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                 schedule_timeout_uninterruptible(cfs_time_seconds(1));
1753         }
1754
1755         cfs_restore_sigs(blocked);
1756 }
1757
1758 static void
1759 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1760 {
1761         struct lnet_ni          *ni;
1762         struct lnet_net         *net;
1763         struct lnet_ni_status *ns;
1764         int                     i;
1765         int                     rc;
1766
1767         i = 0;
1768         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1769                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1770                         LASSERT(i < pbuf->pb_nnis);
1771
1772                         ns = &pbuf->pb_info.pi_ni[i];
1773
1774                         ns->ns_nid = ni->ni_nid;
1775
1776                         lnet_ni_lock(ni);
1777                         ns->ns_status = (ni->ni_status != NULL) ?
1778                                          ni->ni_status->ns_status :
1779                                                 LNET_NI_STATUS_UP;
1780                         ni->ni_status = ns;
1781                         lnet_ni_unlock(ni);
1782
1783                         i++;
1784                 }
1785         }
1786         /*
1787          * We (ab)use the ns_status of the loopback interface to
1788          * transmit the sequence number. The first interface listed
1789          * must be the loopback interface.
1790          */
1791         rc = lnet_ping_info_validate(&pbuf->pb_info);
1792         if (rc) {
1793                 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1794                 LBUG();
1795         }
1796         LNET_PING_BUFFER_SEQNO(pbuf) =
1797                 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1798 }
1799
1800 static void
1801 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1802                         struct lnet_handle_md ping_mdh)
1803 {
1804         struct lnet_ping_buffer *old_pbuf = NULL;
1805         struct lnet_handle_md old_ping_md;
1806
1807         /* switch the NIs to point to the new ping info created */
1808         lnet_net_lock(LNET_LOCK_EX);
1809
1810         if (!the_lnet.ln_routing)
1811                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1812         if (!lnet_peer_discovery_disabled)
1813                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1814
1815         /* Ensure only known feature bits have been set. */
1816         LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1817         LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1818
1819         lnet_ping_target_install_locked(pbuf);
1820
1821         if (the_lnet.ln_ping_target) {
1822                 old_pbuf = the_lnet.ln_ping_target;
1823                 old_ping_md = the_lnet.ln_ping_target_md;
1824         }
1825         the_lnet.ln_ping_target_md = ping_mdh;
1826         the_lnet.ln_ping_target = pbuf;
1827
1828         lnet_net_unlock(LNET_LOCK_EX);
1829
1830         if (old_pbuf) {
1831                 /* unlink and free the old ping info */
1832                 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1833                 lnet_ping_buffer_decref(old_pbuf);
1834         }
1835
1836         lnet_push_update_to_peers(0);
1837 }
1838
1839 static void
1840 lnet_ping_target_fini(void)
1841 {
1842         int             rc;
1843
1844         lnet_ping_md_unlink(the_lnet.ln_ping_target,
1845                             &the_lnet.ln_ping_target_md);
1846
1847         rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1848         LASSERT(rc == 0);
1849
1850         lnet_ping_target_destroy();
1851 }
1852
1853 /* Resize the push target. */
1854 int lnet_push_target_resize(void)
1855 {
1856         struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
1857         struct lnet_md md = { NULL };
1858         struct lnet_me *me;
1859         struct lnet_handle_md mdh;
1860         struct lnet_handle_md old_mdh;
1861         struct lnet_ping_buffer *pbuf;
1862         struct lnet_ping_buffer *old_pbuf;
1863         int nnis = the_lnet.ln_push_target_nnis;
1864         int rc;
1865
1866         if (nnis <= 0) {
1867                 rc = -EINVAL;
1868                 goto fail_return;
1869         }
1870 again:
1871         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1872         if (!pbuf) {
1873                 rc = -ENOMEM;
1874                 goto fail_return;
1875         }
1876
1877         me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1878                           LNET_PROTO_PING_MATCHBITS, 0,
1879                           LNET_UNLINK, LNET_INS_AFTER);
1880
1881         if (IS_ERR(me)) {
1882                 rc = PTR_ERR(me);
1883                 CERROR("Can't create push target ME: %d\n", rc);
1884                 goto fail_decref_pbuf;
1885         }
1886
1887         /* initialize md content */
1888         md.start     = &pbuf->pb_info;
1889         md.length    = LNET_PING_INFO_SIZE(nnis);
1890         md.threshold = LNET_MD_THRESH_INF;
1891         md.max_size  = 0;
1892         md.options   = LNET_MD_OP_PUT | LNET_MD_TRUNCATE |
1893                        LNET_MD_MANAGE_REMOTE;
1894         md.user_ptr  = pbuf;
1895         md.eq_handle = the_lnet.ln_push_target_eq;
1896
1897         rc = LNetMDAttach(me, md, LNET_RETAIN, &mdh);
1898         if (rc) {
1899                 CERROR("Can't attach push MD: %d\n", rc);
1900                 goto fail_unlink_me;
1901         }
1902         lnet_ping_buffer_addref(pbuf);
1903
1904         lnet_net_lock(LNET_LOCK_EX);
1905         old_pbuf = the_lnet.ln_push_target;
1906         old_mdh = the_lnet.ln_push_target_md;
1907         the_lnet.ln_push_target = pbuf;
1908         the_lnet.ln_push_target_md = mdh;
1909         lnet_net_unlock(LNET_LOCK_EX);
1910
1911         if (old_pbuf) {
1912                 LNetMDUnlink(old_mdh);
1913                 lnet_ping_buffer_decref(old_pbuf);
1914         }
1915
1916         if (nnis < the_lnet.ln_push_target_nnis)
1917                 goto again;
1918
1919         CDEBUG(D_NET, "nnis %d success\n", nnis);
1920
1921         return 0;
1922
1923 fail_unlink_me:
1924         LNetMEUnlink(me);
1925 fail_decref_pbuf:
1926         lnet_ping_buffer_decref(pbuf);
1927 fail_return:
1928         CDEBUG(D_NET, "nnis %d error %d\n", nnis, rc);
1929         return rc;
1930 }
1931
1932 static void lnet_push_target_event_handler(struct lnet_event *ev)
1933 {
1934         struct lnet_ping_buffer *pbuf = ev->md.user_ptr;
1935
1936         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
1937                 lnet_swap_pinginfo(pbuf);
1938
1939         lnet_peer_push_event(ev);
1940         if (ev->unlinked)
1941                 lnet_ping_buffer_decref(pbuf);
1942 }
1943
1944 /* Initialize the push target. */
1945 static int lnet_push_target_init(void)
1946 {
1947         int rc;
1948
1949         if (the_lnet.ln_push_target)
1950                 return -EALREADY;
1951
1952         rc = LNetEQAlloc(0, lnet_push_target_event_handler,
1953                          &the_lnet.ln_push_target_eq);
1954         if (rc) {
1955                 CERROR("Can't allocated push target EQ: %d\n", rc);
1956                 return rc;
1957         }
1958
1959         /* Start at the required minimum, we'll enlarge if required. */
1960         the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
1961
1962         rc = lnet_push_target_resize();
1963
1964         if (rc) {
1965                 LNetEQFree(the_lnet.ln_push_target_eq);
1966                 LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1967         }
1968
1969         return rc;
1970 }
1971
1972 /* Clean up the push target. */
1973 static void lnet_push_target_fini(void)
1974 {
1975         if (!the_lnet.ln_push_target)
1976                 return;
1977
1978         /* Unlink and invalidate to prevent new references. */
1979         LNetMDUnlink(the_lnet.ln_push_target_md);
1980         LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
1981
1982         /* Wait for the unlink to complete. */
1983         while (atomic_read(&the_lnet.ln_push_target->pb_refcnt) > 1) {
1984                 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1985                 schedule_timeout_uninterruptible(cfs_time_seconds(1));
1986         }
1987
1988         lnet_ping_buffer_decref(the_lnet.ln_push_target);
1989         the_lnet.ln_push_target = NULL;
1990         the_lnet.ln_push_target_nnis = 0;
1991
1992         LNetEQFree(the_lnet.ln_push_target_eq);
1993         LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1994 }
1995
1996 static int
1997 lnet_ni_tq_credits(struct lnet_ni *ni)
1998 {
1999         int     credits;
2000
2001         LASSERT(ni->ni_ncpts >= 1);
2002
2003         if (ni->ni_ncpts == 1)
2004                 return ni->ni_net->net_tunables.lct_max_tx_credits;
2005
2006         credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2007         credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2008         credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2009
2010         return credits;
2011 }
2012
2013 static void
2014 lnet_ni_unlink_locked(struct lnet_ni *ni)
2015 {
2016         /* move it to zombie list and nobody can find it anymore */
2017         LASSERT(!list_empty(&ni->ni_netlist));
2018         list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2019         lnet_ni_decref_locked(ni, 0);
2020 }
2021
2022 static void
2023 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2024 {
2025         int             i;
2026         int             islo;
2027         struct lnet_ni  *ni;
2028         struct list_head *zombie_list = &net->net_ni_zombie;
2029
2030         /*
2031          * Now wait for the NIs I just nuked to show up on the zombie
2032          * list and shut them down in guaranteed thread context
2033          */
2034         i = 2;
2035         while (!list_empty(zombie_list)) {
2036                 int     *ref;
2037                 int     j;
2038
2039                 ni = list_entry(zombie_list->next,
2040                                 struct lnet_ni, ni_netlist);
2041                 list_del_init(&ni->ni_netlist);
2042                 /* the ni should be in deleting state. If it's not it's
2043                  * a bug */
2044                 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2045                 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2046                         if (*ref == 0)
2047                                 continue;
2048                         /* still busy, add it back to zombie list */
2049                         list_add(&ni->ni_netlist, zombie_list);
2050                         break;
2051                 }
2052
2053                 if (!list_empty(&ni->ni_netlist)) {
2054                         lnet_net_unlock(LNET_LOCK_EX);
2055                         ++i;
2056                         if ((i & (-i)) == i) {
2057                                 CDEBUG(D_WARNING,
2058                                        "Waiting for zombie LNI %s\n",
2059                                        libcfs_nid2str(ni->ni_nid));
2060                         }
2061                         schedule_timeout_uninterruptible(cfs_time_seconds(1));
2062                         lnet_net_lock(LNET_LOCK_EX);
2063                         continue;
2064                 }
2065
2066                 lnet_net_unlock(LNET_LOCK_EX);
2067
2068                 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2069
2070                 LASSERT(!in_interrupt());
2071                 /* Holding the mutex makes it safe for lnd_shutdown
2072                  * to call module_put(). Module unload cannot finish
2073                  * until lnet_unregister_lnd() completes, and that
2074                  * requires the mutex.
