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 #ifdef HAVE_KERNEL_PARAM_OPS
 128 static struct kernel_param_ops param_ops_interfaces_max = {
 129         .set = intf_max_set,
 130         .get = param_get_int,
 131 };
 132
 133 #define param_check_interfaces_max(name, p) \
 134                 __param_check(name, p, int)
 135
 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                   &lnet_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 #ifdef HAVE_KERNEL_PARAM_OPS
 148 static struct kernel_param_ops param_ops_discovery_disabled = {
 149         .set = discovery_set,
 150         .get = param_get_int,
 151 };
 152
 153 #define param_check_discovery_disabled(name, p) \
 154                 __param_check(name, p, int)
 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                   &lnet_peer_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 #ifdef HAVE_KERNEL_PARAM_OPS
 167 static struct kernel_param_ops param_ops_drop_asym_route = {
 168         .set = drop_asym_route_set,
 169         .get = param_get_int,
 170 };
 171
 172 #define param_check_drop_asym_route(name, p)    \
 173         __param_check(name, p, int)
 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                   &lnet_drop_asym_route, 0644);
 178 #endif
 179 MODULE_PARM_DESC(lnet_drop_asym_route,
 180                  "Set to 1 to drop asymmetrical route messages.");
 181
 182 #define LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT 50
 183 #define LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT 50
 184
 185 unsigned lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
 186 static int transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp);
 187 #ifdef HAVE_KERNEL_PARAM_OPS
 188 static struct kernel_param_ops param_ops_transaction_timeout = {
 189         .set = transaction_to_set,
 190         .get = param_get_int,
 191 };
 192
 193 #define param_check_transaction_timeout(name, p) \
 194                 __param_check(name, p, int)
 195 module_param(lnet_transaction_timeout, transaction_timeout, S_IRUGO|S_IWUSR);
 196 #else
 197 module_param_call(lnet_transaction_timeout, transaction_to_set, param_get_int,
 198                   &lnet_transaction_timeout, S_IRUGO|S_IWUSR);
 199 #endif
 200 MODULE_PARM_DESC(lnet_transaction_timeout,
 201                 "Maximum number of seconds to wait for a peer response.");
 202
 203 #define LNET_RETRY_COUNT_HEALTH_DEFAULT 2
 204 unsigned lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
 205 static int retry_count_set(const char *val, cfs_kernel_param_arg_t *kp);
 206 #ifdef HAVE_KERNEL_PARAM_OPS
 207 static struct kernel_param_ops param_ops_retry_count = {
 208         .set = retry_count_set,
 209         .get = param_get_int,
 210 };
 211
 212 #define param_check_retry_count(name, p) \
 213                 __param_check(name, p, int)
 214 module_param(lnet_retry_count, retry_count, S_IRUGO|S_IWUSR);
 215 #else
 216 module_param_call(lnet_retry_count, retry_count_set, param_get_int,
 217                   &lnet_retry_count, S_IRUGO|S_IWUSR);
 218 #endif
 219 MODULE_PARM_DESC(lnet_retry_count,
 220                  "Maximum number of times to retry transmitting a message");
 221
 222
 223 unsigned lnet_lnd_timeout = LNET_LND_DEFAULT_TIMEOUT;
 224
 225 /*
 226  * This sequence number keeps track of how many times DLC was used to
 227  * update the local NIs. It is incremented when a NI is added or
 228  * removed and checked when sending a message to determine if there is
 229  * a need to re-run the selection algorithm. See lnet_select_pathway()
 230  * for more details on its usage.
 231  */
 232 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
 233
 234 static int lnet_ping(struct lnet_process_id id, signed long timeout,
 235                      struct lnet_process_id __user *ids, int n_ids);
 236
 237 static int lnet_discover(struct lnet_process_id id, __u32 force,
 238                          struct lnet_process_id __user *ids, int n_ids);
 239
 240 static int
 241 sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
 242 {
 243         int rc;
 244         unsigned *sensitivity = (unsigned *)kp->arg;
 245         unsigned long value;
 246
 247         rc = kstrtoul(val, 0, &value);
 248         if (rc) {
 249                 CERROR("Invalid module parameter value for 'lnet_health_sensitivity'\n");
 250                 return rc;
 251         }
 252
 253         /*
 254          * The purpose of locking the api_mutex here is to ensure that
 255          * the correct value ends up stored properly.
 256          */
 257         mutex_lock(&the_lnet.ln_api_mutex);
 258
 259         if (value > LNET_MAX_HEALTH_VALUE) {
 260                 mutex_unlock(&the_lnet.ln_api_mutex);
 261                 CERROR("Invalid health value. Maximum: %d value = %lu\n",
 262                        LNET_MAX_HEALTH_VALUE, value);
 263                 return -EINVAL;
 264         }
 265
 266         /*
 267          * if we're turning on health then use the health timeout
 268          * defaults.
 269          */
 270         if (*sensitivity == 0 && value != 0) {
 271                 lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
 272                 lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
 273         /*
 274          * if we're turning off health then use the no health timeout
 275          * default.
 276          */
 277         } else if (*sensitivity != 0 && value == 0) {
 278                 lnet_transaction_timeout =
 279                         LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT;
 280                 lnet_retry_count = 0;
 281         }
 282
 283         *sensitivity = value;
 284
 285         mutex_unlock(&the_lnet.ln_api_mutex);
 286
 287         return 0;
 288 }
 289
 290 static int
 291 recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp)
 292 {
 293         int rc;
 294         unsigned *interval = (unsigned *)kp->arg;
 295         unsigned long value;
 296
 297         rc = kstrtoul(val, 0, &value);
 298         if (rc) {
 299                 CERROR("Invalid module parameter value for 'lnet_recovery_interval'\n");
 300                 return rc;
 301         }
 302
 303         if (value < 1) {
 304                 CERROR("lnet_recovery_interval must be at least 1 second\n");
 305                 return -EINVAL;
 306         }
 307
 308         /*
 309          * The purpose of locking the api_mutex here is to ensure that
 310          * the correct value ends up stored properly.
 311          */
 312         mutex_lock(&the_lnet.ln_api_mutex);
 313
 314         *interval = value;
 315
 316         mutex_unlock(&the_lnet.ln_api_mutex);
 317
 318         return 0;
 319 }
 320
 321 static int
 322 discovery_set(const char *val, cfs_kernel_param_arg_t *kp)
 323 {
 324         int rc;
 325         unsigned *discovery = (unsigned *)kp->arg;
 326         unsigned long value;
 327         struct lnet_ping_buffer *pbuf;
 328
 329         rc = kstrtoul(val, 0, &value);
 330         if (rc) {
 331                 CERROR("Invalid module parameter value for 'lnet_peer_discovery_disabled'\n");
 332                 return rc;
 333         }
 334
 335         value = (value) ? 1 : 0;
 336
 337         /*
 338          * The purpose of locking the api_mutex here is to ensure that
 339          * the correct value ends up stored properly.
 340          */
 341         mutex_lock(&the_lnet.ln_api_mutex);
 342
 343         if (value == *discovery) {
 344                 mutex_unlock(&the_lnet.ln_api_mutex);
 345                 return 0;
 346         }
 347
 348         *discovery = value;
 349
 350         if (the_lnet.ln_state != LNET_STATE_RUNNING) {
 351                 mutex_unlock(&the_lnet.ln_api_mutex);
 352                 return 0;
 353         }
 354
 355         /* tell peers that discovery setting has changed */
 356         lnet_net_lock(LNET_LOCK_EX);
 357         pbuf = the_lnet.ln_ping_target;
 358         if (value)
 359                 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
 360         else
 361                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
 362         lnet_net_unlock(LNET_LOCK_EX);
 363
 364         lnet_push_update_to_peers(1);
 365
 366         mutex_unlock(&the_lnet.ln_api_mutex);
 367
 368         return 0;
 369 }
 370
 371 static int
 372 drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp)
 373 {
 374         int rc;
 375         unsigned int *drop_asym_route = (unsigned int *)kp->arg;
 376         unsigned long value;
 377
 378         rc = kstrtoul(val, 0, &value);
 379         if (rc) {
 380                 CERROR("Invalid module parameter value for "
 381                        "'lnet_drop_asym_route'\n");
 382                 return rc;
 383         }
 384
 385         /*
 386          * The purpose of locking the api_mutex here is to ensure that
 387          * the correct value ends up stored properly.
 388          */
 389         mutex_lock(&the_lnet.ln_api_mutex);
 390
 391         if (value == *drop_asym_route) {
 392                 mutex_unlock(&the_lnet.ln_api_mutex);
 393                 return 0;
 394         }
 395
 396         *drop_asym_route = value;
 397
 398         mutex_unlock(&the_lnet.ln_api_mutex);
 399
 400         return 0;
 401 }
 402
 403 static int
 404 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
 405 {
 406         int rc;
 407         unsigned *transaction_to = (unsigned *)kp->arg;
 408         unsigned long value;
 409
 410         rc = kstrtoul(val, 0, &value);
 411         if (rc) {
 412                 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
 413                 return rc;
 414         }
 415
 416         /*
 417          * The purpose of locking the api_mutex here is to ensure that
 418          * the correct value ends up stored properly.
 419          */
 420         mutex_lock(&the_lnet.ln_api_mutex);
 421
 422         if (value < lnet_retry_count || value == 0) {
 423                 mutex_unlock(&the_lnet.ln_api_mutex);
 424                 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
 425                        "Has to be greater than lnet_retry_count (%u)\n",
 426                        value, lnet_retry_count);
 427                 return -EINVAL;
 428         }
 429
 430         if (value == *transaction_to) {
 431                 mutex_unlock(&the_lnet.ln_api_mutex);
 432                 return 0;
 433         }
 434
 435         *transaction_to = value;
 436         if (lnet_retry_count == 0)
 437                 lnet_lnd_timeout = value;
 438         else
 439                 lnet_lnd_timeout = value / lnet_retry_count;
 440
 441         mutex_unlock(&the_lnet.ln_api_mutex);
 442
 443         return 0;
 444 }
 445
 446 static int
 447 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
 448 {
 449         int rc;
 450         unsigned *retry_count = (unsigned *)kp->arg;
 451         unsigned long value;
 452
 453         rc = kstrtoul(val, 0, &value);
 454         if (rc) {
 455                 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
 456                 return rc;
 457         }
 458
 459         /*
 460          * The purpose of locking the api_mutex here is to ensure that
 461          * the correct value ends up stored properly.
 462          */
 463         mutex_lock(&the_lnet.ln_api_mutex);
 464
 465         if (lnet_health_sensitivity == 0) {
 466                 mutex_unlock(&the_lnet.ln_api_mutex);
 467                 CERROR("Can not set retry_count when health feature is turned off\n");
 468                 return -EINVAL;
 469         }
 470
 471         if (value > lnet_transaction_timeout) {
 472                 mutex_unlock(&the_lnet.ln_api_mutex);
 473                 CERROR("Invalid value for lnet_retry_count (%lu). "
 474                        "Has to be smaller than lnet_transaction_timeout (%u)\n",
 475                        value, lnet_transaction_timeout);
 476                 return -EINVAL;
 477         }
 478
 479         *retry_count = value;
 480
 481         if (value == 0)
 482                 lnet_lnd_timeout = lnet_transaction_timeout;
 483         else
 484                 lnet_lnd_timeout = lnet_transaction_timeout / value;
 485
 486         mutex_unlock(&the_lnet.ln_api_mutex);
 487
 488         return 0;
 489 }
 490
 491 static int
 492 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
 493 {
 494         int value, rc;
 495
 496         rc = kstrtoint(val, 0, &value);
 497         if (rc) {
 498                 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
 499                 return rc;
 500         }
 501
 502         if (value < LNET_INTERFACES_MIN) {
 503                 CWARN("max interfaces provided are too small, setting to %d\n",
 504                       LNET_INTERFACES_MAX_DEFAULT);
 505                 value = LNET_INTERFACES_MAX_DEFAULT;
 506         }
 507
 508         *(int *)kp->arg = value;
 509
 510         return 0;
 511 }
 512
 513 static char *
 514 lnet_get_routes(void)
 515 {
 516         return routes;
 517 }
 518
 519 static char *
 520 lnet_get_networks(void)
 521 {
 522         char   *nets;
 523         int     rc;
 524
 525         if (*networks != 0 && *ip2nets != 0) {
 526                 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
 527                                    "'ip2nets' but not both at once\n");
 528                 return NULL;
 529         }
 530
 531         if (*ip2nets != 0) {
 532                 rc = lnet_parse_ip2nets(&nets, ip2nets);
 533                 return (rc == 0) ? nets : NULL;
 534         }
 535
 536         if (*networks != 0)
 537                 return networks;
 538
 539         return "tcp";
 540 }
 541
 542 static void
 543 lnet_init_locks(void)
 544 {
 545         spin_lock_init(&the_lnet.ln_eq_wait_lock);
 546         spin_lock_init(&the_lnet.ln_msg_resend_lock);
 547         init_waitqueue_head(&the_lnet.ln_eq_waitq);
 548         init_waitqueue_head(&the_lnet.ln_mt_waitq);
 549         mutex_init(&the_lnet.ln_lnd_mutex);
 550 }
 551
 552 static void
 553 lnet_fini_locks(void)
 554 {
 555 }
 556
 557 struct kmem_cache *lnet_mes_cachep;        /* MEs kmem_cache */
 558 struct kmem_cache *lnet_small_mds_cachep;  /* <= LNET_SMALL_MD_SIZE bytes
 559                                             *  MDs kmem_cache */
 560
 561 static int
 562 lnet_descriptor_setup(void)
 563 {
 564         /* create specific kmem_cache for MEs and small MDs (i.e., originally
 565          * allocated in <size-xxx> kmem_cache).
