutex_unlock(&the_lnet.ln_api_mutex);
 368
 369         return 0;
 370 }
 371
 372 static int
 373 drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp)
 374 {
 375         int rc;
 376         unsigned int *drop_asym_route = (unsigned int *)kp->arg;
 377         unsigned long value;
 378
 379         rc = kstrtoul(val, 0, &value);
 380         if (rc) {
 381                 CERROR("Invalid module parameter value for "
 382                        "'lnet_drop_asym_route'\n");
 383                 return rc;
 384         }
 385
 386         /*
 387          * The purpose of locking the api_mutex here is to ensure that
 388          * the correct value ends up stored properly.
 389          */
 390         mutex_lock(&the_lnet.ln_api_mutex);
 391
 392         if (value == *drop_asym_route) {
 393                 mutex_unlock(&the_lnet.ln_api_mutex);
 394                 return 0;
 395         }
 396
 397         *drop_asym_route = value;
 398
 399         mutex_unlock(&the_lnet.ln_api_mutex);
 400
 401         return 0;
 402 }
 403
 404 static int
 405 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
 406 {
 407         int rc;
 408         unsigned *transaction_to = (unsigned *)kp->arg;
 409         unsigned long value;
 410
 411         rc = kstrtoul(val, 0, &value);
 412         if (rc) {
 413                 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
 414                 return rc;
 415         }
 416
 417         /*
 418          * The purpose of locking the api_mutex here is to ensure that
 419          * the correct value ends up stored properly.
 420          */
 421         mutex_lock(&the_lnet.ln_api_mutex);
 422
 423         if (value < lnet_retry_count || value == 0) {
 424                 mutex_unlock(&the_lnet.ln_api_mutex);
 425                 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
 426                        "Has to be greater than lnet_retry_count (%u)\n",
 427                        value, lnet_retry_count);
 428                 return -EINVAL;
 429         }
 430
 431         if (value == *transaction_to) {
 432                 mutex_unlock(&the_lnet.ln_api_mutex);
 433                 return 0;
 434         }
 435
 436         *transaction_to = value;
 437         if (lnet_retry_count == 0)
 438                 lnet_lnd_timeout = value;
 439         else
 440                 lnet_lnd_timeout = value / lnet_retry_count;
 441
 442         mutex_unlock(&the_lnet.ln_api_mutex);
 443
 444         return 0;
 445 }
 446
 447 static int
 448 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
 449 {
 450         int rc;
 451         unsigned *retry_count = (unsigned *)kp->arg;
 452         unsigned long value;
 453
 454         rc = kstrtoul(val, 0, &value);
 455         if (rc) {
 456                 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
 457                 return rc;
 458         }
 459
 460         /*
 461          * The purpose of locking the api_mutex here is to ensure that
 462          * the correct value ends up stored properly.
 463          */
 464         mutex_lock(&the_lnet.ln_api_mutex);
 465
 466         if (lnet_health_sensitivity == 0) {
 467                 mutex_unlock(&the_lnet.ln_api_mutex);
 468                 CERROR("Can not set retry_count when health feature is turned off\n");
 469                 return -EINVAL;
 470         }
 471
 472         if (value > lnet_transaction_timeout) {
 473                 mutex_unlock(&the_lnet.ln_api_mutex);
 474                 CERROR("Invalid value for lnet_retry_count (%lu). "
 475                        "Has to be smaller than lnet_transaction_timeout (%u)\n",
 476                        value, lnet_transaction_timeout);
 477                 return -EINVAL;
 478         }
 479
 480         *retry_count = value;
 481
 482         if (value == 0)
 483                 lnet_lnd_timeout = lnet_transaction_timeout;
 484         else
 485                 lnet_lnd_timeout = lnet_transaction_timeout / value;
 486
 487         mutex_unlock(&the_lnet.ln_api_mutex);
 488
 489         return 0;
 490 }
 491
 492 static int
 493 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
 494 {
 495         int value, rc;
 496
 497         rc = kstrtoint(val, 0, &value);
 498         if (rc) {
 499                 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
 500                 return rc;
 501         }
 502
 503         if (value < LNET_INTERFACES_MIN) {
 504                 CWARN("max interfaces provided are too small, setting to %d\n",
 505                       LNET_INTERFACES_MAX_DEFAULT);
 506                 value = LNET_INTERFACES_MAX_DEFAULT;
 507         }
 508
 509         *(int *)kp->arg = value;
 510
 511         return 0;
 512 }
 513
 514 static char *
 515 lnet_get_routes(void)
 516 {
 517         return routes;
 518 }
 519
 520 static char *
 521 lnet_get_networks(void)
 522 {
 523         char   *nets;
 524         int     rc;
 525
 526         if (*networks != 0 && *ip2nets != 0) {
 527                 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
 528                                    "'ip2nets' but not both at once\n");
 529                 return NULL;
 530         }
 531
 532         if (*ip2nets != 0) {
 533                 rc = lnet_parse_ip2nets(&nets, ip2nets);
 534                 return (rc == 0) ? nets : NULL;
 535         }
 536
 537         if (*networks != 0)
 538                 return networks;
 539
 540         return "tcp";
 541 }
 542
 543 static void
 544 lnet_init_locks(void)
 545 {
 546         spin_lock_init(&the_lnet.ln_eq_wait_lock);
 547         spin_lock_init(&the_lnet.ln_msg_resend_lock);
 548         init_waitqueue_head(&the_lnet.ln_eq_waitq);
 549         init_completion(&the_lnet.ln_mt_wait_complete);
 550         mutex_init(&the_lnet.ln_lnd_mutex);
 551 }
 552
 553 struct kmem_cache *lnet_mes_cachep;        /* MEs kmem_cache */
 554 struct kmem_cache *lnet_small_mds_cachep;  /* <= LNET_SMALL_MD_SIZE bytes
 555                                             *  MDs kmem_cache */
 556 struct kmem_cache *lnet_rspt_cachep;       /* response tracker cache */
 557 struct kmem_cache *lnet_msg_cachep;
 558
 559 static int
 560 lnet_slab_setup(void)
 561 {
 562         /* create specific kmem_cache for MEs and small MDs (i.e., originally
 563          * allocated in <size-xxx> kmem_cache).
 564          */
 565         lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
 566                                             0, 0, NULL);
 567         if (!lnet_mes_cachep)
 568                 return -ENOMEM;
 569
 570         lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
 571                                                   LNET_SMALL_MD_SIZE, 0, 0,
 572                                                   NULL);
 573         if (!lnet_small_mds_cachep)
 574                 return -ENOMEM;
 575
 576         lnet_rspt_cachep = kmem_cache_create("lnet_rspt", sizeof(struct lnet_rsp_tracker),
 577                                             0, 0, NULL);
 578         if (!lnet_rspt_cachep)
 579                 return -ENOMEM;
 580
 581         lnet_msg_cachep = kmem_cache_create("lnet_msg", sizeof(struct lnet_msg),
 582                                             0, 0, NULL);
 583         if (!lnet_msg_cachep)
 584                 return -ENOMEM;
 585
 586         return 0;
 587 }
 588
 589 static void
 590 lnet_slab_cleanup(void)
 591 {
 592         if (lnet_msg_cachep) {
 593                 kmem_cache_destroy(lnet_msg_cachep);
 594                 lnet_msg_cachep = NULL;
 595         }
 596
 597
 598         if (lnet_rspt_cachep) {
 599                 kmem_cache_destroy(lnet_rspt_cachep);
 600                 lnet_rspt_cachep = NULL;
 601         }
 602
 603         if (lnet_small_mds_cachep) {
 604                 kmem_cache_destroy(lnet_small_mds_cachep);
 605                 lnet_small_mds_cachep = NULL;
 606         }
 607
 608         if (lnet_mes_cachep) {
 609                 kmem_cache_destroy(lnet_mes_cachep);
 610                 lnet_mes_cachep = NULL;
 611         }
 612 }
 613
 614 static int
 615 lnet_create_remote_nets_table(void)
 616 {
 617         int               i;
 618         struct list_head *hash;
 619
 620         LASSERT(the_lnet.ln_remote_nets_hash == NULL);
 621         LASSERT(the_lnet.ln_remote_nets_hbits > 0);
 622         LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
 623         if (hash == NULL) {
 624                 CERROR("Failed to create remote nets hash table\n");
 625                 return -ENOMEM;
 626         }
 627
 628         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 629                 INIT_LIST_HEAD(&hash[i]);
 630         the_lnet.ln_remote_nets_hash = hash;
 631         return 0;
 632 }
 633
 634 static void
 635 lnet_destroy_remote_nets_table(void)
 636 {
 637         int i;
 638
 639         if (the_lnet.ln_remote_nets_hash == NULL)
 640                 return;
 641
 642         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 643                 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
 644
 645         LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
 646                     LNET_REMOTE_NETS_HASH_SIZE *
 647                     sizeof(the_lnet.ln_remote_nets_hash[0]));
 648         the_lnet.ln_remote_nets_hash = NULL;
 649 }
 650
 651 static void
 652 lnet_destroy_locks(void)
 653 {
 654         if (the_lnet.ln_res_lock != NULL) {
 655                 cfs_percpt_lock_free(the_lnet.ln_res_lock);
 656                 the_lnet.ln_res_lock = NULL;
 657         }
 658
 659         if (the_lnet.ln_net_lock != NULL) {
 660                 cfs_percpt_lock_free(the_lnet.ln_net_lock);
 661                 the_lnet.ln_net_lock = NULL;
 662         }
 663 }
 664
 665 static int
 666 lnet_create_locks(void)
 667 {
 668         lnet_init_locks();
 669
 670         the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 671         if (the_lnet.ln_res_lock == NULL)
 672                 goto failed;
 673
 674         the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 675         if (the_lnet.ln_net_lock == NULL)
 676                 goto failed;
 677
 678         return 0;
 679
 680  failed:
 681         lnet_destroy_locks();
 682         return -ENOMEM;
 683 }
 684
 685 static void lnet_assert_wire_constants(void)
 686 {
 687         /* Wire protocol assertions generated by 'wirecheck'
 688          * running on Linux robert.bartonsoftware.com 2.6.8-1.521
 689          * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
 690          * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
 691
 692         /* Constants... */
 693         BUILD_BUG_ON(LNET_PROTO_TCP_MAGIC != 0xeebc0ded);
 694         BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MAJOR != 1);
 695         BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MINOR != 0);
 696         BUILD_BUG_ON(LNET_MSG_ACK != 0);
 697         BUILD_BUG_ON(LNET_MSG_PUT != 1);
 698         BUILD_BUG_ON(LNET_MSG_GET != 2);
 699         BUILD_BUG_ON(LNET_MSG_REPLY != 3);
 700         BUILD_BUG_ON(LNET_MSG_HELLO != 4);
 701
 702         /* Checks for struct lnet_handle_wire */
 703         BUILD_BUG_ON((int)sizeof(struct lnet_handle_wire) != 16);
 704         BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
 705                                    wh_interface_cookie) != 0);
 706         BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) != 8);
 707         BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
 708                                    wh_object_cookie) != 8);
 709         BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) != 8);
 710
 711         /* Checks for struct struct lnet_magicversion */
 712         BUILD_BUG_ON((int)sizeof(struct lnet_magicversion) != 8);
 713         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, magic) != 0);
 714         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->magic) != 4);
 715         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, version_major) != 4);
 716         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_major) != 2);
 717         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion,
 718                                    version_minor) != 6);
 719         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_minor) != 2);
 720
 721         /* Checks for struct struct lnet_hdr */
 722         BUILD_BUG_ON((int)sizeof(struct lnet_hdr) != 72);
 723         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_nid) != 0);
 724         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_nid) != 8);
 725         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_nid) != 8);
 726         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_nid) != 8);
 727         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_pid) != 16);
 728         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_pid) != 4);
 729         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_pid) != 20);
 730         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_pid) != 4);
 731         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, type) != 24);
 732         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->type) != 4);
 733         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, payload_length) != 28);
 734         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->payload_length) != 4);
 735         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg) != 32);
 736         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg) != 40);
 737
 738         /* Ack */
 739         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) != 32);
 740         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) != 16);
 741         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.match_bits) != 48);
 742         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) != 8);
 743         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.mlength) != 56);
 744         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) != 4);
 745
 746         /* Put */
 747         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) != 32);
 748         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) != 16);
 749         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.match_bits) != 48);
 750         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) != 8);
 751         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.hdr_data) != 56);
 752         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) != 8);
 753         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ptl_index) != 64);
 754         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) != 4);
 755         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.offset) != 68);
 756         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) != 4);
 757
 758         /* Get */
 759         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.return_wmd) != 32);
 760         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) != 16);
 761         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.match_bits) != 48);
 762         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) != 8);
 763         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.ptl_index) != 56);
 764         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) != 4);
 765         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.src_offset) != 60);
 766         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) != 4);
 767         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.sink_length) != 64);
 768         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) != 4);
 769
 770         /* Reply */
 771         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) != 32);
 772         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) != 16);
 773
 774         /* Hello */
 775         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.incarnation) != 32);
 776         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) != 8);
 777         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.type) != 40);
 778         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) != 4);
 779
 780         /* Checks for struct lnet_ni_status and related constants */
 781         BUILD_BUG_ON(LNET_NI_STATUS_INVALID != 0x00000000);
 782         BUILD_BUG_ON(LNET_NI_STATUS_UP != 0x15aac0de);
 783         BUILD_BUG_ON(LNET_NI_STATUS_DOWN != 0xdeadface);
 784
 785         /* Checks for struct lnet_ni_status */
 786         BUILD_BUG_ON((int)sizeof(struct lnet_ni_status) != 16);
 787         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_nid) != 0);
 788         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) != 8);
 789         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_status) != 8);
 790         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_status) != 4);
 791         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_unused) != 12);
 792         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) != 4);
 793
 794         /* Checks for struct lnet_ping_info and related constants */
 795         BUILD_BUG_ON(LNET_PROTO_PING_MAGIC != 0x70696E67);
 796         BUILD_BUG_ON(LNET_PING_FEAT_INVAL != 0);
 797         BUILD_BUG_ON(LNET_PING_FEAT_BASE != 1);
 798         BUILD_BUG_ON(LNET_PING_FEAT_NI_STATUS != 2);
 799         BUILD_BUG_ON(LNET_PING_FEAT_RTE_DISABLED != 4);
 800         BUILD_BUG_ON(LNET_PING_FEAT_MULTI_RAIL != 8);
 801         BUILD_BUG_ON(LNET_PING_FEAT_DISCOVERY != 16);
 802         BUILD_BUG_ON(LNET_PING_FEAT_BITS != 31);
 803
 804         /* Checks for struct lnet_ping_info */
 805         BUILD_BUG_ON((int)sizeof(struct lnet_ping_info) != 16);
 806         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_magic) != 0);
 807         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) != 4);
 808         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_features) != 4);
 809         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_features) != 4);
 810         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_pid) != 8);
 811         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) != 4);
 812         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_nnis) != 12);
 813         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) != 4);
 814         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_ni) != 16);
 815         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) != 0);
 816 }
 817
 818 static const struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
 819 {
 820         const struct lnet_lnd *lnd;
 821
 822         /* holding lnd mutex */
 823         if (type >= NUM_LNDS)
 824                 return NULL;
 825         lnd = the_lnet.ln_lnds[type];
 826         LASSERT(!lnd || lnd->lnd_type == type);