2075                  */
2076                 mutex_lock(&the_lnet.ln_lnd_mutex);
2077                 (net->net_lnd->lnd_shutdown)(ni);
2078                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2079
2080                 if (!islo)
2081                         CDEBUG(D_LNI, "Removed LNI %s\n",
2082                               libcfs_nid2str(ni->ni_nid));
2083
2084                 lnet_ni_free(ni);
2085                 i = 2;
2086                 lnet_net_lock(LNET_LOCK_EX);
2087         }
2088 }
2089
2090 /* shutdown down the NI and release refcount */
2091 static void
2092 lnet_shutdown_lndni(struct lnet_ni *ni)
2093 {
2094         int i;
2095         struct lnet_net *net = ni->ni_net;
2096
2097         lnet_net_lock(LNET_LOCK_EX);
2098         lnet_ni_lock(ni);
2099         ni->ni_state = LNET_NI_STATE_DELETING;
2100         lnet_ni_unlock(ni);
2101         lnet_ni_unlink_locked(ni);
2102         lnet_incr_dlc_seq();
2103         lnet_net_unlock(LNET_LOCK_EX);
2104
2105         /* clear messages for this NI on the lazy portal */
2106         for (i = 0; i < the_lnet.ln_nportals; i++)
2107                 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2108
2109         lnet_net_lock(LNET_LOCK_EX);
2110         lnet_clear_zombies_nis_locked(net);
2111         lnet_net_unlock(LNET_LOCK_EX);
2112 }
2113
2114 static void
2115 lnet_shutdown_lndnet(struct lnet_net *net)
2116 {
2117         struct lnet_ni *ni;
2118
2119         lnet_net_lock(LNET_LOCK_EX);
2120
2121         list_del_init(&net->net_list);
2122
2123         while (!list_empty(&net->net_ni_list)) {
2124                 ni = list_entry(net->net_ni_list.next,
2125                                 struct lnet_ni, ni_netlist);
2126                 lnet_net_unlock(LNET_LOCK_EX);
2127                 lnet_shutdown_lndni(ni);
2128                 lnet_net_lock(LNET_LOCK_EX);
2129         }
2130
2131         lnet_net_unlock(LNET_LOCK_EX);
2132
2133         /* Do peer table cleanup for this net */
2134         lnet_peer_tables_cleanup(net);
2135
2136         lnet_net_free(net);
2137 }
2138
2139 static void
2140 lnet_shutdown_lndnets(void)
2141 {
2142         struct lnet_net *net;
2143         LIST_HEAD(resend);
2144         struct lnet_msg *msg, *tmp;
2145
2146         /* NB called holding the global mutex */
2147
2148         /* All quiet on the API front */
2149         LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2150         LASSERT(the_lnet.ln_refcount == 0);
2151
2152         lnet_net_lock(LNET_LOCK_EX);
2153         the_lnet.ln_state = LNET_STATE_STOPPING;
2154
2155         /*
2156          * move the nets to the zombie list to avoid them being
2157          * picked up for new work. LONET is also included in the
2158          * Nets that will be moved to the zombie list
2159          */
2160         list_splice_init(&the_lnet.ln_nets, &the_lnet.ln_net_zombie);
2161
2162         /* Drop the cached loopback Net. */
2163         if (the_lnet.ln_loni != NULL) {
2164                 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2165                 the_lnet.ln_loni = NULL;
2166         }
2167         lnet_net_unlock(LNET_LOCK_EX);
2168
2169         /* iterate through the net zombie list and delete each net */
2170         while (!list_empty(&the_lnet.ln_net_zombie)) {
2171                 net = list_entry(the_lnet.ln_net_zombie.next,
2172                                  struct lnet_net, net_list);
2173                 lnet_shutdown_lndnet(net);
2174         }
2175
2176         spin_lock(&the_lnet.ln_msg_resend_lock);
2177         list_splice(&the_lnet.ln_msg_resend, &resend);
2178         spin_unlock(&the_lnet.ln_msg_resend_lock);
2179
2180         list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2181                 list_del_init(&msg->msg_list);
2182                 msg->msg_no_resend = true;
2183                 lnet_finalize(msg, -ECANCELED);
2184         }
2185
2186         lnet_net_lock(LNET_LOCK_EX);
2187         the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2188         lnet_net_unlock(LNET_LOCK_EX);
2189 }
2190
2191 static int
2192 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2193 {
2194         int                     rc = -EINVAL;
2195         struct lnet_tx_queue    *tq;
2196         int                     i;
2197         struct lnet_net         *net = ni->ni_net;
2198
2199         mutex_lock(&the_lnet.ln_lnd_mutex);
2200
2201         if (tun) {
2202                 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2203                 ni->ni_lnd_tunables_set = true;
2204         }
2205
2206         rc = (net->net_lnd->lnd_startup)(ni);
2207
2208         mutex_unlock(&the_lnet.ln_lnd_mutex);
2209
2210         if (rc != 0) {
2211                 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2212                                    rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2213                 goto failed0;
2214         }
2215
2216         lnet_ni_lock(ni);
2217         ni->ni_state = LNET_NI_STATE_ACTIVE;
2218         lnet_ni_unlock(ni);
2219
2220         /* We keep a reference on the loopback net through the loopback NI */
2221         if (net->net_lnd->lnd_type == LOLND) {
2222                 lnet_ni_addref(ni);
2223                 LASSERT(the_lnet.ln_loni == NULL);
2224                 the_lnet.ln_loni = ni;
2225                 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2226                 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2227                 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2228                 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2229                 return 0;
2230         }
2231
2232         if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2233             ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2234                 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2235                                    libcfs_lnd2str(net->net_lnd->lnd_type),
2236                                    ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2237                                         "" : "per-peer ");
2238                 /* shutdown the NI since if we get here then it must've already
2239                  * been started
2240                  */
2241                 lnet_shutdown_lndni(ni);
2242                 return -EINVAL;
2243         }
2244
2245         cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2246                 tq->tq_credits_min =
2247                 tq->tq_credits_max =
2248                 tq->tq_credits = lnet_ni_tq_credits(ni);
2249         }
2250
2251         atomic_set(&ni->ni_tx_credits,
2252                    lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2253         atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2254
2255         CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2256                 libcfs_nid2str(ni->ni_nid),
2257                 ni->ni_net->net_tunables.lct_peer_tx_credits,
2258                 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2259                 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2260                 ni->ni_net->net_tunables.lct_peer_timeout);
2261
2262         return 0;
2263 failed0:
2264         lnet_ni_free(ni);
2265         return rc;
2266 }
2267
2268 static int
2269 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2270 {
2271         struct lnet_ni *ni;
2272         struct lnet_net *net_l = NULL;
2273         LIST_HEAD(local_ni_list);
2274         int rc;
2275         int ni_count = 0;
2276         __u32 lnd_type;
2277         const struct lnet_lnd  *lnd;
2278         int peer_timeout =
2279                 net->net_tunables.lct_peer_timeout;
2280         int maxtxcredits =
2281                 net->net_tunables.lct_max_tx_credits;
2282         int peerrtrcredits =
2283                 net->net_tunables.lct_peer_rtr_credits;
2284
2285         /*
2286          * make sure that this net is unique. If it isn't then
2287          * we are adding interfaces to an already existing network, and
2288          * 'net' is just a convenient way to pass in the list.
2289          * if it is unique we need to find the LND and load it if
2290          * necessary.
2291          */
2292         if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2293                 lnd_type = LNET_NETTYP(net->net_id);
2294
2295                 mutex_lock(&the_lnet.ln_lnd_mutex);
2296                 lnd = lnet_find_lnd_by_type(lnd_type);
2297
2298                 if (lnd == NULL) {
2299                         mutex_unlock(&the_lnet.ln_lnd_mutex);
2300                         rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2301                         mutex_lock(&the_lnet.ln_lnd_mutex);
2302
2303                         lnd = lnet_find_lnd_by_type(lnd_type);
2304                         if (lnd == NULL) {
2305                                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2306                                 CERROR("Can't load LND %s, module %s, rc=%d\n",
2307                                 libcfs_lnd2str(lnd_type),
2308                                 libcfs_lnd2modname(lnd_type), rc);
2309 #ifndef HAVE_MODULE_LOADING_SUPPORT
2310                                 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2311                                                 "compiled with kernel module "
2312                                                 "loading support.");
2313 #endif
2314                                 rc = -EINVAL;
2315                                 goto failed0;
2316                         }
2317                 }
2318
2319                 net->net_lnd = lnd;
2320
2321                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2322
2323                 net_l = net;
2324         }
2325
2326         /*
2327          * net_l: if the network being added is unique then net_l
2328          *        will point to that network
2329          *        if the network being added is not unique then
2330          *        net_l points to the existing network.
2331          *
2332          * When we enter the loop below, we'll pick NIs off he
2333          * network beign added and start them up, then add them to
2334          * a local ni list. Once we've successfully started all
2335          * the NIs then we join the local NI list (of started up
2336          * networks) with the net_l->net_ni_list, which should
2337          * point to the correct network to add the new ni list to
2338          *
2339          * If any of the new NIs fail to start up, then we want to
2340          * iterate through the local ni list, which should include
2341          * any NIs which were successfully started up, and shut
2342          * them down.
2343          *
2344          * After than we want to delete the network being added,
2345          * to avoid a memory leak.
2346          */
2347
2348         /*
2349          * When a network uses TCP bonding then all its interfaces
2350          * must be specified when the network is first defined: the
2351          * TCP bonding code doesn't allow for interfaces to be added
2352          * or removed.