 566          */
 567         lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
 568                                             0, 0, NULL);
 569         if (!lnet_mes_cachep)
 570                 return -ENOMEM;
 571
 572         lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
 573                                                   LNET_SMALL_MD_SIZE, 0, 0,
 574                                                   NULL);
 575         if (!lnet_small_mds_cachep)
 576                 return -ENOMEM;
 577
 578         return 0;
 579 }
 580
 581 static void
 582 lnet_descriptor_cleanup(void)
 583 {
 584
 585         if (lnet_small_mds_cachep) {
 586                 kmem_cache_destroy(lnet_small_mds_cachep);
 587                 lnet_small_mds_cachep = NULL;
 588         }
 589
 590         if (lnet_mes_cachep) {
 591                 kmem_cache_destroy(lnet_mes_cachep);
 592                 lnet_mes_cachep = NULL;
 593         }
 594 }
 595
 596 static int
 597 lnet_create_remote_nets_table(void)
 598 {
 599         int               i;
 600         struct list_head *hash;
 601
 602         LASSERT(the_lnet.ln_remote_nets_hash == NULL);
 603         LASSERT(the_lnet.ln_remote_nets_hbits > 0);
 604         LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
 605         if (hash == NULL) {
 606                 CERROR("Failed to create remote nets hash table\n");
 607                 return -ENOMEM;
 608         }
 609
 610         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 611                 INIT_LIST_HEAD(&hash[i]);
 612         the_lnet.ln_remote_nets_hash = hash;
 613         return 0;
 614 }
 615
 616 static void
 617 lnet_destroy_remote_nets_table(void)
 618 {
 619         int i;
 620
 621         if (the_lnet.ln_remote_nets_hash == NULL)
 622                 return;
 623
 624         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 625                 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
 626
 627         LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
 628                     LNET_REMOTE_NETS_HASH_SIZE *
 629                     sizeof(the_lnet.ln_remote_nets_hash[0]));
 630         the_lnet.ln_remote_nets_hash = NULL;
 631 }
 632
 633 static void
 634 lnet_destroy_locks(void)
 635 {
 636         if (the_lnet.ln_res_lock != NULL) {
 637                 cfs_percpt_lock_free(the_lnet.ln_res_lock);
 638                 the_lnet.ln_res_lock = NULL;
 639         }
 640
 641         if (the_lnet.ln_net_lock != NULL) {
 642                 cfs_percpt_lock_free(the_lnet.ln_net_lock);
 643                 the_lnet.ln_net_lock = NULL;
 644         }
 645
 646         lnet_fini_locks();
 647 }
 648
 649 static int
 650 lnet_create_locks(void)
 651 {
 652         lnet_init_locks();
 653
 654         the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 655         if (the_lnet.ln_res_lock == NULL)
 656                 goto failed;
 657
 658         the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 659         if (the_lnet.ln_net_lock == NULL)
 660                 goto failed;
 661
 662         return 0;
 663
 664  failed:
 665         lnet_destroy_locks();
 666         return -ENOMEM;
 667 }
 668
 669 static void lnet_assert_wire_constants(void)
 670 {
 671         /* Wire protocol assertions generated by 'wirecheck'
 672          * running on Linux robert.bartonsoftware.com 2.6.8-1.521
 673          * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
 674          * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
 675
 676         /* Constants... */
 677         CLASSERT(LNET_PROTO_TCP_MAGIC == 0xeebc0ded);
 678         CLASSERT(LNET_PROTO_TCP_VERSION_MAJOR == 1);
 679         CLASSERT(LNET_PROTO_TCP_VERSION_MINOR == 0);
 680         CLASSERT(LNET_MSG_ACK == 0);
 681         CLASSERT(LNET_MSG_PUT == 1);
 682         CLASSERT(LNET_MSG_GET == 2);
 683         CLASSERT(LNET_MSG_REPLY == 3);
 684         CLASSERT(LNET_MSG_HELLO == 4);
 685
 686         /* Checks for struct lnet_handle_wire */
 687         CLASSERT((int)sizeof(struct lnet_handle_wire) == 16);
 688         CLASSERT((int)offsetof(struct lnet_handle_wire, wh_interface_cookie) == 0);
 689         CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) == 8);
 690         CLASSERT((int)offsetof(struct lnet_handle_wire, wh_object_cookie) == 8);
 691         CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) == 8);
 692
 693         /* Checks for struct struct lnet_magicversion */
 694         CLASSERT((int)sizeof(struct lnet_magicversion) == 8);
 695         CLASSERT((int)offsetof(struct lnet_magicversion, magic) == 0);
 696         CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->magic) == 4);
 697         CLASSERT((int)offsetof(struct lnet_magicversion, version_major) == 4);
 698         CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->version_major) == 2);
 699         CLASSERT((int)offsetof(struct lnet_magicversion, version_minor) == 6);
 700         CLASSERT((int)sizeof(((struct lnet_magicversion *)0)->version_minor) == 2);
 701
 702         /* Checks for struct struct lnet_hdr */
 703         CLASSERT((int)sizeof(struct lnet_hdr) == 72);
 704         CLASSERT((int)offsetof(struct lnet_hdr, dest_nid) == 0);
 705         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->dest_nid) == 8);
 706         CLASSERT((int)offsetof(struct lnet_hdr, src_nid) == 8);
 707         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->src_nid) == 8);
 708         CLASSERT((int)offsetof(struct lnet_hdr, dest_pid) == 16);
 709         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->dest_pid) == 4);
 710         CLASSERT((int)offsetof(struct lnet_hdr, src_pid) == 20);
 711         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->src_pid) == 4);
 712         CLASSERT((int)offsetof(struct lnet_hdr, type) == 24);
 713         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->type) == 4);
 714         CLASSERT((int)offsetof(struct lnet_hdr, payload_length) == 28);
 715         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->payload_length) == 4);
 716         CLASSERT((int)offsetof(struct lnet_hdr, msg) == 32);
 717         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg) == 40);
 718
 719         /* Ack */
 720         CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) == 32);
 721         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) == 16);
 722         CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.match_bits) == 48);
 723         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) == 8);
 724         CLASSERT((int)offsetof(struct lnet_hdr, msg.ack.mlength) == 56);
 725         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) == 4);
 726
 727         /* Put */
 728         CLASSERT((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) == 32);
 729         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) == 16);
 730         CLASSERT((int)offsetof(struct lnet_hdr, msg.put.match_bits) == 48);
 731         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) == 8);
 732         CLASSERT((int)offsetof(struct lnet_hdr, msg.put.hdr_data) == 56);
 733         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) == 8);
 734         CLASSERT((int)offsetof(struct lnet_hdr, msg.put.ptl_index) == 64);
 735         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) == 4);
 736         CLASSERT((int)offsetof(struct lnet_hdr, msg.put.offset) == 68);
 737         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) == 4);
 738
 739         /* Get */
 740         CLASSERT((int)offsetof(struct lnet_hdr, msg.get.return_wmd) == 32);
 741         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) == 16);
 742         CLASSERT((int)offsetof(struct lnet_hdr, msg.get.match_bits) == 48);
 743         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) == 8);
 744         CLASSERT((int)offsetof(struct lnet_hdr, msg.get.ptl_index) == 56);
 745         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) == 4);
 746         CLASSERT((int)offsetof(struct lnet_hdr, msg.get.src_offset) == 60);
 747         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) == 4);
 748         CLASSERT((int)offsetof(struct lnet_hdr, msg.get.sink_length) == 64);
 749         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) == 4);
 750
 751         /* Reply */
 752         CLASSERT((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) == 32);
 753         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) == 16);
 754
 755         /* Hello */
 756         CLASSERT((int)offsetof(struct lnet_hdr, msg.hello.incarnation) == 32);
 757         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) == 8);
 758         CLASSERT((int)offsetof(struct lnet_hdr, msg.hello.type) == 40);
 759         CLASSERT((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) == 4);
 760
 761         /* Checks for struct lnet_ni_status and related constants */
 762         CLASSERT(LNET_NI_STATUS_INVALID == 0x00000000);
 763         CLASSERT(LNET_NI_STATUS_UP == 0x15aac0de);
 764         CLASSERT(LNET_NI_STATUS_DOWN == 0xdeadface);
 765
 766         /* Checks for struct lnet_ni_status */
 767         CLASSERT((int)sizeof(struct lnet_ni_status) == 16);
 768         CLASSERT((int)offsetof(struct lnet_ni_status, ns_nid) == 0);
 769         CLASSERT((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) == 8);
 770         CLASSERT((int)offsetof(struct lnet_ni_status, ns_status) == 8);
 771         CLASSERT((int)sizeof(((struct lnet_ni_status *)0)->ns_status) == 4);
 772         CLASSERT((int)offsetof(struct lnet_ni_status, ns_unused) == 12);
 773         CLASSERT((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) == 4);
 774
 775         /* Checks for struct lnet_ping_info and related constants */
 776         CLASSERT(LNET_PROTO_PING_MAGIC == 0x70696E67);
 777         CLASSERT(LNET_PING_FEAT_INVAL == 0);
 778         CLASSERT(LNET_PING_FEAT_BASE == 1);
 779         CLASSERT(LNET_PING_FEAT_NI_STATUS == 2);
 780         CLASSERT(LNET_PING_FEAT_RTE_DISABLED == 4);
 781         CLASSERT(LNET_PING_FEAT_MULTI_RAIL == 8);
 782         CLASSERT(LNET_PING_FEAT_DISCOVERY == 16);
 783         CLASSERT(LNET_PING_FEAT_BITS == 31);
 784
 785         /* Checks for struct lnet_ping_info */
 786         CLASSERT((int)sizeof(struct lnet_ping_info) == 16);
 787         CLASSERT((int)offsetof(struct lnet_ping_info, pi_magic) == 0);
 788         CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) == 4);
 789         CLASSERT((int)offsetof(struct lnet_ping_info, pi_features) == 4);
 790         CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_features) == 4);
 791         CLASSERT((int)offsetof(struct lnet_ping_info, pi_pid) == 8);
 792         CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) == 4);
 793         CLASSERT((int)offsetof(struct lnet_ping_info, pi_nnis) == 12);
 794         CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) == 4);
 795         CLASSERT((int)offsetof(struct lnet_ping_info, pi_ni) == 16);
 796         CLASSERT((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) == 0);
 797 }
 798
 799 static struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
 800 {
 801         struct lnet_lnd *lnd;
 802         struct list_head *tmp;
 803
 804         /* holding lnd mutex */
 805         list_for_each(tmp, &the_lnet.ln_lnds) {
 806                 lnd = list_entry(tmp, struct lnet_lnd, lnd_list);
 807
 808                 if (lnd->lnd_type == type)
 809                         return lnd;
 810         }
 811         return NULL;
 812 }
 813
 814 unsigned int
 815 lnet_get_lnd_timeout(void)
 816 {
 817         return lnet_lnd_timeout;
 818 }
 819 EXPORT_SYMBOL(lnet_get_lnd_timeout);
 820
 821 void
 822 lnet_register_lnd(struct lnet_lnd *lnd)
 823 {
 824         mutex_lock(&the_lnet.ln_lnd_mutex);
 825
 826         LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
 827         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
 828
 829         list_add_tail(&lnd->lnd_list, &the_lnet.ln_lnds);
 830         lnd->lnd_refcount = 0;
 831
 832         CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
 833
 834         mutex_unlock(&the_lnet.ln_lnd_mutex);
 835 }
 836 EXPORT_SYMBOL(lnet_register_lnd);
 837
 838 void
 839 lnet_unregister_lnd(struct lnet_lnd *lnd)
 840 {
 841         mutex_lock(&the_lnet.ln_lnd_mutex);
 842
 843         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
 844         LASSERT(lnd->lnd_refcount == 0);
 845
 846         list_del(&lnd->lnd_list);
 847         CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
 848
 849         mutex_unlock(&the_lnet.ln_lnd_mutex);
 850 }
 851 EXPORT_SYMBOL(lnet_unregister_lnd);
 852
 853 void
 854 lnet_counters_get_common(struct lnet_counters_common *common)
 855 {
 856         struct lnet_counters *ctr;
 857         int i;
 858
 859         memset(common, 0, sizeof(*common));
 860
 861         lnet_net_lock(LNET_LOCK_EX);
 862
 863         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
 864                 common->lcc_msgs_max     += ctr->lct_common.lcc_msgs_max;
 865                 common->lcc_msgs_alloc   += ctr->lct_common.lcc_msgs_alloc;
 866                 common->lcc_errors       += ctr->lct_common.lcc_errors;
 867                 common->lcc_send_count   += ctr->lct_common.lcc_send_count;
 868                 common->lcc_recv_count   += ctr->lct_common.lcc_recv_count;
 869                 common->lcc_route_count  += ctr->lct_common.lcc_route_count;
 870                 common->lcc_drop_count   += ctr->lct_common.lcc_drop_count;
 871                 common->lcc_send_length  += ctr->lct_common.lcc_send_length;
 872                 common->lcc_recv_length  += ctr->lct_common.lcc_recv_length;
 873                 common->lcc_route_length += ctr->lct_common.lcc_route_length;
 874                 common->lcc_drop_length  += ctr->lct_common.lcc_drop_length;
 875         }
 876         lnet_net_unlock(LNET_LOCK_EX);
 877 }
 878 EXPORT_SYMBOL(lnet_counters_get_common);
 879
 880 void
 881 lnet_counters_get(struct lnet_counters *counters)
 882 {
 883         struct lnet_counters *ctr;
 884         struct lnet_counters_health *health = &counters->lct_health;
 885         int             i;
 886
 887         memset(counters, 0, sizeof(*counters));
 888
 889         lnet_counters_get_common(&counters->lct_common);
 890
 891         lnet_net_lock(LNET_LOCK_EX);
 892
 893         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
 894                 health->lch_rst_alloc    += ctr->lct_health.lch_rst_alloc;
 895                 health->lch_resend_count += ctr->lct_health.lch_resend_count;
 896                 health->lch_response_timeout_count +=
 897                                 ctr->lct_health.lch_response_timeout_count;
 898                 health->lch_local_interrupt_count +=
 899                                 ctr->lct_health.lch_local_interrupt_count;
 900                 health->lch_local_dropped_count +=
 901                                 ctr->lct_health.lch_local_dropped_count;
 902                 health->lch_local_aborted_count +=
 903                                 ctr->lct_health.lch_local_aborted_count;
 904                 health->lch_local_no_route_count +=
 905                                 ctr->lct_health.lch_local_no_route_count;
 906                 health->lch_local_timeout_count +=
 907                                 ctr->lct_health.lch_local_timeout_count;
 908                 health->lch_local_error_count +=
 909                                 ctr->lct_health.lch_local_error_count;
 910                 health->lch_remote_dropped_count +=
 911                                 ctr->lct_health.lch_remote_dropped_count;
 912                 health->lch_remote_error_count +=
 913                                 ctr->lct_health.lch_remote_error_count;
 914                 health->lch_remote_timeout_count +=
 915                                 ctr->lct_health.lch_remote_timeout_count;
 916                 health->lch_network_timeout_count +=
 917                                 ctr->lct_health.lch_network_timeout_count;
 918         }
 919         lnet_net_unlock(LNET_LOCK_EX);
 920 }
 921 EXPORT_SYMBOL(lnet_counters_get);
 922
 923 void
 924 lnet_counters_reset(void)
 925 {
 926         struct lnet_counters *counters;
 927         int             i;
 928
 929         lnet_net_lock(LNET_LOCK_EX);
 930
 931         cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
 932                 memset(counters, 0, sizeof(struct lnet_counters));
 933
 934         lnet_net_unlock(LNET_LOCK_EX);
 935 }
 936
 937 static char *
 938 lnet_res_type2str(int type)
 939 {
 940         switch (type) {
 941         default:
 942                 LBUG();
 943         case LNET_COOKIE_TYPE_MD:
 944                 return "MD";
 945         case LNET_COOKIE_TYPE_ME:
 946                 return "ME";
 947         case LNET_COOKIE_TYPE_EQ:
 948                 return "EQ";
 949         }
 950 }
 951
 952 static void
 953 lnet_res_container_cleanup(struct lnet_res_container *rec)
 954 {
 955         int     count = 0;
 956
 957         if (rec->rec_type == 0) /* not set yet, it's uninitialized */
 958                 return;
 959
 960         while (!list_empty(&rec->rec_active)) {
 961                 struct list_head *e = rec->rec_active.next;
 962
 963                 list_del_init(e);
 964                 if (rec->rec_type == LNET_COOKIE_TYPE_EQ) {
 965                         lnet_eq_free(list_entry(e, struct lnet_eq, eq_list));
 966
 967                 } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
 968                         lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
 969
 970                 } else { /* NB: Active MEs should be attached on portals */
 971                         LBUG();
 972                 }
 973                 count++;
 974         }
 975
 976         if (count > 0) {
 977                 /* Found alive MD/ME/EQ, user really should unlink/free
 978                  * all of them before finalize LNet, but if someone didn't,
 979                  * we have to recycle garbage for him */
 980                 CERROR("%d active elements on exit of %s container\n",
 981                        count, lnet_res_type2str(rec->rec_type));
 982         }
 983
 984         if (rec->rec_lh_hash != NULL) {
 985                 LIBCFS_FREE(rec->rec_lh_hash,
 986                             LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
 987                 rec->rec_lh_hash = NULL;
 988         }
 989
 990         rec->rec_type = 0; /* mark it as finalized */
 991 }
 992
 993 static int
 994 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
 995 {
 996         int     rc = 0;
 997         int     i;
 998
 999         LASSERT(rec->rec_type == 0);
1000
1001         rec->rec_type = type;
1002         INIT_LIST_HEAD(&rec->rec_active);
1003
1004         rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1005
1006         /* Arbitrary choice of hash table size */
1007         LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1008                          LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1009         if (rec->rec_lh_hash == NULL) {
1010                 rc = -ENOMEM;
1011                 goto out;
1012         }
1013
1014         for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1015                 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1016
1017         return 0;
1018
1019 out:
1020         CERROR("Failed to setup %s resource container\n",
1021                lnet_res_type2str(type));
1022         lnet_res_container_cleanup(rec);
1023         return rc;
1024 }
1025
1026 static void
1027 lnet_res_containers_destroy(struct lnet_res_container **recs)
1028 {
1029         struct lnet_res_container       *rec;
1030         int                             i;
1031
1032         cfs_percpt_for_each(rec, i, recs)
1033                 lnet_res_container_cleanup(rec);
1034
1035         cfs_percpt_free(recs);
1036 }
1037
1038 static struct lnet_res_container **
1039 lnet_res_containers_create(int type)
1040 {
1041         struct lnet_res_container       **recs;
1042         struct lnet_res_container       *rec;
1043         int                             rc;
1044         int                             i;
1045
1046         recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1047         if (recs == NULL) {
1048                 CERROR("Failed to allocate %s resource containers\n",
1049                        lnet_res_type2str(type));
1050                 return NULL;
1051         }
1052
1053         cfs_percpt_for_each(rec, i, recs) {
1054                 rc = lnet_res_container_setup(rec, i, type);
1055                 if (rc != 0) {
1056                         lnet_res_containers_destroy(recs);
1057                         return NULL;
1058                 }
1059         }
1060
1061         return recs;
1062 }
1063
1064 struct lnet_libhandle *
1065 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1066 {
1067         /* ALWAYS called with lnet_res_lock held */
1068         struct list_head        *head;
1069         struct lnet_libhandle   *lh;
1070         unsigned int            hash;
1071
1072         if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1073                 return NULL;
1074
1075         hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1076         head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1077
1078         list_for_each_entry(lh, head, lh_hash_chain) {
1079                 if (lh->lh_cookie == cookie)
1080                         return lh;
1081         }
1082
1083         return NULL;
1084 }
1085
1086 void
1087 lnet_res_lh_initialize(struct lnet_res_container *rec,
1088                        struct lnet_libhandle *lh)
1089 {
1090         /* ALWAYS called with lnet_res_lock held */
1091         unsigned int    ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1092         unsigned int    hash;
1093
1094         lh->lh_cookie = rec->rec_lh_cookie;
1095         rec->rec_lh_cookie += 1 << ibits;
1096
1097         hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1098
1099         list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1100 }
1101
1102 struct list_head **
1103 lnet_create_array_of_queues(void)
1104 {
1105         struct list_head **qs;
1106         struct list_head *q;
1107         int i;
1108
1109         qs = cfs_percpt_alloc(lnet_cpt_table(),
1110                               sizeof(struct list_head));
1111         if (!qs) {
1112                 CERROR("Failed to allocate queues\n");
1113                 return NULL;
1114         }
1115
1116         cfs_percpt_for_each(q, i, qs)
1117                 INIT_LIST_HEAD(q);
1118
1119         return qs;
1120 }
1121
1122 static int lnet_unprepare(void);
1123
1124 static int
1125 lnet_prepare(lnet_pid_t requested_pid)
1126 {
1127         /* Prepare to bring up the network */
1128         struct lnet_res_container **recs;
1129         int                       rc = 0;
1130
1131         if (requested_pid == LNET_PID_ANY) {
1132                 /* Don't instantiate LNET just for me */
1133                 return -ENETDOWN;
1134         }
1135
1136         LASSERT(the_lnet.ln_refcount == 0);
1137
1138         the_lnet.ln_routing = 0;
1139
1140         LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1141         the_lnet.ln_pid = requested_pid;
1142
1143         INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1144         INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1145         INIT_LIST_HEAD(&the_lnet.ln_nets);
1146         INIT_LIST_HEAD(&the_lnet.ln_routers);
1147         INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1148         INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1149         INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1150         INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1151         INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1152         INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1153         INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1154         init_waitqueue_head(&the_lnet.ln_dc_waitq);
1155         LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
1156
1157         rc = lnet_descriptor_setup();
1158         if (rc != 0)
1159                 goto failed;
1160
1161         rc = lnet_create_remote_nets_table();
1162         if (rc != 0)
1163                 goto failed;
1164
1165         /*
1166          * NB the interface cookie in wire handles guards against delayed
1167          * replies and ACKs appearing valid after reboot.