 827
 828         return lnd;
 829 }
 830
 831 unsigned int
 832 lnet_get_lnd_timeout(void)
 833 {
 834         return lnet_lnd_timeout;
 835 }
 836 EXPORT_SYMBOL(lnet_get_lnd_timeout);
 837
 838 void
 839 lnet_register_lnd(const struct lnet_lnd *lnd)
 840 {
 841         mutex_lock(&the_lnet.ln_lnd_mutex);
 842
 843         LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
 844         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
 845
 846         the_lnet.ln_lnds[lnd->lnd_type] = lnd;
 847
 848         CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
 849
 850         mutex_unlock(&the_lnet.ln_lnd_mutex);
 851 }
 852 EXPORT_SYMBOL(lnet_register_lnd);
 853
 854 void
 855 lnet_unregister_lnd(const struct lnet_lnd *lnd)
 856 {
 857         mutex_lock(&the_lnet.ln_lnd_mutex);
 858
 859         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
 860
 861         the_lnet.ln_lnds[lnd->lnd_type] = NULL;
 862         CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
 863
 864         mutex_unlock(&the_lnet.ln_lnd_mutex);
 865 }
 866 EXPORT_SYMBOL(lnet_unregister_lnd);
 867
 868 void
 869 lnet_counters_get_common(struct lnet_counters_common *common)
 870 {
 871         struct lnet_counters *ctr;
 872         int i;
 873
 874         memset(common, 0, sizeof(*common));
 875
 876         lnet_net_lock(LNET_LOCK_EX);
 877
 878         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
 879                 common->lcc_msgs_max     += ctr->lct_common.lcc_msgs_max;
 880                 common->lcc_msgs_alloc   += ctr->lct_common.lcc_msgs_alloc;
 881                 common->lcc_errors       += ctr->lct_common.lcc_errors;
 882                 common->lcc_send_count   += ctr->lct_common.lcc_send_count;
 883                 common->lcc_recv_count   += ctr->lct_common.lcc_recv_count;
 884                 common->lcc_route_count  += ctr->lct_common.lcc_route_count;
 885                 common->lcc_drop_count   += ctr->lct_common.lcc_drop_count;
 886                 common->lcc_send_length  += ctr->lct_common.lcc_send_length;
 887                 common->lcc_recv_length  += ctr->lct_common.lcc_recv_length;
 888                 common->lcc_route_length += ctr->lct_common.lcc_route_length;
 889                 common->lcc_drop_length  += ctr->lct_common.lcc_drop_length;
 890         }
 891         lnet_net_unlock(LNET_LOCK_EX);
 892 }
 893 EXPORT_SYMBOL(lnet_counters_get_common);
 894
 895 void
 896 lnet_counters_get(struct lnet_counters *counters)
 897 {
 898         struct lnet_counters *ctr;
 899         struct lnet_counters_health *health = &counters->lct_health;
 900         int             i;
 901
 902         memset(counters, 0, sizeof(*counters));
 903
 904         lnet_counters_get_common(&counters->lct_common);
 905
 906         lnet_net_lock(LNET_LOCK_EX);
 907
 908         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
 909                 health->lch_rst_alloc    += ctr->lct_health.lch_rst_alloc;
 910                 health->lch_resend_count += ctr->lct_health.lch_resend_count;
 911                 health->lch_response_timeout_count +=
 912                                 ctr->lct_health.lch_response_timeout_count;
 913                 health->lch_local_interrupt_count +=
 914                                 ctr->lct_health.lch_local_interrupt_count;
 915                 health->lch_local_dropped_count +=
 916                                 ctr->lct_health.lch_local_dropped_count;
 917                 health->lch_local_aborted_count +=
 918                                 ctr->lct_health.lch_local_aborted_count;
 919                 health->lch_local_no_route_count +=
 920                                 ctr->lct_health.lch_local_no_route_count;
 921                 health->lch_local_timeout_count +=
 922                                 ctr->lct_health.lch_local_timeout_count;
 923                 health->lch_local_error_count +=
 924                                 ctr->lct_health.lch_local_error_count;
 925                 health->lch_remote_dropped_count +=
 926                                 ctr->lct_health.lch_remote_dropped_count;
 927                 health->lch_remote_error_count +=
 928                                 ctr->lct_health.lch_remote_error_count;
 929                 health->lch_remote_timeout_count +=
 930                                 ctr->lct_health.lch_remote_timeout_count;
 931                 health->lch_network_timeout_count +=
 932                                 ctr->lct_health.lch_network_timeout_count;
 933         }
 934         lnet_net_unlock(LNET_LOCK_EX);
 935 }
 936 EXPORT_SYMBOL(lnet_counters_get);
 937
 938 void
 939 lnet_counters_reset(void)
 940 {
 941         struct lnet_counters *counters;
 942         int             i;
 943
 944         lnet_net_lock(LNET_LOCK_EX);
 945
 946         cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
 947                 memset(counters, 0, sizeof(struct lnet_counters));
 948
 949         lnet_net_unlock(LNET_LOCK_EX);
 950 }
 951
 952 static char *
 953 lnet_res_type2str(int type)
 954 {
 955         switch (type) {
 956         default:
 957                 LBUG();
 958         case LNET_COOKIE_TYPE_MD:
 959                 return "MD";
 960         case LNET_COOKIE_TYPE_ME:
 961                 return "ME";
 962         case LNET_COOKIE_TYPE_EQ:
 963                 return "EQ";
 964         }
 965 }
 966
 967 static void
 968 lnet_res_container_cleanup(struct lnet_res_container *rec)
 969 {
 970         int     count = 0;
 971
 972         if (rec->rec_type == 0) /* not set yet, it's uninitialized */
 973                 return;
 974
 975         while (!list_empty(&rec->rec_active)) {
 976                 struct list_head *e = rec->rec_active.next;
 977
 978                 list_del_init(e);
 979                 if (rec->rec_type == LNET_COOKIE_TYPE_EQ) {
 980                         lnet_eq_free(list_entry(e, struct lnet_eq, eq_list));
 981
 982                 } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
 983                         lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
 984
 985                 } else { /* NB: Active MEs should be attached on portals */
 986                         LBUG();
 987                 }
 988                 count++;
 989         }
 990
 991         if (count > 0) {
 992                 /* Found alive MD/ME/EQ, user really should unlink/free
 993                  * all of them before finalize LNet, but if someone didn't,
 994                  * we have to recycle garbage for him */
 995                 CERROR("%d active elements on exit of %s container\n",
 996                        count, lnet_res_type2str(rec->rec_type));
 997         }
 998
 999         if (rec->rec_lh_hash != NULL) {
1000                 LIBCFS_FREE(rec->rec_lh_hash,
1001                             LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1002                 rec->rec_lh_hash = NULL;
1003         }
1004
1005         rec->rec_type = 0; /* mark it as finalized */
1006 }
1007
1008 static int
1009 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
1010 {
1011         int     rc = 0;
1012         int     i;
1013
1014         LASSERT(rec->rec_type == 0);
1015
1016         rec->rec_type = type;
1017         INIT_LIST_HEAD(&rec->rec_active);
1018
1019         rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1020
1021         /* Arbitrary choice of hash table size */
1022         LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1023                          LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1024         if (rec->rec_lh_hash == NULL) {
1025                 rc = -ENOMEM;
1026                 goto out;
1027         }
1028
1029         for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1030                 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1031
1032         return 0;
1033
1034 out:
1035         CERROR("Failed to setup %s resource container\n",
1036                lnet_res_type2str(type));
1037         lnet_res_container_cleanup(rec);
1038         return rc;
1039 }
1040
1041 static void
1042 lnet_res_containers_destroy(struct lnet_res_container **recs)
1043 {
1044         struct lnet_res_container       *rec;
1045         int                             i;
1046
1047         cfs_percpt_for_each(rec, i, recs)
1048                 lnet_res_container_cleanup(rec);
1049
1050         cfs_percpt_free(recs);
1051 }
1052
1053 static struct lnet_res_container **
1054 lnet_res_containers_create(int type)
1055 {
1056         struct lnet_res_container       **recs;
1057         struct lnet_res_container       *rec;
1058         int                             rc;
1059         int                             i;
1060
1061         recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1062         if (recs == NULL) {
1063                 CERROR("Failed to allocate %s resource containers\n",
1064                        lnet_res_type2str(type));
1065                 return NULL;
1066         }
1067
1068         cfs_percpt_for_each(rec, i, recs) {
1069                 rc = lnet_res_container_setup(rec, i, type);
1070                 if (rc != 0) {
1071                         lnet_res_containers_destroy(recs);
1072                         return NULL;
1073                 }
1074         }
1075
1076         return recs;
1077 }
1078
1079 struct lnet_libhandle *
1080 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1081 {
1082         /* ALWAYS called with lnet_res_lock held */
1083         struct list_head        *head;
1084         struct lnet_libhandle   *lh;
1085         unsigned int            hash;
1086
1087         if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1088                 return NULL;
1089
1090         hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1091         head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1092
1093         list_for_each_entry(lh, head, lh_hash_chain) {
1094                 if (lh->lh_cookie == cookie)
1095                         return lh;
1096         }
1097
1098         return NULL;
1099 }
1100
1101 void
1102 lnet_res_lh_initialize(struct lnet_res_container *rec,
1103                        struct lnet_libhandle *lh)
1104 {
1105         /* ALWAYS called with lnet_res_lock held */
1106         unsigned int    ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1107         unsigned int    hash;
1108
1109         lh->lh_cookie = rec->rec_lh_cookie;
1110         rec->rec_lh_cookie += 1 << ibits;
1111
1112         hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1113
1114         list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1115 }
1116
1117 struct list_head **
1118 lnet_create_array_of_queues(void)
1119 {
1120         struct list_head **qs;
1121         struct list_head *q;
1122         int i;
1123
1124         qs = cfs_percpt_alloc(lnet_cpt_table(),
1125                               sizeof(struct list_head));
1126         if (!qs) {
1127                 CERROR("Failed to allocate queues\n");
1128                 return NULL;
1129         }
1130
1131         cfs_percpt_for_each(q, i, qs)
1132                 INIT_LIST_HEAD(q);
1133
1134         return qs;
1135 }
1136
1137 static int lnet_unprepare(void);
1138
1139 static int
1140 lnet_prepare(lnet_pid_t requested_pid)
1141 {
1142         /* Prepare to bring up the network */
1143         struct lnet_res_container **recs;
1144         int                       rc = 0;
1145
1146         if (requested_pid == LNET_PID_ANY) {
1147                 /* Don't instantiate LNET just for me */
1148                 return -ENETDOWN;
1149         }
1150
1151         LASSERT(the_lnet.ln_refcount == 0);
1152
1153         the_lnet.ln_routing = 0;
1154
1155         LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1156         the_lnet.ln_pid = requested_pid;
1157
1158         INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1159         INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1160         INIT_LIST_HEAD(&the_lnet.ln_nets);
1161         INIT_LIST_HEAD(&the_lnet.ln_routers);
1162         INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1163         INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1164         INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1165         INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1166         INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1167         INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1168         INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1169         init_waitqueue_head(&the_lnet.ln_dc_waitq);
1170         LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
1171         init_completion(&the_lnet.ln_started);
1172
1173         rc = lnet_slab_setup();
1174         if (rc != 0)
1175                 goto failed;
1176
1177         rc = lnet_create_remote_nets_table();
1178         if (rc != 0)
1179                 goto failed;
1180
1181         /*
1182          * NB the interface cookie in wire handles guards against delayed
1183          * replies and ACKs appearing valid after reboot.
1184          */
1185         the_lnet.ln_interface_cookie = ktime_get_real_ns();
1186
1187         the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1188                                                 sizeof(struct lnet_counters));
1189         if (the_lnet.ln_counters == NULL) {
1190                 CERROR("Failed to allocate counters for LNet\n");
1191                 rc = -ENOMEM;
1192                 goto failed;
1193         }
1194
1195         rc = lnet_peer_tables_create();
1196         if (rc != 0)
1197                 goto failed;
1198
1199         rc = lnet_msg_containers_create();
1200         if (rc != 0)
1201                 goto failed;
1202
1203         rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1204                                       LNET_COOKIE_TYPE_EQ);
1205         if (rc != 0)
1206                 goto failed;
1207
1208         recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1209         if (recs == NULL) {
1210                 rc = -ENOMEM;
1211                 goto failed;
1212         }
1213
1214         the_lnet.ln_md_containers = recs;
1215
1216         rc = lnet_portals_create();
1217         if (rc != 0) {
1218                 CERROR("Failed to create portals for LNet: %d\n", rc);
1219                 goto failed;
1220         }
1221
1222         the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1223         if (!the_lnet.ln_mt_zombie_rstqs) {
1224                 rc = -ENOMEM;
1225                 goto failed;
1226         }
1227
1228         return 0;
1229
1230  failed:
1231         lnet_unprepare();
1232         return rc;
1233 }
1234
1235 static int
1236 lnet_unprepare (void)
1237 {
1238         int rc;
1239
1240         /* NB no LNET_LOCK since this is the last reference.  All LND instances
1241          * have shut down already, so it is safe to unlink and free all
1242          * descriptors, even those that appear committed to a network op (eg MD
1243          * with non-zero pending count) */
1244
1245         lnet_fail_nid(LNET_NID_ANY, 0);
1246
1247         LASSERT(the_lnet.ln_refcount == 0);
1248         LASSERT(list_empty(&the_lnet.ln_test_peers));
1249         LASSERT(list_empty(&the_lnet.ln_nets));
1250
1251         if (the_lnet.ln_mt_zombie_rstqs) {
1252                 lnet_clean_zombie_rstqs();
1253                 the_lnet.ln_mt_zombie_rstqs = NULL;
1254         }
1255
1256         if (!LNetEQHandleIsInvalid(the_lnet.ln_mt_eqh)) {
1257                 rc = LNetEQFree(the_lnet.ln_mt_eqh);
1258                 LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
1259                 LASSERT(rc == 0);
1260         }
1261
1262         lnet_portals_destroy();
1263
1264         if (the_lnet.ln_md_containers != NULL) {
1265                 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1266                 the_lnet.ln_md_containers = NULL;
1267         }
1268
1269         lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1270
1271         lnet_msg_containers_destroy();
1272         lnet_peer_uninit();
1273         lnet_rtrpools_free(0);
1274
1275         if (the_lnet.ln_counters != NULL) {
1276                 cfs_percpt_free(the_lnet.ln_counters);
1277                 the_lnet.ln_counters = NULL;
1278         }
1279         lnet_destroy_remote_nets_table();
1280         lnet_slab_cleanup();
1281
1282         return 0;
1283 }
1284
1285 struct lnet_ni  *
1286 lnet_net2ni_locked(__u32 net_id, int cpt)
1287 {
1288         struct lnet_ni   *ni;
1289         struct lnet_net  *net;
1290
1291         LASSERT(cpt != LNET_LOCK_EX);
1292
1293         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1294                 if (net->net_id == net_id) {
1295                         ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1296                                         ni_netlist);
1297                         return ni;
1298                 }
1299         }
1300
1301         return NULL;
1302 }
1303
1304 struct lnet_ni *
1305 lnet_net2ni_addref(__u32 net)
1306 {
1307         struct lnet_ni *ni;
1308
1309         lnet_net_lock(0);
1310         ni = lnet_net2ni_locked(net, 0);
1311         if (ni)
1312                 lnet_ni_addref_locked(ni, 0);
1313         lnet_net_unlock(0);
1314
1315         return ni;
1316 }
1317 EXPORT_SYMBOL(lnet_net2ni_addref);
1318
1319 struct lnet_net *
1320 lnet_get_net_locked(__u32 net_id)
1321 {
1322         struct lnet_net  *net;
1323
1324         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1325                 if (net->net_id == net_id)
1326                         return net;
1327         }
1328
1329         return NULL;
1330 }
1331
1332 unsigned int
1333 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1334 {
1335         __u64           key = nid;
1336         unsigned int    val;
1337
1338         LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1339
1340         if (number == 1)
1341                 return 0;
1342
1343         val = hash_long(key, LNET_CPT_BITS);
1344         /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1345         if (val < number)
1346                 return val;
1347
1348         return (unsigned int)(key + val + (val >> 1)) % number;
1349 }
1350
1351 int
1352 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1353 {
1354         struct lnet_net *net;
1355
1356         /* must called with hold of lnet_net_lock */
1357         if (LNET_CPT_NUMBER == 1)
1358                 return 0; /* the only one */
1359
1360         /*
1361          * If NI is provided then use the CPT identified in the NI cpt
1362          * list if one exists. If one doesn't exist, then that NI is
1363          * associated with all CPTs and it follows that the net it belongs
1364          * to is implicitly associated with all CPTs, so just hash the nid
1365          * and return that.
