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