1168          */
1169         the_lnet.ln_interface_cookie = ktime_get_real_ns();
1170
1171         the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1172                                                 sizeof(struct lnet_counters));
1173         if (the_lnet.ln_counters == NULL) {
1174                 CERROR("Failed to allocate counters for LNet\n");
1175                 rc = -ENOMEM;
1176                 goto failed;
1177         }
1178
1179         rc = lnet_peer_tables_create();
1180         if (rc != 0)
1181                 goto failed;
1182
1183         rc = lnet_msg_containers_create();
1184         if (rc != 0)
1185                 goto failed;
1186
1187         rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1188                                       LNET_COOKIE_TYPE_EQ);
1189         if (rc != 0)
1190                 goto failed;
1191
1192         recs = lnet_res_containers_create(LNET_COOKIE_TYPE_ME);
1193         if (recs == NULL) {
1194                 rc = -ENOMEM;
1195                 goto failed;
1196         }
1197
1198         the_lnet.ln_me_containers = recs;
1199
1200         recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1201         if (recs == NULL) {
1202                 rc = -ENOMEM;
1203                 goto failed;
1204         }
1205
1206         the_lnet.ln_md_containers = recs;
1207
1208         rc = lnet_portals_create();
1209         if (rc != 0) {
1210                 CERROR("Failed to create portals for LNet: %d\n", rc);
1211                 goto failed;
1212         }
1213
1214         the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1215         if (!the_lnet.ln_mt_zombie_rstqs) {
1216                 rc = -ENOMEM;
1217                 goto failed;
1218         }
1219
1220         return 0;
1221
1222  failed:
1223         lnet_unprepare();
1224         return rc;
1225 }
1226
1227 static int
1228 lnet_unprepare (void)
1229 {
1230         int rc;
1231
1232         /* NB no LNET_LOCK since this is the last reference.  All LND instances
1233          * have shut down already, so it is safe to unlink and free all
1234          * descriptors, even those that appear committed to a network op (eg MD
1235          * with non-zero pending count) */
1236
1237         lnet_fail_nid(LNET_NID_ANY, 0);
1238
1239         LASSERT(the_lnet.ln_refcount == 0);
1240         LASSERT(list_empty(&the_lnet.ln_test_peers));
1241         LASSERT(list_empty(&the_lnet.ln_nets));
1242
1243         if (the_lnet.ln_mt_zombie_rstqs) {
1244                 lnet_clean_zombie_rstqs();
1245                 the_lnet.ln_mt_zombie_rstqs = NULL;
1246         }
1247
1248         if (!LNetEQHandleIsInvalid(the_lnet.ln_mt_eqh)) {
1249                 rc = LNetEQFree(the_lnet.ln_mt_eqh);
1250                 LNetInvalidateEQHandle(&the_lnet.ln_mt_eqh);
1251                 LASSERT(rc == 0);
1252         }
1253
1254         lnet_portals_destroy();
1255
1256         if (the_lnet.ln_md_containers != NULL) {
1257                 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1258                 the_lnet.ln_md_containers = NULL;
1259         }
1260
1261         if (the_lnet.ln_me_containers != NULL) {
1262                 lnet_res_containers_destroy(the_lnet.ln_me_containers);
1263                 the_lnet.ln_me_containers = NULL;
1264         }
1265
1266         lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1267
1268         lnet_msg_containers_destroy();
1269         lnet_peer_uninit();
1270         lnet_rtrpools_free(0);
1271
1272         if (the_lnet.ln_counters != NULL) {
1273                 cfs_percpt_free(the_lnet.ln_counters);
1274                 the_lnet.ln_counters = NULL;
1275         }
1276         lnet_destroy_remote_nets_table();
1277         lnet_descriptor_cleanup();
1278
1279         return 0;
1280 }
1281
1282 struct lnet_ni  *
1283 lnet_net2ni_locked(__u32 net_id, int cpt)
1284 {
1285         struct lnet_ni   *ni;
1286         struct lnet_net  *net;
1287
1288         LASSERT(cpt != LNET_LOCK_EX);
1289
1290         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1291                 if (net->net_id == net_id) {
1292                         ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1293                                         ni_netlist);
1294                         return ni;
1295                 }
1296         }
1297
1298         return NULL;
1299 }
1300
1301 struct lnet_ni *
1302 lnet_net2ni_addref(__u32 net)
1303 {
1304         struct lnet_ni *ni;
1305
1306         lnet_net_lock(0);
1307         ni = lnet_net2ni_locked(net, 0);
1308         if (ni)
1309                 lnet_ni_addref_locked(ni, 0);
1310         lnet_net_unlock(0);
1311
1312         return ni;
1313 }
1314 EXPORT_SYMBOL(lnet_net2ni_addref);
1315
1316 struct lnet_net *
1317 lnet_get_net_locked(__u32 net_id)
1318 {
1319         struct lnet_net  *net;
1320
1321         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1322                 if (net->net_id == net_id)
1323                         return net;
1324         }
1325
1326         return NULL;
1327 }
1328
1329 unsigned int
1330 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1331 {
1332         __u64           key = nid;
1333         unsigned int    val;
1334
1335         LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1336
1337         if (number == 1)
1338                 return 0;
1339
1340         val = hash_long(key, LNET_CPT_BITS);
1341         /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1342         if (val < number)
1343                 return val;
1344
1345         return (unsigned int)(key + val + (val >> 1)) % number;
1346 }
1347
1348 int
1349 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1350 {
1351         struct lnet_net *net;
1352
1353         /* must called with hold of lnet_net_lock */
1354         if (LNET_CPT_NUMBER == 1)
1355                 return 0; /* the only one */
1356
1357         /*
1358          * If NI is provided then use the CPT identified in the NI cpt
1359          * list if one exists. If one doesn't exist, then that NI is
1360          * associated with all CPTs and it follows that the net it belongs
1361          * to is implicitly associated with all CPTs, so just hash the nid
1362          * and return that.
1363          */
1364         if (ni != NULL) {
1365                 if (ni->ni_cpts != NULL)
1366                         return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1367                                                              ni->ni_ncpts)];
1368                 else
1369                         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1370         }
1371
1372         /* no NI provided so look at the net */
1373         net = lnet_get_net_locked(LNET_NIDNET(nid));
1374
1375         if (net != NULL && net->net_cpts != NULL) {
1376                 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1377         }
1378
1379         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1380 }
1381
1382 int
1383 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1384 {
1385         int     cpt;
1386         int     cpt2;
1387
1388         if (LNET_CPT_NUMBER == 1)
1389                 return 0; /* the only one */
1390
1391         cpt = lnet_net_lock_current();
1392
1393         cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1394
1395         lnet_net_unlock(cpt);
1396
1397         return cpt2;
1398 }
1399 EXPORT_SYMBOL(lnet_cpt_of_nid);
1400
1401 int
1402 lnet_islocalnet(__u32 net_id)
1403 {
1404         struct lnet_net *net;
1405         int             cpt;
1406         bool            local;
1407
1408         cpt = lnet_net_lock_current();
1409
1410         net = lnet_get_net_locked(net_id);
1411
1412         local = net != NULL;
1413
1414         lnet_net_unlock(cpt);
1415
1416         return local;
1417 }
1418
1419 struct lnet_ni  *
1420 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
1421 {
1422         struct lnet_net  *net;
1423         struct lnet_ni   *ni;
1424
1425         LASSERT(cpt != LNET_LOCK_EX);
1426
1427         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1428                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1429                         if (ni->ni_nid == nid)
1430                                 return ni;
1431                 }
1432         }
1433
1434         return NULL;
1435 }
1436
1437 struct lnet_ni *
1438 lnet_nid2ni_addref(lnet_nid_t nid)
1439 {
1440         struct lnet_ni *ni;
1441
1442         lnet_net_lock(0);
1443         ni = lnet_nid2ni_locked(nid, 0);
1444         if (ni)
1445                 lnet_ni_addref_locked(ni, 0);
1446         lnet_net_unlock(0);
1447
1448         return ni;
1449 }
1450 EXPORT_SYMBOL(lnet_nid2ni_addref);
1451
1452 int
1453 lnet_islocalnid(lnet_nid_t nid)
1454 {
1455         struct lnet_ni  *ni;
1456         int             cpt;
1457
1458         cpt = lnet_net_lock_current();
1459         ni = lnet_nid2ni_locked(nid, cpt);
1460         lnet_net_unlock(cpt);
1461
1462         return ni != NULL;
1463 }
1464
1465 int
1466 lnet_count_acceptor_nets(void)
1467 {
1468         /* Return the # of NIs that need the acceptor. */
1469         int              count = 0;
1470         struct lnet_net  *net;
1471         int              cpt;
1472
1473         cpt = lnet_net_lock_current();
1474         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1475                 /* all socklnd type networks should have the acceptor
1476                  * thread started */
1477                 if (net->net_lnd->lnd_accept != NULL)
1478                         count++;
1479         }
1480
1481         lnet_net_unlock(cpt);
1482
1483         return count;
1484 }
1485
1486 struct lnet_ping_buffer *
1487 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1488 {
1489         struct lnet_ping_buffer *pbuf;
1490
1491         LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1492         if (pbuf) {
1493                 pbuf->pb_nnis = nnis;
1494                 atomic_set(&pbuf->pb_refcnt, 1);
1495         }
1496
1497         return pbuf;
1498 }
1499
1500 void
1501 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1502 {
1503         LASSERT(lnet_ping_buffer_numref(pbuf) == 0);
1504         LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1505 }
1506
1507 static struct lnet_ping_buffer *
1508 lnet_ping_target_create(int nnis)
1509 {
1510         struct lnet_ping_buffer *pbuf;
1511
1512         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1513         if (pbuf == NULL) {
1514                 CERROR("Can't allocate ping source [%d]\n", nnis);
1515                 return NULL;
1516         }
1517
1518         pbuf->pb_info.pi_nnis = nnis;
1519         pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1520         pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1521         pbuf->pb_info.pi_features =
1522                 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1523
1524         return pbuf;
1525 }
1526
1527 static inline int
1528 lnet_get_net_ni_count_locked(struct lnet_net *net)
1529 {
1530         struct lnet_ni  *ni;
1531         int             count = 0;
1532
1533         list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1534                 count++;
1535
1536         return count;
1537 }
1538
1539 static inline int
1540 lnet_get_net_ni_count_pre(struct lnet_net *net)
1541 {
1542         struct lnet_ni  *ni;
1543         int             count = 0;
1544
1545         list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1546                 count++;
1547
1548         return count;
1549 }
1550
1551 static inline int
1552 lnet_get_ni_count(void)
1553 {
1554         struct lnet_ni  *ni;
1555         struct lnet_net *net;
1556         int             count = 0;
1557
1558         lnet_net_lock(0);
1559
1560         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1561                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1562                         count++;
1563         }
1564
1565         lnet_net_unlock(0);
1566
1567         return count;
1568 }
1569
1570 int
1571 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1572 {
1573         if (!pinfo)
1574                 return -EINVAL;
1575         if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1576                 return -EPROTO;
1577         if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1578                 return -EPROTO;
1579         /* Loopback is guaranteed to be present */
1580         if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1581                 return -ERANGE;
1582         if (LNET_PING_INFO_LONI(pinfo) != LNET_NID_LO_0)
1583                 return -EPROTO;
1584         return 0;
1585 }
1586
1587 static void
1588 lnet_ping_target_destroy(void)
1589 {
1590         struct lnet_net *net;
1591         struct lnet_ni  *ni;
1592
1593         lnet_net_lock(LNET_LOCK_EX);
1594
1595         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1596                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1597                         lnet_ni_lock(ni);
1598                         ni->ni_status = NULL;
1599                         lnet_ni_unlock(ni);
1600                 }
1601         }
1602
1603         lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1604         the_lnet.ln_ping_target = NULL;
1605
1606         lnet_net_unlock(LNET_LOCK_EX);
1607 }
1608
1609 static void
1610 lnet_ping_target_event_handler(struct lnet_event *event)
1611 {
1612         struct lnet_ping_buffer *pbuf = event->md.user_ptr;
1613
1614         if (event->unlinked)
1615                 lnet_ping_buffer_decref(pbuf);
1616 }
1617
1618 static int
1619 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1620                        struct lnet_handle_md *ping_mdh,
1621                        int ni_count, bool set_eq)
1622 {
1623         struct lnet_process_id id = {
1624                 .nid = LNET_NID_ANY,
1625                 .pid = LNET_PID_ANY
1626         };
1627         struct lnet_handle_me me_handle;
1628         struct lnet_md md = { NULL };
1629         int rc, rc2;
1630
1631         if (set_eq) {
1632                 rc = LNetEQAlloc(0, lnet_ping_target_event_handler,
1633                                  &the_lnet.ln_ping_target_eq);
1634                 if (rc != 0) {
1635                         CERROR("Can't allocate ping buffer EQ: %d\n", rc);
1636                         return rc;
1637                 }
1638         }
1639
1640         *ppbuf = lnet_ping_target_create(ni_count);
1641         if (*ppbuf == NULL) {
1642                 rc = -ENOMEM;
1643                 goto fail_free_eq;
1644         }
1645
1646         /* Ping target ME/MD */
1647         rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1648                           LNET_PROTO_PING_MATCHBITS, 0,
1649                           LNET_UNLINK, LNET_INS_AFTER,
1650                           &me_handle);
1651         if (rc != 0) {
1652                 CERROR("Can't create ping target ME: %d\n", rc);
1653                 goto fail_decref_ping_buffer;
1654         }
1655
1656         /* initialize md content */
1657         md.start     = &(*ppbuf)->pb_info;
1658         md.length    = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1659         md.threshold = LNET_MD_THRESH_INF;
1660         md.max_size  = 0;
1661         md.options   = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1662                        LNET_MD_MANAGE_REMOTE;
1663         md.eq_handle = the_lnet.ln_ping_target_eq;
1664         md.user_ptr  = *ppbuf;
1665
1666         rc = LNetMDAttach(me_handle, md, LNET_RETAIN, ping_mdh);
1667         if (rc != 0) {
1668                 CERROR("Can't attach ping target MD: %d\n", rc);
1669                 goto fail_unlink_ping_me;
1670         }
1671         lnet_ping_buffer_addref(*ppbuf);
1672
1673         return 0;
1674
1675 fail_unlink_ping_me:
1676         rc2 = LNetMEUnlink(me_handle);
1677         LASSERT(rc2 == 0);
1678 fail_decref_ping_buffer:
1679         LASSERT(lnet_ping_buffer_numref(*ppbuf) == 1);
1680         lnet_ping_buffer_decref(*ppbuf);
1681         *ppbuf = NULL;
1682 fail_free_eq:
1683         if (set_eq) {
1684                 rc2 = LNetEQFree(the_lnet.ln_ping_target_eq);
1685                 LASSERT(rc2 == 0);
1686         }
1687         return rc;
1688 }
1689
1690 static void
1691 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1692                     struct lnet_handle_md *ping_mdh)
1693 {
1694         sigset_t        blocked = cfs_block_allsigs();
1695
1696         LNetMDUnlink(*ping_mdh);
1697         LNetInvalidateMDHandle(ping_mdh);
1698
1699         /* NB the MD could be busy; this just starts the unlink */
1700         while (lnet_ping_buffer_numref(pbuf) > 1) {
1701                 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1702                 set_current_state(TASK_UNINTERRUPTIBLE);
1703                 schedule_timeout(cfs_time_seconds(1));
1704         }
1705
1706         cfs_restore_sigs(blocked);
1707 }
1708
1709 static void
1710 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1711 {
1712         struct lnet_ni          *ni;
1713         struct lnet_net         *net;
1714         struct lnet_ni_status *ns;
1715         int                     i;
1716         int                     rc;
1717
1718         i = 0;
1719         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1720                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1721                         LASSERT(i < pbuf->pb_nnis);
1722
1723                         ns = &pbuf->pb_info.pi_ni[i];
1724
1725                         ns->ns_nid = ni->ni_nid;
1726
1727                         lnet_ni_lock(ni);
1728                         ns->ns_status = (ni->ni_status != NULL) ?