1366          */
1367         if (ni != NULL) {
1368                 if (ni->ni_cpts != NULL)
1369                         return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1370                                                              ni->ni_ncpts)];
1371                 else
1372                         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1373         }
1374
1375         /* no NI provided so look at the net */
1376         net = lnet_get_net_locked(LNET_NIDNET(nid));
1377
1378         if (net != NULL && net->net_cpts != NULL) {
1379                 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1380         }
1381
1382         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1383 }
1384
1385 int
1386 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1387 {
1388         int     cpt;
1389         int     cpt2;
1390
1391         if (LNET_CPT_NUMBER == 1)
1392                 return 0; /* the only one */
1393
1394         cpt = lnet_net_lock_current();
1395
1396         cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1397
1398         lnet_net_unlock(cpt);
1399
1400         return cpt2;
1401 }
1402 EXPORT_SYMBOL(lnet_cpt_of_nid);
1403
1404 int
1405 lnet_islocalnet_locked(__u32 net_id)
1406 {
1407         struct lnet_net *net;
1408         bool local;
1409
1410         net = lnet_get_net_locked(net_id);
1411
1412         local = net != NULL;
1413
1414         return local;
1415 }
1416
1417 int
1418 lnet_islocalnet(__u32 net_id)
1419 {
1420         int cpt;
1421         bool local;
1422
1423         cpt = lnet_net_lock_current();
1424
1425         local = lnet_islocalnet_locked(net_id);
1426
1427         lnet_net_unlock(cpt);
1428
1429         return local;
1430 }
1431
1432 struct lnet_ni  *
1433 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
1434 {
1435         struct lnet_net  *net;
1436         struct lnet_ni   *ni;
1437
1438         LASSERT(cpt != LNET_LOCK_EX);
1439
1440         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1441                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1442                         if (ni->ni_nid == nid)
1443                                 return ni;
1444                 }
1445         }
1446
1447         return NULL;
1448 }
1449
1450 struct lnet_ni *
1451 lnet_nid2ni_addref(lnet_nid_t nid)
1452 {
1453         struct lnet_ni *ni;
1454
1455         lnet_net_lock(0);
1456         ni = lnet_nid2ni_locked(nid, 0);
1457         if (ni)
1458                 lnet_ni_addref_locked(ni, 0);
1459         lnet_net_unlock(0);
1460
1461         return ni;
1462 }
1463 EXPORT_SYMBOL(lnet_nid2ni_addref);
1464
1465 int
1466 lnet_islocalnid(lnet_nid_t nid)
1467 {
1468         struct lnet_ni  *ni;
1469         int             cpt;
1470
1471         cpt = lnet_net_lock_current();
1472         ni = lnet_nid2ni_locked(nid, cpt);
1473         lnet_net_unlock(cpt);
1474
1475         return ni != NULL;
1476 }
1477
1478 int
1479 lnet_count_acceptor_nets(void)
1480 {
1481         /* Return the # of NIs that need the acceptor. */
1482         int              count = 0;
1483         struct lnet_net  *net;
1484         int              cpt;
1485
1486         cpt = lnet_net_lock_current();
1487         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1488                 /* all socklnd type networks should have the acceptor
1489                  * thread started */
1490                 if (net->net_lnd->lnd_accept != NULL)
1491                         count++;
1492         }
1493
1494         lnet_net_unlock(cpt);
1495
1496         return count;
1497 }
1498
1499 struct lnet_ping_buffer *
1500 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1501 {
1502         struct lnet_ping_buffer *pbuf;
1503
1504         LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1505         if (pbuf) {
1506                 pbuf->pb_nnis = nnis;
1507                 atomic_set(&pbuf->pb_refcnt, 1);
1508         }
1509
1510         return pbuf;
1511 }
1512
1513 void
1514 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1515 {
1516         LASSERT(atomic_read(&pbuf->pb_refcnt) == 0);
1517         LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1518 }
1519
1520 static struct lnet_ping_buffer *
1521 lnet_ping_target_create(int nnis)
1522 {
1523         struct lnet_ping_buffer *pbuf;
1524
1525         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1526         if (pbuf == NULL) {
1527                 CERROR("Can't allocate ping source [%d]\n", nnis);
1528                 return NULL;
1529         }
1530
1531         pbuf->pb_info.pi_nnis = nnis;
1532         pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1533         pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1534         pbuf->pb_info.pi_features =
1535                 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1536
1537         return pbuf;
1538 }
1539
1540 static inline int
1541 lnet_get_net_ni_count_locked(struct lnet_net *net)
1542 {
1543         struct lnet_ni  *ni;
1544         int             count = 0;
1545
1546         list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1547                 count++;
1548
1549         return count;
1550 }
1551
1552 static inline int
1553 lnet_get_net_ni_count_pre(struct lnet_net *net)
1554 {
1555         struct lnet_ni  *ni;
1556         int             count = 0;
1557
1558         list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1559                 count++;
1560
1561         return count;
1562 }
1563
1564 static inline int
1565 lnet_get_ni_count(void)
1566 {
1567         struct lnet_ni  *ni;
1568         struct lnet_net *net;
1569         int             count = 0;
1570
1571         lnet_net_lock(0);
1572
1573         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1574                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1575                         count++;
1576         }
1577
1578         lnet_net_unlock(0);
1579
1580         return count;
1581 }
1582
1583 int
1584 lnet_get_net_count(void)
1585 {
1586         struct lnet_net *net;
1587         int count = 0;
1588
1589         lnet_net_lock(0);
1590
1591         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1592                 count++;
1593         }
1594
1595         lnet_net_unlock(0);
1596
1597         return count;
1598 }
1599
1600 void
1601 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1602 {
1603         struct lnet_ni_status *stat;
1604         int nnis;
1605         int i;
1606
1607         __swab32s(&pbuf->pb_info.pi_magic);
1608         __swab32s(&pbuf->pb_info.pi_features);
1609         __swab32s(&pbuf->pb_info.pi_pid);
1610         __swab32s(&pbuf->pb_info.pi_nnis);
1611         nnis = pbuf->pb_info.pi_nnis;
1612         if (nnis > pbuf->pb_nnis)
1613                 nnis = pbuf->pb_nnis;
1614         for (i = 0; i < nnis; i++) {
1615                 stat = &pbuf->pb_info.pi_ni[i];
1616                 __swab64s(&stat->ns_nid);
1617                 __swab32s(&stat->ns_status);
1618         }
1619 }
1620
1621 int
1622 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1623 {
1624         if (!pinfo)
1625                 return -EINVAL;
1626         if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1627                 return -EPROTO;
1628         if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1629                 return -EPROTO;
1630         /* Loopback is guaranteed to be present */
1631         if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1632                 return -ERANGE;
1633         if (LNET_NETTYP(LNET_NIDNET(LNET_PING_INFO_LONI(pinfo))) != LOLND)
1634                 return -EPROTO;
1635         return 0;
1636 }
1637
1638 static void
1639 lnet_ping_target_destroy(void)
1640 {
1641         struct lnet_net *net;
1642         struct lnet_ni  *ni;
1643
1644         lnet_net_lock(LNET_LOCK_EX);
1645
1646         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1647                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1648                         lnet_ni_lock(ni);
1649                         ni->ni_status = NULL;
1650                         lnet_ni_unlock(ni);
1651                 }
1652         }
1653
1654         lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1655         the_lnet.ln_ping_target = NULL;
1656
1657         lnet_net_unlock(LNET_LOCK_EX);
1658 }
1659
1660 static void
1661 lnet_ping_target_event_handler(struct lnet_event *event)
1662 {
1663         struct lnet_ping_buffer *pbuf = event->md.user_ptr;
1664
1665         if (event->unlinked)
1666                 lnet_ping_buffer_decref(pbuf);
1667 }
1668
1669 static int
1670 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1671                        struct lnet_handle_md *ping_mdh,
1672                        int ni_count, bool set_eq)
1673 {
1674         struct lnet_process_id id = {
1675                 .nid = LNET_NID_ANY,
1676                 .pid = LNET_PID_ANY
1677         };
1678         struct lnet_me *me;
1679         struct lnet_md md = { NULL };
1680         int rc, rc2;
1681
1682         if (set_eq) {
1683                 rc = LNetEQAlloc(0, lnet_ping_target_event_handler,
1684                                  &the_lnet.ln_ping_target_eq);
1685                 if (rc != 0) {
1686                         CERROR("Can't allocate ping buffer EQ: %d\n", rc);
1687                         return rc;
1688                 }
1689         }
1690
1691         *ppbuf = lnet_ping_target_create(ni_count);
1692         if (*ppbuf == NULL) {
1693                 rc = -ENOMEM;
1694                 goto fail_free_eq;
1695         }
1696
1697         /* Ping target ME/MD */
1698         me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1699                           LNET_PROTO_PING_MATCHBITS, 0,
1700                           LNET_UNLINK, LNET_INS_AFTER);
1701         if (IS_ERR(me)) {
1702                 rc = PTR_ERR(me);
1703                 CERROR("Can't create ping target ME: %d\n", rc);
1704                 goto fail_decref_ping_buffer;
1705         }
1706
1707         /* initialize md content */
1708         md.start     = &(*ppbuf)->pb_info;
1709         md.length    = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1710         md.threshold = LNET_MD_THRESH_INF;
1711         md.max_size  = 0;
1712         md.options   = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1713                        LNET_MD_MANAGE_REMOTE;
1714         md.eq_handle = the_lnet.ln_ping_target_eq;
1715         md.user_ptr  = *ppbuf;
1716
1717         rc = LNetMDAttach(me, md, LNET_RETAIN, ping_mdh);
1718         if (rc != 0) {
1719                 CERROR("Can't attach ping target MD: %d\n", rc);
1720                 goto fail_unlink_ping_me;
1721         }
1722         lnet_ping_buffer_addref(*ppbuf);
1723
1724         return 0;
1725
1726 fail_unlink_ping_me:
1727         LNetMEUnlink(me);
1728 fail_decref_ping_buffer:
1729         LASSERT(atomic_read(&(*ppbuf)->pb_refcnt) == 1);
1730         lnet_ping_buffer_decref(*ppbuf);
1731         *ppbuf = NULL;
1732 fail_free_eq:
1733         if (set_eq) {
1734                 rc2 = LNetEQFree(the_lnet.ln_ping_target_eq);
1735                 LASSERT(rc2 == 0);
1736         }
1737         return rc;
1738 }
1739
1740 static void
1741 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1742                     struct lnet_handle_md *ping_mdh)
1743 {
1744         sigset_t        blocked = cfs_block_allsigs();
1745
1746         LNetMDUnlink(*ping_mdh);
1747         LNetInvalidateMDHandle(ping_mdh);
1748
1749         /* NB the MD could be busy; this just starts the unlink */
1750         while (atomic_read(&pbuf->pb_refcnt) > 1) {
1751                 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1752                 set_current_state(TASK_UNINTERRUPTIBLE);
1753                 schedule_timeout(cfs_time_seconds(1));
1754         }
1755
1756         cfs_restore_sigs(blocked);
1757 }
1758
1759 static void
1760 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1761 {
1762         struct lnet_ni          *ni;
1763         struct lnet_net         *net;
1764         struct lnet_ni_status *ns;
1765         int                     i;
1766         int                     rc;
1767
1768         i = 0;
1769         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1770                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1771                         LASSERT(i < pbuf->pb_nnis);
1772
1773                         ns = &pbuf->pb_info.pi_ni[i];
1774
1775                         ns->ns_nid = ni->ni_nid;
1776
1777                         lnet_ni_lock(ni);
1778                         ns->ns_status = (ni->ni_status != NULL) ?
1779                                          ni->ni_status->ns_status :
1780                                                 LNET_NI_STATUS_UP;
1781                         ni->ni_status = ns;
1782                         lnet_ni_unlock(ni);
1783
1784                         i++;
1785                 }
1786         }
1787         /*
1788          * We (ab)use the ns_status of the loopback interface to
1789          * transmit the sequence number. The first interface listed
1790          * must be the loopback interface.