2353          */
2354         if (net_l != net && net_l != NULL && use_tcp_bonding &&
2355             LNET_NETTYP(net_l->net_id) == SOCKLND) {
2356                 rc = -EINVAL;
2357                 goto failed0;
2358         }
2359
2360         while (!list_empty(&net->net_ni_added)) {
2361                 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2362                                 ni_netlist);
2363                 list_del_init(&ni->ni_netlist);
2364
2365                 /* make sure that the the NI we're about to start
2366                  * up is actually unique. if it's not fail. */
2367                 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2368                                         ni->ni_interfaces[0])) {
2369                         rc = -EEXIST;
2370                         goto failed1;
2371                 }
2372
2373                 /* adjust the pointer the parent network, just in case it
2374                  * the net is a duplicate */
2375                 ni->ni_net = net_l;
2376
2377                 rc = lnet_startup_lndni(ni, tun);
2378
2379                 if (rc < 0)
2380                         goto failed1;
2381
2382                 lnet_ni_addref(ni);
2383                 list_add_tail(&ni->ni_netlist, &local_ni_list);
2384
2385                 ni_count++;
2386         }
2387
2388         lnet_net_lock(LNET_LOCK_EX);
2389         list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2390         lnet_incr_dlc_seq();
2391         lnet_net_unlock(LNET_LOCK_EX);
2392
2393         /* if the network is not unique then we don't want to keep
2394          * it around after we're done. Free it. Otherwise add that
2395          * net to the global the_lnet.ln_nets */
2396         if (net_l != net && net_l != NULL) {
2397                 /*
2398                  * TODO - note. currently the tunables can not be updated
2399                  * once added
2400                  */
2401                 lnet_net_free(net);
2402         } else {
2403                 /*
2404                  * restore tunables after it has been overwitten by the
2405                  * lnd
2406                  */
2407                 if (peer_timeout != -1)
2408                         net->net_tunables.lct_peer_timeout = peer_timeout;
2409                 if (maxtxcredits != -1)
2410                         net->net_tunables.lct_max_tx_credits = maxtxcredits;
2411                 if (peerrtrcredits != -1)
2412                         net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2413
2414                 lnet_net_lock(LNET_LOCK_EX);
2415                 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2416                 lnet_net_unlock(LNET_LOCK_EX);
2417         }
2418
2419         /* update net count */
2420         lnet_current_net_count = lnet_get_net_count();
2421
2422         return ni_count;
2423
2424 failed1:
2425         /*
2426          * shutdown the new NIs that are being started up
2427          * free the NET being started
2428          */
2429         while (!list_empty(&local_ni_list)) {
2430                 ni = list_entry(local_ni_list.next, struct lnet_ni,
2431                                 ni_netlist);
2432
2433                 lnet_shutdown_lndni(ni);
2434         }
2435
2436 failed0:
2437         lnet_net_free(net);
2438
2439         return rc;
2440 }
2441
2442 static int
2443 lnet_startup_lndnets(struct list_head *netlist)
2444 {
2445         struct lnet_net         *net;
2446         int                     rc;
2447         int                     ni_count = 0;
2448
2449         /*
2450          * Change to running state before bringing up the LNDs. This
2451          * allows lnet_shutdown_lndnets() to assert that we've passed
2452          * through here.
2453          */
2454         lnet_net_lock(LNET_LOCK_EX);
2455         the_lnet.ln_state = LNET_STATE_RUNNING;
2456         lnet_net_unlock(LNET_LOCK_EX);
2457
2458         while (!list_empty(netlist)) {
2459                 net = list_entry(netlist->next, struct lnet_net, net_list);
2460                 list_del_init(&net->net_list);
2461
2462                 rc = lnet_startup_lndnet(net, NULL);
2463
2464                 if (rc < 0)
2465                         goto failed;
2466
2467                 ni_count += rc;
2468         }
2469
2470         return ni_count;
2471 failed:
2472         lnet_shutdown_lndnets();
2473
2474         return rc;
2475 }
2476
2477 /**
2478  * Initialize LNet library.
2479  *
2480  * Automatically called at module loading time. Caller has to call
2481  * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2482  * latter returned 0. It must be called exactly once.
2483  *
2484  * \retval 0 on success
2485  * \retval -ve on failures.
2486  */
2487 int lnet_lib_init(void)
2488 {
2489         int rc;
2490
2491         lnet_assert_wire_constants();
2492
2493         /* refer to global cfs_cpt_table for now */
2494         the_lnet.ln_cpt_table   = cfs_cpt_table;
2495         the_lnet.ln_cpt_number  = cfs_cpt_number(cfs_cpt_table);
2496
2497         LASSERT(the_lnet.ln_cpt_number > 0);
2498         if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2499                 /* we are under risk of consuming all lh_cookie */
2500                 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2501                        "please change setting of CPT-table and retry\n",
2502                        the_lnet.ln_cpt_number, LNET_CPT_MAX);
2503                 return -E2BIG;
2504         }
2505
2506         while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2507                 the_lnet.ln_cpt_bits++;
2508
2509         rc = lnet_create_locks();
2510         if (rc != 0) {
2511                 CERROR("Can't create LNet global locks: %d\n", rc);
2512                 return rc;
2513         }
2514
2515         the_lnet.ln_refcount = 0;
2516         INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2517         INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2518
2519         /* The hash table size is the number of bits it takes to express the set
2520          * ln_num_routes, minus 1 (better to under estimate than over so we
2521          * don't waste memory). */
2522         if (rnet_htable_size <= 0)
2523                 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2524         else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2525                 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2526         the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2527                                            order_base_2(rnet_htable_size) - 1);
2528
2529         /* All LNDs apart from the LOLND are in separate modules.  They
2530          * register themselves when their module loads, and unregister
2531          * themselves when their module is unloaded. */
2532         lnet_register_lnd(&the_lolnd);
2533         return 0;
2534 }
2535
2536 /**
2537  * Finalize LNet library.
2538  *
2539  * \pre lnet_lib_init() called with success.
2540  * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2541  *
2542  * As this happens at module-unload, all lnds must already be unloaded,
2543  * so they must already be unregistered.
2544  */
2545 void lnet_lib_exit(void)
2546 {
2547         int i;
2548
2549         LASSERT(the_lnet.ln_refcount == 0);
2550         lnet_unregister_lnd(&the_lolnd);
2551         for (i = 0; i < NUM_LNDS; i++)
2552                 LASSERT(!the_lnet.ln_lnds[i]);
2553         lnet_destroy_locks();
2554 }
2555
2556 /**
2557  * Set LNet PID and start LNet interfaces, routing, and forwarding.
2558  *
2559  * Users must call this function at least once before any other functions.
2560  * For each successful call there must be a corresponding call to
2561  * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2562  * ignored.
2563  *
2564  * The PID used by LNet may be different from the one requested.
2565  * See LNetGetId().
2566  *
2567  * \param requested_pid PID requested by the caller.
2568  *
2569  * \return >= 0 on success, and < 0 error code on failures.
2570  */
2571 int
2572 LNetNIInit(lnet_pid_t requested_pid)
2573 {
2574         int                     im_a_router = 0;
2575         int                     rc;
2576         int                     ni_count;
2577         struct lnet_ping_buffer *pbuf;
2578         struct lnet_handle_md   ping_mdh;
2579         LIST_HEAD(net_head);
2580         struct lnet_net         *net;
2581
2582         mutex_lock(&the_lnet.ln_api_mutex);
2583
2584         CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2585
2586         if (the_lnet.ln_refcount > 0) {
2587                 rc = the_lnet.ln_refcount++;
2588                 mutex_unlock(&the_lnet.ln_api_mutex);
2589                 return rc;
2590         }
2591
2592         rc = lnet_prepare(requested_pid);
2593         if (rc != 0) {
2594                 mutex_unlock(&the_lnet.ln_api_mutex);
2595                 return rc;
2596         }
2597
2598         /* create a network for Loopback network */
2599         net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2600         if (net == NULL) {
2601                 rc = -ENOMEM;
2602                 goto err_empty_list;
2603         }
2604
2605         /* Add in the loopback NI */
2606         if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2607                 rc = -ENOMEM;
2608                 goto err_empty_list;
2609         }
2610
2611         /* If LNet is being initialized via DLC it is possible
2612          * that the user requests not to load module parameters (ones which
2613          * are supported by DLC) on initialization.  Therefore, make sure not
2614          * to load networks, routes and forwarding from module parameters
2615          * in this case.  On cleanup in case of failure only clean up
2616          * routes if it has been loaded */
2617         if (!the_lnet.ln_nis_from_mod_params) {
2618                 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
2619                                          use_tcp_bonding);
2620                 if (rc < 0)
2621                         goto err_empty_list;
2622         }
2623
2624         ni_count = lnet_startup_lndnets(&net_head);
2625         if (ni_count < 0) {
2626                 rc = ni_count;
2627                 goto err_empty_list;
2628         }
2629
2630         if (!the_lnet.ln_nis_from_mod_params) {
2631                 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2632                 if (rc != 0)
2633                         goto err_shutdown_lndnis;
2634
2635                 rc = lnet_rtrpools_alloc(im_a_router);
2636                 if (rc != 0)
2637                         goto err_destroy_routes;
2638         }
2639
2640         rc = lnet_acceptor_start();
2641         if (rc != 0)
2642                 goto err_destroy_routes;
2643
2644         the_lnet.ln_refcount = 1;
2645         /* Now I may use my own API functions... */
2646
2647         rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2648         if (rc != 0)
2649                 goto err_acceptor_stop;
2650
2651         lnet_ping_target_update(pbuf, ping_mdh);
2652
2653         rc = LNetEQAlloc(0, lnet_mt_event_handler, &the_lnet.ln_mt_eqh);
2654         if (rc != 0) {
2655                 CERROR("Can't allocate monitor thread EQ: %d\n", rc);
2656                 goto err_stop_ping;
2657         }
2658
2659         rc = lnet_push_target_init();
2660         if (rc != 0)
2661                 goto err_stop_ping;
2662
2663         rc = lnet_peer_discovery_start();
2664         if (rc != 0)
2665                 goto err_destroy_push_target;
2666
2667         rc = lnet_monitor_thr_start();
2668         if (rc != 0)
2669                 goto err_stop_discovery_thr;
2670
2671         lnet_fault_init();
2672         lnet_router_debugfs_init();
2673
2674         mutex_unlock(&the_lnet.ln_api_mutex);
2675
2676         complete_all(&the_lnet.ln_started);
2677
2678         /* wait for all routers to start */
2679         lnet_wait_router_start();
2680
2681         return 0;
2682
2683 err_stop_discovery_thr:
2684         lnet_peer_discovery_stop();
2685 err_destroy_push_target:
2686         lnet_push_target_fini();
2687 err_stop_ping:
2688         lnet_ping_target_fini();
2689 err_acceptor_stop:
2690         the_lnet.ln_refcount = 0;
2691         lnet_acceptor_stop();
2692 err_destroy_routes:
2693         if (!the_lnet.ln_nis_from_mod_params)
2694                 lnet_destroy_routes();
2695 err_shutdown_lndnis:
2696         lnet_shutdown_lndnets();
2697 err_empty_list:
2698         lnet_unprepare();
2699         LASSERT(rc < 0);
2700         mutex_unlock(&the_lnet.ln_api_mutex);
2701         while (!list_empty(&net_head)) {
2702                 struct lnet_net *net;
2703
2704                 net = list_entry(net_head.next, struct lnet_net, net_list);
2705                 list_del_init(&net->net_list);
2706                 lnet_net_free(net);
2707         }
2708         return rc;
2709 }
2710 EXPORT_SYMBOL(LNetNIInit);
2711
2712 /**
2713  * Stop LNet interfaces, routing, and forwarding.
2714  *
2715  * Users must call this function once for each successful call to LNetNIInit().
2716  * Once the LNetNIFini() operation has been started, the results of pending
2717  * API operations are undefined.
2718  *
2719  * \return always 0 for current implementation.