1729                                          ni->ni_status->ns_status :
1730                                                 LNET_NI_STATUS_UP;
1731                         ni->ni_status = ns;
1732                         lnet_ni_unlock(ni);
1733
1734                         i++;
1735                 }
1736         }
1737         /*
1738          * We (ab)use the ns_status of the loopback interface to
1739          * transmit the sequence number. The first interface listed
1740          * must be the loopback interface.
1741          */
1742         rc = lnet_ping_info_validate(&pbuf->pb_info);
1743         if (rc) {
1744                 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1745                 LBUG();
1746         }
1747         LNET_PING_BUFFER_SEQNO(pbuf) =
1748                 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1749 }
1750
1751 static void
1752 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1753                         struct lnet_handle_md ping_mdh)
1754 {
1755         struct lnet_ping_buffer *old_pbuf = NULL;
1756         struct lnet_handle_md old_ping_md;
1757
1758         /* switch the NIs to point to the new ping info created */
1759         lnet_net_lock(LNET_LOCK_EX);
1760
1761         if (!the_lnet.ln_routing)
1762                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1763         if (!lnet_peer_discovery_disabled)
1764                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1765
1766         /* Ensure only known feature bits have been set. */
1767         LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1768         LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1769
1770         lnet_ping_target_install_locked(pbuf);
1771
1772         if (the_lnet.ln_ping_target) {
1773                 old_pbuf = the_lnet.ln_ping_target;
1774                 old_ping_md = the_lnet.ln_ping_target_md;
1775         }
1776         the_lnet.ln_ping_target_md = ping_mdh;
1777         the_lnet.ln_ping_target = pbuf;
1778
1779         lnet_net_unlock(LNET_LOCK_EX);
1780
1781         if (old_pbuf) {
1782                 /* unlink and free the old ping info */
1783                 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1784                 lnet_ping_buffer_decref(old_pbuf);
1785         }
1786
1787         lnet_push_update_to_peers(0);
1788 }
1789
1790 static void
1791 lnet_ping_target_fini(void)
1792 {
1793         int             rc;
1794
1795         lnet_ping_md_unlink(the_lnet.ln_ping_target,
1796                             &the_lnet.ln_ping_target_md);
1797
1798         rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1799         LASSERT(rc == 0);
1800
1801         lnet_ping_target_destroy();
1802 }
1803
1804 /* Resize the push target. */
1805 int lnet_push_target_resize(void)
1806 {
1807         struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
1808         struct lnet_md md = { NULL };
1809         struct lnet_handle_me meh;
1810         struct lnet_handle_md mdh;
1811         struct lnet_handle_md old_mdh;
1812         struct lnet_ping_buffer *pbuf;
1813         struct lnet_ping_buffer *old_pbuf;
1814         int nnis = the_lnet.ln_push_target_nnis;
1815         int rc;
1816
1817         if (nnis <= 0) {
1818                 rc = -EINVAL;
1819                 goto fail_return;
1820         }
1821 again:
1822         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1823         if (!pbuf) {
1824                 rc = -ENOMEM;
1825                 goto fail_return;
1826         }
1827
1828         rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1829                           LNET_PROTO_PING_MATCHBITS, 0,
1830                           LNET_UNLINK, LNET_INS_AFTER,
1831                           &meh);
1832         if (rc) {
1833                 CERROR("Can't create push target ME: %d\n", rc);
1834                 goto fail_decref_pbuf;
1835         }
1836
1837         /* initialize md content */
1838         md.start     = &pbuf->pb_info;
1839         md.length    = LNET_PING_INFO_SIZE(nnis);
1840         md.threshold = LNET_MD_THRESH_INF;
1841         md.max_size  = 0;
1842         md.options   = LNET_MD_OP_PUT | LNET_MD_TRUNCATE |
1843                        LNET_MD_MANAGE_REMOTE;
1844         md.user_ptr  = pbuf;
1845         md.eq_handle = the_lnet.ln_push_target_eq;
1846
1847         rc = LNetMDAttach(meh, md, LNET_RETAIN, &mdh);
1848         if (rc) {
1849                 CERROR("Can't attach push MD: %d\n", rc);
1850                 goto fail_unlink_meh;
1851         }
1852         lnet_ping_buffer_addref(pbuf);
1853
1854         lnet_net_lock(LNET_LOCK_EX);
1855         old_pbuf = the_lnet.ln_push_target;
1856         old_mdh = the_lnet.ln_push_target_md;
1857         the_lnet.ln_push_target = pbuf;
1858         the_lnet.ln_push_target_md = mdh;
1859         lnet_net_unlock(LNET_LOCK_EX);
1860
1861         if (old_pbuf) {
1862                 LNetMDUnlink(old_mdh);
1863                 lnet_ping_buffer_decref(old_pbuf);
1864         }
1865
1866         if (nnis < the_lnet.ln_push_target_nnis)
1867                 goto again;
1868
1869         CDEBUG(D_NET, "nnis %d success\n", nnis);
1870
1871         return 0;
1872
1873 fail_unlink_meh:
1874         LNetMEUnlink(meh);
1875 fail_decref_pbuf:
1876         lnet_ping_buffer_decref(pbuf);
1877 fail_return:
1878         CDEBUG(D_NET, "nnis %d error %d\n", nnis, rc);
1879         return rc;
1880 }
1881
1882 static void lnet_push_target_event_handler(struct lnet_event *ev)
1883 {
1884         struct lnet_ping_buffer *pbuf = ev->md.user_ptr;
1885
1886         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
1887                 lnet_swap_pinginfo(pbuf);
1888
1889         lnet_peer_push_event(ev);
1890         if (ev->unlinked)
1891                 lnet_ping_buffer_decref(pbuf);
1892 }
1893
1894 /* Initialize the push target. */
1895 static int lnet_push_target_init(void)
1896 {
1897         int rc;
1898
1899         if (the_lnet.ln_push_target)
1900                 return -EALREADY;
1901
1902         rc = LNetEQAlloc(0, lnet_push_target_event_handler,
1903                          &the_lnet.ln_push_target_eq);
1904         if (rc) {
1905                 CERROR("Can't allocated push target EQ: %d\n", rc);
1906                 return rc;
1907         }
1908
1909         /* Start at the required minimum, we'll enlarge if required. */
1910         the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
1911
1912         rc = lnet_push_target_resize();
1913
1914         if (rc) {
1915                 LNetEQFree(the_lnet.ln_push_target_eq);
1916                 LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1917         }
1918
1919         return rc;
1920 }
1921
1922 /* Clean up the push target. */
1923 static void lnet_push_target_fini(void)
1924 {
1925         if (!the_lnet.ln_push_target)
1926                 return;
1927
1928         /* Unlink and invalidate to prevent new references. */
1929         LNetMDUnlink(the_lnet.ln_push_target_md);
1930         LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
1931
1932         /* Wait for the unlink to complete. */
1933         while (lnet_ping_buffer_numref(the_lnet.ln_push_target) > 1) {
1934                 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1935                 set_current_state(TASK_UNINTERRUPTIBLE);
1936                 schedule_timeout(cfs_time_seconds(1));
1937         }
1938
1939         lnet_ping_buffer_decref(the_lnet.ln_push_target);
1940         the_lnet.ln_push_target = NULL;
1941         the_lnet.ln_push_target_nnis = 0;
1942
1943         LNetEQFree(the_lnet.ln_push_target_eq);
1944         LNetInvalidateEQHandle(&the_lnet.ln_push_target_eq);
1945 }
1946
1947 static int
1948 lnet_ni_tq_credits(struct lnet_ni *ni)
1949 {
1950         int     credits;
1951
1952         LASSERT(ni->ni_ncpts >= 1);
1953
1954         if (ni->ni_ncpts == 1)
1955                 return ni->ni_net->net_tunables.lct_max_tx_credits;
1956
1957         credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
1958         credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
1959         credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
1960
1961         return credits;
1962 }
1963
1964 static void
1965 lnet_ni_unlink_locked(struct lnet_ni *ni)
1966 {
1967         /* move it to zombie list and nobody can find it anymore */
1968         LASSERT(!list_empty(&ni->ni_netlist));
1969         list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
1970         lnet_ni_decref_locked(ni, 0);
1971 }
1972
1973 static void
1974 lnet_clear_zombies_nis_locked(struct lnet_net *net)
1975 {
1976         int             i;
1977         int             islo;
1978         struct lnet_ni  *ni;
1979         struct list_head *zombie_list = &net->net_ni_zombie;
1980
1981         /*
1982          * Now wait for the NIs I just nuked to show up on the zombie
1983          * list and shut them down in guaranteed thread context
1984          */
1985         i = 2;
1986         while (!list_empty(zombie_list)) {
1987                 int     *ref;
1988                 int     j;
1989
1990                 ni = list_entry(zombie_list->next,
1991                                 struct lnet_ni, ni_netlist);
1992                 list_del_init(&ni->ni_netlist);
1993                 /* the ni should be in deleting state. If it's not it's
1994                  * a bug */
1995                 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
1996                 cfs_percpt_for_each(ref, j, ni->ni_refs) {
1997                         if (*ref == 0)
1998                                 continue;
1999                         /* still busy, add it back to zombie list */
2000                         list_add(&ni->ni_netlist, zombie_list);
2001                         break;
2002                 }
2003
2004                 if (!list_empty(&ni->ni_netlist)) {
2005                         /* Unlock mutex while waiting to allow other
2006                          * threads to read the LNet state and fall through
2007                          * to avoid deadlock
2008                          */
2009                         lnet_net_unlock(LNET_LOCK_EX);
2010                         mutex_unlock(&the_lnet.ln_api_mutex);
2011
2012                         ++i;
2013                         if ((i & (-i)) == i) {
2014                                 CDEBUG(D_WARNING,
2015                                        "Waiting for zombie LNI %s\n",
2016                                        libcfs_nid2str(ni->ni_nid));
2017                         }
2018                         set_current_state(TASK_UNINTERRUPTIBLE);
2019                         schedule_timeout(cfs_time_seconds(1));
2020
2021                         mutex_lock(&the_lnet.ln_api_mutex);
2022                         lnet_net_lock(LNET_LOCK_EX);
2023                         continue;
2024                 }
2025
2026                 lnet_net_unlock(LNET_LOCK_EX);
2027
2028                 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2029
2030                 LASSERT(!in_interrupt());
2031                 (net->net_lnd->lnd_shutdown)(ni);
2032
2033                 if (!islo)
2034                         CDEBUG(D_LNI, "Removed LNI %s\n",
2035                               libcfs_nid2str(ni->ni_nid));
2036
2037                 lnet_ni_free(ni);
2038                 i = 2;
2039                 lnet_net_lock(LNET_LOCK_EX);
2040         }
2041 }
2042
2043 /* shutdown down the NI and release refcount */
2044 static void
2045 lnet_shutdown_lndni(struct lnet_ni *ni)
2046 {
2047         int i;
2048         struct lnet_net *net = ni->ni_net;
2049
2050         lnet_net_lock(LNET_LOCK_EX);
2051         lnet_ni_lock(ni);
2052         ni->ni_state = LNET_NI_STATE_DELETING;
2053         lnet_ni_unlock(ni);
2054         lnet_ni_unlink_locked(ni);
2055         lnet_incr_dlc_seq();
2056         lnet_net_unlock(LNET_LOCK_EX);
2057
2058         /* clear messages for this NI on the lazy portal */
2059         for (i = 0; i < the_lnet.ln_nportals; i++)
2060                 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2061
2062         lnet_net_lock(LNET_LOCK_EX);
2063         lnet_clear_zombies_nis_locked(net);
2064         lnet_net_unlock(LNET_LOCK_EX);
2065 }
2066
2067 static void
2068 lnet_shutdown_lndnet(struct lnet_net *net)
2069 {
2070         struct lnet_ni *ni;
2071
2072         lnet_net_lock(LNET_LOCK_EX);
2073
2074         net->net_state = LNET_NET_STATE_DELETING;
2075
2076         list_del_init(&net->net_list);
2077
2078         while (!list_empty(&net->net_ni_list)) {
2079                 ni = list_entry(net->net_ni_list.next,
2080                                 struct lnet_ni, ni_netlist);
2081                 lnet_net_unlock(LNET_LOCK_EX);
2082                 lnet_shutdown_lndni(ni);
2083                 lnet_net_lock(LNET_LOCK_EX);
2084         }
2085
2086         lnet_net_unlock(LNET_LOCK_EX);
2087
2088         /* Do peer table cleanup for this net */
2089         lnet_peer_tables_cleanup(net);
2090
2091         lnet_net_lock(LNET_LOCK_EX);
2092         /*
2093          * decrement ref count on lnd only when the entire network goes
2094          * away
2095          */
2096         net->net_lnd->lnd_refcount--;
2097
2098         lnet_net_unlock(LNET_LOCK_EX);
2099
2100         lnet_net_free(net);
2101 }
2102
2103 static void
2104 lnet_shutdown_lndnets(void)
2105 {
2106         struct lnet_net *net;
2107         struct list_head resend;
2108         struct lnet_msg *msg, *tmp;
2109
2110         INIT_LIST_HEAD(&resend);
2111
2112         /* NB called holding the global mutex */
2113
2114         /* All quiet on the API front */
2115         LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2116         LASSERT(the_lnet.ln_refcount == 0);
2117
2118         lnet_net_lock(LNET_LOCK_EX);
2119         the_lnet.ln_state = LNET_STATE_STOPPING;
2120
2121         while (!list_empty(&the_lnet.ln_nets)) {
2122                 /*
2123                  * move the nets to the zombie list to avoid them being
2124                  * picked up for new work. LONET is also included in the
2125                  * Nets that will be moved to the zombie list
2126                  */
2127                 net = list_entry(the_lnet.ln_nets.next,
2128                                  struct lnet_net, net_list);
2129                 list_move(&net->net_list, &the_lnet.ln_net_zombie);
2130         }
2131
2132         /* Drop the cached loopback Net. */
2133         if (the_lnet.ln_loni != NULL) {
2134                 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2135                 the_lnet.ln_loni = NULL;
2136         }
2137         lnet_net_unlock(LNET_LOCK_EX);
2138
2139         /* iterate through the net zombie list and delete each net */
2140         while (!list_empty(&the_lnet.ln_net_zombie)) {
2141                 net = list_entry(the_lnet.ln_net_zombie.next,
2142                                  struct lnet_net, net_list);
2143                 lnet_shutdown_lndnet(net);
2144         }
2145
2146         spin_lock(&the_lnet.ln_msg_resend_lock);
2147         list_splice(&the_lnet.ln_msg_resend, &resend);
2148         spin_unlock(&the_lnet.ln_msg_resend_lock);
2149
2150         list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2151                 list_del_init(&msg->msg_list);
2152                 msg->msg_no_resend = true;
2153                 lnet_finalize(msg, -ECANCELED);
2154         }
2155
2156         lnet_net_lock(LNET_LOCK_EX);
2157         the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2158         lnet_net_unlock(LNET_LOCK_EX);
2159 }
2160
2161 static int
2162 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2163 {
2164         int                     rc = -EINVAL;
2165         struct lnet_tx_queue    *tq;
2166         int                     i;
2167         struct lnet_net         *net = ni->ni_net;
2168
2169         mutex_lock(&the_lnet.ln_lnd_mutex);
2170
2171         if (tun) {
2172                 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2173                 ni->ni_lnd_tunables_set = true;
2174         }
2175
2176         rc = (net->net_lnd->lnd_startup)(ni);
2177
2178         mutex_unlock(&the_lnet.ln_lnd_mutex);
2179
2180         if (rc != 0) {
2181                 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2182                                    rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2183                 lnet_net_lock(LNET_LOCK_EX);
2184                 net->net_lnd->lnd_refcount--;
2185                 lnet_net_unlock(LNET_LOCK_EX);
2186                 goto failed0;
2187         }
2188
2189         lnet_ni_lock(ni);
2190         ni->ni_state = LNET_NI_STATE_ACTIVE;
2191         lnet_ni_unlock(ni);
2192
2193         /* We keep a reference on the loopback net through the loopback NI */
2194         if (net->net_lnd->lnd_type == LOLND) {
2195                 lnet_ni_addref(ni);
2196                 LASSERT(the_lnet.ln_loni == NULL);
2197                 the_lnet.ln_loni = ni;
2198                 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2199                 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2200                 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2201                 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2202                 return 0;
2203         }
2204
2205         if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2206             ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2207                 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2208                                    libcfs_lnd2str(net->net_lnd->lnd_type),
2209                                    ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2210                                         "" : "per-peer ");
2211                 /* shutdown the NI since if we get here then it must've already
2212                  * been started
2213                  */
2214                 lnet_shutdown_lndni(ni);
2215                 return -EINVAL;
2216         }
2217
2218         cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2219                 tq->tq_credits_min =
2220                 tq->tq_credits_max =
2221                 tq->tq_credits = lnet_ni_tq_credits(ni);
2222         }
2223
2224         atomic_set(&ni->ni_tx_credits,
2225                    lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2226         atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2227
2228         CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2229                 libcfs_nid2str(ni->ni_nid),
2230                 ni->ni_net->net_tunables.lct_peer_tx_credits,
2231                 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2232                 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2233                 ni->ni_net->net_tunables.lct_peer_timeout);
2234
2235         return 0;
2236 failed0:
2237         lnet_ni_free(ni);
2238         return rc;
2239 }
2240
2241 static int
2242 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2243 {
2244         struct lnet_ni *ni;
2245         struct lnet_net *net_l = NULL;
2246         struct list_head        local_ni_list;
2247         int                     rc;
2248         int                     ni_count = 0;
2249         __u32                   lnd_type;
2250         struct lnet_lnd *lnd;
2251         int                     peer_timeout =
2252                 net->net_tunables.lct_peer_timeout;
2253         int                     maxtxcredits =
2254                 net->net_tunables.lct_max_tx_credits;
2255         int                     peerrtrcredits =
2256                 net->net_tunables.lct_peer_rtr_credits;
2257
2258         INIT_LIST_HEAD(&local_ni_list);
2259
2260         /*
2261          * make sure that this net is unique. If it isn't then
2262          * we are adding interfaces to an already existing network, and
2263          * 'net' is just a convenient way to pass in the list.