1791          */
1792         rc = lnet_ping_info_validate(&pbuf->pb_info);
1793         if (rc) {
1794                 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1795                 LBUG();
1796         }
1797         LNET_PING_BUFFER_SEQNO(pbuf) =
1798                 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1799 }
1800
1801 static void
1802 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1803                         struct lnet_handle_md ping_mdh)
1804 {
1805         struct lnet_ping_buffer *old_pbuf = NULL;
1806         struct lnet_handle_md old_ping_md;
1807
1808         /* switch the NIs to point to the new ping info created */
1809         lnet_net_lock(LNET_LOCK_EX);
1810
1811         if (!the_lnet.ln_routing)
1812                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1813         if (!lnet_peer_discovery_disabled)
1814                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1815
1816         /* Ensure only known feature bits have been set. */
1817         LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1818         LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1819
1820         lnet_ping_target_install_locked(pbuf);
1821
1822         if (the_lnet.ln_ping_target) {
1823                 old_pbuf = the_lnet.ln_ping_target;
1824                 old_ping_md = the_lnet.ln_ping_target_md;
1825         }
1826         the_lnet.ln_ping_target_md = ping_mdh;
1827         the_lnet.ln_ping_target = pbuf;
1828
1829         lnet_net_unlock(LNET_LOCK_EX);
1830
1831         if (old_pbuf) {
1832                 /* unlink and free the old ping info */
1833                 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1834                 lnet_ping_buffer_decref(old_pbuf);
1835         }
1836
1837         lnet_push_update_to_peers(0);
1838 }
1839
1840 static void
1841 lnet_ping_target_fini(void)
1842 {
1843         int             rc;
1844
1845         lnet_ping_md_unlink(the_lnet.ln_ping_target,
1846                             &the_lnet.ln_ping_target_md);
1847
1848         rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1849         LASSERT(rc == 0);
1850
1851         lnet_ping_target_destroy();
1852 }
1853
1854 /* Resize the push target. */
1855 int lnet_push_target_resize(void)
1856 {
1857         struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
1858         struct lnet_md md = { NULL };
1859         struct lnet_me *me;
1860         struct lnet_handle_md mdh;
1861         struct lnet_handle_md old_mdh;
1862         struct lnet_ping_buffer *pbuf;
1863         struct lnet_ping_buffer *old_pbuf;
1864         int nnis = the_lnet.ln_push_target_nnis;
1865         int rc;
1866
1867         if (nnis <= 0) {
1868                 rc = -EINVAL;
1869                 goto fail_return;
1870         }
1871 again:
1872         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1873         if (!pbuf) {
1874                 rc = -ENOMEM;
1875                 goto fail_return;
1876         }
1877
1878         me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1879                           LNET_PROTO_PING_MATCHBITS, 0,
1880                           LNET_UNLINK, LNET_INS_AFTER);
1881
1882         if (IS_ERR(me)) {
1883                 rc = PTR_ERR(me);
1884                 CERROR("Can't create push target ME: %d\n", rc);
1885                 goto fail_decref_pbuf;
1886         }
1887
1888         /* initialize md content */
1889         md.start     = &pbuf->pb_info;
1890         md.length    = LNET_PING_INFO_SIZE(nnis);
1891         md.threshold = LNET_MD_THRESH_INF;
1892         md.max_size  = 0;
1893         md.options   = LNET_MD_OP_PUT | LNET_MD_TRUNCATE |
1894                        LNET_MD_MANAGE_REMOTE;
1895         md.user_ptr  = pbuf;
1896         md.eq_handle = the_lnet.ln_push_target_eq;
1897
1898         rc = LNetMDAttach(me, md, LNET_RETAIN, &mdh);
1899         if (rc) {
1900                 CERROR("Can't attach push MD: %d\n", rc);
1901                 goto fail_unlink_me;
1902         }
1903         lnet_ping_buffer_addref(pbuf);
1904
1905         lnet_net_lock(LNET_LOCK_EX);
1906         old_pbuf = the_lnet.ln_push_target;
1907         old_mdh = the_lnet.ln_push_target_md;
1908         the_lnet.ln_push_target = pbuf;
1909         the_lnet.ln_push_target_md = mdh;
1910         lnet_net_unlock(LNET_LOCK_EX);
1911
1912         if (old_pbuf) {
1913                 LNetMDUnlink(old_mdh);
1914                 lnet_ping_buffer_decref(old_pbuf);
1915         }
1916
1917         if (nnis < the_lnet.ln_push_target_nnis)
1918                 goto again;
1919
1920         CDEBUG(D_NET, "nnis %d success\n", nnis);
1921
1922         return 0;
1923
1924 fail_unlink_me:
1925         LNetMEUnlink(me);
1926 fail_decref_pbuf:
1927         lnet_ping_buffer_decref(pbuf);
1928 fail_return:
1929         CDEBUG(D_NET, "nnis %d error %d\n", nnis, rc);
1930         return rc;
1931 }
1932
1933 static void lnet_push_target_event_handler(struct lnet_event *ev)
1934 {
1935         struct lnet_ping_buffer *pbuf = ev->md.user_ptr;
1936
1937         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
1938                 lnet_swap_pinginfo(pbuf);
1939
1940         lnet_peer_push_event(ev);
1941         if (ev->unlinked)
1942                 lnet_ping_buffer_decref(pbuf);
1943 }
1944
1945 /* Initialize the push target. */
1946 static int lnet_push_target_init(void)
1947 {
1948         int rc;
1949
1950         if (the_lnet.ln_push_target)
1951                 return -EALREADY;
1952
1953         rc = LNetEQAlloc(0, lnet_push_target_event_handler,
1954                          &the_lnet.ln_push_target_eq);
1955         if (rc) {
1956                 CERROR("Can't allocated push target EQ: %d\n", rc);
1957                 return rc;
1958         }
1959
1960         /* Start at the required minimum, we'll enlarge if required. */
1961         the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
1962
1963         rc = lnet_push_target_resize();
1964
1965         if (rc) {
1966                 LNetEQFree(the_lnet.ln_push_target_eq);
1967                 LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1968         }
1969
1970         return rc;
1971 }
1972
1973 /* Clean up the push target. */
1974 static void lnet_push_target_fini(void)
1975 {
1976         if (!the_lnet.ln_push_target)
1977                 return;
1978
1979         /* Unlink and invalidate to prevent new references. */
1980         LNetMDUnlink(the_lnet.ln_push_target_md);
1981         LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
1982
1983         /* Wait for the unlink to complete. */
1984         while (atomic_read(&the_lnet.ln_push_target->pb_refcnt) > 1) {
1985                 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1986                 set_current_state(TASK_UNINTERRUPTIBLE);
1987                 schedule_timeout(cfs_time_seconds(1));
1988         }
1989
1990         lnet_ping_buffer_decref(the_lnet.ln_push_target);
1991         the_lnet.ln_push_target = NULL;
1992         the_lnet.ln_push_target_nnis = 0;
1993
1994         LNetEQFree(the_lnet.ln_push_target_eq);
1995         LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1996 }
1997
1998 static int
1999 lnet_ni_tq_credits(struct lnet_ni *ni)
2000 {
2001         int     credits;
2002
2003         LASSERT(ni->ni_ncpts >= 1);
2004
2005         if (ni->ni_ncpts == 1)
2006                 return ni->ni_net->net_tunables.lct_max_tx_credits;
2007
2008         credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2009         credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2010         credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2011
2012         return credits;
2013 }
2014
2015 static void
2016 lnet_ni_unlink_locked(struct lnet_ni *ni)
2017 {
2018         /* move it to zombie list and nobody can find it anymore */
2019         LASSERT(!list_empty(&ni->ni_netlist));
2020         list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2021         lnet_ni_decref_locked(ni, 0);
2022 }
2023
2024 static void
2025 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2026 {
2027         int             i;
2028         int             islo;
2029         struct lnet_ni  *ni;
2030         struct list_head *zombie_list = &net->net_ni_zombie;
2031
2032         /*
2033          * Now wait for the NIs I just nuked to show up on the zombie
2034          * list and shut them down in guaranteed thread context
2035          */
2036         i = 2;
2037         while (!list_empty(zombie_list)) {
2038                 int     *ref;
2039                 int     j;
2040
2041                 ni = list_entry(zombie_list->next,
2042                                 struct lnet_ni, ni_netlist);
2043                 list_del_init(&ni->ni_netlist);
2044                 /* the ni should be in deleting state. If it's not it's
2045                  * a bug */
2046                 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2047                 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2048                         if (*ref == 0)
2049                                 continue;
2050                         /* still busy, add it back to zombie list */
2051                         list_add(&ni->ni_netlist, zombie_list);
2052                         break;
2053                 }
2054
2055                 if (!list_empty(&ni->ni_netlist)) {
2056                         lnet_net_unlock(LNET_LOCK_EX);
2057                         ++i;
2058                         if ((i & (-i)) == i) {
2059                                 CDEBUG(D_WARNING,
2060                                        "Waiting for zombie LNI %s\n",
2061                                        libcfs_nid2str(ni->ni_nid));
2062                         }
2063                         set_current_state(TASK_UNINTERRUPTIBLE);
2064                         schedule_timeout(cfs_time_seconds(1));
2065                         lnet_net_lock(LNET_LOCK_EX);
2066                         continue;
2067                 }
2068
2069                 lnet_net_unlock(LNET_LOCK_EX);
2070
2071                 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2072
2073                 LASSERT(!in_interrupt());
2074                 /* Holding the mutex makes it safe for lnd_shutdown
2075                  * to call module_put(). Module unload cannot finish
2076                  * until lnet_unregister_lnd() completes, and that
2077                  * requires the mutex.
2078                  */
2079                 mutex_lock(&the_lnet.ln_lnd_mutex);
2080                 (net->net_lnd->lnd_shutdown)(ni);
2081                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2082
2083                 if (!islo)
2084                         CDEBUG(D_LNI, "Removed LNI %s\n",
2085                               libcfs_nid2str(ni->ni_nid));
2086
2087                 lnet_ni_free(ni);
2088                 i = 2;
2089                 lnet_net_lock(LNET_LOCK_EX);
2090         }
2091 }
2092
2093 /* shutdown down the NI and release refcount */
2094 static void
2095 lnet_shutdown_lndni(struct lnet_ni *ni)
2096 {
2097         int i;
2098         struct lnet_net *net = ni->ni_net;
2099
2100         lnet_net_lock(LNET_LOCK_EX);
2101         lnet_ni_lock(ni);
2102         ni->ni_state = LNET_NI_STATE_DELETING;
2103         lnet_ni_unlock(ni);
2104         lnet_ni_unlink_locked(ni);
2105         lnet_incr_dlc_seq();
2106         lnet_net_unlock(LNET_LOCK_EX);
2107
2108         /* clear messages for this NI on the lazy portal */
2109         for (i = 0; i < the_lnet.ln_nportals; i++)
2110                 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2111
2112         lnet_net_lock(LNET_LOCK_EX);
2113         lnet_clear_zombies_nis_locked(net);
2114         lnet_net_unlock(LNET_LOCK_EX);
2115 }
2116
2117 static void
2118 lnet_shutdown_lndnet(struct lnet_net *net)
2119 {
2120         struct lnet_ni *ni;
2121
2122         lnet_net_lock(LNET_LOCK_EX);
2123
2124         list_del_init(&net->net_list);
2125
2126         while (!list_empty(&net->net_ni_list)) {
2127                 ni = list_entry(net->net_ni_list.next,
2128                                 struct lnet_ni, ni_netlist);
2129                 lnet_net_unlock(LNET_LOCK_EX);
2130                 lnet_shutdown_lndni(ni);
2131                 lnet_net_lock(LNET_LOCK_EX);
2132         }
2133
2134         lnet_net_unlock(LNET_LOCK_EX);
2135
2136         /* Do peer table cleanup for this net */
2137         lnet_peer_tables_cleanup(net);
2138
2139         lnet_net_free(net);
2140 }
2141
2142 static void
2143 lnet_shutdown_lndnets(void)
2144 {
2145         struct lnet_net *net;
2146         LIST_HEAD(resend);
2147         struct lnet_msg *msg, *tmp;
2148
2149         /* NB called holding the global mutex */
2150
2151         /* All quiet on the API front */
2152         LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2153         LASSERT(the_lnet.ln_refcount == 0);
2154
2155         lnet_net_lock(LNET_LOCK_EX);
2156         the_lnet.ln_state = LNET_STATE_STOPPING;
2157
2158         while (!list_empty(&the_lnet.ln_nets)) {
2159                 /*
2160                  * move the nets to the zombie list to avoid them being
2161                  * picked up for new work. LONET is also included in the
2162                  * Nets that will be moved to the zombie list
2163                  */
2164                 net = list_entry(the_lnet.ln_nets.next,
2165                                  struct lnet_net, net_list);
2166                 list_move(&net->net_list, &the_lnet.ln_net_zombie);
2167         }
2168
2169         /* Drop the cached loopback Net. */
2170         if (the_lnet.ln_loni != NULL) {
2171                 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2172                 the_lnet.ln_loni = NULL;
2173         }
2174         lnet_net_unlock(LNET_LOCK_EX);
2175
2176         /* iterate through the net zombie list and delete each net */
2177         while (!list_empty(&the_lnet.ln_net_zombie)) {
2178                 net = list_entry(the_lnet.ln_net_zombie.next,
2179                                  struct lnet_net, net_list);
2180                 lnet_shutdown_lndnet(net);
2181         }
2182
2183         spin_lock(&the_lnet.ln_msg_resend_lock);
2184         list_splice(&the_lnet.ln_msg_resend, &resend);
2185         spin_unlock(&the_lnet.ln_msg_resend_lock);
2186
2187         list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2188                 list_del_init(&msg->msg_list);
2189                 msg->msg_no_resend = true;
2190                 lnet_finalize(msg, -ECANCELED);
2191         }
2192
2193         lnet_net_lock(LNET_LOCK_EX);
2194         the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2195         lnet_net_unlock(LNET_LOCK_EX);
2196 }
2197
2198 static int
2199 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2200 {
2201         int                     rc = -EINVAL;
2202         struct lnet_tx_queue    *tq;
2203         int                     i;
2204         struct lnet_net         *net = ni->ni_net;
2205
2206         mutex_lock(&the_lnet.ln_lnd_mutex);
2207
2208         if (tun) {
2209                 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2210                 ni->ni_lnd_tunables_set = true;
2211         }
2212
2213         rc = (net->net_lnd->lnd_startup)(ni);
2214
2215         mutex_unlock(&the_lnet.ln_lnd_mutex);
2216
2217         if (rc != 0) {
2218                 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2219                                    rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2220                 goto failed0;
2221         }
2222
2223         lnet_ni_lock(ni);
2224         ni->ni_state = LNET_NI_STATE_ACTIVE;
2225         lnet_ni_unlock(ni);
2226
2227         /* We keep a reference on the loopback net through the loopback NI */
2228         if (net->net_lnd->lnd_type == LOLND) {
2229                 lnet_ni_addref(ni);
2230                 LASSERT(the_lnet.ln_loni == NULL);
2231                 the_lnet.ln_loni = ni;
2232                 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2233                 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2234                 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2235                 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2236                 return 0;
2237         }
2238
2239         if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2240             ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2241                 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2242                                    libcfs_lnd2str(net->net_lnd->lnd_type),
2243                                    ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2244                                         "" : "per-peer ");
2245                 /* shutdown the NI since if we get here then it must've already
2246                  * been started
2247                  */
2248                 lnet_shutdown_lndni(ni);
2249                 return -EINVAL;
2250         }
2251
2252         cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2253                 tq->tq_credits_min =
2254                 tq->tq_credits_max =
2255                 tq->tq_credits = lnet_ni_tq_credits(ni);
2256         }
2257
2258         atomic_set(&ni->ni_tx_credits,
2259                    lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2260         atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2261
2262         CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2263                 libcfs_nid2str(ni->ni_nid),
2264                 ni->ni_net->net_tunables.lct_peer_tx_credits,
2265                 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2266                 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2267                 ni->ni_net->net_tunables.lct_peer_timeout);
2268
2269         return 0;
2270 failed0:
2271         lnet_ni_free(ni);
2272         return rc;
2273 }
2274
2275 static int
2276 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2277 {
2278         struct lnet_ni *ni;
2279         struct lnet_net *net_l = NULL;
2280         LIST_HEAD(local_ni_list);
2281         int rc;
2282         int ni_count = 0;
2283         __u32 lnd_type;
2284         const struct lnet_lnd  *lnd;
2285         int peer_timeout =
2286                 net->net_tunables.lct_peer_timeout;
2287         int maxtxcredits =
2288                 net->net_tunables.lct_max_tx_credits;
2289         int peerrtrcredits =
2290                 net->net_tunables.lct_peer_rtr_credits;
2291
2292         /*
2293          * make sure that this net is unique. If it isn't then
2294          * we are adding interfaces to an already existing network, and
2295          * 'net' is just a convenient way to pass in the list.
2296          * if it is unique we need to find the LND and load it if
2297          * necessary.
2298          */
2299         if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2300                 lnd_type = LNET_NETTYP(net->net_id);
2301
2302                 mutex_lock(&the_lnet.ln_lnd_mutex);
2303                 lnd = lnet_find_lnd_by_type(lnd_type);
2304
2305                 if (lnd == NULL) {
2306                         mutex_unlock(&the_lnet.ln_lnd_mutex);
2307                         rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2308                         mutex_lock(&the_lnet.ln_lnd_mutex);
2309
2310                         lnd = lnet_find_lnd_by_type(lnd_type);
2311                         if (lnd == NULL) {
2312                                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2313                                 CERROR("Can't load LND %s, module %s, rc=%d\n",
2314                                 libcfs_lnd2str(lnd_type),
2315                                 libcfs_lnd2modname(lnd_type), rc);
2316 #ifndef HAVE_MODULE_LOADING_SUPPORT
2317                                 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2318                                                 "compiled with kernel module "
2319                                                 "loading support.");
2320 #endif
2321                                 rc = -EINVAL;
2322                                 goto failed0;
2323                         }
2324                 }
2325
2326                 net->net_lnd = lnd;
2327
2328                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2329
2330                 net_l = net;
2331         }
2332
2333         /*
2334          * net_l: if the network being added is unique then net_l
2335          *        will point to that network
2336          *        if the network being added is not unique then
2337          *        net_l points to the existing network.
2338          *
2339          * When we enter the loop below, we'll pick NIs off he
2340          * network beign added and start them up, then add them to
2341          * a local ni list. Once we've successfully started all
2342          * the NIs then we join the local NI list (of started up
2343          * networks) with the net_l->net_ni_list, which should
2344          * point to the correct network to add the new ni list to
2345          *
2346          * If any of the new NIs fail to start up, then we want to
2347          * iterate through the local ni list, which should include
2348          * any NIs which were successfully started up, and shut
2349          * them down.
2350          *
2351          * After than we want to delete the network being added,
2352          * to avoid a memory leak.
2353          */
2354
2355         /*
2356          * When a network uses TCP bonding then all its interfaces
2357          * must be specified when the network is first defined: the
2358          * TCP bonding code doesn't allow for interfaces to be added
2359          * or removed.