2720  */
2721 int
2722 LNetNIFini(void)
2723 {
2724         mutex_lock(&the_lnet.ln_api_mutex);
2725
2726         LASSERT(the_lnet.ln_refcount > 0);
2727
2728         if (the_lnet.ln_refcount != 1) {
2729                 the_lnet.ln_refcount--;
2730         } else {
2731                 LASSERT(!the_lnet.ln_niinit_self);
2732
2733                 lnet_fault_fini();
2734
2735                 lnet_router_debugfs_fini();
2736                 lnet_monitor_thr_stop();
2737                 lnet_peer_discovery_stop();
2738                 lnet_push_target_fini();
2739                 lnet_ping_target_fini();
2740
2741                 /* Teardown fns that use my own API functions BEFORE here */
2742                 the_lnet.ln_refcount = 0;
2743
2744                 lnet_acceptor_stop();
2745                 lnet_destroy_routes();
2746                 lnet_shutdown_lndnets();
2747                 lnet_unprepare();
2748         }
2749
2750         mutex_unlock(&the_lnet.ln_api_mutex);
2751         return 0;
2752 }
2753 EXPORT_SYMBOL(LNetNIFini);
2754
2755 /**
2756  * Grabs the ni data from the ni structure and fills the out
2757  * parameters
2758  *
2759  * \param[in] ni network        interface structure
2760  * \param[out] cfg_ni           NI config information
2761  * \param[out] tun              network and LND tunables
2762  */
2763 static void
2764 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2765                    struct lnet_ioctl_config_lnd_tunables *tun,
2766                    struct lnet_ioctl_element_stats *stats,
2767                    __u32 tun_size)
2768 {
2769         size_t min_size = 0;
2770         int i;
2771
2772         if (!ni || !cfg_ni || !tun)
2773                 return;
2774
2775         if (ni->ni_interfaces[0] != NULL) {
2776                 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2777                         if (ni->ni_interfaces[i] != NULL) {
2778                                 strncpy(cfg_ni->lic_ni_intf[i],
2779                                         ni->ni_interfaces[i],
2780                                         sizeof(cfg_ni->lic_ni_intf[i]));
2781                         }
2782                 }
2783         }
2784
2785         cfg_ni->lic_nid = ni->ni_nid;
2786         if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2787                 cfg_ni->lic_status = LNET_NI_STATUS_UP;
2788         else
2789                 cfg_ni->lic_status = ni->ni_status->ns_status;
2790         cfg_ni->lic_tcp_bonding = use_tcp_bonding;
2791         cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
2792
2793         memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
2794
2795         if (stats) {
2796                 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
2797                                                        LNET_STATS_TYPE_SEND);
2798                 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
2799                                                        LNET_STATS_TYPE_RECV);
2800                 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
2801                                                        LNET_STATS_TYPE_DROP);
2802         }
2803
2804         /*
2805          * tun->lt_tun will always be present, but in order to be
2806          * backwards compatible, we need to deal with the cases when
2807          * tun->lt_tun is smaller than what the kernel has, because it
2808          * comes from an older version of a userspace program, then we'll
2809          * need to copy as much information as we have available space.
2810          */
2811         min_size = tun_size - sizeof(tun->lt_cmn);
2812         memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
2813
2814         /* copy over the cpts */
2815         if (ni->ni_ncpts == LNET_CPT_NUMBER &&
2816             ni->ni_cpts == NULL)  {
2817                 for (i = 0; i < ni->ni_ncpts; i++)
2818                         cfg_ni->lic_cpts[i] = i;
2819         } else {
2820                 for (i = 0;
2821                      ni->ni_cpts != NULL && i < ni->ni_ncpts &&
2822                      i < LNET_MAX_SHOW_NUM_CPT;
2823                      i++)
2824                         cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2825         }
2826         cfg_ni->lic_ncpts = ni->ni_ncpts;
2827 }
2828
2829 /**
2830  * NOTE: This is a legacy function left in the code to be backwards
2831  * compatible with older userspace programs. It should eventually be
2832  * removed.
2833  *
2834  * Grabs the ni data from the ni structure and fills the out
2835  * parameters
2836  *
2837  * \param[in] ni network        interface structure
2838  * \param[out] config           config information
2839  */
2840 static void
2841 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
2842                          struct lnet_ioctl_config_data *config)
2843 {
2844         struct lnet_ioctl_net_config *net_config;
2845         struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
2846         size_t min_size, tunable_size = 0;
2847         int i;
2848
2849         if (!ni || !config)
2850                 return;
2851
2852         net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
2853         if (!net_config)
2854                 return;
2855
2856         BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
2857                      ARRAY_SIZE(net_config->ni_interfaces));
2858
2859         for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2860                 if (!ni->ni_interfaces[i])
2861                         break;
2862
2863                 strncpy(net_config->ni_interfaces[i],
2864                         ni->ni_interfaces[i],
2865                         sizeof(net_config->ni_interfaces[i]));
2866         }
2867
2868         config->cfg_nid = ni->ni_nid;
2869         config->cfg_config_u.cfg_net.net_peer_timeout =
2870                 ni->ni_net->net_tunables.lct_peer_timeout;
2871         config->cfg_config_u.cfg_net.net_max_tx_credits =
2872                 ni->ni_net->net_tunables.lct_max_tx_credits;
2873         config->cfg_config_u.cfg_net.net_peer_tx_credits =
2874                 ni->ni_net->net_tunables.lct_peer_tx_credits;
2875         config->cfg_config_u.cfg_net.net_peer_rtr_credits =
2876                 ni->ni_net->net_tunables.lct_peer_rtr_credits;
2877
2878         if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2879                 net_config->ni_status = LNET_NI_STATUS_UP;
2880         else
2881                 net_config->ni_status = ni->ni_status->ns_status;
2882
2883         if (ni->ni_cpts) {
2884                 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
2885
2886                 for (i = 0; i < num_cpts; i++)
2887                         net_config->ni_cpts[i] = ni->ni_cpts[i];
2888
2889                 config->cfg_ncpts = num_cpts;
2890         }
2891
2892         /*
2893          * See if user land tools sent in a newer and larger version
2894          * of struct lnet_tunables than what the kernel uses.
2895          */
2896         min_size = sizeof(*config) + sizeof(*net_config);
2897
2898         if (config->cfg_hdr.ioc_len > min_size)
2899                 tunable_size = config->cfg_hdr.ioc_len - min_size;
2900
2901         /* Don't copy too much data to user space */
2902         min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
2903         lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
2904
2905         if (lnd_cfg && min_size) {
2906                 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
2907                 config->cfg_config_u.cfg_net.net_interface_count = 1;
2908
2909                 /* Tell user land that kernel side has less data */
2910                 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
2911                         min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
2912                         config->cfg_hdr.ioc_len -= min_size;
2913                 }
2914         }
2915 }
2916
2917 struct lnet_ni *
2918 lnet_get_ni_idx_locked(int idx)
2919 {
2920         struct lnet_ni          *ni;
2921         struct lnet_net         *net;
2922
2923         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2924                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2925                         if (idx-- == 0)
2926                                 return ni;
2927                 }
2928         }
2929
2930         return NULL;
2931 }
2932
2933 struct lnet_ni *
2934 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2935 {
2936         struct lnet_ni          *ni;
2937         struct lnet_net         *net = mynet;
2938
2939         /*
2940          * It is possible that the net has been cleaned out while there is
2941          * a message being sent. This function accessed the net without
2942          * checking if the list is empty
2943          */
2944         if (prev == NULL) {
2945                 if (net == NULL)
2946                         net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2947                                         net_list);
2948                 if (list_empty(&net->net_ni_list))
2949                         return NULL;
2950                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2951                                 ni_netlist);
2952
2953                 return ni;
2954         }
2955
2956         if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
2957                 /* if you reached the end of the ni list and the net is
2958                  * specified, then there are no more nis in that net */
2959                 if (net != NULL)
2960                         return NULL;
2961
2962                 /* we reached the end of this net ni list. move to the
2963                  * next net */
2964                 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
2965                         /* no more nets and no more NIs. */
2966                         return NULL;
2967
2968                 /* get the next net */
2969                 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
2970                                  net_list);
2971                 if (list_empty(&net->net_ni_list))
2972                         return NULL;
2973                 /* get the ni on it */
2974                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2975                                 ni_netlist);
2976
2977                 return ni;
2978         }
2979
2980         if (list_empty(&prev->ni_netlist))
2981                 return NULL;
2982
2983         /* there are more nis left */
2984         ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
2985
2986         return ni;
2987 }
2988
2989 int
2990 lnet_get_net_config(struct lnet_ioctl_config_data *config)
2991 {
2992         struct lnet_ni *ni;
2993         int cpt;
2994         int rc = -ENOENT;
2995         int idx = config->cfg_count;
2996
2997         cpt = lnet_net_lock_current();
2998
2999         ni = lnet_get_ni_idx_locked(idx);
3000
3001         if (ni != NULL) {
3002                 rc = 0;
3003                 lnet_ni_lock(ni);
3004                 lnet_fill_ni_info_legacy(ni, config);
3005                 lnet_ni_unlock(ni);
3006         }
3007
3008         lnet_net_unlock(cpt);
3009         return rc;
3010 }
3011
3012 int
3013 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3014                    struct lnet_ioctl_config_lnd_tunables *tun,
3015                    struct lnet_ioctl_element_stats *stats,
3016                    __u32 tun_size)
3017 {
3018         struct lnet_ni          *ni;
3019         int                     cpt;
3020         int                     rc = -ENOENT;
3021
3022         if (!cfg_ni || !tun || !stats)
3023                 return -EINVAL;
3024
3025         cpt = lnet_net_lock_current();
3026
3027         ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3028
3029         if (ni) {
3030                 rc = 0;
3031                 lnet_ni_lock(ni);
3032                 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3033                 lnet_ni_unlock(ni);
3034         }
3035
3036         lnet_net_unlock(cpt);
3037         return rc;
3038 }
3039
3040 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3041 {
3042         struct lnet_ni *ni;
3043         int cpt;
3044         int rc = -ENOENT;
3045
3046         if (!msg_stats)
3047                 return -EINVAL;
3048
3049         cpt = lnet_net_lock_current();
3050
3051         ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3052
3053         if (ni) {
3054                 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3055                 rc = 0;
3056         }
3057
3058         lnet_net_unlock(cpt);
3059
3060         return rc;
3061 }
3062
3063 static int lnet_add_net_common(struct lnet_net *net,
3064                                struct lnet_ioctl_config_lnd_tunables *tun)
3065 {
3066         __u32                   net_id;
3067         struct lnet_ping_buffer *pbuf;
3068         struct lnet_handle_md   ping_mdh;
3069         int                     rc;
3070         struct lnet_remotenet *rnet;
3071         int                     net_ni_count;
3072         int                     num_acceptor_nets;
3073
3074         lnet_net_lock(LNET_LOCK_EX);
3075         rnet = lnet_find_rnet_locked(net->net_id);
3076         lnet_net_unlock(LNET_LOCK_EX);
3077         /*
3078          * make sure that the net added doesn't invalidate the current
3079          * configuration LNet is keeping
3080          */
3081         if (rnet) {
3082                 CERROR("Adding net %s will invalidate routing configuration\n",
3083                        libcfs_net2str(net->net_id));
3084                 lnet_net_free(net);
3085                 return -EUSERS;
3086         }
3087
3088         /*
3089          * make sure you calculate the correct number of slots in the ping
3090          * buffer. Since the ping info is a flattened list of all the NIs,
3091          * we should allocate enough slots to accomodate the number of NIs
3092          * which will be added.