2264          * if it is unique we need to find the LND and load it if
2265          * necessary.
2266          */
2267         if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2268                 lnd_type = LNET_NETTYP(net->net_id);
2269
2270                 mutex_lock(&the_lnet.ln_lnd_mutex);
2271                 lnd = lnet_find_lnd_by_type(lnd_type);
2272
2273                 if (lnd == NULL) {
2274                         mutex_unlock(&the_lnet.ln_lnd_mutex);
2275                         rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2276                         mutex_lock(&the_lnet.ln_lnd_mutex);
2277
2278                         lnd = lnet_find_lnd_by_type(lnd_type);
2279                         if (lnd == NULL) {
2280                                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2281                                 CERROR("Can't load LND %s, module %s, rc=%d\n",
2282                                 libcfs_lnd2str(lnd_type),
2283                                 libcfs_lnd2modname(lnd_type), rc);
2284 #ifndef HAVE_MODULE_LOADING_SUPPORT
2285                                 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2286                                                 "compiled with kernel module "
2287                                                 "loading support.");
2288 #endif
2289                                 rc = -EINVAL;
2290                                 goto failed0;
2291                         }
2292                 }
2293
2294                 lnet_net_lock(LNET_LOCK_EX);
2295                 lnd->lnd_refcount++;
2296                 lnet_net_unlock(LNET_LOCK_EX);
2297
2298                 net->net_lnd = lnd;
2299
2300                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2301
2302                 net_l = net;
2303         }
2304
2305         /*
2306          * net_l: if the network being added is unique then net_l
2307          *        will point to that network
2308          *        if the network being added is not unique then
2309          *        net_l points to the existing network.
2310          *
2311          * When we enter the loop below, we'll pick NIs off he
2312          * network beign added and start them up, then add them to
2313          * a local ni list. Once we've successfully started all
2314          * the NIs then we join the local NI list (of started up
2315          * networks) with the net_l->net_ni_list, which should
2316          * point to the correct network to add the new ni list to
2317          *
2318          * If any of the new NIs fail to start up, then we want to
2319          * iterate through the local ni list, which should include
2320          * any NIs which were successfully started up, and shut
2321          * them down.
2322          *
2323          * After than we want to delete the network being added,
2324          * to avoid a memory leak.
2325          */
2326
2327         /*
2328          * When a network uses TCP bonding then all its interfaces
2329          * must be specified when the network is first defined: the
2330          * TCP bonding code doesn't allow for interfaces to be added
2331          * or removed.
2332          */
2333         if (net_l != net && net_l != NULL && use_tcp_bonding &&
2334             LNET_NETTYP(net_l->net_id) == SOCKLND) {
2335                 rc = -EINVAL;
2336                 goto failed0;
2337         }
2338
2339         while (!list_empty(&net->net_ni_added)) {
2340                 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2341                                 ni_netlist);
2342                 list_del_init(&ni->ni_netlist);
2343
2344                 /* make sure that the the NI we're about to start
2345                  * up is actually unique. if it's not fail. */
2346                 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2347                                         ni->ni_interfaces[0])) {
2348                         rc = -EEXIST;
2349                         goto failed1;
2350                 }
2351
2352                 /* adjust the pointer the parent network, just in case it
2353                  * the net is a duplicate */
2354                 ni->ni_net = net_l;
2355
2356                 rc = lnet_startup_lndni(ni, tun);
2357
2358                 LASSERT(ni->ni_net->net_tunables.lct_peer_timeout <= 0 ||
2359                         ni->ni_net->net_lnd->lnd_query != NULL);
2360
2361                 if (rc < 0)
2362                         goto failed1;
2363
2364                 lnet_ni_addref(ni);
2365                 list_add_tail(&ni->ni_netlist, &local_ni_list);
2366
2367                 ni_count++;
2368         }
2369
2370         lnet_net_lock(LNET_LOCK_EX);
2371         list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2372         lnet_incr_dlc_seq();
2373         lnet_net_unlock(LNET_LOCK_EX);
2374
2375         /* if the network is not unique then we don't want to keep
2376          * it around after we're done. Free it. Otherwise add that
2377          * net to the global the_lnet.ln_nets */
2378         if (net_l != net && net_l != NULL) {
2379                 /*
2380                  * TODO - note. currently the tunables can not be updated
2381                  * once added
2382                  */
2383                 lnet_net_free(net);
2384         } else {
2385                 net->net_state = LNET_NET_STATE_ACTIVE;
2386                 /*
2387                  * restore tunables after it has been overwitten by the
2388                  * lnd
2389                  */
2390                 if (peer_timeout != -1)
2391                         net->net_tunables.lct_peer_timeout = peer_timeout;
2392                 if (maxtxcredits != -1)
2393                         net->net_tunables.lct_max_tx_credits = maxtxcredits;
2394                 if (peerrtrcredits != -1)
2395                         net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2396
2397                 lnet_net_lock(LNET_LOCK_EX);
2398                 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2399                 lnet_net_unlock(LNET_LOCK_EX);
2400         }
2401
2402         return ni_count;
2403
2404 failed1:
2405         /*
2406          * shutdown the new NIs that are being started up
2407          * free the NET being started
2408          */
2409         while (!list_empty(&local_ni_list)) {
2410                 ni = list_entry(local_ni_list.next, struct lnet_ni,
2411                                 ni_netlist);
2412
2413                 lnet_shutdown_lndni(ni);
2414         }
2415
2416 failed0:
2417         lnet_net_free(net);
2418
2419         return rc;
2420 }
2421
2422 static int
2423 lnet_startup_lndnets(struct list_head *netlist)
2424 {
2425         struct lnet_net         *net;
2426         int                     rc;
2427         int                     ni_count = 0;
2428
2429         /*
2430          * Change to running state before bringing up the LNDs. This
2431          * allows lnet_shutdown_lndnets() to assert that we've passed
2432          * through here.
2433          */
2434         lnet_net_lock(LNET_LOCK_EX);
2435         the_lnet.ln_state = LNET_STATE_RUNNING;
2436         lnet_net_unlock(LNET_LOCK_EX);
2437
2438         while (!list_empty(netlist)) {
2439                 net = list_entry(netlist->next, struct lnet_net, net_list);
2440                 list_del_init(&net->net_list);
2441
2442                 rc = lnet_startup_lndnet(net, NULL);
2443
2444                 if (rc < 0)
2445                         goto failed;
2446
2447                 ni_count += rc;
2448         }
2449
2450         return ni_count;
2451 failed:
2452         lnet_shutdown_lndnets();
2453
2454         return rc;
2455 }
2456
2457 /**
2458  * Initialize LNet library.
2459  *
2460  * Automatically called at module loading time. Caller has to call
2461  * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2462  * latter returned 0. It must be called exactly once.
2463  *
2464  * \retval 0 on success
2465  * \retval -ve on failures.
2466  */
2467 int lnet_lib_init(void)
2468 {
2469         int rc;
2470
2471         lnet_assert_wire_constants();
2472
2473         /* refer to global cfs_cpt_table for now */
2474         the_lnet.ln_cpt_table   = cfs_cpt_table;
2475         the_lnet.ln_cpt_number  = cfs_cpt_number(cfs_cpt_table);
2476
2477         LASSERT(the_lnet.ln_cpt_number > 0);
2478         if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2479                 /* we are under risk of consuming all lh_cookie */
2480                 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2481                        "please change setting of CPT-table and retry\n",
2482                        the_lnet.ln_cpt_number, LNET_CPT_MAX);
2483                 return -E2BIG;
2484         }
2485
2486         while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2487                 the_lnet.ln_cpt_bits++;
2488
2489         rc = lnet_create_locks();
2490         if (rc != 0) {
2491                 CERROR("Can't create LNet global locks: %d\n", rc);
2492                 return rc;
2493         }
2494
2495         the_lnet.ln_refcount = 0;
2496         LNetInvalidateEQHandle(&the_lnet.ln_rc_eqh);
2497         INIT_LIST_HEAD(&the_lnet.ln_lnds);
2498         INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2499         INIT_LIST_HEAD(&the_lnet.ln_rcd_zombie);
2500         INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2501         INIT_LIST_HEAD(&the_lnet.ln_rcd_deathrow);
2502
2503         /* The hash table size is the number of bits it takes to express the set
2504          * ln_num_routes, minus 1 (better to under estimate than over so we
2505          * don't waste memory). */
2506         if (rnet_htable_size <= 0)
2507                 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2508         else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2509                 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2510         the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2511                                            order_base_2(rnet_htable_size) - 1);
2512
2513         /* All LNDs apart from the LOLND are in separate modules.  They
2514          * register themselves when their module loads, and unregister
2515          * themselves when their module is unloaded. */
2516         lnet_register_lnd(&the_lolnd);
2517         return 0;
2518 }
2519
2520 /**
2521  * Finalize LNet library.
2522  *
2523  * \pre lnet_lib_init() called with success.
2524  * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2525  */
2526 void lnet_lib_exit(void)
2527 {
2528         LASSERT(the_lnet.ln_refcount == 0);
2529
2530         while (!list_empty(&the_lnet.ln_lnds))
2531                 lnet_unregister_lnd(list_entry(the_lnet.ln_lnds.next,
2532                                                struct lnet_lnd, lnd_list));
2533         lnet_destroy_locks();
2534 }
2535
2536 /**
2537  * Set LNet PID and start LNet interfaces, routing, and forwarding.
2538  *
2539  * Users must call this function at least once before any other functions.
2540  * For each successful call there must be a corresponding call to
2541  * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2542  * ignored.
2543  *
2544  * The PID used by LNet may be different from the one requested.
2545  * See LNetGetId().
2546  *
2547  * \param requested_pid PID requested by the caller.
2548  *
2549  * \return >= 0 on success, and < 0 error code on failures.
2550  */
2551 int
2552 LNetNIInit(lnet_pid_t requested_pid)
2553 {
2554         int                     im_a_router = 0;
2555         int                     rc;
2556         int                     ni_count;
2557         struct lnet_ping_buffer *pbuf;
2558         struct lnet_handle_md   ping_mdh;
2559         struct list_head        net_head;
2560         struct lnet_net         *net;
2561
2562         INIT_LIST_HEAD(&net_head);
2563
2564         mutex_lock(&the_lnet.ln_api_mutex);
2565
2566         CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2567
2568         if (the_lnet.ln_refcount > 0) {
2569                 rc = the_lnet.ln_refcount++;
2570                 mutex_unlock(&the_lnet.ln_api_mutex);
2571                 return rc;
2572         }
2573
2574         rc = lnet_prepare(requested_pid);
2575         if (rc != 0) {
2576                 mutex_unlock(&the_lnet.ln_api_mutex);
2577                 return rc;
2578         }
2579
2580         /* create a network for Loopback network */
2581         net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2582         if (net == NULL) {
2583                 rc = -ENOMEM;
2584                 goto err_empty_list;
2585         }
2586
2587         /* Add in the loopback NI */
2588         if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2589                 rc = -ENOMEM;
2590                 goto err_empty_list;
2591         }
2592
2593         /* If LNet is being initialized via DLC it is possible
2594          * that the user requests not to load module parameters (ones which
2595          * are supported by DLC) on initialization.  Therefore, make sure not
2596          * to load networks, routes and forwarding from module parameters
2597          * in this case.  On cleanup in case of failure only clean up
2598          * routes if it has been loaded */
2599         if (!the_lnet.ln_nis_from_mod_params) {
2600                 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
2601                                          use_tcp_bonding);
2602                 if (rc < 0)
2603                         goto err_empty_list;
2604         }
2605
2606         ni_count = lnet_startup_lndnets(&net_head);
2607         if (ni_count < 0) {
2608                 rc = ni_count;
2609                 goto err_empty_list;
2610         }
2611
2612         if (!the_lnet.ln_nis_from_mod_params) {
2613                 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2614                 if (rc != 0)
2615                         goto err_shutdown_lndnis;
2616
2617                 rc = lnet_check_routes();
2618                 if (rc != 0)
2619                         goto err_destroy_routes;
2620
2621                 rc = lnet_rtrpools_alloc(im_a_router);
2622                 if (rc != 0)
2623                         goto err_destroy_routes;
2624         }
2625
2626         rc = lnet_acceptor_start();
2627         if (rc != 0)
2628                 goto err_destroy_routes;
2629
2630         the_lnet.ln_refcount = 1;
2631         /* Now I may use my own API functions... */
2632
2633         rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2634         if (rc != 0)
2635                 goto err_acceptor_stop;
2636
2637         lnet_ping_target_update(pbuf, ping_mdh);
2638
2639         rc = LNetEQAlloc(0, lnet_mt_event_handler, &the_lnet.ln_mt_eqh);
2640         if (rc != 0) {
2641                 CERROR("Can't allocate monitor thread EQ: %d\n", rc);
2642                 goto err_stop_ping;
2643         }
2644
2645         rc = lnet_monitor_thr_start();
2646         if (rc != 0)
2647                 goto err_stop_ping;
2648
2649         rc = lnet_push_target_init();
2650         if (rc != 0)
2651                 goto err_stop_monitor_thr;
2652
2653         rc = lnet_peer_discovery_start();
2654         if (rc != 0)
2655                 goto err_destroy_push_target;
2656
2657         lnet_fault_init();
2658         lnet_router_debugfs_init();
2659
2660         mutex_unlock(&the_lnet.ln_api_mutex);
2661
2662         return 0;
2663
2664 err_destroy_push_target:
2665         lnet_push_target_fini();
2666 err_stop_monitor_thr:
2667         lnet_monitor_thr_stop();
2668 err_stop_ping:
2669         lnet_ping_target_fini();
2670 err_acceptor_stop:
2671         the_lnet.ln_refcount = 0;
2672         lnet_acceptor_stop();
2673 err_destroy_routes:
2674         if (!the_lnet.ln_nis_from_mod_params)
2675                 lnet_destroy_routes();
2676 err_shutdown_lndnis:
2677         lnet_shutdown_lndnets();
2678 err_empty_list:
2679         lnet_unprepare();
2680         LASSERT(rc < 0);
2681         mutex_unlock(&the_lnet.ln_api_mutex);
2682         while (!list_empty(&net_head)) {
2683                 struct lnet_net *net;
2684
2685                 net = list_entry(net_head.next, struct lnet_net, net_list);
2686                 list_del_init(&net->net_list);
2687                 lnet_net_free(net);
2688         }
2689         return rc;
2690 }
2691 EXPORT_SYMBOL(LNetNIInit);
2692
2693 /**
2694  * Stop LNet interfaces, routing, and forwarding.
2695  *
2696  * Users must call this function once for each successful call to LNetNIInit().
2697  * Once the LNetNIFini() operation has been started, the results of pending
2698  * API operations are undefined.
2699  *
2700  * \return always 0 for current implementation.