2360          */
2361         if (net_l != net && net_l != NULL && use_tcp_bonding &&
2362             LNET_NETTYP(net_l->net_id) == SOCKLND) {
2363                 rc = -EINVAL;
2364                 goto failed0;
2365         }
2366
2367         while (!list_empty(&net->net_ni_added)) {
2368                 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2369                                 ni_netlist);
2370                 list_del_init(&ni->ni_netlist);
2371
2372                 /* make sure that the the NI we're about to start
2373                  * up is actually unique. if it's not fail. */
2374                 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2375                                         ni->ni_interfaces[0])) {
2376                         rc = -EEXIST;
2377                         goto failed1;
2378                 }
2379
2380                 /* adjust the pointer the parent network, just in case it
2381                  * the net is a duplicate */
2382                 ni->ni_net = net_l;
2383
2384                 rc = lnet_startup_lndni(ni, tun);
2385
2386                 if (rc < 0)
2387                         goto failed1;
2388
2389                 lnet_ni_addref(ni);
2390                 list_add_tail(&ni->ni_netlist, &local_ni_list);
2391
2392                 ni_count++;
2393         }
2394
2395         lnet_net_lock(LNET_LOCK_EX);
2396         list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2397         lnet_incr_dlc_seq();
2398         lnet_net_unlock(LNET_LOCK_EX);
2399
2400         /* if the network is not unique then we don't want to keep
2401          * it around after we're done. Free it. Otherwise add that
2402          * net to the global the_lnet.ln_nets */
2403         if (net_l != net && net_l != NULL) {
2404                 /*
2405                  * TODO - note. currently the tunables can not be updated
2406                  * once added
2407                  */
2408                 lnet_net_free(net);
2409         } else {
2410                 /*
2411                  * restore tunables after it has been overwitten by the
2412                  * lnd
2413                  */
2414                 if (peer_timeout != -1)
2415                         net->net_tunables.lct_peer_timeout = peer_timeout;
2416                 if (maxtxcredits != -1)
2417                         net->net_tunables.lct_max_tx_credits = maxtxcredits;
2418                 if (peerrtrcredits != -1)
2419                         net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2420
2421                 lnet_net_lock(LNET_LOCK_EX);
2422                 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2423                 lnet_net_unlock(LNET_LOCK_EX);
2424         }
2425
2426         /* update net count */
2427         lnet_current_net_count = lnet_get_net_count();
2428
2429         return ni_count;
2430
2431 failed1:
2432         /*
2433          * shutdown the new NIs that are being started up
2434          * free the NET being started
2435          */
2436         while (!list_empty(&local_ni_list)) {
2437                 ni = list_entry(local_ni_list.next, struct lnet_ni,
2438                                 ni_netlist);
2439
2440                 lnet_shutdown_lndni(ni);
2441         }
2442
2443 failed0:
2444         lnet_net_free(net);
2445
2446         return rc;
2447 }
2448
2449 static int
2450 lnet_startup_lndnets(struct list_head *netlist)
2451 {
2452         struct lnet_net         *net;
2453         int                     rc;
2454         int                     ni_count = 0;
2455
2456         /*
2457          * Change to running state before bringing up the LNDs. This
2458          * allows lnet_shutdown_lndnets() to assert that we've passed
2459          * through here.
2460          */
2461         lnet_net_lock(LNET_LOCK_EX);
2462         the_lnet.ln_state = LNET_STATE_RUNNING;
2463         lnet_net_unlock(LNET_LOCK_EX);
2464
2465         while (!list_empty(netlist)) {
2466                 net = list_entry(netlist->next, struct lnet_net, net_list);
2467                 list_del_init(&net->net_list);
2468
2469                 rc = lnet_startup_lndnet(net, NULL);
2470
2471                 if (rc < 0)
2472                         goto failed;
2473
2474                 ni_count += rc;
2475         }
2476
2477         return ni_count;
2478 failed:
2479         lnet_shutdown_lndnets();
2480
2481         return rc;
2482 }
2483
2484 /**
2485  * Initialize LNet library.
2486  *
2487  * Automatically called at module loading time. Caller has to call
2488  * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2489  * latter returned 0. It must be called exactly once.
2490  *
2491  * \retval 0 on success
2492  * \retval -ve on failures.
2493  */
2494 int lnet_lib_init(void)
2495 {
2496         int rc;
2497
2498         lnet_assert_wire_constants();
2499
2500         /* refer to global cfs_cpt_table for now */
2501         the_lnet.ln_cpt_table   = cfs_cpt_table;
2502         the_lnet.ln_cpt_number  = cfs_cpt_number(cfs_cpt_table);
2503
2504         LASSERT(the_lnet.ln_cpt_number > 0);
2505         if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2506                 /* we are under risk of consuming all lh_cookie */
2507                 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2508                        "please change setting of CPT-table and retry\n",
2509                        the_lnet.ln_cpt_number, LNET_CPT_MAX);
2510                 return -E2BIG;
2511         }
2512
2513         while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2514                 the_lnet.ln_cpt_bits++;
2515
2516         rc = lnet_create_locks();
2517         if (rc != 0) {
2518                 CERROR("Can't create LNet global locks: %d\n", rc);
2519                 return rc;
2520         }
2521
2522         the_lnet.ln_refcount = 0;
2523         INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2524         INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2525
2526         /* The hash table size is the number of bits it takes to express the set
2527          * ln_num_routes, minus 1 (better to under estimate than over so we
2528          * don't waste memory). */
2529         if (rnet_htable_size <= 0)
2530                 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2531         else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2532                 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2533         the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2534                                            order_base_2(rnet_htable_size) - 1);
2535
2536         /* All LNDs apart from the LOLND are in separate modules.  They
2537          * register themselves when their module loads, and unregister
2538          * themselves when their module is unloaded. */
2539         lnet_register_lnd(&the_lolnd);
2540         return 0;
2541 }
2542
2543 /**
2544  * Finalize LNet library.
2545  *
2546  * \pre lnet_lib_init() called with success.
2547  * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2548  *
2549  * As this happens at module-unload, all lnds must already be unloaded,
2550  * so they must already be unregistered.
2551  */
2552 void lnet_lib_exit(void)
2553 {
2554         int i;
2555
2556         LASSERT(the_lnet.ln_refcount == 0);
2557         lnet_unregister_lnd(&the_lolnd);
2558         for (i = 0; i < NUM_LNDS; i++)
2559                 LASSERT(!the_lnet.ln_lnds[i]);
2560         lnet_destroy_locks();
2561 }
2562
2563 /**
2564  * Set LNet PID and start LNet interfaces, routing, and forwarding.
2565  *
2566  * Users must call this function at least once before any other functions.
2567  * For each successful call there must be a corresponding call to
2568  * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2569  * ignored.
2570  *
2571  * The PID used by LNet may be different from the one requested.
2572  * See LNetGetId().
2573  *
2574  * \param requested_pid PID requested by the caller.
2575  *
2576  * \return >= 0 on success, and < 0 error code on failures.
2577  */
2578 int
2579 LNetNIInit(lnet_pid_t requested_pid)
2580 {
2581         int                     im_a_router = 0;
2582         int                     rc;
2583         int                     ni_count;
2584         struct lnet_ping_buffer *pbuf;
2585         struct lnet_handle_md   ping_mdh;
2586         LIST_HEAD(net_head);
2587         struct lnet_net         *net;
2588
2589         mutex_lock(&the_lnet.ln_api_mutex);
2590
2591         CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2592
2593         if (the_lnet.ln_refcount > 0) {
2594                 rc = the_lnet.ln_refcount++;
2595                 mutex_unlock(&the_lnet.ln_api_mutex);
2596                 return rc;
2597         }
2598
2599         rc = lnet_prepare(requested_pid);
2600         if (rc != 0) {
2601                 mutex_unlock(&the_lnet.ln_api_mutex);
2602                 return rc;
2603         }
2604
2605         /* create a network for Loopback network */
2606         net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2607         if (net == NULL) {
2608                 rc = -ENOMEM;
2609                 goto err_empty_list;
2610         }
2611
2612         /* Add in the loopback NI */
2613         if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2614                 rc = -ENOMEM;
2615                 goto err_empty_list;
2616         }
2617
2618         /* If LNet is being initialized via DLC it is possible
2619          * that the user requests not to load module parameters (ones which
2620          * are supported by DLC) on initialization.  Therefore, make sure not
2621          * to load networks, routes and forwarding from module parameters
2622          * in this case.  On cleanup in case of failure only clean up
2623          * routes if it has been loaded */
2624         if (!the_lnet.ln_nis_from_mod_params) {
2625                 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
2626                                          use_tcp_bonding);
2627                 if (rc < 0)
2628                         goto err_empty_list;
2629         }
2630
2631         ni_count = lnet_startup_lndnets(&net_head);
2632         if (ni_count < 0) {
2633                 rc = ni_count;
2634                 goto err_empty_list;
2635         }
2636
2637         if (!the_lnet.ln_nis_from_mod_params) {
2638                 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2639                 if (rc != 0)
2640                         goto err_shutdown_lndnis;
2641
2642                 rc = lnet_rtrpools_alloc(im_a_router);
2643                 if (rc != 0)
2644                         goto err_destroy_routes;
2645         }
2646
2647         rc = lnet_acceptor_start();
2648         if (rc != 0)
2649                 goto err_destroy_routes;
2650
2651         the_lnet.ln_refcount = 1;
2652         /* Now I may use my own API functions... */
2653
2654         rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2655         if (rc != 0)
2656                 goto err_acceptor_stop;
2657
2658         lnet_ping_target_update(pbuf, ping_mdh);
2659
2660         rc = LNetEQAlloc(0, lnet_mt_event_handler, &the_lnet.ln_mt_eqh);
2661         if (rc != 0) {
2662                 CERROR("Can't allocate monitor thread EQ: %d\n", rc);
2663                 goto err_stop_ping;
2664         }
2665
2666         rc = lnet_push_target_init();
2667         if (rc != 0)
2668                 goto err_stop_ping;
2669
2670         rc = lnet_peer_discovery_start();
2671         if (rc != 0)
2672                 goto err_destroy_push_target;
2673
2674         rc = lnet_monitor_thr_start();
2675         if (rc != 0)
2676                 goto err_stop_discovery_thr;
2677
2678         lnet_fault_init();
2679         lnet_router_debugfs_init();
2680
2681         mutex_unlock(&the_lnet.ln_api_mutex);
2682
2683         complete_all(&the_lnet.ln_started);
2684
2685         /* wait for all routers to start */
2686         lnet_wait_router_start();
2687
2688         return 0;
2689
2690 err_stop_discovery_thr:
2691         lnet_peer_discovery_stop();
2692 err_destroy_push_target:
2693         lnet_push_target_fini();
2694 err_stop_ping:
2695         lnet_ping_target_fini();
2696 err_acceptor_stop:
2697         the_lnet.ln_refcount = 0;
2698         lnet_acceptor_stop();
2699 err_destroy_routes:
2700         if (!the_lnet.ln_nis_from_mod_params)
2701                 lnet_destroy_routes();
2702 err_shutdown_lndnis:
2703         lnet_shutdown_lndnets();
2704 err_empty_list:
2705         lnet_unprepare();
2706         LASSERT(rc < 0);
2707         mutex_unlock(&the_lnet.ln_api_mutex);
2708         while (!list_empty(&net_head)) {
2709                 struct lnet_net *net;
2710
2711                 net = list_entry(net_head.next, struct lnet_net, net_list);
2712                 list_del_init(&net->net_list);
2713                 lnet_net_free(net);
2714         }
2715         return rc;
2716 }
2717 EXPORT_SYMBOL(LNetNIInit);
2718
2719 /**
2720  * Stop LNet interfaces, routing, and forwarding.
2721  *
2722  * Users must call this function once for each successful call to LNetNIInit().
2723  * Once the LNetNIFini() operation has been started, the results of pending
2724  * API operations are undefined.
2725  *
2726  * \return always 0 for current implementation.
2727  */
2728 int
2729 LNetNIFini(void)
2730 {
2731         mutex_lock(&the_lnet.ln_api_mutex);
2732
2733         LASSERT(the_lnet.ln_refcount > 0);
2734
2735         if (the_lnet.ln_refcount != 1) {
2736                 the_lnet.ln_refcount--;
2737         } else {
2738                 LASSERT(!the_lnet.ln_niinit_self);
2739
2740                 lnet_fault_fini();
2741
2742                 lnet_router_debugfs_fini();
2743                 lnet_monitor_thr_stop();
2744                 lnet_peer_discovery_stop();
2745                 lnet_push_target_fini();
2746                 lnet_ping_target_fini();
2747
2748                 /* Teardown fns that use my own API functions BEFORE here */
2749                 the_lnet.ln_refcount = 0;
2750
2751                 lnet_acceptor_stop();
2752                 lnet_destroy_routes();
2753                 lnet_shutdown_lndnets();
2754                 lnet_unprepare();
2755         }
2756
2757         mutex_unlock(&the_lnet.ln_api_mutex);
2758         return 0;
2759 }
2760 EXPORT_SYMBOL(LNetNIFini);
2761
2762 /**
2763  * Grabs the ni data from the ni structure and fills the out
2764  * parameters
2765  *
2766  * \param[in] ni network        interface structure
2767  * \param[out] cfg_ni           NI config information
2768  * \param[out] tun              network and LND tunables
2769  */
2770 static void
2771 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2772                    struct lnet_ioctl_config_lnd_tunables *tun,
2773                    struct lnet_ioctl_element_stats *stats,
2774                    __u32 tun_size)
2775 {
2776         size_t min_size = 0;
2777         int i;
2778
2779         if (!ni || !cfg_ni || !tun)
2780                 return;
2781
2782         if (ni->ni_interfaces[0] != NULL) {
2783                 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2784                         if (ni->ni_interfaces[i] != NULL) {
2785                                 strncpy(cfg_ni->lic_ni_intf[i],
2786                                         ni->ni_interfaces[i],
2787                                         sizeof(cfg_ni->lic_ni_intf[i]));
2788                         }
2789                 }
2790         }
2791
2792         cfg_ni->lic_nid = ni->ni_nid;
2793         if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2794                 cfg_ni->lic_status = LNET_NI_STATUS_UP;
2795         else
2796                 cfg_ni->lic_status = ni->ni_status->ns_status;
2797         cfg_ni->lic_tcp_bonding = use_tcp_bonding;
2798         cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
2799
2800         memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
2801
2802         if (stats) {
2803                 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
2804                                                        LNET_STATS_TYPE_SEND);
2805                 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
2806                                                        LNET_STATS_TYPE_RECV);
2807                 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
2808                                                        LNET_STATS_TYPE_DROP);
2809         }
2810
2811         /*
2812          * tun->lt_tun will always be present, but in order to be
2813          * backwards compatible, we need to deal with the cases when
2814          * tun->lt_tun is smaller than what the kernel has, because it
2815          * comes from an older version of a userspace program, then we'll
2816          * need to copy as much information as we have available space.
2817          */
2818         min_size = tun_size - sizeof(tun->lt_cmn);
2819         memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
2820
2821         /* copy over the cpts */
2822         if (ni->ni_ncpts == LNET_CPT_NUMBER &&
2823             ni->ni_cpts == NULL)  {
2824                 for (i = 0; i < ni->ni_ncpts; i++)
2825                         cfg_ni->lic_cpts[i] = i;
2826         } else {
2827                 for (i = 0;
2828                      ni->ni_cpts != NULL && i < ni->ni_ncpts &&
2829                      i < LNET_MAX_SHOW_NUM_CPT;
2830                      i++)
2831                         cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2832         }
2833         cfg_ni->lic_ncpts = ni->ni_ncpts;
2834 }
2835
2836 /**
2837  * NOTE: This is a legacy function left in the code to be backwards
2838  * compatible with older userspace programs. It should eventually be
2839  * removed.
2840  *
2841  * Grabs the ni data from the ni structure and fills the out
2842  * parameters
2843  *
2844  * \param[in] ni network        interface structure
2845  * \param[out] config           config information
2846  */
2847 static void
2848 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
2849                          struct lnet_ioctl_config_data *config)
2850 {
2851         struct lnet_ioctl_net_config *net_config;
2852         struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
2853         size_t min_size, tunable_size = 0;
2854         int i;
2855
2856         if (!ni || !config)
2857                 return;
2858
2859         net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
2860         if (!net_config)
2861                 return;
2862
2863         BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
2864                      ARRAY_SIZE(net_config->ni_interfaces));
2865
2866         for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2867                 if (!ni->ni_interfaces[i])
2868                         break;
2869
2870                 strncpy(net_config->ni_interfaces[i],
2871                         ni->ni_interfaces[i],
2872                         sizeof(net_config->ni_interfaces[i]));
2873         }
2874
2875         config->cfg_nid = ni->ni_nid;
2876         config->cfg_config_u.cfg_net.net_peer_timeout =
2877                 ni->ni_net->net_tunables.lct_peer_timeout;
2878         config->cfg_config_u.cfg_net.net_max_tx_credits =
2879                 ni->ni_net->net_tunables.lct_max_tx_credits;
2880         config->cfg_config_u.cfg_net.net_peer_tx_credits =
2881                 ni->ni_net->net_tunables.lct_peer_tx_credits;
2882         config->cfg_config_u.cfg_net.net_peer_rtr_credits =
2883                 ni->ni_net->net_tunables.lct_peer_rtr_credits;
2884
2885         if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2886                 net_config->ni_status = LNET_NI_STATUS_UP;
2887         else
2888                 net_config->ni_status = ni->ni_status->ns_status;
2889
2890         if (ni->ni_cpts) {
2891                 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
2892
2893                 for (i = 0; i < num_cpts; i++)
2894                         net_config->ni_cpts[i] = ni->ni_cpts[i];
2895
2896                 config->cfg_ncpts = num_cpts;
2897         }
2898
2899         /*
2900          * See if user land tools sent in a newer and larger version
2901          * of struct lnet_tunables than what the kernel uses.