3093          *
3094          * since ni hasn't been configured yet, use
3095          * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3096          */
3097         net_ni_count = lnet_get_net_ni_count_pre(net);
3098
3099         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3100                                     net_ni_count + lnet_get_ni_count(),
3101                                     false);
3102         if (rc < 0) {
3103                 lnet_net_free(net);
3104                 return rc;
3105         }
3106
3107         if (tun)
3108                 memcpy(&net->net_tunables,
3109                        &tun->lt_cmn, sizeof(net->net_tunables));
3110         else
3111                 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3112
3113         /*
3114          * before starting this network get a count of the current TCP
3115          * networks which require the acceptor thread running. If that
3116          * count is == 0 before we start up this network, then we'd want to
3117          * start up the acceptor thread after starting up this network
3118          */
3119         num_acceptor_nets = lnet_count_acceptor_nets();
3120
3121         net_id = net->net_id;
3122
3123         rc = lnet_startup_lndnet(net,
3124                                  (tun) ? &tun->lt_tun : NULL);
3125         if (rc < 0)
3126                 goto failed;
3127
3128         lnet_net_lock(LNET_LOCK_EX);
3129         net = lnet_get_net_locked(net_id);
3130         lnet_net_unlock(LNET_LOCK_EX);
3131
3132         LASSERT(net);
3133
3134         /*
3135          * Start the acceptor thread if this is the first network
3136          * being added that requires the thread.
3137          */
3138         if (net->net_lnd->lnd_accept && num_acceptor_nets == 0) {
3139                 rc = lnet_acceptor_start();
3140                 if (rc < 0) {
3141                         /* shutdown the net that we just started */
3142                         CERROR("Failed to start up acceptor thread\n");
3143                         lnet_shutdown_lndnet(net);
3144                         goto failed;
3145                 }
3146         }
3147
3148         lnet_net_lock(LNET_LOCK_EX);
3149         lnet_peer_net_added(net);
3150         lnet_net_unlock(LNET_LOCK_EX);
3151
3152         lnet_ping_target_update(pbuf, ping_mdh);
3153
3154         return 0;
3155
3156 failed:
3157         lnet_ping_md_unlink(pbuf, &ping_mdh);
3158         lnet_ping_buffer_decref(pbuf);
3159         return rc;
3160 }
3161
3162 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3163                                       struct lnet_ioctl_config_lnd_tunables *tun)
3164 {
3165         struct lnet_net *net;
3166         char *nets;
3167         int rc;
3168         LIST_HEAD(net_head);
3169
3170         rc = lnet_parse_ip2nets(&nets, ip2nets);
3171         if (rc < 0)
3172                 return rc;
3173
3174         rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3175         if (rc < 0)
3176                 return rc;
3177
3178         mutex_lock(&the_lnet.ln_api_mutex);
3179         while (!list_empty(&net_head)) {
3180                 net = list_entry(net_head.next, struct lnet_net, net_list);
3181                 list_del_init(&net->net_list);
3182                 rc = lnet_add_net_common(net, tun);
3183                 if (rc < 0)
3184                         goto out;
3185         }
3186
3187 out:
3188         mutex_unlock(&the_lnet.ln_api_mutex);
3189
3190         while (!list_empty(&net_head)) {
3191                 net = list_entry(net_head.next, struct lnet_net, net_list);
3192                 list_del_init(&net->net_list);
3193                 lnet_net_free(net);
3194         }
3195         return rc;
3196 }
3197
3198 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3199 {
3200         struct lnet_net *net;
3201         struct lnet_ni *ni;
3202         struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3203         int rc, i;
3204         __u32 net_id, lnd_type;
3205
3206         /* get the tunables if they are available */
3207         if (conf->lic_cfg_hdr.ioc_len >=
3208             sizeof(*conf) + sizeof(*tun))
3209                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3210                         conf->lic_bulk;
3211
3212         /* handle legacy ip2nets from DLC */
3213         if (conf->lic_legacy_ip2nets[0] != '\0')
3214                 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3215                                                   tun);
3216
3217         net_id = LNET_NIDNET(conf->lic_nid);
3218         lnd_type = LNET_NETTYP(net_id);
3219
3220         if (!libcfs_isknown_lnd(lnd_type)) {
3221                 CERROR("No valid net and lnd information provided\n");
3222                 return -EINVAL;
3223         }
3224
3225         net = lnet_net_alloc(net_id, NULL);
3226         if (!net)
3227                 return -ENOMEM;
3228
3229         for (i = 0; i < conf->lic_ncpts; i++) {
3230                 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3231                         return -EINVAL;
3232         }
3233
3234         ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3235                                        conf->lic_ni_intf[0]);
3236         if (!ni)
3237                 return -ENOMEM;
3238
3239         mutex_lock(&the_lnet.ln_api_mutex);
3240
3241         rc = lnet_add_net_common(net, tun);
3242
3243         mutex_unlock(&the_lnet.ln_api_mutex);
3244
3245         return rc;
3246 }
3247
3248 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3249 {
3250         struct lnet_net  *net;
3251         struct lnet_ni *ni;
3252         __u32 net_id = LNET_NIDNET(conf->lic_nid);
3253         struct lnet_ping_buffer *pbuf;
3254         struct lnet_handle_md  ping_mdh;
3255         int               rc;
3256         int               net_count;
3257         __u32             addr;
3258
3259         /* don't allow userspace to shutdown the LOLND */
3260         if (LNET_NETTYP(net_id) == LOLND)
3261                 return -EINVAL;
3262
3263         mutex_lock(&the_lnet.ln_api_mutex);
3264
3265         lnet_net_lock(0);
3266
3267         net = lnet_get_net_locked(net_id);
3268         if (!net) {
3269                 CERROR("net %s not found\n",
3270                        libcfs_net2str(net_id));
3271                 rc = -ENOENT;
3272                 goto unlock_net;
3273         }
3274
3275         addr = LNET_NIDADDR(conf->lic_nid);
3276         if (addr == 0) {
3277                 /* remove the entire net */
3278                 net_count = lnet_get_net_ni_count_locked(net);
3279
3280                 lnet_net_unlock(0);
3281
3282                 /* create and link a new ping info, before removing the old one */
3283                 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3284                                         lnet_get_ni_count() - net_count,
3285                                         false);
3286                 if (rc != 0)
3287                         goto unlock_api_mutex;
3288
3289                 lnet_shutdown_lndnet(net);
3290
3291                 if (lnet_count_acceptor_nets() == 0)
3292                         lnet_acceptor_stop();
3293
3294                 lnet_ping_target_update(pbuf, ping_mdh);
3295
3296                 goto unlock_api_mutex;
3297         }
3298
3299         ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3300         if (!ni) {
3301                 CERROR("nid %s not found\n",
3302                        libcfs_nid2str(conf->lic_nid));
3303                 rc = -ENOENT;
3304                 goto unlock_net;
3305         }
3306
3307         net_count = lnet_get_net_ni_count_locked(net);
3308
3309         lnet_net_unlock(0);
3310
3311         /* create and link a new ping info, before removing the old one */
3312         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3313                                   lnet_get_ni_count() - 1, false);
3314         if (rc != 0)
3315                 goto unlock_api_mutex;
3316
3317         lnet_shutdown_lndni(ni);
3318
3319         if (lnet_count_acceptor_nets() == 0)
3320                 lnet_acceptor_stop();
3321
3322         lnet_ping_target_update(pbuf, ping_mdh);
3323
3324         /* check if the net is empty and remove it if it is */
3325         if (net_count == 1)
3326                 lnet_shutdown_lndnet(net);
3327
3328         goto unlock_api_mutex;
3329
3330 unlock_net:
3331         lnet_net_unlock(0);
3332 unlock_api_mutex:
3333         mutex_unlock(&the_lnet.ln_api_mutex);
3334
3335         return rc;
3336 }
3337
3338 /*
3339  * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3340  * They are only expected to be called for unique networks.