2701  */
2702 int
2703 LNetNIFini()
2704 {
2705         mutex_lock(&the_lnet.ln_api_mutex);
2706
2707         LASSERT(the_lnet.ln_refcount > 0);
2708
2709         if (the_lnet.ln_refcount != 1) {
2710                 the_lnet.ln_refcount--;
2711         } else {
2712                 LASSERT(!the_lnet.ln_niinit_self);
2713
2714                 lnet_fault_fini();
2715
2716                 lnet_router_debugfs_fini();
2717                 lnet_peer_discovery_stop();
2718                 lnet_push_target_fini();
2719                 lnet_monitor_thr_stop();
2720                 lnet_ping_target_fini();
2721
2722                 /* Teardown fns that use my own API functions BEFORE here */
2723                 the_lnet.ln_refcount = 0;
2724
2725                 lnet_acceptor_stop();
2726                 lnet_destroy_routes();
2727                 lnet_shutdown_lndnets();
2728                 lnet_unprepare();
2729         }
2730
2731         mutex_unlock(&the_lnet.ln_api_mutex);
2732         return 0;
2733 }
2734 EXPORT_SYMBOL(LNetNIFini);
2735
2736 /**
2737  * Grabs the ni data from the ni structure and fills the out
2738  * parameters
2739  *
2740  * \param[in] ni network        interface structure
2741  * \param[out] cfg_ni           NI config information
2742  * \param[out] tun              network and LND tunables
2743  */
2744 static void
2745 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2746                    struct lnet_ioctl_config_lnd_tunables *tun,
2747                    struct lnet_ioctl_element_stats *stats,
2748                    __u32 tun_size)
2749 {
2750         size_t min_size = 0;
2751         int i;
2752
2753         if (!ni || !cfg_ni || !tun)
2754                 return;
2755
2756         if (ni->ni_interfaces[0] != NULL) {
2757                 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2758                         if (ni->ni_interfaces[i] != NULL) {
2759                                 strncpy(cfg_ni->lic_ni_intf[i],
2760                                         ni->ni_interfaces[i],
2761                                         sizeof(cfg_ni->lic_ni_intf[i]));
2762                         }
2763                 }
2764         }
2765
2766         cfg_ni->lic_nid = ni->ni_nid;
2767         if (ni->ni_nid == LNET_NID_LO_0)
2768                 cfg_ni->lic_status = LNET_NI_STATUS_UP;
2769         else
2770                 cfg_ni->lic_status = ni->ni_status->ns_status;
2771         cfg_ni->lic_tcp_bonding = use_tcp_bonding;
2772         cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
2773
2774         memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
2775
2776         if (stats) {
2777                 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
2778                                                        LNET_STATS_TYPE_SEND);
2779                 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
2780                                                        LNET_STATS_TYPE_RECV);
2781                 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
2782                                                        LNET_STATS_TYPE_DROP);
2783         }
2784
2785         /*
2786          * tun->lt_tun will always be present, but in order to be
2787          * backwards compatible, we need to deal with the cases when
2788          * tun->lt_tun is smaller than what the kernel has, because it
2789          * comes from an older version of a userspace program, then we'll
2790          * need to copy as much information as we have available space.
2791          */
2792         min_size = tun_size - sizeof(tun->lt_cmn);
2793         memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
2794
2795         /* copy over the cpts */
2796         if (ni->ni_ncpts == LNET_CPT_NUMBER &&
2797             ni->ni_cpts == NULL)  {
2798                 for (i = 0; i < ni->ni_ncpts; i++)
2799                         cfg_ni->lic_cpts[i] = i;
2800         } else {
2801                 for (i = 0;
2802                      ni->ni_cpts != NULL && i < ni->ni_ncpts &&
2803                      i < LNET_MAX_SHOW_NUM_CPT;
2804                      i++)
2805                         cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2806         }
2807         cfg_ni->lic_ncpts = ni->ni_ncpts;
2808 }
2809
2810 /**
2811  * NOTE: This is a legacy function left in the code to be backwards
2812  * compatible with older userspace programs. It should eventually be
2813  * removed.
2814  *
2815  * Grabs the ni data from the ni structure and fills the out
2816  * parameters
2817  *
2818  * \param[in] ni network        interface structure
2819  * \param[out] config           config information
2820  */
2821 static void
2822 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
2823                          struct lnet_ioctl_config_data *config)
2824 {
2825         struct lnet_ioctl_net_config *net_config;
2826         struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
2827         size_t min_size, tunable_size = 0;
2828         int i;
2829
2830         if (!ni || !config)
2831                 return;
2832
2833         net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
2834         if (!net_config)
2835                 return;
2836
2837         BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
2838                      ARRAY_SIZE(net_config->ni_interfaces));
2839
2840         for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2841                 if (!ni->ni_interfaces[i])
2842                         break;
2843
2844                 strncpy(net_config->ni_interfaces[i],
2845                         ni->ni_interfaces[i],
2846                         sizeof(net_config->ni_interfaces[i]));
2847         }
2848
2849         config->cfg_nid = ni->ni_nid;
2850         config->cfg_config_u.cfg_net.net_peer_timeout =
2851                 ni->ni_net->net_tunables.lct_peer_timeout;
2852         config->cfg_config_u.cfg_net.net_max_tx_credits =
2853                 ni->ni_net->net_tunables.lct_max_tx_credits;
2854         config->cfg_config_u.cfg_net.net_peer_tx_credits =
2855                 ni->ni_net->net_tunables.lct_peer_tx_credits;
2856         config->cfg_config_u.cfg_net.net_peer_rtr_credits =
2857                 ni->ni_net->net_tunables.lct_peer_rtr_credits;
2858
2859         if (ni->ni_nid == LNET_NID_LO_0)
2860                 net_config->ni_status = LNET_NI_STATUS_UP;
2861         else
2862                 net_config->ni_status = ni->ni_status->ns_status;
2863
2864         if (ni->ni_cpts) {
2865                 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
2866
2867                 for (i = 0; i < num_cpts; i++)
2868                         net_config->ni_cpts[i] = ni->ni_cpts[i];
2869
2870                 config->cfg_ncpts = num_cpts;
2871         }
2872
2873         /*
2874          * See if user land tools sent in a newer and larger version
2875          * of struct lnet_tunables than what the kernel uses.
2876          */
2877         min_size = sizeof(*config) + sizeof(*net_config);
2878
2879         if (config->cfg_hdr.ioc_len > min_size)
2880                 tunable_size = config->cfg_hdr.ioc_len - min_size;
2881
2882         /* Don't copy too much data to user space */
2883         min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
2884         lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
2885
2886         if (lnd_cfg && min_size) {
2887                 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
2888                 config->cfg_config_u.cfg_net.net_interface_count = 1;
2889
2890                 /* Tell user land that kernel side has less data */
2891                 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
2892                         min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
2893                         config->cfg_hdr.ioc_len -= min_size;
2894                 }
2895         }
2896 }
2897
2898 struct lnet_ni *
2899 lnet_get_ni_idx_locked(int idx)
2900 {
2901         struct lnet_ni          *ni;
2902         struct lnet_net         *net;
2903
2904         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2905                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2906                         if (idx-- == 0)
2907                                 return ni;
2908                 }
2909         }
2910
2911         return NULL;
2912 }
2913
2914 struct lnet_ni *
2915 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2916 {
2917         struct lnet_ni          *ni;
2918         struct lnet_net         *net = mynet;
2919
2920         /*
2921          * It is possible that the net has been cleaned out while there is
2922          * a message being sent. This function accessed the net without
2923          * checking if the list is empty
2924          */
2925         if (prev == NULL) {
2926                 if (net == NULL)
2927                         net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2928                                         net_list);
2929                 if (list_empty(&net->net_ni_list))
2930                         return NULL;
2931                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2932                                 ni_netlist);
2933
2934                 return ni;
2935         }
2936
2937         if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
2938                 /* if you reached the end of the ni list and the net is
2939                  * specified, then there are no more nis in that net */
2940                 if (net != NULL)
2941                         return NULL;
2942
2943                 /* we reached the end of this net ni list. move to the
2944                  * next net */
2945                 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
2946                         /* no more nets and no more NIs. */
2947                         return NULL;
2948
2949                 /* get the next net */
2950                 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
2951                                  net_list);
2952                 if (list_empty(&net->net_ni_list))
2953                         return NULL;
2954                 /* get the ni on it */
2955                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2956                                 ni_netlist);
2957
2958                 return ni;
2959         }
2960
2961         if (list_empty(&prev->ni_netlist))
2962                 return NULL;
2963
2964         /* there are more nis left */
2965         ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
2966
2967         return ni;
2968 }
2969
2970 int
2971 lnet_get_net_config(struct lnet_ioctl_config_data *config)
2972 {
2973         struct lnet_ni *ni;
2974         int cpt;
2975         int rc = -ENOENT;
2976         int idx = config->cfg_count;
2977
2978         cpt = lnet_net_lock_current();
2979
2980         ni = lnet_get_ni_idx_locked(idx);
2981
2982         if (ni != NULL) {
2983                 rc = 0;
2984                 lnet_ni_lock(ni);
2985                 lnet_fill_ni_info_legacy(ni, config);
2986                 lnet_ni_unlock(ni);
2987         }
2988
2989         lnet_net_unlock(cpt);
2990         return rc;
2991 }
2992
2993 int
2994 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
2995                    struct lnet_ioctl_config_lnd_tunables *tun,
2996                    struct lnet_ioctl_element_stats *stats,
2997                    __u32 tun_size)
2998 {
2999         struct lnet_ni          *ni;
3000         int                     cpt;
3001         int                     rc = -ENOENT;
3002
3003         if (!cfg_ni || !tun || !stats)
3004                 return -EINVAL;
3005
3006         cpt = lnet_net_lock_current();
3007
3008         ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3009
3010         if (ni) {
3011                 rc = 0;
3012                 lnet_ni_lock(ni);
3013                 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3014                 lnet_ni_unlock(ni);
3015         }
3016
3017         lnet_net_unlock(cpt);
3018         return rc;
3019 }
3020
3021 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3022 {
3023         struct lnet_ni *ni;
3024         int cpt;
3025         int rc = -ENOENT;
3026
3027         if (!msg_stats)
3028                 return -EINVAL;
3029
3030         cpt = lnet_net_lock_current();
3031
3032         ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3033
3034         if (ni) {
3035                 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3036                 rc = 0;
3037         }
3038
3039         lnet_net_unlock(cpt);
3040
3041         return rc;
3042 }
3043
3044 static int lnet_add_net_common(struct lnet_net *net,
3045                                struct lnet_ioctl_config_lnd_tunables *tun)
3046 {
3047         __u32                   net_id;
3048         struct lnet_ping_buffer *pbuf;
3049         struct lnet_handle_md   ping_mdh;
3050         int                     rc;
3051         struct lnet_remotenet *rnet;
3052         int                     net_ni_count;
3053         int                     num_acceptor_nets;
3054
3055         lnet_net_lock(LNET_LOCK_EX);
3056         rnet = lnet_find_rnet_locked(net->net_id);
3057         lnet_net_unlock(LNET_LOCK_EX);
3058         /*
3059          * make sure that the net added doesn't invalidate the current
3060          * configuration LNet is keeping
3061          */
3062         if (rnet) {
3063                 CERROR("Adding net %s will invalidate routing configuration\n",
3064                        libcfs_net2str(net->net_id));
3065                 lnet_net_free(net);
3066                 return -EUSERS;
3067         }
3068
3069         /*
3070          * make sure you calculate the correct number of slots in the ping
3071          * buffer. Since the ping info is a flattened list of all the NIs,
3072          * we should allocate enough slots to accomodate the number of NIs
3073          * which will be added.
3074          *
3075          * since ni hasn't been configured yet, use
3076          * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3077          */
3078         net_ni_count = lnet_get_net_ni_count_pre(net);
3079
3080         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3081                                     net_ni_count + lnet_get_ni_count(),
3082                                     false);
3083         if (rc < 0) {
3084                 lnet_net_free(net);
3085                 return rc;
3086         }
3087
3088         if (tun)
3089                 memcpy(&net->net_tunables,
3090                        &tun->lt_cmn, sizeof(net->net_tunables));
3091         else
3092                 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3093
3094         /*
3095          * before starting this network get a count of the current TCP
3096          * networks which require the acceptor thread running. If that
3097          * count is == 0 before we start up this network, then we'd want to
3098          * start up the acceptor thread after starting up this network
3099          */
3100         num_acceptor_nets = lnet_count_acceptor_nets();
3101
3102         net_id = net->net_id;
3103
3104         rc = lnet_startup_lndnet(net,
3105                                  (tun) ? &tun->lt_tun : NULL);
3106         if (rc < 0)
3107                 goto failed;
3108
3109         lnet_net_lock(LNET_LOCK_EX);
3110         net = lnet_get_net_locked(net_id);
3111         lnet_net_unlock(LNET_LOCK_EX);
3112
3113         LASSERT(net);
3114
3115         /*
3116          * Start the acceptor thread if this is the first network
3117          * being added that requires the thread.
3118          */
3119         if (net->net_lnd->lnd_accept && num_acceptor_nets == 0) {
3120                 rc = lnet_acceptor_start();
3121                 if (rc < 0) {
3122                         /* shutdown the net that we just started */
3123                         CERROR("Failed to start up acceptor thread\n");
3124                         lnet_shutdown_lndnet(net);
3125                         goto failed;
3126                 }
3127         }
3128
3129         lnet_net_lock(LNET_LOCK_EX);
3130         lnet_peer_net_added(net);
3131         lnet_net_unlock(LNET_LOCK_EX);
3132
3133         lnet_ping_target_update(pbuf, ping_mdh);
3134
3135         return 0;
3136
3137 failed:
3138         lnet_ping_md_unlink(pbuf, &ping_mdh);
3139         lnet_ping_buffer_decref(pbuf);
3140         return rc;
3141 }
3142
3143 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3144                                       struct lnet_ioctl_config_lnd_tunables *tun)
3145 {
3146         struct lnet_net *net;
3147         char *nets;
3148         int rc;
3149         struct list_head net_head;
3150
3151         INIT_LIST_HEAD(&net_head);
3152
3153         rc = lnet_parse_ip2nets(&nets, ip2nets);
3154         if (rc < 0)
3155                 return rc;
3156
3157         rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3158         if (rc < 0)
3159                 return rc;
3160
3161         mutex_lock(&the_lnet.ln_api_mutex);
3162         while (!list_empty(&net_head)) {
3163                 net = list_entry(net_head.next, struct lnet_net, net_list);
3164                 list_del_init(&net->net_list);
3165                 rc = lnet_add_net_common(net, tun);
3166                 if (rc < 0)
3167                         goto out;
3168         }
3169
3170 out:
3171         mutex_unlock(&the_lnet.ln_api_mutex);
3172
3173         while (!list_empty(&net_head)) {
3174                 net = list_entry(net_head.next, struct lnet_net, net_list);
3175                 list_del_init(&net->net_list);
3176                 lnet_net_free(net);
3177         }
3178         return rc;
3179 }
3180
3181 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3182 {
3183         struct lnet_net *net;
3184         struct lnet_ni *ni;
3185         struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3186         int rc, i;
3187         __u32 net_id, lnd_type;
3188
3189         /* get the tunables if they are available */
3190         if (conf->lic_cfg_hdr.ioc_len >=
3191             sizeof(*conf) + sizeof(*tun))
3192                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3193                         conf->lic_bulk;
3194
3195         /* handle legacy ip2nets from DLC */
3196         if (conf->lic_legacy_ip2nets[0] != '\0')
3197                 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3198                                                   tun);
3199
3200         net_id = LNET_NIDNET(conf->lic_nid);
3201         lnd_type = LNET_NETTYP(net_id);
3202
3203         if (!libcfs_isknown_lnd(lnd_type)) {
3204                 CERROR("No valid net and lnd information provided\n");
3205                 return -EINVAL;
3206         }
3207
3208         net = lnet_net_alloc(net_id, NULL);
3209         if (!net)
3210                 return -ENOMEM;
3211
3212         for (i = 0; i < conf->lic_ncpts; i++) {
3213                 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3214                         return -EINVAL;
3215         }
3216
3217         ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3218                                        conf->lic_ni_intf[0]);
3219         if (!ni)
3220                 return -ENOMEM;
3221
3222         mutex_lock(&the_lnet.ln_api_mutex);
3223
3224         rc = lnet_add_net_common(net, tun);
3225
3226         mutex_unlock(&the_lnet.ln_api_mutex);
3227
3228         return rc;
3229 }
3230
3231 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3232 {
3233         struct lnet_net  *net;
3234         struct lnet_ni *ni;
3235         __u32 net_id = LNET_NIDNET(conf->lic_nid);
3236         struct lnet_ping_buffer *pbuf;
3237         struct lnet_handle_md  ping_mdh;
3238         int               rc;
3239         int               net_count;
3240         __u32             addr;
3241
3242         /* don't allow userspace to shutdown the LOLND */
3243         if (LNET_NETTYP(net_id) == LOLND)
3244                 return -EINVAL;
3245
3246         mutex_lock(&the_lnet.ln_api_mutex);
3247
3248         lnet_net_lock(0);
3249
3250         net = lnet_get_net_locked(net_id);
3251         if (!net) {
3252                 CERROR("net %s not found\n",
3253                        libcfs_net2str(net_id));
3254                 rc = -ENOENT;
3255                 goto unlock_net;
3256         }
3257
3258         addr = LNET_NIDADDR(conf->lic_nid);
3259         if (addr == 0) {
3260                 /* remove the entire net */
3261                 net_count = lnet_get_net_ni_count_locked(net);
3262
3263                 lnet_net_unlock(0);
3264
3265                 /* create and link a new ping info, before removing the old one */
3266                 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3267                                         lnet_get_ni_count() - net_count,
3268                                         false);
3269                 if (rc != 0)
3270                         goto unlock_api_mutex;
3271
3272                 lnet_shutdown_lndnet(net);
3273
3274                 if (lnet_count_acceptor_nets() == 0)
3275                         lnet_acceptor_stop();
3276
3277                 lnet_ping_target_update(pbuf, ping_mdh);
3278
3279                 goto unlock_api_mutex;
3280         }
3281
3282         ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3283         if (!ni) {
3284                 CERROR("nid %s not found\n",
3285                        libcfs_nid2str(conf->lic_nid));
3286                 rc = -ENOENT;
3287                 goto unlock_net;
3288         }
3289
3290         net_count = lnet_get_net_ni_count_locked(net);
3291
3292         lnet_net_unlock(0);
3293
3294         /* create and link a new ping info, before removing the old one */
3295         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3296                                   lnet_get_ni_count() - 1, false);
3297         if (rc != 0)
3298                 goto unlock_api_mutex;
3299
3300         lnet_shutdown_lndni(ni);
3301
3302         if (lnet_count_acceptor_nets() == 0)
3303                 lnet_acceptor_stop();
3304
3305         lnet_ping_target_update(pbuf, ping_mdh);
3306
3307         /* check if the net is empty and remove it if it is */
3308         if (net_count == 1)
3309                 lnet_shutdown_lndnet(net);
3310
3311         goto unlock_api_mutex;
3312
3313 unlock_net:
3314         lnet_net_unlock(0);
3315 unlock_api_mutex:
3316         mutex_unlock(&the_lnet.ln_api_mutex);
3317
3318         return rc;
3319 }
3320
3321 /*
3322  * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3323  * They are only expected to be called for unique networks.