2902          */
2903         min_size = sizeof(*config) + sizeof(*net_config);
2904
2905         if (config->cfg_hdr.ioc_len > min_size)
2906                 tunable_size = config->cfg_hdr.ioc_len - min_size;
2907
2908         /* Don't copy too much data to user space */
2909         min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
2910         lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
2911
2912         if (lnd_cfg && min_size) {
2913                 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
2914                 config->cfg_config_u.cfg_net.net_interface_count = 1;
2915
2916                 /* Tell user land that kernel side has less data */
2917                 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
2918                         min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
2919                         config->cfg_hdr.ioc_len -= min_size;
2920                 }
2921         }
2922 }
2923
2924 struct lnet_ni *
2925 lnet_get_ni_idx_locked(int idx)
2926 {
2927         struct lnet_ni          *ni;
2928         struct lnet_net         *net;
2929
2930         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2931                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2932                         if (idx-- == 0)
2933                                 return ni;
2934                 }
2935         }
2936
2937         return NULL;
2938 }
2939
2940 struct lnet_ni *
2941 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2942 {
2943         struct lnet_ni          *ni;
2944         struct lnet_net         *net = mynet;
2945
2946         /*
2947          * It is possible that the net has been cleaned out while there is
2948          * a message being sent. This function accessed the net without
2949          * checking if the list is empty
2950          */
2951         if (prev == NULL) {
2952                 if (net == NULL)
2953                         net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2954                                         net_list);
2955                 if (list_empty(&net->net_ni_list))
2956                         return NULL;
2957                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2958                                 ni_netlist);
2959
2960                 return ni;
2961         }
2962
2963         if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
2964                 /* if you reached the end of the ni list and the net is
2965                  * specified, then there are no more nis in that net */
2966                 if (net != NULL)
2967                         return NULL;
2968
2969                 /* we reached the end of this net ni list. move to the
2970                  * next net */
2971                 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
2972                         /* no more nets and no more NIs. */
2973                         return NULL;
2974
2975                 /* get the next net */
2976                 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
2977                                  net_list);
2978                 if (list_empty(&net->net_ni_list))
2979                         return NULL;
2980                 /* get the ni on it */
2981                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2982                                 ni_netlist);
2983
2984                 return ni;
2985         }
2986
2987         if (list_empty(&prev->ni_netlist))
2988                 return NULL;
2989
2990         /* there are more nis left */
2991         ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
2992
2993         return ni;
2994 }
2995
2996 int
2997 lnet_get_net_config(struct lnet_ioctl_config_data *config)
2998 {
2999         struct lnet_ni *ni;
3000         int cpt;
3001         int rc = -ENOENT;
3002         int idx = config->cfg_count;
3003
3004         cpt = lnet_net_lock_current();
3005
3006         ni = lnet_get_ni_idx_locked(idx);
3007
3008         if (ni != NULL) {
3009                 rc = 0;
3010                 lnet_ni_lock(ni);
3011                 lnet_fill_ni_info_legacy(ni, config);
3012                 lnet_ni_unlock(ni);
3013         }
3014
3015         lnet_net_unlock(cpt);
3016         return rc;
3017 }
3018
3019 int
3020 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3021                    struct lnet_ioctl_config_lnd_tunables *tun,
3022                    struct lnet_ioctl_element_stats *stats,
3023                    __u32 tun_size)
3024 {
3025         struct lnet_ni          *ni;
3026         int                     cpt;
3027         int                     rc = -ENOENT;
3028
3029         if (!cfg_ni || !tun || !stats)
3030                 return -EINVAL;
3031
3032         cpt = lnet_net_lock_current();
3033
3034         ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3035
3036         if (ni) {
3037                 rc = 0;
3038                 lnet_ni_lock(ni);
3039                 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3040                 lnet_ni_unlock(ni);
3041         }
3042
3043         lnet_net_unlock(cpt);
3044         return rc;
3045 }
3046
3047 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3048 {
3049         struct lnet_ni *ni;
3050         int cpt;
3051         int rc = -ENOENT;
3052
3053         if (!msg_stats)
3054                 return -EINVAL;
3055
3056         cpt = lnet_net_lock_current();
3057
3058         ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3059
3060         if (ni) {
3061                 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3062                 rc = 0;
3063         }
3064
3065         lnet_net_unlock(cpt);
3066
3067         return rc;
3068 }
3069
3070 static int lnet_add_net_common(struct lnet_net *net,
3071                                struct lnet_ioctl_config_lnd_tunables *tun)
3072 {
3073         __u32                   net_id;
3074         struct lnet_ping_buffer *pbuf;
3075         struct lnet_handle_md   ping_mdh;
3076         int                     rc;
3077         struct lnet_remotenet *rnet;
3078         int                     net_ni_count;
3079         int                     num_acceptor_nets;
3080
3081         lnet_net_lock(LNET_LOCK_EX);
3082         rnet = lnet_find_rnet_locked(net->net_id);
3083         lnet_net_unlock(LNET_LOCK_EX);
3084         /*
3085          * make sure that the net added doesn't invalidate the current
3086          * configuration LNet is keeping
3087          */
3088         if (rnet) {
3089                 CERROR("Adding net %s will invalidate routing configuration\n",
3090                        libcfs_net2str(net->net_id));
3091                 lnet_net_free(net);
3092                 return -EUSERS;
3093         }
3094
3095         /*
3096          * make sure you calculate the correct number of slots in the ping
3097          * buffer. Since the ping info is a flattened list of all the NIs,
3098          * we should allocate enough slots to accomodate the number of NIs
3099          * which will be added.
3100          *
3101          * since ni hasn't been configured yet, use
3102          * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3103          */
3104         net_ni_count = lnet_get_net_ni_count_pre(net);
3105
3106         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3107                                     net_ni_count + lnet_get_ni_count(),
3108                                     false);
3109         if (rc < 0) {
3110                 lnet_net_free(net);
3111                 return rc;
3112         }
3113
3114         if (tun)
3115                 memcpy(&net->net_tunables,
3116                        &tun->lt_cmn, sizeof(net->net_tunables));
3117         else
3118                 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3119
3120         /*
3121          * before starting this network get a count of the current TCP
3122          * networks which require the acceptor thread running. If that
3123          * count is == 0 before we start up this network, then we'd want to
3124          * start up the acceptor thread after starting up this network
3125          */
3126         num_acceptor_nets = lnet_count_acceptor_nets();
3127
3128         net_id = net->net_id;
3129
3130         rc = lnet_startup_lndnet(net,
3131                                  (tun) ? &tun->lt_tun : NULL);
3132         if (rc < 0)
3133                 goto failed;
3134
3135         lnet_net_lock(LNET_LOCK_EX);
3136         net = lnet_get_net_locked(net_id);
3137         lnet_net_unlock(LNET_LOCK_EX);
3138
3139         LASSERT(net);
3140
3141         /*
3142          * Start the acceptor thread if this is the first network
3143          * being added that requires the thread.
3144          */
3145         if (net->net_lnd->lnd_accept && num_acceptor_nets == 0) {
3146                 rc = lnet_acceptor_start();
3147                 if (rc < 0) {
3148                         /* shutdown the net that we just started */
3149                         CERROR("Failed to start up acceptor thread\n");
3150                         lnet_shutdown_lndnet(net);
3151                         goto failed;
3152                 }
3153         }
3154
3155         lnet_net_lock(LNET_LOCK_EX);
3156         lnet_peer_net_added(net);
3157         lnet_net_unlock(LNET_LOCK_EX);
3158
3159         lnet_ping_target_update(pbuf, ping_mdh);
3160
3161         return 0;
3162
3163 failed:
3164         lnet_ping_md_unlink(pbuf, &ping_mdh);
3165         lnet_ping_buffer_decref(pbuf);
3166         return rc;
3167 }
3168
3169 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3170                                       struct lnet_ioctl_config_lnd_tunables *tun)
3171 {
3172         struct lnet_net *net;
3173         char *nets;
3174         int rc;
3175         LIST_HEAD(net_head);
3176
3177         rc = lnet_parse_ip2nets(&nets, ip2nets);
3178         if (rc < 0)
3179                 return rc;
3180
3181         rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3182         if (rc < 0)
3183                 return rc;
3184
3185         mutex_lock(&the_lnet.ln_api_mutex);
3186         while (!list_empty(&net_head)) {
3187                 net = list_entry(net_head.next, struct lnet_net, net_list);
3188                 list_del_init(&net->net_list);
3189                 rc = lnet_add_net_common(net, tun);
3190                 if (rc < 0)
3191                         goto out;
3192         }
3193
3194 out:
3195         mutex_unlock(&the_lnet.ln_api_mutex);
3196
3197         while (!list_empty(&net_head)) {
3198                 net = list_entry(net_head.next, struct lnet_net, net_list);
3199                 list_del_init(&net->net_list);
3200                 lnet_net_free(net);
3201         }
3202         return rc;
3203 }
3204
3205 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3206 {
3207         struct lnet_net *net;
3208         struct lnet_ni *ni;
3209         struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3210         int rc, i;
3211         __u32 net_id, lnd_type;
3212
3213         /* get the tunables if they are available */
3214         if (conf->lic_cfg_hdr.ioc_len >=
3215             sizeof(*conf) + sizeof(*tun))
3216                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3217                         conf->lic_bulk;
3218
3219         /* handle legacy ip2nets from DLC */
3220         if (conf->lic_legacy_ip2nets[0] != '\0')
3221                 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3222                                                   tun);
3223
3224         net_id = LNET_NIDNET(conf->lic_nid);
3225         lnd_type = LNET_NETTYP(net_id);
3226
3227         if (!libcfs_isknown_lnd(lnd_type)) {
3228                 CERROR("No valid net and lnd information provided\n");
3229                 return -EINVAL;
3230         }
3231
3232         net = lnet_net_alloc(net_id, NULL);
3233         if (!net)
3234                 return -ENOMEM;
3235
3236         for (i = 0; i < conf->lic_ncpts; i++) {
3237                 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3238                         return -EINVAL;
3239         }
3240
3241         ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3242                                        conf->lic_ni_intf[0]);
3243         if (!ni)
3244                 return -ENOMEM;
3245
3246         mutex_lock(&the_lnet.ln_api_mutex);
3247
3248         rc = lnet_add_net_common(net, tun);
3249
3250         mutex_unlock(&the_lnet.ln_api_mutex);
3251
3252         return rc;
3253 }
3254
3255 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3256 {
3257         struct lnet_net  *net;
3258         struct lnet_ni *ni;
3259         __u32 net_id = LNET_NIDNET(conf->lic_nid);
3260         struct lnet_ping_buffer *pbuf;
3261         struct lnet_handle_md  ping_mdh;
3262         int               rc;
3263         int               net_count;
3264         __u32             addr;
3265
3266         /* don't allow userspace to shutdown the LOLND */
3267         if (LNET_NETTYP(net_id) == LOLND)
3268                 return -EINVAL;
3269
3270         mutex_lock(&the_lnet.ln_api_mutex);
3271
3272         lnet_net_lock(0);
3273
3274         net = lnet_get_net_locked(net_id);
3275         if (!net) {
3276                 CERROR("net %s not found\n",
3277                        libcfs_net2str(net_id));
3278                 rc = -ENOENT;
3279                 goto unlock_net;
3280         }
3281
3282         addr = LNET_NIDADDR(conf->lic_nid);
3283         if (addr == 0) {
3284                 /* remove the entire net */
3285                 net_count = lnet_get_net_ni_count_locked(net);
3286
3287                 lnet_net_unlock(0);
3288
3289                 /* create and link a new ping info, before removing the old one */
3290                 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3291                                         lnet_get_ni_count() - net_count,
3292                                         false);
3293                 if (rc != 0)
3294                         goto unlock_api_mutex;
3295
3296                 lnet_shutdown_lndnet(net);
3297
3298                 if (lnet_count_acceptor_nets() == 0)
3299                         lnet_acceptor_stop();
3300
3301                 lnet_ping_target_update(pbuf, ping_mdh);
3302
3303                 goto unlock_api_mutex;
3304         }
3305
3306         ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3307         if (!ni) {
3308                 CERROR("nid %s not found\n",
3309                        libcfs_nid2str(conf->lic_nid));
3310                 rc = -ENOENT;
3311                 goto unlock_net;
3312         }
3313
3314         net_count = lnet_get_net_ni_count_locked(net);
3315
3316         lnet_net_unlock(0);
3317
3318         /* create and link a new ping info, before removing the old one */
3319         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3320                                   lnet_get_ni_count() - 1, false);
3321         if (rc != 0)
3322                 goto unlock_api_mutex;
3323
3324         lnet_shutdown_lndni(ni);
3325
3326         if (lnet_count_acceptor_nets() == 0)
3327                 lnet_acceptor_stop();
3328
3329         lnet_ping_target_update(pbuf, ping_mdh);
3330
3331         /* check if the net is empty and remove it if it is */
3332         if (net_count == 1)
3333                 lnet_shutdown_lndnet(net);
3334
3335         goto unlock_api_mutex;
3336
3337 unlock_net:
3338         lnet_net_unlock(0);
3339 unlock_api_mutex:
3340         mutex_unlock(&the_lnet.ln_api_mutex);
3341
3342         return rc;
3343 }
3344
3345 /*
3346  * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3347  * They are only expected to be called for unique networks.