3341  * That can be as a result of older DLC library
3342  * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3343  */
3344 int
3345 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3346 {
3347         struct lnet_net *net;
3348         LIST_HEAD(net_head);
3349         int rc;
3350         struct lnet_ioctl_config_lnd_tunables tun;
3351         char *nets = conf->cfg_config_u.cfg_net.net_intf;
3352
3353         /* Create a net/ni structures for the network string */
3354         rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3355         if (rc <= 0)
3356                 return rc == 0 ? -EINVAL : rc;
3357
3358         mutex_lock(&the_lnet.ln_api_mutex);
3359
3360         if (rc > 1) {
3361                 rc = -EINVAL; /* only add one network per call */
3362                 goto out_unlock_clean;
3363         }
3364
3365         net = list_entry(net_head.next, struct lnet_net, net_list);
3366         list_del_init(&net->net_list);
3367
3368         LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3369
3370         memset(&tun, 0, sizeof(tun));
3371
3372         tun.lt_cmn.lct_peer_timeout =
3373           conf->cfg_config_u.cfg_net.net_peer_timeout;
3374         tun.lt_cmn.lct_peer_tx_credits =
3375           conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3376         tun.lt_cmn.lct_peer_rtr_credits =
3377           conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3378         tun.lt_cmn.lct_max_tx_credits =
3379           conf->cfg_config_u.cfg_net.net_max_tx_credits;
3380
3381         rc = lnet_add_net_common(net, &tun);
3382
3383 out_unlock_clean:
3384         mutex_unlock(&the_lnet.ln_api_mutex);
3385         while (!list_empty(&net_head)) {
3386                 /* net_head list is empty in success case */
3387                 net = list_entry(net_head.next, struct lnet_net, net_list);
3388                 list_del_init(&net->net_list);
3389                 lnet_net_free(net);
3390         }
3391         return rc;
3392 }
3393
3394 int
3395 lnet_dyn_del_net(__u32 net_id)
3396 {
3397         struct lnet_net  *net;
3398         struct lnet_ping_buffer *pbuf;
3399         struct lnet_handle_md ping_mdh;
3400         int               rc;
3401         int               net_ni_count;
3402
3403         /* don't allow userspace to shutdown the LOLND */
3404         if (LNET_NETTYP(net_id) == LOLND)
3405                 return -EINVAL;
3406
3407         mutex_lock(&the_lnet.ln_api_mutex);
3408
3409         lnet_net_lock(0);
3410
3411         net = lnet_get_net_locked(net_id);
3412         if (net == NULL) {
3413                 lnet_net_unlock(0);
3414                 rc = -EINVAL;
3415                 goto out;
3416         }
3417
3418         net_ni_count = lnet_get_net_ni_count_locked(net);
3419
3420         lnet_net_unlock(0);
3421
3422         /* create and link a new ping info, before removing the old one */
3423         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3424                                     lnet_get_ni_count() - net_ni_count, false);
3425         if (rc != 0)
3426                 goto out;
3427
3428         lnet_shutdown_lndnet(net);
3429
3430         if (lnet_count_acceptor_nets() == 0)
3431                 lnet_acceptor_stop();
3432
3433         lnet_ping_target_update(pbuf, ping_mdh);
3434
3435 out:
3436         mutex_unlock(&the_lnet.ln_api_mutex);
3437
3438         return rc;
3439 }
3440
3441 void lnet_incr_dlc_seq(void)
3442 {
3443         atomic_inc(&lnet_dlc_seq_no);
3444 }
3445
3446 __u32 lnet_get_dlc_seq_locked(void)
3447 {
3448         return atomic_read(&lnet_dlc_seq_no);
3449 }
3450
3451 static void
3452 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3453 {
3454         struct lnet_net *net;
3455         struct lnet_ni *ni;
3456
3457         lnet_net_lock(LNET_LOCK_EX);
3458         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3459                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3460                         if (ni->ni_nid == nid || all) {
3461                                 atomic_set(&ni->ni_healthv, value);
3462                                 if (list_empty(&ni->ni_recovery) &&
3463                                     value < LNET_MAX_HEALTH_VALUE) {
3464                                         CERROR("manually adding local NI %s to recovery\n",
3465                                                libcfs_nid2str(ni->ni_nid));
3466                                         list_add_tail(&ni->ni_recovery,
3467                                                       &the_lnet.ln_mt_localNIRecovq);
3468                                         lnet_ni_addref_locked(ni, 0);
3469                                 }
3470                                 if (!all) {
3471                                         lnet_net_unlock(LNET_LOCK_EX);
3472                                         return;
3473                                 }
3474                         }
3475                 }
3476         }
3477         lnet_net_unlock(LNET_LOCK_EX);
3478 }
3479
3480 static int
3481 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3482 {
3483         int cpt, rc = 0;
3484         struct lnet_ni *ni;
3485         lnet_nid_t nid = stats->hlni_nid;
3486
3487         cpt = lnet_net_lock_current();
3488         ni = lnet_nid2ni_locked(nid, cpt);
3489
3490         if (!ni) {
3491                 rc = -ENOENT;
3492                 goto unlock;
3493         }
3494
3495         stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3496         stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3497         stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3498         stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3499         stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3500         stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3501         stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3502
3503 unlock:
3504         lnet_net_unlock(cpt);
3505
3506         return rc;
3507 }
3508
3509 static int
3510 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3511 {
3512         struct lnet_ni *ni;
3513         int i = 0;
3514
3515         lnet_net_lock(LNET_LOCK_EX);
3516         list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3517                 list->rlst_nid_array[i] = ni->ni_nid;
3518                 i++;
3519                 if (i >= LNET_MAX_SHOW_NUM_NID)
3520                         break;
3521         }
3522         lnet_net_unlock(LNET_LOCK_EX);
3523         list->rlst_num_nids = i;
3524
3525         return 0;
3526 }
3527
3528 static int
3529 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3530 {
3531         struct lnet_peer_ni *lpni;
3532         int i = 0;
3533
3534         lnet_net_lock(LNET_LOCK_EX);
3535         list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3536                 list->rlst_nid_array[i] = lpni->lpni_nid;
3537                 i++;
3538                 if (i >= LNET_MAX_SHOW_NUM_NID)
3539                         break;
3540         }
3541         lnet_net_unlock(LNET_LOCK_EX);
3542         list->rlst_num_nids = i;
3543
3544         return 0;
3545 }
3546
3547 /**
3548  * LNet ioctl handler.
3549  *
3550  */
3551 int
3552 LNetCtl(unsigned int cmd, void *arg)
3553 {
3554         struct libcfs_ioctl_data *data = arg;
3555         struct lnet_ioctl_config_data *config;
3556         struct lnet_process_id    id = {0};
3557         struct lnet_ni           *ni;
3558         int                       rc;
3559
3560         BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3561                      sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3562
3563         switch (cmd) {
3564         case IOC_LIBCFS_GET_NI:
3565                 rc = LNetGetId(data->ioc_count, &id);
3566                 data->ioc_nid = id.nid;
3567                 return rc;
3568
3569         case IOC_LIBCFS_FAIL_NID:
3570                 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3571
3572         case IOC_LIBCFS_ADD_ROUTE: {
3573                 /* default router sensitivity to 1 */
3574                 unsigned int sensitivity = 1;
3575                 config = arg;
3576
3577                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3578                         return -EINVAL;
3579
3580                 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3581                         sensitivity =
3582                           config->cfg_config_u.cfg_route.rtr_sensitivity;
3583                 }
3584
3585                 mutex_lock(&the_lnet.ln_api_mutex);
3586                 rc = lnet_add_route(config->cfg_net,
3587                                     config->cfg_config_u.cfg_route.rtr_hop,
3588                                     config->cfg_nid,
3589                                     config->cfg_config_u.cfg_route.
3590                                         rtr_priority, sensitivity);
3591                 mutex_unlock(&the_lnet.ln_api_mutex);
3592                 return rc;
3593         }
3594
3595         case IOC_LIBCFS_DEL_ROUTE:
3596                 config = arg;
3597
3598                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3599                         return -EINVAL;
3600
3601                 mutex_lock(&the_lnet.ln_api_mutex);
3602                 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3603                 mutex_unlock(&the_lnet.ln_api_mutex);
3604                 return rc;
3605
3606         case IOC_LIBCFS_GET_ROUTE:
3607                 config = arg;
3608
3609                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3610                         return -EINVAL;
3611
3612                 mutex_lock(&the_lnet.ln_api_mutex);
3613                 rc = lnet_get_route(config->cfg_count,
3614                                     &config->cfg_net,
3615                                     &config->cfg_config_u.cfg_route.rtr_hop,
3616                                     &config->cfg_nid,
3617                                     &config->cfg_config_u.cfg_route.rtr_flags,
3618                                     &config->cfg_config_u.cfg_route.
3619                                         rtr_priority,
3620                                     &config->cfg_config_u.cfg_route.
3621                                         rtr_sensitivity);
3622                 mutex_unlock(&the_lnet.ln_api_mutex);
3623                 return rc;
3624
3625         case IOC_LIBCFS_GET_LOCAL_NI: {
3626                 struct lnet_ioctl_config_ni *cfg_ni;
3627                 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3628                 struct lnet_ioctl_element_stats *stats;
3629                 __u32 tun_size;
3630
3631                 cfg_ni = arg;
3632
3633                 /* get the tunables if they are available */
3634                 if (cfg_ni->lic_cfg_hdr.ioc_len <
3635                     sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3636                         return -EINVAL;
3637
3638                 stats = (struct lnet_ioctl_element_stats *)
3639                         cfg_ni->lic_bulk;
3640                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3641                                 (cfg_ni->lic_bulk + sizeof(*stats));
3642
3643                 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3644                         sizeof(*stats);
3645
3646                 mutex_lock(&the_lnet.ln_api_mutex);
3647                 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3648                 mutex_unlock(&the_lnet.ln_api_mutex);
3649                 return rc;
3650         }
3651
3652         case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3653                 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3654
3655                 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3656                         return -EINVAL;
3657
3658                 mutex_lock(&the_lnet.ln_api_mutex);
3659                 rc = lnet_get_ni_stats(msg_stats);
3660                 mutex_unlock(&the_lnet.ln_api_mutex);
3661
3662                 return rc;
3663         }
3664
3665         case IOC_LIBCFS_GET_NET: {
3666                 size_t total = sizeof(*config) +
3667                                sizeof(struct lnet_ioctl_net_config);
3668                 config = arg;
3669
3670                 if (config->cfg_hdr.ioc_len < total)
3671                         return -EINVAL;
3672
3673                 mutex_lock(&the_lnet.ln_api_mutex);
3674                 rc = lnet_get_net_config(config);
3675                 mutex_unlock(&the_lnet.ln_api_mutex);
3676                 return rc;
3677         }
3678
3679         case IOC_LIBCFS_GET_LNET_STATS:
3680         {
3681                 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
3682
3683                 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
3684                         return -EINVAL;
3685
3686                 mutex_lock(&the_lnet.ln_api_mutex);
3687                 lnet_counters_get(&lnet_stats->st_cntrs);
3688                 mutex_unlock(&the_lnet.ln_api_mutex);
3689                 return 0;
3690         }
3691
3692         case IOC_LIBCFS_CONFIG_RTR:
3693                 config = arg;
3694
3695                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3696                         return -EINVAL;
3697
3698                 mutex_lock(&the_lnet.ln_api_mutex);
3699                 if (config->cfg_config_u.cfg_buffers.buf_enable) {
3700                         rc = lnet_rtrpools_enable();
3701                         mutex_unlock(&the_lnet.ln_api_mutex);
3702                         return rc;
3703                 }
3704                 lnet_rtrpools_disable();
3705                 mutex_unlock(&the_lnet.ln_api_mutex);
3706                 return 0;
3707
3708         case IOC_LIBCFS_ADD_BUF:
3709                 config = arg;
3710
3711                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3712                         return -EINVAL;
3713
3714                 mutex_lock(&the_lnet.ln_api_mutex);
3715                 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
3716                                                 buf_tiny,
3717                                           config->cfg_config_u.cfg_buffers.
3718                                                 buf_small,
3719                                           config->cfg_config_u.cfg_buffers.