3324  * That can be as a result of older DLC library
3325  * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3326  */
3327 int
3328 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3329 {
3330         struct lnet_net         *net;
3331         struct list_head        net_head;
3332         int                     rc;
3333         struct lnet_ioctl_config_lnd_tunables tun;
3334         char *nets = conf->cfg_config_u.cfg_net.net_intf;
3335
3336         INIT_LIST_HEAD(&net_head);
3337
3338         /* Create a net/ni structures for the network string */
3339         rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3340         if (rc <= 0)
3341                 return rc == 0 ? -EINVAL : rc;
3342
3343         mutex_lock(&the_lnet.ln_api_mutex);
3344
3345         if (rc > 1) {
3346                 rc = -EINVAL; /* only add one network per call */
3347                 goto out_unlock_clean;
3348         }
3349
3350         net = list_entry(net_head.next, struct lnet_net, net_list);
3351         list_del_init(&net->net_list);
3352
3353         LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3354
3355         memset(&tun, 0, sizeof(tun));
3356
3357         tun.lt_cmn.lct_peer_timeout =
3358           conf->cfg_config_u.cfg_net.net_peer_timeout;
3359         tun.lt_cmn.lct_peer_tx_credits =
3360           conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3361         tun.lt_cmn.lct_peer_rtr_credits =
3362           conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3363         tun.lt_cmn.lct_max_tx_credits =
3364           conf->cfg_config_u.cfg_net.net_max_tx_credits;
3365
3366         rc = lnet_add_net_common(net, &tun);
3367
3368 out_unlock_clean:
3369         mutex_unlock(&the_lnet.ln_api_mutex);
3370         while (!list_empty(&net_head)) {
3371                 /* net_head list is empty in success case */
3372                 net = list_entry(net_head.next, struct lnet_net, net_list);
3373                 list_del_init(&net->net_list);
3374                 lnet_net_free(net);
3375         }
3376         return rc;
3377 }
3378
3379 int
3380 lnet_dyn_del_net(__u32 net_id)
3381 {
3382         struct lnet_net  *net;
3383         struct lnet_ping_buffer *pbuf;
3384         struct lnet_handle_md ping_mdh;
3385         int               rc;
3386         int               net_ni_count;
3387
3388         /* don't allow userspace to shutdown the LOLND */
3389         if (LNET_NETTYP(net_id) == LOLND)
3390                 return -EINVAL;
3391
3392         mutex_lock(&the_lnet.ln_api_mutex);
3393
3394         lnet_net_lock(0);
3395
3396         net = lnet_get_net_locked(net_id);
3397         if (net == NULL) {
3398                 lnet_net_unlock(0);
3399                 rc = -EINVAL;
3400                 goto out;
3401         }
3402
3403         net_ni_count = lnet_get_net_ni_count_locked(net);
3404
3405         lnet_net_unlock(0);
3406
3407         /* create and link a new ping info, before removing the old one */
3408         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3409                                     lnet_get_ni_count() - net_ni_count, false);
3410         if (rc != 0)
3411                 goto out;
3412
3413         lnet_shutdown_lndnet(net);
3414
3415         if (lnet_count_acceptor_nets() == 0)
3416                 lnet_acceptor_stop();
3417
3418         lnet_ping_target_update(pbuf, ping_mdh);
3419
3420 out:
3421         mutex_unlock(&the_lnet.ln_api_mutex);
3422
3423         return rc;
3424 }
3425
3426 void lnet_incr_dlc_seq(void)
3427 {
3428         atomic_inc(&lnet_dlc_seq_no);
3429 }
3430
3431 __u32 lnet_get_dlc_seq_locked(void)
3432 {
3433         return atomic_read(&lnet_dlc_seq_no);
3434 }
3435
3436 static void
3437 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3438 {
3439         struct lnet_net *net;
3440         struct lnet_ni *ni;
3441
3442         lnet_net_lock(LNET_LOCK_EX);
3443         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3444                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3445                         if (ni->ni_nid == nid || all) {
3446                                 atomic_set(&ni->ni_healthv, value);
3447                                 if (list_empty(&ni->ni_recovery) &&
3448                                     value < LNET_MAX_HEALTH_VALUE) {
3449                                         CERROR("manually adding local NI %s to recovery\n",
3450                                                libcfs_nid2str(ni->ni_nid));
3451                                         list_add_tail(&ni->ni_recovery,
3452                                                       &the_lnet.ln_mt_localNIRecovq);
3453                                         lnet_ni_addref_locked(ni, 0);
3454                                 }
3455                                 if (!all) {
3456                                         lnet_net_unlock(LNET_LOCK_EX);
3457                                         return;
3458                                 }
3459                         }
3460                 }
3461         }
3462         lnet_net_unlock(LNET_LOCK_EX);
3463 }
3464
3465 static int
3466 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3467 {
3468         int cpt, rc = 0;
3469         struct lnet_ni *ni;
3470         lnet_nid_t nid = stats->hlni_nid;
3471
3472         cpt = lnet_net_lock_current();
3473         ni = lnet_nid2ni_locked(nid, cpt);
3474
3475         if (!ni) {
3476                 rc = -ENOENT;
3477                 goto unlock;
3478         }
3479
3480         stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3481         stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3482         stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3483         stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3484         stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3485         stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3486         stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3487
3488 unlock:
3489         lnet_net_unlock(cpt);
3490
3491         return rc;
3492 }
3493
3494 static int
3495 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3496 {
3497         struct lnet_ni *ni;
3498         int i = 0;
3499
3500         lnet_net_lock(LNET_LOCK_EX);
3501         list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3502                 list->rlst_nid_array[i] = ni->ni_nid;
3503                 i++;
3504                 if (i >= LNET_MAX_SHOW_NUM_NID)
3505                         break;
3506         }
3507         lnet_net_unlock(LNET_LOCK_EX);
3508         list->rlst_num_nids = i;
3509
3510         return 0;
3511 }
3512
3513 static int
3514 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3515 {
3516         struct lnet_peer_ni *lpni;
3517         int i = 0;
3518
3519         lnet_net_lock(LNET_LOCK_EX);
3520         list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3521                 list->rlst_nid_array[i] = lpni->lpni_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 /**
3533  * LNet ioctl handler.
3534  *
3535  */
3536 int
3537 LNetCtl(unsigned int cmd, void *arg)
3538 {
3539         struct libcfs_ioctl_data *data = arg;
3540         struct lnet_ioctl_config_data *config;
3541         struct lnet_process_id    id = {0};
3542         struct lnet_ni           *ni;
3543         int                       rc;
3544
3545         BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3546                      sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3547
3548         switch (cmd) {
3549         case IOC_LIBCFS_GET_NI:
3550                 rc = LNetGetId(data->ioc_count, &id);
3551                 data->ioc_nid = id.nid;
3552                 return rc;
3553
3554         case IOC_LIBCFS_FAIL_NID:
3555                 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3556
3557         case IOC_LIBCFS_ADD_ROUTE:
3558                 config = arg;
3559
3560                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3561                         return -EINVAL;
3562
3563                 mutex_lock(&the_lnet.ln_api_mutex);
3564                 rc = lnet_add_route(config->cfg_net,
3565                                     config->cfg_config_u.cfg_route.rtr_hop,
3566                                     config->cfg_nid,
3567                                     config->cfg_config_u.cfg_route.
3568                                         rtr_priority);
3569                 if (rc == 0) {
3570                         rc = lnet_check_routes();
3571                         if (rc != 0)
3572                                 lnet_del_route(config->cfg_net,
3573                                                config->cfg_nid);
3574                 }
3575                 mutex_unlock(&the_lnet.ln_api_mutex);
3576                 return rc;
3577
3578         case IOC_LIBCFS_DEL_ROUTE:
3579                 config = arg;
3580
3581                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3582                         return -EINVAL;
3583
3584                 mutex_lock(&the_lnet.ln_api_mutex);
3585                 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3586                 mutex_unlock(&the_lnet.ln_api_mutex);
3587                 return rc;
3588
3589         case IOC_LIBCFS_GET_ROUTE:
3590                 config = arg;
3591
3592                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3593                         return -EINVAL;
3594
3595                 mutex_lock(&the_lnet.ln_api_mutex);
3596                 rc = lnet_get_route(config->cfg_count,
3597                                     &config->cfg_net,
3598                                     &config->cfg_config_u.cfg_route.rtr_hop,
3599                                     &config->cfg_nid,
3600                                     &config->cfg_config_u.cfg_route.rtr_flags,
3601                                     &config->cfg_config_u.cfg_route.
3602                                         rtr_priority);
3603                 mutex_unlock(&the_lnet.ln_api_mutex);
3604                 return rc;
3605
3606         case IOC_LIBCFS_GET_LOCAL_NI: {
3607                 struct lnet_ioctl_config_ni *cfg_ni;
3608                 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3609                 struct lnet_ioctl_element_stats *stats;
3610                 __u32 tun_size;
3611
3612                 cfg_ni = arg;
3613
3614                 /* get the tunables if they are available */
3615                 if (cfg_ni->lic_cfg_hdr.ioc_len <
3616                     sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3617                         return -EINVAL;
3618
3619                 stats = (struct lnet_ioctl_element_stats *)
3620                         cfg_ni->lic_bulk;
3621                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3622                                 (cfg_ni->lic_bulk + sizeof(*stats));
3623
3624                 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3625                         sizeof(*stats);
3626
3627                 mutex_lock(&the_lnet.ln_api_mutex);
3628                 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3629                 mutex_unlock(&the_lnet.ln_api_mutex);
3630                 return rc;
3631         }
3632
3633         case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3634                 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3635
3636                 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3637                         return -EINVAL;
3638
3639                 mutex_lock(&the_lnet.ln_api_mutex);
3640                 rc = lnet_get_ni_stats(msg_stats);
3641                 mutex_unlock(&the_lnet.ln_api_mutex);
3642
3643                 return rc;
3644         }
3645
3646         case IOC_LIBCFS_GET_NET: {
3647                 size_t total = sizeof(*config) +
3648                                sizeof(struct lnet_ioctl_net_config);
3649                 config = arg;
3650
3651                 if (config->cfg_hdr.ioc_len < total)
3652                         return -EINVAL;
3653
3654                 mutex_lock(&the_lnet.ln_api_mutex);
3655                 rc = lnet_get_net_config(config);
3656                 mutex_unlock(&the_lnet.ln_api_mutex);
3657                 return rc;
3658         }
3659
3660         case IOC_LIBCFS_GET_LNET_STATS:
3661         {
3662                 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
3663
3664                 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
3665                         return -EINVAL;
3666
3667                 mutex_lock(&the_lnet.ln_api_mutex);
3668                 lnet_counters_get(&lnet_stats->st_cntrs);
3669                 mutex_unlock(&the_lnet.ln_api_mutex);
3670                 return 0;
3671         }
3672
3673         case IOC_LIBCFS_CONFIG_RTR:
3674                 config = arg;
3675
3676                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3677                         return -EINVAL;
3678
3679                 mutex_lock(&the_lnet.ln_api_mutex);
3680                 if (config->cfg_config_u.cfg_buffers.buf_enable) {
3681                         rc = lnet_rtrpools_enable();
3682                         mutex_unlock(&the_lnet.ln_api_mutex);
3683                         return rc;
3684                 }
3685                 lnet_rtrpools_disable();
3686                 mutex_unlock(&the_lnet.ln_api_mutex);
3687                 return 0;
3688
3689         case IOC_LIBCFS_ADD_BUF:
3690                 config = arg;
3691
3692                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3693                         return -EINVAL;
3694
3695                 mutex_lock(&the_lnet.ln_api_mutex);
3696                 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
3697                                                 buf_tiny,
3698                                           config->cfg_config_u.cfg_buffers.
3699                                                 buf_small,
3700                                           config->cfg_config_u.cfg_buffers.
3701                                                 buf_large);
3702                 mutex_unlock(&the_lnet.ln_api_mutex);
3703                 return rc;
3704
3705         case IOC_LIBCFS_SET_NUMA_RANGE: {
3706                 struct lnet_ioctl_set_value *numa;
3707                 numa = arg;
3708                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3709                         return -EINVAL;
3710                 lnet_net_lock(LNET_LOCK_EX);
3711                 lnet_numa_range = numa->sv_value;
3712                 lnet_net_unlock(LNET_LOCK_EX);
3713                 return 0;
3714         }
3715
3716         case IOC_LIBCFS_GET_NUMA_RANGE: {
3717                 struct lnet_ioctl_set_value *numa;
3718                 numa = arg;
3719                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3720                         return -EINVAL;
3721                 numa->sv_value = lnet_numa_range;
3722                 return 0;
3723         }
3724
3725         case IOC_LIBCFS_GET_BUF: {
3726                 struct lnet_ioctl_pool_cfg *pool_cfg;
3727                 size_t total = sizeof(*config) + sizeof(*pool_cfg);
3728
3729                 config = arg;
3730
3731                 if (config->cfg_hdr.ioc_len < total)
3732                         return -EINVAL;
3733
3734                 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
3735
3736                 mutex_lock(&the_lnet.ln_api_mutex);
3737                 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
3738                 mutex_unlock(&the_lnet.ln_api_mutex);
3739                 return rc;
3740         }
3741
3742         case IOC_LIBCFS_GET_LOCAL_HSTATS: {
3743                 struct lnet_ioctl_local_ni_hstats *stats = arg;
3744
3745                 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
3746                         return -EINVAL;
3747
3748                 mutex_lock(&the_lnet.ln_api_mutex);
3749                 rc = lnet_get_local_ni_hstats(stats);
3750                 mutex_unlock(&the_lnet.ln_api_mutex);
3751
3752                 return rc;
3753         }
3754
3755         case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
3756                 struct lnet_ioctl_recovery_list *list = arg;
3757                 if (list->rlst_hdr.ioc_len < sizeof(*list))
3758                         return -EINVAL;
3759
3760                 mutex_lock(&the_lnet.ln_api_mutex);
3761                 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
3762                         rc = lnet_get_local_ni_recovery_list(list);
3763                 else
3764                         rc = lnet_get_peer_ni_recovery_list(list);
3765                 mutex_unlock(&the_lnet.ln_api_mutex);
3766                 return rc;
3767         }
3768
3769         case IOC_LIBCFS_ADD_PEER_NI: {
3770                 struct lnet_ioctl_peer_cfg *cfg = arg;
3771
3772                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3773                         return -EINVAL;
3774
3775                 mutex_lock(&the_lnet.ln_api_mutex);
3776                 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
3777                                       cfg->prcfg_cfg_nid,
3778                                       cfg->prcfg_mr);
3779                 mutex_unlock(&the_lnet.ln_api_mutex);
3780                 return rc;
3781         }
3782
3783         case IOC_LIBCFS_DEL_PEER_NI: {
3784                 struct lnet_ioctl_peer_cfg *cfg = arg;
3785
3786                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3787                         return -EINVAL;
3788
3789                 mutex_lock(&the_lnet.ln_api_mutex);
3790                 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
3791                                       cfg->prcfg_cfg_nid);
3792                 mutex_unlock(&the_lnet.ln_api_mutex);
3793                 return rc;
3794         }
3795
3796         case IOC_LIBCFS_GET_PEER_INFO: {
3797                 struct lnet_ioctl_peer *peer_info = arg;
3798
3799                 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
3800                         return -EINVAL;
3801
3802                 mutex_lock(&the_lnet.ln_api_mutex);
3803                 rc = lnet_get_peer_ni_info(
3804                    peer_info->pr_count,
3805                    &peer_info->pr_nid,
3806                    peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
3807                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
3808                    &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
3809                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
3810                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
3811                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
3812                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
3813                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
3814                 mutex_unlock(&the_lnet.ln_api_mutex);
3815                 return rc;
3816         }
3817
3818         case IOC_LIBCFS_GET_PEER_NI: {
3819                 struct lnet_ioctl_peer_cfg *cfg = arg;
3820
3821                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3822                         return -EINVAL;
3823
3824                 mutex_lock(&the_lnet.ln_api_mutex);
3825                 rc = lnet_get_peer_info(cfg,
3826                                         (void __user *)cfg->prcfg_bulk);
3827                 mutex_unlock(&the_lnet.ln_api_mutex);
3828                 return rc;
3829         }
3830
3831         case IOC_LIBCFS_GET_PEER_LIST: {
3832                 struct lnet_ioctl_peer_cfg *cfg = arg;
3833
3834                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3835                         return -EINVAL;
3836
3837                 mutex_lock(&the_lnet.ln_api_mutex);
3838                 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
3839                                 (struct lnet_process_id __user *)cfg->prcfg_bulk);
3840                 mutex_unlock(&the_lnet.ln_api_mutex);
3841                 return rc;
3842         }
3843
3844         case IOC_LIBCFS_SET_HEALHV: {
3845                 struct lnet_ioctl_reset_health_cfg *cfg = arg;
3846                 int value;
3847                 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
3848                         return -EINVAL;
3849                 if (cfg->rh_value < 0 ||
3850                     cfg->rh_value > LNET_MAX_HEALTH_VALUE)
3851                         value = LNET_MAX_HEALTH_VALUE;
3852                 else
3853                         value = cfg->rh_value;
3854                 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
3855                        value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
3856                        "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
3857                 mutex_lock(&the_lnet.ln_api_mutex);
3858                 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
3859                         lnet_ni_set_healthv(cfg->rh_nid, value,
3860                                              cfg->rh_all);
3861                 else
3862                         lnet_peer_ni_set_healthv(cfg->rh_nid, value,
3863                                                   cfg->rh_all);
3864                 mutex_unlock(&the_lnet.ln_api_mutex);
3865                 return 0;
3866         }
3867
3868         case IOC_LIBCFS_NOTIFY_ROUTER: {
3869                 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
3870
3871                 /* The deadline passed in by the user should be some time in
3872                  * seconds in the future since the UNIX epoch. We have to map
3873                  * that deadline to the wall clock.