3348  * That can be as a result of older DLC library
3349  * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3350  */
3351 int
3352 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3353 {
3354         struct lnet_net *net;
3355         LIST_HEAD(net_head);
3356         int rc;
3357         struct lnet_ioctl_config_lnd_tunables tun;
3358         char *nets = conf->cfg_config_u.cfg_net.net_intf;
3359
3360         /* Create a net/ni structures for the network string */
3361         rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3362         if (rc <= 0)
3363                 return rc == 0 ? -EINVAL : rc;
3364
3365         mutex_lock(&the_lnet.ln_api_mutex);
3366
3367         if (rc > 1) {
3368                 rc = -EINVAL; /* only add one network per call */
3369                 goto out_unlock_clean;
3370         }
3371
3372         net = list_entry(net_head.next, struct lnet_net, net_list);
3373         list_del_init(&net->net_list);
3374
3375         LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3376
3377         memset(&tun, 0, sizeof(tun));
3378
3379         tun.lt_cmn.lct_peer_timeout =
3380           conf->cfg_config_u.cfg_net.net_peer_timeout;
3381         tun.lt_cmn.lct_peer_tx_credits =
3382           conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3383         tun.lt_cmn.lct_peer_rtr_credits =
3384           conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3385         tun.lt_cmn.lct_max_tx_credits =
3386           conf->cfg_config_u.cfg_net.net_max_tx_credits;
3387
3388         rc = lnet_add_net_common(net, &tun);
3389
3390 out_unlock_clean:
3391         mutex_unlock(&the_lnet.ln_api_mutex);
3392         while (!list_empty(&net_head)) {
3393                 /* net_head list is empty in success case */
3394                 net = list_entry(net_head.next, struct lnet_net, net_list);
3395                 list_del_init(&net->net_list);
3396                 lnet_net_free(net);
3397         }
3398         return rc;
3399 }
3400
3401 int
3402 lnet_dyn_del_net(__u32 net_id)
3403 {
3404         struct lnet_net  *net;
3405         struct lnet_ping_buffer *pbuf;
3406         struct lnet_handle_md ping_mdh;
3407         int               rc;
3408         int               net_ni_count;
3409
3410         /* don't allow userspace to shutdown the LOLND */
3411         if (LNET_NETTYP(net_id) == LOLND)
3412                 return -EINVAL;
3413
3414         mutex_lock(&the_lnet.ln_api_mutex);
3415
3416         lnet_net_lock(0);
3417
3418         net = lnet_get_net_locked(net_id);
3419         if (net == NULL) {
3420                 lnet_net_unlock(0);
3421                 rc = -EINVAL;
3422                 goto out;
3423         }
3424
3425         net_ni_count = lnet_get_net_ni_count_locked(net);
3426
3427         lnet_net_unlock(0);
3428
3429         /* create and link a new ping info, before removing the old one */
3430         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3431                                     lnet_get_ni_count() - net_ni_count, false);
3432         if (rc != 0)
3433                 goto out;
3434
3435         lnet_shutdown_lndnet(net);
3436
3437         if (lnet_count_acceptor_nets() == 0)
3438                 lnet_acceptor_stop();
3439
3440         lnet_ping_target_update(pbuf, ping_mdh);
3441
3442 out:
3443         mutex_unlock(&the_lnet.ln_api_mutex);
3444
3445         return rc;
3446 }
3447
3448 void lnet_incr_dlc_seq(void)
3449 {
3450         atomic_inc(&lnet_dlc_seq_no);
3451 }
3452
3453 __u32 lnet_get_dlc_seq_locked(void)
3454 {
3455         return atomic_read(&lnet_dlc_seq_no);
3456 }
3457
3458 static void
3459 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3460 {
3461         struct lnet_net *net;
3462         struct lnet_ni *ni;
3463
3464         lnet_net_lock(LNET_LOCK_EX);
3465         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3466                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3467                         if (ni->ni_nid == nid || all) {
3468                                 atomic_set(&ni->ni_healthv, value);
3469                                 if (list_empty(&ni->ni_recovery) &&
3470                                     value < LNET_MAX_HEALTH_VALUE) {
3471                                         CERROR("manually adding local NI %s to recovery\n",
3472                                                libcfs_nid2str(ni->ni_nid));
3473                                         list_add_tail(&ni->ni_recovery,
3474                                                       &the_lnet.ln_mt_localNIRecovq);
3475                                         lnet_ni_addref_locked(ni, 0);
3476                                 }
3477                                 if (!all) {
3478                                         lnet_net_unlock(LNET_LOCK_EX);
3479                                         return;
3480                                 }
3481                         }
3482                 }
3483         }
3484         lnet_net_unlock(LNET_LOCK_EX);
3485 }
3486
3487 static int
3488 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3489 {
3490         int cpt, rc = 0;
3491         struct lnet_ni *ni;
3492         lnet_nid_t nid = stats->hlni_nid;
3493
3494         cpt = lnet_net_lock_current();
3495         ni = lnet_nid2ni_locked(nid, cpt);
3496
3497         if (!ni) {
3498                 rc = -ENOENT;
3499                 goto unlock;
3500         }
3501
3502         stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3503         stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3504         stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3505         stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3506         stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3507         stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3508         stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3509
3510 unlock:
3511         lnet_net_unlock(cpt);
3512
3513         return rc;
3514 }
3515
3516 static int
3517 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3518 {
3519         struct lnet_ni *ni;
3520         int i = 0;
3521
3522         lnet_net_lock(LNET_LOCK_EX);
3523         list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3524                 list->rlst_nid_array[i] = ni->ni_nid;
3525                 i++;
3526                 if (i >= LNET_MAX_SHOW_NUM_NID)
3527                         break;
3528         }
3529         lnet_net_unlock(LNET_LOCK_EX);
3530         list->rlst_num_nids = i;
3531
3532         return 0;
3533 }
3534
3535 static int
3536 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3537 {
3538         struct lnet_peer_ni *lpni;
3539         int i = 0;
3540
3541         lnet_net_lock(LNET_LOCK_EX);
3542         list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3543                 list->rlst_nid_array[i] = lpni->lpni_nid;
3544                 i++;
3545                 if (i >= LNET_MAX_SHOW_NUM_NID)
3546                         break;
3547         }
3548         lnet_net_unlock(LNET_LOCK_EX);
3549         list->rlst_num_nids = i;
3550
3551         return 0;
3552 }
3553
3554 /**
3555  * LNet ioctl handler.
3556  *
3557  */
3558 int
3559 LNetCtl(unsigned int cmd, void *arg)
3560 {
3561         struct libcfs_ioctl_data *data = arg;
3562         struct lnet_ioctl_config_data *config;
3563         struct lnet_process_id    id = {0};
3564         struct lnet_ni           *ni;
3565         int                       rc;
3566
3567         BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3568                      sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3569
3570         switch (cmd) {
3571         case IOC_LIBCFS_GET_NI:
3572                 rc = LNetGetId(data->ioc_count, &id);
3573                 data->ioc_nid = id.nid;
3574                 return rc;
3575
3576         case IOC_LIBCFS_FAIL_NID:
3577                 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3578
3579         case IOC_LIBCFS_ADD_ROUTE: {
3580                 /* default router sensitivity to 1 */
3581                 unsigned int sensitivity = 1;
3582                 config = arg;
3583
3584                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3585                         return -EINVAL;
3586
3587                 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3588                         sensitivity =
3589                           config->cfg_config_u.cfg_route.rtr_sensitivity;
3590                 }
3591
3592                 mutex_lock(&the_lnet.ln_api_mutex);
3593                 rc = lnet_add_route(config->cfg_net,
3594                                     config->cfg_config_u.cfg_route.rtr_hop,
3595                                     config->cfg_nid,
3596                                     config->cfg_config_u.cfg_route.
3597                                         rtr_priority, sensitivity);
3598                 mutex_unlock(&the_lnet.ln_api_mutex);
3599                 return rc;
3600         }
3601
3602         case IOC_LIBCFS_DEL_ROUTE:
3603                 config = arg;
3604
3605                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3606                         return -EINVAL;
3607
3608                 mutex_lock(&the_lnet.ln_api_mutex);
3609                 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3610                 mutex_unlock(&the_lnet.ln_api_mutex);
3611                 return rc;
3612
3613         case IOC_LIBCFS_GET_ROUTE:
3614                 config = arg;
3615
3616                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3617                         return -EINVAL;
3618
3619                 mutex_lock(&the_lnet.ln_api_mutex);
3620                 rc = lnet_get_route(config->cfg_count,
3621                                     &config->cfg_net,
3622                                     &config->cfg_config_u.cfg_route.rtr_hop,
3623                                     &config->cfg_nid,
3624                                     &config->cfg_config_u.cfg_route.rtr_flags,
3625                                     &config->cfg_config_u.cfg_route.
3626                                         rtr_priority,
3627                                     &config->cfg_config_u.cfg_route.
3628                                         rtr_sensitivity);
3629                 mutex_unlock(&the_lnet.ln_api_mutex);
3630                 return rc;
3631
3632         case IOC_LIBCFS_GET_LOCAL_NI: {
3633                 struct lnet_ioctl_config_ni *cfg_ni;
3634                 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3635                 struct lnet_ioctl_element_stats *stats;
3636                 __u32 tun_size;
3637
3638                 cfg_ni = arg;
3639
3640                 /* get the tunables if they are available */
3641                 if (cfg_ni->lic_cfg_hdr.ioc_len <
3642                     sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3643                         return -EINVAL;
3644
3645                 stats = (struct lnet_ioctl_element_stats *)
3646                         cfg_ni->lic_bulk;
3647                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3648                                 (cfg_ni->lic_bulk + sizeof(*stats));
3649
3650                 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3651                         sizeof(*stats);
3652
3653                 mutex_lock(&the_lnet.ln_api_mutex);
3654                 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3655                 mutex_unlock(&the_lnet.ln_api_mutex);
3656                 return rc;
3657         }
3658
3659         case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3660                 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3661
3662                 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3663                         return -EINVAL;
3664
3665                 mutex_lock(&the_lnet.ln_api_mutex);
3666                 rc = lnet_get_ni_stats(msg_stats);
3667                 mutex_unlock(&the_lnet.ln_api_mutex);
3668
3669                 return rc;
3670         }
3671
3672         case IOC_LIBCFS_GET_NET: {
3673                 size_t total = sizeof(*config) +
3674                                sizeof(struct lnet_ioctl_net_config);
3675                 config = arg;
3676
3677                 if (config->cfg_hdr.ioc_len < total)
3678                         return -EINVAL;
3679
3680                 mutex_lock(&the_lnet.ln_api_mutex);
3681                 rc = lnet_get_net_config(config);
3682                 mutex_unlock(&the_lnet.ln_api_mutex);
3683                 return rc;
3684         }
3685
3686         case IOC_LIBCFS_GET_LNET_STATS:
3687         {
3688                 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
3689
3690                 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
3691                         return -EINVAL;
3692
3693                 mutex_lock(&the_lnet.ln_api_mutex);
3694                 lnet_counters_get(&lnet_stats->st_cntrs);
3695                 mutex_unlock(&the_lnet.ln_api_mutex);
3696                 return 0;
3697         }
3698
3699         case IOC_LIBCFS_CONFIG_RTR:
3700                 config = arg;
3701
3702                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3703                         return -EINVAL;
3704
3705                 mutex_lock(&the_lnet.ln_api_mutex);
3706                 if (config->cfg_config_u.cfg_buffers.buf_enable) {
3707                         rc = lnet_rtrpools_enable();
3708                         mutex_unlock(&the_lnet.ln_api_mutex);
3709                         return rc;
3710                 }
3711                 lnet_rtrpools_disable();
3712                 mutex_unlock(&the_lnet.ln_api_mutex);
3713                 return 0;
3714
3715         case IOC_LIBCFS_ADD_BUF:
3716                 config = arg;
3717
3718                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3719                         return -EINVAL;
3720
3721                 mutex_lock(&the_lnet.ln_api_mutex);
3722                 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
3723                                                 buf_tiny,
3724                                           config->cfg_config_u.cfg_buffers.
3725                                                 buf_small,
3726                                           config->cfg_config_u.cfg_buffers.
3727                                                 buf_large);
3728                 mutex_unlock(&the_lnet.ln_api_mutex);
3729                 return rc;
3730
3731         case IOC_LIBCFS_SET_NUMA_RANGE: {
3732                 struct lnet_ioctl_set_value *numa;
3733                 numa = arg;
3734                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3735                         return -EINVAL;
3736                 lnet_net_lock(LNET_LOCK_EX);
3737                 lnet_numa_range = numa->sv_value;
3738                 lnet_net_unlock(LNET_LOCK_EX);
3739                 return 0;
3740         }
3741
3742         case IOC_LIBCFS_GET_NUMA_RANGE: {
3743                 struct lnet_ioctl_set_value *numa;
3744                 numa = arg;
3745                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3746                         return -EINVAL;
3747                 numa->sv_value = lnet_numa_range;
3748                 return 0;
3749         }
3750
3751         case IOC_LIBCFS_GET_BUF: {
3752                 struct lnet_ioctl_pool_cfg *pool_cfg;
3753                 size_t total = sizeof(*config) + sizeof(*pool_cfg);
3754
3755                 config = arg;
3756
3757                 if (config->cfg_hdr.ioc_len < total)
3758                         return -EINVAL;
3759
3760                 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
3761
3762                 mutex_lock(&the_lnet.ln_api_mutex);
3763                 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
3764                 mutex_unlock(&the_lnet.ln_api_mutex);
3765                 return rc;
3766         }
3767
3768         case IOC_LIBCFS_GET_LOCAL_HSTATS: {
3769                 struct lnet_ioctl_local_ni_hstats *stats = arg;
3770
3771                 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
3772                         return -EINVAL;
3773
3774                 mutex_lock(&the_lnet.ln_api_mutex);
3775                 rc = lnet_get_local_ni_hstats(stats);
3776                 mutex_unlock(&the_lnet.ln_api_mutex);
3777
3778                 return rc;
3779         }
3780
3781         case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
3782                 struct lnet_ioctl_recovery_list *list = arg;
3783                 if (list->rlst_hdr.ioc_len < sizeof(*list))
3784                         return -EINVAL;
3785
3786                 mutex_lock(&the_lnet.ln_api_mutex);
3787                 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
3788                         rc = lnet_get_local_ni_recovery_list(list);
3789                 else
3790                         rc = lnet_get_peer_ni_recovery_list(list);
3791                 mutex_unlock(&the_lnet.ln_api_mutex);
3792                 return rc;
3793         }
3794
3795         case IOC_LIBCFS_ADD_PEER_NI: {
3796                 struct lnet_ioctl_peer_cfg *cfg = arg;
3797
3798                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3799                         return -EINVAL;
3800
3801                 mutex_lock(&the_lnet.ln_api_mutex);
3802                 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
3803                                       cfg->prcfg_cfg_nid,
3804                                       cfg->prcfg_mr);
3805                 mutex_unlock(&the_lnet.ln_api_mutex);
3806                 return rc;
3807         }
3808
3809         case IOC_LIBCFS_DEL_PEER_NI: {
3810                 struct lnet_ioctl_peer_cfg *cfg = arg;
3811
3812                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3813                         return -EINVAL;
3814
3815                 mutex_lock(&the_lnet.ln_api_mutex);
3816                 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
3817                                       cfg->prcfg_cfg_nid);
3818                 mutex_unlock(&the_lnet.ln_api_mutex);
3819                 return rc;
3820         }
3821
3822         case IOC_LIBCFS_GET_PEER_INFO: {
3823                 struct lnet_ioctl_peer *peer_info = arg;
3824
3825                 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
3826                         return -EINVAL;
3827
3828                 mutex_lock(&the_lnet.ln_api_mutex);
3829                 rc = lnet_get_peer_ni_info(
3830                    peer_info->pr_count,
3831                    &peer_info->pr_nid,
3832                    peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
3833                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
3834                    &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
3835                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
3836                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
3837                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
3838                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
3839                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
3840                 mutex_unlock(&the_lnet.ln_api_mutex);
3841                 return rc;
3842         }
3843
3844         case IOC_LIBCFS_GET_PEER_NI: {
3845                 struct lnet_ioctl_peer_cfg *cfg = arg;
3846
3847                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3848                         return -EINVAL;
3849
3850                 mutex_lock(&the_lnet.ln_api_mutex);
3851                 rc = lnet_get_peer_info(cfg,
3852                                         (void __user *)cfg->prcfg_bulk);
3853                 mutex_unlock(&the_lnet.ln_api_mutex);
3854                 return rc;
3855         }
3856
3857         case IOC_LIBCFS_GET_PEER_LIST: {
3858                 struct lnet_ioctl_peer_cfg *cfg = arg;
3859
3860                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3861                         return -EINVAL;
3862
3863                 mutex_lock(&the_lnet.ln_api_mutex);
3864                 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
3865                                 (struct lnet_process_id __user *)cfg->prcfg_bulk);
3866                 mutex_unlock(&the_lnet.ln_api_mutex);
3867                 return rc;
3868         }
3869
3870         case IOC_LIBCFS_SET_HEALHV: {
3871                 struct lnet_ioctl_reset_health_cfg *cfg = arg;
3872                 int value;
3873                 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
3874                         return -EINVAL;
3875                 if (cfg->rh_value < 0 ||
3876                     cfg->rh_value > LNET_MAX_HEALTH_VALUE)
3877                         value = LNET_MAX_HEALTH_VALUE;
3878                 else
3879                         value = cfg->rh_value;
3880                 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
3881                        value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
3882                        "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
3883                 mutex_lock(&the_lnet.ln_api_mutex);
3884                 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
3885                         lnet_ni_set_healthv(cfg->rh_nid, value,
3886                                              cfg->rh_all);
3887                 else
3888                         lnet_peer_ni_set_healthv(cfg->rh_nid, value,
3889                                                   cfg->rh_all);
3890                 mutex_unlock(&the_lnet.ln_api_mutex);
3891                 return 0;
3892         }
3893
3894         case IOC_LIBCFS_NOTIFY_ROUTER: {
3895                 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
3896
3897                 /* The deadline passed in by the user should be some time in
3898                  * seconds in the future since the UNIX epoch. We have to map
3899                  * that deadline to the wall clock.