3720                                                 buf_large);
3721                 mutex_unlock(&the_lnet.ln_api_mutex);
3722                 return rc;
3723
3724         case IOC_LIBCFS_SET_NUMA_RANGE: {
3725                 struct lnet_ioctl_set_value *numa;
3726                 numa = arg;
3727                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3728                         return -EINVAL;
3729                 lnet_net_lock(LNET_LOCK_EX);
3730                 lnet_numa_range = numa->sv_value;
3731                 lnet_net_unlock(LNET_LOCK_EX);
3732                 return 0;
3733         }
3734
3735         case IOC_LIBCFS_GET_NUMA_RANGE: {
3736                 struct lnet_ioctl_set_value *numa;
3737                 numa = arg;
3738                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3739                         return -EINVAL;
3740                 numa->sv_value = lnet_numa_range;
3741                 return 0;
3742         }
3743
3744         case IOC_LIBCFS_GET_BUF: {
3745                 struct lnet_ioctl_pool_cfg *pool_cfg;
3746                 size_t total = sizeof(*config) + sizeof(*pool_cfg);
3747
3748                 config = arg;
3749
3750                 if (config->cfg_hdr.ioc_len < total)
3751                         return -EINVAL;
3752
3753                 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
3754
3755                 mutex_lock(&the_lnet.ln_api_mutex);
3756                 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
3757                 mutex_unlock(&the_lnet.ln_api_mutex);
3758                 return rc;
3759         }
3760
3761         case IOC_LIBCFS_GET_LOCAL_HSTATS: {
3762                 struct lnet_ioctl_local_ni_hstats *stats = arg;
3763
3764                 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
3765                         return -EINVAL;
3766
3767                 mutex_lock(&the_lnet.ln_api_mutex);
3768                 rc = lnet_get_local_ni_hstats(stats);
3769                 mutex_unlock(&the_lnet.ln_api_mutex);
3770
3771                 return rc;
3772         }
3773
3774         case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
3775                 struct lnet_ioctl_recovery_list *list = arg;
3776                 if (list->rlst_hdr.ioc_len < sizeof(*list))
3777                         return -EINVAL;
3778
3779                 mutex_lock(&the_lnet.ln_api_mutex);
3780                 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
3781                         rc = lnet_get_local_ni_recovery_list(list);
3782                 else
3783                         rc = lnet_get_peer_ni_recovery_list(list);
3784                 mutex_unlock(&the_lnet.ln_api_mutex);
3785                 return rc;
3786         }
3787
3788         case IOC_LIBCFS_ADD_PEER_NI: {
3789                 struct lnet_ioctl_peer_cfg *cfg = arg;
3790
3791                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3792                         return -EINVAL;
3793
3794                 mutex_lock(&the_lnet.ln_api_mutex);
3795                 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
3796                                       cfg->prcfg_cfg_nid,
3797                                       cfg->prcfg_mr);
3798                 mutex_unlock(&the_lnet.ln_api_mutex);
3799                 return rc;
3800         }
3801
3802         case IOC_LIBCFS_DEL_PEER_NI: {
3803                 struct lnet_ioctl_peer_cfg *cfg = arg;
3804
3805                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3806                         return -EINVAL;
3807
3808                 mutex_lock(&the_lnet.ln_api_mutex);
3809                 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
3810                                       cfg->prcfg_cfg_nid);
3811                 mutex_unlock(&the_lnet.ln_api_mutex);
3812                 return rc;
3813         }
3814
3815         case IOC_LIBCFS_GET_PEER_INFO: {
3816                 struct lnet_ioctl_peer *peer_info = arg;
3817
3818                 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
3819                         return -EINVAL;
3820
3821                 mutex_lock(&the_lnet.ln_api_mutex);
3822                 rc = lnet_get_peer_ni_info(
3823                    peer_info->pr_count,
3824                    &peer_info->pr_nid,
3825                    peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
3826                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
3827                    &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
3828                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
3829                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
3830                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
3831                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
3832                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
3833                 mutex_unlock(&the_lnet.ln_api_mutex);
3834                 return rc;
3835         }
3836
3837         case IOC_LIBCFS_GET_PEER_NI: {
3838                 struct lnet_ioctl_peer_cfg *cfg = arg;
3839
3840                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3841                         return -EINVAL;
3842
3843                 mutex_lock(&the_lnet.ln_api_mutex);
3844                 rc = lnet_get_peer_info(cfg,
3845                                         (void __user *)cfg->prcfg_bulk);
3846                 mutex_unlock(&the_lnet.ln_api_mutex);
3847                 return rc;
3848         }
3849
3850         case IOC_LIBCFS_GET_PEER_LIST: {
3851                 struct lnet_ioctl_peer_cfg *cfg = arg;
3852
3853                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3854                         return -EINVAL;
3855
3856                 mutex_lock(&the_lnet.ln_api_mutex);
3857                 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
3858                                 (struct lnet_process_id __user *)cfg->prcfg_bulk);
3859                 mutex_unlock(&the_lnet.ln_api_mutex);
3860                 return rc;
3861         }
3862
3863         case IOC_LIBCFS_SET_HEALHV: {
3864                 struct lnet_ioctl_reset_health_cfg *cfg = arg;
3865                 int value;
3866                 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
3867                         return -EINVAL;
3868                 if (cfg->rh_value < 0 ||
3869                     cfg->rh_value > LNET_MAX_HEALTH_VALUE)
3870                         value = LNET_MAX_HEALTH_VALUE;
3871                 else
3872                         value = cfg->rh_value;
3873                 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
3874                        value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
3875                        "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
3876                 mutex_lock(&the_lnet.ln_api_mutex);
3877                 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
3878                         lnet_ni_set_healthv(cfg->rh_nid, value,
3879                                              cfg->rh_all);
3880                 else
3881                         lnet_peer_ni_set_healthv(cfg->rh_nid, value,
3882                                                   cfg->rh_all);
3883                 mutex_unlock(&the_lnet.ln_api_mutex);
3884                 return 0;
3885         }
3886
3887         case IOC_LIBCFS_NOTIFY_ROUTER: {
3888                 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
3889
3890                 /* The deadline passed in by the user should be some time in
3891                  * seconds in the future since the UNIX epoch. We have to map
3892                  * that deadline to the wall clock.
3893                  */
3894                 deadline += ktime_get_seconds();
3895                 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
3896                                    deadline);
3897         }
3898
3899         case IOC_LIBCFS_LNET_DIST:
3900                 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
3901                 if (rc < 0 && rc != -EHOSTUNREACH)
3902                         return rc;
3903
3904                 data->ioc_u32[0] = rc;
3905                 return 0;
3906
3907         case IOC_LIBCFS_TESTPROTOCOMPAT:
3908                 the_lnet.ln_testprotocompat = data->ioc_flags;
3909                 return 0;
3910
3911         case IOC_LIBCFS_LNET_FAULT:
3912                 return lnet_fault_ctl(data->ioc_flags, data);
3913
3914         case IOC_LIBCFS_PING: {
3915                 signed long timeout;
3916
3917                 id.nid = data->ioc_nid;
3918                 id.pid = data->ioc_u32[0];
3919
3920                 /* If timeout is negative then set default of 3 minutes */
3921                 if (((s32)data->ioc_u32[1] <= 0) ||
3922                     data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3923                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3924                 else
3925                         timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
3926
3927                 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
3928                                data->ioc_plen1 / sizeof(struct lnet_process_id));
3929
3930                 if (rc < 0)
3931                         return rc;
3932
3933                 data->ioc_count = rc;
3934                 return 0;
3935         }
3936
3937         case IOC_LIBCFS_PING_PEER: {
3938                 struct lnet_ioctl_ping_data *ping = arg;
3939                 struct lnet_peer *lp;
3940                 signed long timeout;
3941
3942                 /* If timeout is negative then set default of 3 minutes */
3943                 if (((s32)ping->op_param) <= 0 ||
3944                     ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3945                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3946                 else
3947                         timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
3948
3949                 rc = lnet_ping(ping->ping_id, timeout,
3950                                ping->ping_buf,
3951                                ping->ping_count);
3952                 if (rc < 0)
3953                         return rc;
3954
3955                 mutex_lock(&the_lnet.ln_api_mutex);
3956                 lp = lnet_find_peer(ping->ping_id.nid);
3957                 if (lp) {
3958                         ping->ping_id.nid = lp->lp_primary_nid;
3959                         ping->mr_info = lnet_peer_is_multi_rail(lp);
3960                         lnet_peer_decref_locked(lp);
3961                 }
3962                 mutex_unlock(&the_lnet.ln_api_mutex);
3963
3964                 ping->ping_count = rc;
3965                 return 0;
3966         }
3967
3968         case IOC_LIBCFS_DISCOVER: {
3969                 struct lnet_ioctl_ping_data *discover = arg;
3970                 struct lnet_peer *lp;
3971
3972                 rc = lnet_discover(discover->ping_id, discover->op_param,
3973                                    discover->ping_buf,
3974                                    discover->ping_count);
3975                 if (rc < 0)
3976                         return rc;
3977
3978                 mutex_lock(&the_lnet.ln_api_mutex);
3979                 lp = lnet_find_peer(discover->ping_id.nid);
3980                 if (lp) {
3981                         discover->ping_id.nid = lp->lp_primary_nid;
3982                         discover->mr_info = lnet_peer_is_multi_rail(lp);
3983                         lnet_peer_decref_locked(lp);
3984                 }
3985                 mutex_unlock(&the_lnet.ln_api_mutex);
3986
3987                 discover->ping_count = rc;
3988                 return 0;
3989         }
3990
3991         default:
3992                 ni = lnet_net2ni_addref(data->ioc_net);
3993                 if (ni == NULL)
3994                         return -EINVAL;
3995
3996                 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
3997                         rc = -EINVAL;
3998                 else
3999                         rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
4000
4001                 lnet_ni_decref(ni);
4002                 return rc;
4003         }
4004         /* not reached */
4005 }
4006 EXPORT_SYMBOL(LNetCtl);
4007
4008 void LNetDebugPeer(struct lnet_process_id id)
4009 {
4010         lnet_debug_peer(id.nid);
4011 }
4012 EXPORT_SYMBOL(LNetDebugPeer);
4013
4014 /**
4015  * Determine if the specified peer \a nid is on the local node.
4016  *
4017  * \param nid   peer nid to check
4018  *
4019  * \retval true         If peer NID is on the local node.
4020  * \retval false        If peer NID is not on the local node.
4021  */
4022 bool LNetIsPeerLocal(lnet_nid_t nid)
4023 {
4024         struct lnet_net *net;
4025         struct lnet_ni *ni;
4026         int cpt;
4027
4028         cpt = lnet_net_lock_current();
4029         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4030                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4031                         if (ni->ni_nid == nid) {
4032                                 lnet_net_unlock(cpt);
4033                                 return true;
4034                         }
4035                 }
4036         }
4037         lnet_net_unlock(cpt);
4038
4039         return false;
4040 }
4041 EXPORT_SYMBOL(LNetIsPeerLocal);
4042
4043 /**
4044  * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4045  * Note that all interfaces share a same PID, as requested by LNetNIInit().
4046  *
4047  * \param index Index of the interface to look up.
4048  * \param id On successful return, this location will hold the
4049  * struct lnet_process_id ID of the interface.
4050  *
4051  * \retval 0 If an interface exists at \a index.
4052  * \retval -ENOENT If no interface has been found.
