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