3900                  */
3901                 deadline += ktime_get_seconds();
3902                 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
3903                                    deadline);
3904         }
3905
3906         case IOC_LIBCFS_LNET_DIST:
3907                 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
3908                 if (rc < 0 && rc != -EHOSTUNREACH)
3909                         return rc;
3910
3911                 data->ioc_u32[0] = rc;
3912                 return 0;
3913
3914         case IOC_LIBCFS_TESTPROTOCOMPAT:
3915                 the_lnet.ln_testprotocompat = data->ioc_flags;
3916                 return 0;
3917
3918         case IOC_LIBCFS_LNET_FAULT:
3919                 return lnet_fault_ctl(data->ioc_flags, data);
3920
3921         case IOC_LIBCFS_PING: {
3922                 signed long timeout;
3923
3924                 id.nid = data->ioc_nid;
3925                 id.pid = data->ioc_u32[0];
3926
3927                 /* If timeout is negative then set default of 3 minutes */
3928                 if (((s32)data->ioc_u32[1] <= 0) ||
3929                     data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3930                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3931                 else
3932                         timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
3933
3934                 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
3935                                data->ioc_plen1 / sizeof(struct lnet_process_id));
3936
3937                 if (rc < 0)
3938                         return rc;
3939
3940                 data->ioc_count = rc;
3941                 return 0;
3942         }
3943
3944         case IOC_LIBCFS_PING_PEER: {
3945                 struct lnet_ioctl_ping_data *ping = arg;
3946                 struct lnet_peer *lp;
3947                 signed long timeout;
3948
3949                 /* If timeout is negative then set default of 3 minutes */
3950                 if (((s32)ping->op_param) <= 0 ||
3951                     ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3952                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3953                 else
3954                         timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
3955
3956                 rc = lnet_ping(ping->ping_id, timeout,
3957                                ping->ping_buf,
3958                                ping->ping_count);
3959                 if (rc < 0)
3960                         return rc;
3961
3962                 mutex_lock(&the_lnet.ln_api_mutex);
3963                 lp = lnet_find_peer(ping->ping_id.nid);
3964                 if (lp) {
3965                         ping->ping_id.nid = lp->lp_primary_nid;
3966                         ping->mr_info = lnet_peer_is_multi_rail(lp);
3967                         lnet_peer_decref_locked(lp);
3968                 }
3969                 mutex_unlock(&the_lnet.ln_api_mutex);
3970
3971                 ping->ping_count = rc;
3972                 return 0;
3973         }
3974
3975         case IOC_LIBCFS_DISCOVER: {
3976                 struct lnet_ioctl_ping_data *discover = arg;
3977                 struct lnet_peer *lp;
3978
3979                 rc = lnet_discover(discover->ping_id, discover->op_param,
3980                                    discover->ping_buf,
3981                                    discover->ping_count);
3982                 if (rc < 0)
3983                         return rc;
3984
3985                 mutex_lock(&the_lnet.ln_api_mutex);
3986                 lp = lnet_find_peer(discover->ping_id.nid);
3987                 if (lp) {
3988                         discover->ping_id.nid = lp->lp_primary_nid;
3989                         discover->mr_info = lnet_peer_is_multi_rail(lp);
3990                         lnet_peer_decref_locked(lp);
3991                 }
3992                 mutex_unlock(&the_lnet.ln_api_mutex);
3993
3994                 discover->ping_count = rc;
3995                 return 0;
3996         }
3997
3998         default:
3999                 ni = lnet_net2ni_addref(data->ioc_net);
4000                 if (ni == NULL)
4001                         return -EINVAL;
4002
4003                 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
4004                         rc = -EINVAL;
4005                 else
4006                         rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
4007
4008                 lnet_ni_decref(ni);
4009                 return rc;
4010         }
4011         /* not reached */
4012 }
4013 EXPORT_SYMBOL(LNetCtl);
4014
4015 void LNetDebugPeer(struct lnet_process_id id)
4016 {
4017         lnet_debug_peer(id.nid);
4018 }
4019 EXPORT_SYMBOL(LNetDebugPeer);
4020
4021 /**
4022  * Determine if the specified peer \a nid is on the local node.
4023  *
4024  * \param nid   peer nid to check
4025  *
4026  * \retval true         If peer NID is on the local node.
4027  * \retval false        If peer NID is not on the local node.
4028  */
4029 bool LNetIsPeerLocal(lnet_nid_t nid)
4030 {
4031         struct lnet_net *net;
4032         struct lnet_ni *ni;
4033         int cpt;
4034
4035         cpt = lnet_net_lock_current();
4036         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4037                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4038                         if (ni->ni_nid == nid) {
4039                                 lnet_net_unlock(cpt);
4040                                 return true;
4041                         }
4042                 }
4043         }
4044         lnet_net_unlock(cpt);
4045
4046         return false;
4047 }
4048 EXPORT_SYMBOL(LNetIsPeerLocal);
4049
4050 /**
4051  * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4052  * Note that all interfaces share a same PID, as requested by LNetNIInit().
4053  *
4054  * \param index Index of the interface to look up.
4055  * \param id On successful return, this location will hold the
4056  * struct lnet_process_id ID of the interface.
4057  *
4058  * \retval 0 If an interface exists at \a index.
4059  * \retval -ENOENT If no interface has been found.
4060  */
4061 int
4062 LNetGetId(unsigned int index, struct lnet_process_id *id)
4063 {
4064         struct lnet_ni   *ni;
4065         struct lnet_net  *net;
4066         int               cpt;
4067         int               rc = -ENOENT;
4068
4069         LASSERT(the_lnet.ln_refcount > 0);
4070
4071         cpt = lnet_net_lock_current();
4072
4073         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4074                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4075                         if (index-- != 0)
4076                                 continue;
4077
4078                         id->nid = ni->ni_nid;
4079                         id->pid = the_lnet.ln_pid;
4080                         rc = 0;
4081                         break;
4082                 }
4083         }
4084
4085         lnet_net_unlock(cpt);
4086         return rc;
4087 }
4088 EXPORT_SYMBOL(LNetGetId);
4089
4090 static int lnet_ping(struct lnet_process_id id, signed long timeout,
4091                      struct lnet_process_id __user *ids, int n_ids)
4092 {
4093         struct lnet_handle_eq eqh;
4094         struct lnet_handle_md mdh;
4095         struct lnet_event event;
4096         struct lnet_md md = { NULL };
4097         int which;
4098         int unlinked = 0;
4099         int replied = 0;
4100         const signed long a_long_time = cfs_time_seconds(60);
4101         struct lnet_ping_buffer *pbuf;
4102         struct lnet_process_id tmpid;
4103         int i;
4104         int nob;
4105         int rc;
4106         int rc2;
4107         sigset_t blocked;
4108
4109         /* n_ids limit is arbitrary */
4110         if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4111                 return -EINVAL;
4112
4113         /*
4114          * if the user buffer has more space than the lnet_interfaces_max
4115          * then only fill it up to lnet_interfaces_max
4116          */
4117         if (n_ids > lnet_interfaces_max)
4118                 n_ids = lnet_interfaces_max;
4119
4120         if (id.pid == LNET_PID_ANY)
4121                 id.pid = LNET_PID_LUSTRE;
4122
4123         pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4124         if (!pbuf)
4125                 return -ENOMEM;
4126
4127         /* NB 2 events max (including any unlink event) */
4128         rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &eqh);
4129         if (rc != 0) {
4130                 CERROR("Can't allocate EQ: %d\n", rc);
4131                 goto fail_ping_buffer_decref;
4132         }
4133
4134         /* initialize md content */
4135         md.start     = &pbuf->pb_info;
4136         md.length    = LNET_PING_INFO_SIZE(n_ids);
4137         md.threshold = 2; /* GET/REPLY */
4138         md.max_size  = 0;
4139         md.options   = LNET_MD_TRUNCATE;
4140         md.user_ptr  = NULL;
4141         md.eq_handle = eqh;
4142
4143         rc = LNetMDBind(md, LNET_UNLINK, &mdh);
4144         if (rc != 0) {
4145                 CERROR("Can't bind MD: %d\n", rc);
4146                 goto fail_free_eq;
4147         }
4148
4149         rc = LNetGet(LNET_NID_ANY, mdh, id,
4150                      LNET_RESERVED_PORTAL,
4151                      LNET_PROTO_PING_MATCHBITS, 0, false);
4152
4153         if (rc != 0) {
4154                 /* Don't CERROR; this could be deliberate! */
4155                 rc2 = LNetMDUnlink(mdh);
4156                 LASSERT(rc2 == 0);
4157
4158                 /* NB must wait for the UNLINK event below... */
4159                 unlinked = 1;
4160                 timeout = a_long_time;
4161         }
4162
4163         do {
4164                 /* MUST block for unlink to complete */
4165                 if (unlinked)
4166                         blocked = cfs_block_allsigs();
4167
4168                 rc2 = LNetEQPoll(&eqh, 1, timeout, &event, &which);
4169
4170                 if (unlinked)
4171                         cfs_restore_sigs(blocked);
4172
4173                 CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
4174                        (rc2 <= 0) ? -1 : event.type,
4175                        (rc2 <= 0) ? -1 : event.status,
4176                        (rc2 > 0 && event.unlinked) ? " unlinked" : "");
4177
4178                 LASSERT(rc2 != -EOVERFLOW);     /* can't miss anything */
4179
4180                 if (rc2 <= 0 || event.status != 0) {
4181                         /* timeout or error */
4182                         if (!replied && rc == 0)
4183                                 rc = (rc2 < 0) ? rc2 :
4184                                      (rc2 == 0) ? -ETIMEDOUT :
4185                                      event.status;
4186
4187                         if (!unlinked) {
4188                                 /* Ensure completion in finite time... */
4189                                 LNetMDUnlink(mdh);
4190                                 /* No assertion (racing with network) */
4191                                 unlinked = 1;
4192                                 timeout = a_long_time;
4193                         } else if (rc2 == 0) {
4194                                 /* timed out waiting for unlink */
4195                                 CWARN("ping %s: late network completion\n",
4196                                       libcfs_id2str(id));
4197                         }
4198                 } else if (event.type == LNET_EVENT_REPLY) {
4199                         replied = 1;
4200                         rc = event.mlength;
4201                 }
4202         } while (rc2 <= 0 || !event.unlinked);
4203
4204         if (!replied) {
4205                 if (rc >= 0)
4206                         CWARN("%s: Unexpected rc >= 0 but no reply!\n",
4207                               libcfs_id2str(id));
4208                 rc = -EIO;
4209                 goto fail_free_eq;
4210         }
4211
4212         nob = rc;
4213         LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4214
4215         rc = -EPROTO;           /* if I can't parse... */
4216
4217         if (nob < 8) {
4218                 CERROR("%s: ping info too short %d\n",
4219                        libcfs_id2str(id), nob);
4220                 goto fail_free_eq;
4221         }
4222
4223         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4224                 lnet_swap_pinginfo(pbuf);
4225         } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4226                 CERROR("%s: Unexpected magic %08x\n",
4227                        libcfs_id2str(id), pbuf->pb_info.pi_magic);
4228                 goto fail_free_eq;
4229         }
4230
4231         if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4232                 CERROR("%s: ping w/o NI status: 0x%x\n",
4233                        libcfs_id2str(id), pbuf->pb_info.pi_features);
4234                 goto fail_free_eq;
4235         }
4236
4237         if (nob < LNET_PING_INFO_SIZE(0)) {
4238                 CERROR("%s: Short reply %d(%d min)\n",
4239                        libcfs_id2str(id),
4240                        nob, (int)LNET_PING_INFO_SIZE(0));
4241                 goto fail_free_eq;
4242         }
4243
4244         if (pbuf->pb_info.pi_nnis < n_ids)
4245                 n_ids = pbuf->pb_info.pi_nnis;
4246
4247         if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4248                 CERROR("%s: Short reply %d(%d expected)\n",
4249                        libcfs_id2str(id),
4250                        nob, (int)LNET_PING_INFO_SIZE(n_ids));
4251                 goto fail_free_eq;
4252         }
4253
4254         rc = -EFAULT;           /* if I segv in copy_to_user()... */
4255
4256         memset(&tmpid, 0, sizeof(tmpid));
4257         for (i = 0; i < n_ids; i++) {
4258                 tmpid.pid = pbuf->pb_info.pi_pid;
4259                 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4260                 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4261                         goto fail_free_eq;
4262         }
4263         rc = pbuf->pb_info.pi_nnis;
4264
4265  fail_free_eq:
4266         rc2 = LNetEQFree(eqh);
4267         if (rc2 != 0)
4268                 CERROR("rc2 %d\n", rc2);
4269         LASSERT(rc2 == 0);
4270
4271  fail_ping_buffer_decref:
4272         lnet_ping_buffer_decref(pbuf);
4273         return rc;
4274 }
4275
4276 static int
4277 lnet_discover(struct lnet_process_id id, __u32 force,
4278               struct lnet_process_id __user *ids, int n_ids)
4279 {
4280         struct lnet_peer_ni *lpni;
4281         struct lnet_peer_ni *p;
4282         struct lnet_peer *lp;
4283         struct lnet_process_id *buf;
4284         int cpt;
4285         int i;
4286         int rc;
4287         int max_intf = lnet_interfaces_max;
4288         size_t buf_size;
4289
4290         if (n_ids <= 0 ||
4291             id.nid == LNET_NID_ANY)
4292                 return -EINVAL;
4293
4294         if (id.pid == LNET_PID_ANY)
4295                 id.pid = LNET_PID_LUSTRE;
4296
4297         /*
4298          * if the user buffer has more space than the max_intf
4299          * then only fill it up to max_intf
4300          */
4301         if (n_ids > max_intf)
4302                 n_ids = max_intf;
4303
4304         buf_size = n_ids * sizeof(*buf);
4305
4306         LIBCFS_ALLOC(buf, buf_size);
4307         if (!buf)
4308                 return -ENOMEM;
4309
4310         cpt = lnet_net_lock_current();
4311         lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4312         if (IS_ERR(lpni)) {
4313                 rc = PTR_ERR(lpni);
4314                 goto out;
4315         }
4316
4317         /*
4318          * Clearing the NIDS_UPTODATE flag ensures the peer will
4319          * be discovered, provided discovery has not been disabled.
4320          */
4321         lp = lpni->lpni_peer_net->lpn_peer;
4322         spin_lock(&lp->lp_lock);
4323         lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4324         /* If the force flag is set, force a PING and PUSH as well. */
4325         if (force)
4326                 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4327         spin_unlock(&lp->lp_lock);
4328         rc = lnet_discover_peer_locked(lpni, cpt, true);
4329         if (rc)
4330                 goto out_decref;
4331
4332         /* Peer may have changed. */
4333         lp = lpni->lpni_peer_net->lpn_peer;
4334         if (lp->lp_nnis < n_ids)
4335                 n_ids = lp->lp_nnis;
4336
4337         i = 0;
4338         p = NULL;
4339         while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4340                 buf[i].pid = id.pid;
4341                 buf[i].nid = p->lpni_nid;
4342                 if (++i >= n_ids)
4343                         break;
4344         }
4345
4346         lnet_net_unlock(cpt);
4347
4348         rc = -EFAULT;
4349         if (copy_to_user(ids, buf, n_ids * sizeof(*buf)))
4350                 goto out_relock;
4351         rc = n_ids;
4352 out_relock:
4353         lnet_net_lock(cpt);
4354 out_decref:
4355         lnet_peer_ni_decref_locked(lpni);
4356 out:
4357         lnet_net_unlock(cpt);
4358
4359         LIBCFS_FREE(buf, buf_size);
4360
4361         return rc;
4362 }
4363
4364 /**
4365  * Retrieve peer discovery status.
4366  *
4367  * \retval 1 if lnet_peer_discovery_disabled is 0
4368  * \retval 0 if lnet_peer_discovery_disabled is 1
4369  */
4370 int
4371 LNetGetPeerDiscoveryStatus(void)
4372 {
4373         return !lnet_peer_discovery_disabled;
4374 }
4375 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);