4053  */
4054 int
4055 LNetGetId(unsigned int index, struct lnet_process_id *id)
4056 {
4057         struct lnet_ni   *ni;
4058         struct lnet_net  *net;
4059         int               cpt;
4060         int               rc = -ENOENT;
4061
4062         LASSERT(the_lnet.ln_refcount > 0);
4063
4064         cpt = lnet_net_lock_current();
4065
4066         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4067                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4068                         if (index-- != 0)
4069                                 continue;
4070
4071                         id->nid = ni->ni_nid;
4072                         id->pid = the_lnet.ln_pid;
4073                         rc = 0;
4074                         break;
4075                 }
4076         }
4077
4078         lnet_net_unlock(cpt);
4079         return rc;
4080 }
4081 EXPORT_SYMBOL(LNetGetId);
4082
4083 static int lnet_ping(struct lnet_process_id id, signed long timeout,
4084                      struct lnet_process_id __user *ids, int n_ids)
4085 {
4086         struct lnet_handle_eq eqh;
4087         struct lnet_handle_md mdh;
4088         struct lnet_event event;
4089         struct lnet_md md = { NULL };
4090         int which;
4091         int unlinked = 0;
4092         int replied = 0;
4093         const signed long a_long_time = cfs_time_seconds(60);
4094         struct lnet_ping_buffer *pbuf;
4095         struct lnet_process_id tmpid;
4096         int i;
4097         int nob;
4098         int rc;
4099         int rc2;
4100         sigset_t blocked;
4101
4102         /* n_ids limit is arbitrary */
4103         if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4104                 return -EINVAL;
4105
4106         /*
4107          * if the user buffer has more space than the lnet_interfaces_max
4108          * then only fill it up to lnet_interfaces_max
4109          */
4110         if (n_ids > lnet_interfaces_max)
4111                 n_ids = lnet_interfaces_max;
4112
4113         if (id.pid == LNET_PID_ANY)
4114                 id.pid = LNET_PID_LUSTRE;
4115
4116         pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4117         if (!pbuf)
4118                 return -ENOMEM;
4119
4120         /* NB 2 events max (including any unlink event) */
4121         rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &eqh);
4122         if (rc != 0) {
4123                 CERROR("Can't allocate EQ: %d\n", rc);
4124                 goto fail_ping_buffer_decref;
4125         }
4126
4127         /* initialize md content */
4128         md.start     = &pbuf->pb_info;
4129         md.length    = LNET_PING_INFO_SIZE(n_ids);
4130         md.threshold = 2; /* GET/REPLY */
4131         md.max_size  = 0;
4132         md.options   = LNET_MD_TRUNCATE;
4133         md.user_ptr  = NULL;
4134         md.eq_handle = eqh;
4135
4136         rc = LNetMDBind(md, LNET_UNLINK, &mdh);
4137         if (rc != 0) {
4138                 CERROR("Can't bind MD: %d\n", rc);
4139                 goto fail_free_eq;
4140         }
4141
4142         rc = LNetGet(LNET_NID_ANY, mdh, id,
4143                      LNET_RESERVED_PORTAL,
4144                      LNET_PROTO_PING_MATCHBITS, 0, false);
4145
4146         if (rc != 0) {
4147                 /* Don't CERROR; this could be deliberate! */
4148                 rc2 = LNetMDUnlink(mdh);
4149                 LASSERT(rc2 == 0);
4150
4151                 /* NB must wait for the UNLINK event below... */
4152                 unlinked = 1;
4153                 timeout = a_long_time;
4154         }
4155
4156         do {
4157                 /* MUST block for unlink to complete */
4158                 if (unlinked)
4159                         blocked = cfs_block_allsigs();
4160
4161                 rc2 = LNetEQPoll(&eqh, 1, timeout, &event, &which);
4162
4163                 if (unlinked)
4164                         cfs_restore_sigs(blocked);
4165
4166                 CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
4167                        (rc2 <= 0) ? -1 : event.type,
4168                        (rc2 <= 0) ? -1 : event.status,
4169                        (rc2 > 0 && event.unlinked) ? " unlinked" : "");
4170
4171                 LASSERT(rc2 != -EOVERFLOW);     /* can't miss anything */
4172
4173                 if (rc2 <= 0 || event.status != 0) {
4174                         /* timeout or error */
4175                         if (!replied && rc == 0)
4176                                 rc = (rc2 < 0) ? rc2 :
4177                                      (rc2 == 0) ? -ETIMEDOUT :
4178                                      event.status;
4179
4180                         if (!unlinked) {
4181                                 /* Ensure completion in finite time... */
4182                                 LNetMDUnlink(mdh);
4183                                 /* No assertion (racing with network) */
4184                                 unlinked = 1;
4185                                 timeout = a_long_time;
4186                         } else if (rc2 == 0) {
4187                                 /* timed out waiting for unlink */
4188                                 CWARN("ping %s: late network completion\n",
4189                                       libcfs_id2str(id));
4190                         }
4191                 } else if (event.type == LNET_EVENT_REPLY) {
4192                         replied = 1;
4193                         rc = event.mlength;
4194                 }
4195         } while (rc2 <= 0 || !event.unlinked);
4196
4197         if (!replied) {
4198                 if (rc >= 0)
4199                         CWARN("%s: Unexpected rc >= 0 but no reply!\n",
4200                               libcfs_id2str(id));
4201                 rc = -EIO;
4202                 goto fail_free_eq;
4203         }
4204
4205         nob = rc;
4206         LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4207
4208         rc = -EPROTO;           /* if I can't parse... */
4209
4210         if (nob < 8) {
4211                 CERROR("%s: ping info too short %d\n",
4212                        libcfs_id2str(id), nob);
4213                 goto fail_free_eq;
4214         }
4215
4216         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4217                 lnet_swap_pinginfo(pbuf);
4218         } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4219                 CERROR("%s: Unexpected magic %08x\n",
4220                        libcfs_id2str(id), pbuf->pb_info.pi_magic);
4221                 goto fail_free_eq;
4222         }
4223
4224         if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4225                 CERROR("%s: ping w/o NI status: 0x%x\n",
4226                        libcfs_id2str(id), pbuf->pb_info.pi_features);
4227                 goto fail_free_eq;
4228         }
4229
4230         if (nob < LNET_PING_INFO_SIZE(0)) {
4231                 CERROR("%s: Short reply %d(%d min)\n",
4232                        libcfs_id2str(id),
4233                        nob, (int)LNET_PING_INFO_SIZE(0));
4234                 goto fail_free_eq;
4235         }
4236
4237         if (pbuf->pb_info.pi_nnis < n_ids)
4238                 n_ids = pbuf->pb_info.pi_nnis;
4239
4240         if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4241                 CERROR("%s: Short reply %d(%d expected)\n",
4242                        libcfs_id2str(id),
4243                        nob, (int)LNET_PING_INFO_SIZE(n_ids));
4244                 goto fail_free_eq;
4245         }
4246
4247         rc = -EFAULT;           /* if I segv in copy_to_user()... */
4248
4249         memset(&tmpid, 0, sizeof(tmpid));
4250         for (i = 0; i < n_ids; i++) {
4251                 tmpid.pid = pbuf->pb_info.pi_pid;
4252                 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4253                 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4254                         goto fail_free_eq;
4255         }
4256         rc = pbuf->pb_info.pi_nnis;
4257
4258  fail_free_eq:
4259         rc2 = LNetEQFree(eqh);
4260         if (rc2 != 0)
4261                 CERROR("rc2 %d\n", rc2);
4262         LASSERT(rc2 == 0);
4263
4264  fail_ping_buffer_decref:
4265         lnet_ping_buffer_decref(pbuf);
4266         return rc;
4267 }
4268
4269 static int
4270 lnet_discover(struct lnet_process_id id, __u32 force,
4271               struct lnet_process_id __user *ids, int n_ids)
4272 {
4273         struct lnet_peer_ni *lpni;
4274         struct lnet_peer_ni *p;
4275         struct lnet_peer *lp;
4276         struct lnet_process_id *buf;
4277         int cpt;
4278         int i;
4279         int rc;
4280         int max_intf = lnet_interfaces_max;
4281         size_t buf_size;
4282
4283         if (n_ids <= 0 ||
4284             id.nid == LNET_NID_ANY)
4285                 return -EINVAL;
4286
4287         if (id.pid == LNET_PID_ANY)
4288                 id.pid = LNET_PID_LUSTRE;
4289
4290         /*
4291          * if the user buffer has more space than the max_intf
4292          * then only fill it up to max_intf
4293          */
4294         if (n_ids > max_intf)
4295                 n_ids = max_intf;
4296
4297         buf_size = n_ids * sizeof(*buf);
4298
4299         LIBCFS_ALLOC(buf, buf_size);
4300         if (!buf)
4301                 return -ENOMEM;
4302
4303         cpt = lnet_net_lock_current();
4304         lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4305         if (IS_ERR(lpni)) {
4306                 rc = PTR_ERR(lpni);
4307                 goto out;
4308         }
4309
4310         /*
4311          * Clearing the NIDS_UPTODATE flag ensures the peer will
4312          * be discovered, provided discovery has not been disabled.
4313          */
4314         lp = lpni->lpni_peer_net->lpn_peer;
4315         spin_lock(&lp->lp_lock);
4316         lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4317         /* If the force flag is set, force a PING and PUSH as well. */
4318         if (force)
4319                 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4320         spin_unlock(&lp->lp_lock);
4321         rc = lnet_discover_peer_locked(lpni, cpt, true);
4322         if (rc)
4323                 goto out_decref;
4324
4325         /* Peer may have changed. */
4326         lp = lpni->lpni_peer_net->lpn_peer;
4327         if (lp->lp_nnis < n_ids)
4328                 n_ids = lp->lp_nnis;
4329
4330         i = 0;
4331         p = NULL;
4332         while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4333                 buf[i].pid = id.pid;
4334                 buf[i].nid = p->lpni_nid;
4335                 if (++i >= n_ids)
4336                         break;
4337         }
4338
4339         lnet_net_unlock(cpt);
4340
4341         rc = -EFAULT;
4342         if (copy_to_user(ids, buf, n_ids * sizeof(*buf)))
4343                 goto out_relock;
4344         rc = n_ids;
4345 out_relock:
4346         lnet_net_lock(cpt);
4347 out_decref:
4348         lnet_peer_ni_decref_locked(lpni);
4349 out:
4350         lnet_net_unlock(cpt);
4351
4352         LIBCFS_FREE(buf, buf_size);
4353
4354         return rc;
4355 }
4356
4357 /**
4358  * Retrieve peer discovery status.
4359  *
4360  * \retval 1 if lnet_peer_discovery_disabled is 0
4361  * \retval 0 if lnet_peer_discovery_disabled is 1
4362  */
4363 int
4364 LNetGetPeerDiscoveryStatus(void)
4365 {
4366         return !lnet_peer_discovery_disabled;
4367 }
4368 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);