3874                  */
3875                 deadline += ktime_get_seconds();
3876                 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags,
3877                                    deadline);
3878         }
3879
3880         case IOC_LIBCFS_LNET_DIST:
3881                 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
3882                 if (rc < 0 && rc != -EHOSTUNREACH)
3883                         return rc;
3884
3885                 data->ioc_u32[0] = rc;
3886                 return 0;
3887
3888         case IOC_LIBCFS_TESTPROTOCOMPAT:
3889                 lnet_net_lock(LNET_LOCK_EX);
3890                 the_lnet.ln_testprotocompat = data->ioc_flags;
3891                 lnet_net_unlock(LNET_LOCK_EX);
3892                 return 0;
3893
3894         case IOC_LIBCFS_LNET_FAULT:
3895                 return lnet_fault_ctl(data->ioc_flags, data);
3896
3897         case IOC_LIBCFS_PING: {
3898                 signed long timeout;
3899
3900                 id.nid = data->ioc_nid;
3901                 id.pid = data->ioc_u32[0];
3902
3903                 /* If timeout is negative then set default of 3 minutes */
3904                 if (((s32)data->ioc_u32[1] <= 0) ||
3905                     data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3906                         timeout = msecs_to_jiffies(DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC);
3907                 else
3908                         timeout = msecs_to_jiffies(data->ioc_u32[1]);
3909
3910                 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
3911                                data->ioc_plen1 / sizeof(struct lnet_process_id));
3912
3913                 if (rc < 0)
3914                         return rc;
3915
3916                 data->ioc_count = rc;
3917                 return 0;
3918         }
3919
3920         case IOC_LIBCFS_PING_PEER: {
3921                 struct lnet_ioctl_ping_data *ping = arg;
3922                 struct lnet_peer *lp;
3923                 signed long timeout;
3924
3925                 /* If timeout is negative then set default of 3 minutes */
3926                 if (((s32)ping->op_param) <= 0 ||
3927                     ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3928                         timeout = msecs_to_jiffies(DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC);
3929                 else
3930                         timeout = msecs_to_jiffies(ping->op_param);
3931
3932                 rc = lnet_ping(ping->ping_id, timeout,
3933                                ping->ping_buf,
3934                                ping->ping_count);
3935                 if (rc < 0)
3936                         return rc;
3937
3938                 mutex_lock(&the_lnet.ln_api_mutex);
3939                 lp = lnet_find_peer(ping->ping_id.nid);
3940                 if (lp) {
3941                         ping->ping_id.nid = lp->lp_primary_nid;
3942                         ping->mr_info = lnet_peer_is_multi_rail(lp);
3943                         lnet_peer_decref_locked(lp);
3944                 }
3945                 mutex_unlock(&the_lnet.ln_api_mutex);
3946
3947                 ping->ping_count = rc;
3948                 return 0;
3949         }
3950
3951         case IOC_LIBCFS_DISCOVER: {
3952                 struct lnet_ioctl_ping_data *discover = arg;
3953                 struct lnet_peer *lp;
3954
3955                 rc = lnet_discover(discover->ping_id, discover->op_param,
3956                                    discover->ping_buf,
3957                                    discover->ping_count);
3958                 if (rc < 0)
3959                         return rc;
3960
3961                 mutex_lock(&the_lnet.ln_api_mutex);
3962                 lp = lnet_find_peer(discover->ping_id.nid);
3963                 if (lp) {
3964                         discover->ping_id.nid = lp->lp_primary_nid;
3965                         discover->mr_info = lnet_peer_is_multi_rail(lp);
3966                         lnet_peer_decref_locked(lp);
3967                 }
3968                 mutex_unlock(&the_lnet.ln_api_mutex);
3969
3970                 discover->ping_count = rc;
3971                 return 0;
3972         }
3973
3974         default:
3975                 ni = lnet_net2ni_addref(data->ioc_net);
3976                 if (ni == NULL)
3977                         return -EINVAL;
3978
3979                 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
3980                         rc = -EINVAL;
3981                 else
3982                         rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
3983
3984                 lnet_ni_decref(ni);
3985                 return rc;
3986         }
3987         /* not reached */
3988 }
3989 EXPORT_SYMBOL(LNetCtl);
3990
3991 void LNetDebugPeer(struct lnet_process_id id)
3992 {
3993         lnet_debug_peer(id.nid);
3994 }
3995 EXPORT_SYMBOL(LNetDebugPeer);
3996
3997 /**
3998  * Determine if the specified peer \a nid is on the local node.
3999  *
4000  * \param nid   peer nid to check
4001  *
4002  * \retval true         If peer NID is on the local node.
4003  * \retval false        If peer NID is not on the local node.
4004  */
4005 bool LNetIsPeerLocal(lnet_nid_t nid)
4006 {
4007         struct lnet_net *net;
4008         struct lnet_ni *ni;
4009         int cpt;
4010
4011         cpt = lnet_net_lock_current();
4012         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4013                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4014                         if (ni->ni_nid == nid) {
4015                                 lnet_net_unlock(cpt);
4016                                 return true;
4017                         }
4018                 }
4019         }
4020         lnet_net_unlock(cpt);
4021
4022         return false;
4023 }
4024 EXPORT_SYMBOL(LNetIsPeerLocal);
4025
4026 /**
4027  * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4028  * Note that all interfaces share a same PID, as requested by LNetNIInit().
4029  *
4030  * \param index Index of the interface to look up.
4031  * \param id On successful return, this location will hold the
4032  * struct lnet_process_id ID of the interface.
4033  *
4034  * \retval 0 If an interface exists at \a index.
4035  * \retval -ENOENT If no interface has been found.
4036  */
4037 int
4038 LNetGetId(unsigned int index, struct lnet_process_id *id)
4039 {
4040         struct lnet_ni   *ni;
4041         struct lnet_net  *net;
4042         int               cpt;
4043         int               rc = -ENOENT;
4044
4045         LASSERT(the_lnet.ln_refcount > 0);
4046
4047         cpt = lnet_net_lock_current();
4048
4049         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4050                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4051                         if (index-- != 0)
4052                                 continue;
4053
4054                         id->nid = ni->ni_nid;
4055                         id->pid = the_lnet.ln_pid;
4056                         rc = 0;
4057                         break;
4058                 }
4059         }
4060
4061         lnet_net_unlock(cpt);
4062         return rc;
4063 }
4064 EXPORT_SYMBOL(LNetGetId);
4065
4066 static int lnet_ping(struct lnet_process_id id, signed long timeout,
4067                      struct lnet_process_id __user *ids, int n_ids)
4068 {
4069         struct lnet_handle_eq eqh;
4070         struct lnet_handle_md mdh;
4071         struct lnet_event event;
4072         struct lnet_md md = { NULL };
4073         int which;
4074         int unlinked = 0;
4075         int replied = 0;
4076         const signed long a_long_time = msecs_to_jiffies(60 * MSEC_PER_SEC);
4077         struct lnet_ping_buffer *pbuf;
4078         struct lnet_process_id tmpid;
4079         int i;
4080         int nob;
4081         int rc;
4082         int rc2;
4083         sigset_t blocked;
4084
4085         /* n_ids limit is arbitrary */
4086         if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4087                 return -EINVAL;
4088
4089         /*
4090          * if the user buffer has more space than the lnet_interfaces_max
4091          * then only fill it up to lnet_interfaces_max
4092          */
4093         if (n_ids > lnet_interfaces_max)
4094                 n_ids = lnet_interfaces_max;
4095
4096         if (id.pid == LNET_PID_ANY)
4097                 id.pid = LNET_PID_LUSTRE;
4098
4099         pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4100         if (!pbuf)
4101                 return -ENOMEM;
4102
4103         /* NB 2 events max (including any unlink event) */
4104         rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &eqh);
4105         if (rc != 0) {
4106                 CERROR("Can't allocate EQ: %d\n", rc);
4107                 goto fail_ping_buffer_decref;
4108         }
4109
4110         /* initialize md content */
4111         md.start     = &pbuf->pb_info;
4112         md.length    = LNET_PING_INFO_SIZE(n_ids);
4113         md.threshold = 2; /* GET/REPLY */
4114         md.max_size  = 0;
4115         md.options   = LNET_MD_TRUNCATE;
4116         md.user_ptr  = NULL;
4117         md.eq_handle = eqh;
4118
4119         rc = LNetMDBind(md, LNET_UNLINK, &mdh);
4120         if (rc != 0) {
4121                 CERROR("Can't bind MD: %d\n", rc);
4122                 goto fail_free_eq;
4123         }
4124
4125         rc = LNetGet(LNET_NID_ANY, mdh, id,
4126                      LNET_RESERVED_PORTAL,
4127                      LNET_PROTO_PING_MATCHBITS, 0, false);
4128
4129         if (rc != 0) {
4130                 /* Don't CERROR; this could be deliberate! */
4131                 rc2 = LNetMDUnlink(mdh);
4132                 LASSERT(rc2 == 0);
4133
4134                 /* NB must wait for the UNLINK event below... */
4135                 unlinked = 1;
4136                 timeout = a_long_time;
4137         }
4138
4139         do {
4140                 /* MUST block for unlink to complete */
4141                 if (unlinked)
4142                         blocked = cfs_block_allsigs();
4143
4144                 rc2 = LNetEQPoll(&eqh, 1, timeout, &event, &which);
4145
4146                 if (unlinked)
4147                         cfs_restore_sigs(blocked);
4148
4149                 CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
4150                        (rc2 <= 0) ? -1 : event.type,
4151                        (rc2 <= 0) ? -1 : event.status,
4152                        (rc2 > 0 && event.unlinked) ? " unlinked" : "");
4153
4154                 LASSERT(rc2 != -EOVERFLOW);     /* can't miss anything */
4155
4156                 if (rc2 <= 0 || event.status != 0) {
4157                         /* timeout or error */
4158                         if (!replied && rc == 0)
4159                                 rc = (rc2 < 0) ? rc2 :
4160                                      (rc2 == 0) ? -ETIMEDOUT :
4161                                      event.status;
4162
4163                         if (!unlinked) {
4164                                 /* Ensure completion in finite time... */
4165                                 LNetMDUnlink(mdh);
4166                                 /* No assertion (racing with network) */
4167                                 unlinked = 1;
4168                                 timeout = a_long_time;
4169                         } else if (rc2 == 0) {
4170                                 /* timed out waiting for unlink */
4171                                 CWARN("ping %s: late network completion\n",
4172                                       libcfs_id2str(id));
4173                         }
4174                 } else if (event.type == LNET_EVENT_REPLY) {
4175                         replied = 1;
4176                         rc = event.mlength;
4177                 }
4178         } while (rc2 <= 0 || !event.unlinked);
4179
4180         if (!replied) {
4181                 if (rc >= 0)
4182                         CWARN("%s: Unexpected rc >= 0 but no reply!\n",
4183                               libcfs_id2str(id));
4184                 rc = -EIO;
4185                 goto fail_free_eq;
4186         }
4187
4188         nob = rc;
4189         LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4190
4191         rc = -EPROTO;           /* if I can't parse... */
4192
4193         if (nob < 8) {
4194                 CERROR("%s: ping info too short %d\n",
4195                        libcfs_id2str(id), nob);
4196                 goto fail_free_eq;
4197         }
4198
4199         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4200                 lnet_swap_pinginfo(pbuf);
4201         } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4202                 CERROR("%s: Unexpected magic %08x\n",
4203                        libcfs_id2str(id), pbuf->pb_info.pi_magic);
4204                 goto fail_free_eq;
4205         }
4206
4207         if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4208                 CERROR("%s: ping w/o NI status: 0x%x\n",
4209                        libcfs_id2str(id), pbuf->pb_info.pi_features);
4210                 goto fail_free_eq;
4211         }
4212
4213         if (nob < LNET_PING_INFO_SIZE(0)) {
4214                 CERROR("%s: Short reply %d(%d min)\n",
4215                        libcfs_id2str(id),
4216                        nob, (int)LNET_PING_INFO_SIZE(0));
4217                 goto fail_free_eq;
4218         }
4219
4220         if (pbuf->pb_info.pi_nnis < n_ids)
4221                 n_ids = pbuf->pb_info.pi_nnis;
4222
4223         if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4224                 CERROR("%s: Short reply %d(%d expected)\n",
4225                        libcfs_id2str(id),
4226                        nob, (int)LNET_PING_INFO_SIZE(n_ids));
4227                 goto fail_free_eq;
4228         }
4229
4230         rc = -EFAULT;           /* if I segv in copy_to_user()... */
4231
4232         memset(&tmpid, 0, sizeof(tmpid));
4233         for (i = 0; i < n_ids; i++) {
4234                 tmpid.pid = pbuf->pb_info.pi_pid;
4235                 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4236                 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4237                         goto fail_free_eq;
4238         }
4239         rc = pbuf->pb_info.pi_nnis;
4240
4241  fail_free_eq:
4242         rc2 = LNetEQFree(eqh);
4243         if (rc2 != 0)
4244                 CERROR("rc2 %d\n", rc2);
4245         LASSERT(rc2 == 0);
4246
4247  fail_ping_buffer_decref:
4248         lnet_ping_buffer_decref(pbuf);
4249         return rc;
4250 }
4251
4252 static int
4253 lnet_discover(struct lnet_process_id id, __u32 force,
4254               struct lnet_process_id __user *ids, int n_ids)
4255 {
4256         struct lnet_peer_ni *lpni;
4257         struct lnet_peer_ni *p;
4258         struct lnet_peer *lp;
4259         struct lnet_process_id *buf;
4260         int cpt;
4261         int i;
4262         int rc;
4263         int max_intf = lnet_interfaces_max;
4264         size_t buf_size;
4265
4266         if (n_ids <= 0 ||
4267             id.nid == LNET_NID_ANY)
4268                 return -EINVAL;
4269
4270         if (id.pid == LNET_PID_ANY)
4271                 id.pid = LNET_PID_LUSTRE;
4272
4273         /*
4274          * if the user buffer has more space than the max_intf
4275          * then only fill it up to max_intf
4276          */
4277         if (n_ids > max_intf)
4278                 n_ids = max_intf;
4279
4280         buf_size = n_ids * sizeof(*buf);
4281
4282         LIBCFS_ALLOC(buf, buf_size);
4283         if (!buf)
4284                 return -ENOMEM;
4285
4286         cpt = lnet_net_lock_current();
4287         lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4288         if (IS_ERR(lpni)) {
4289                 rc = PTR_ERR(lpni);
4290                 goto out;
4291         }
4292
4293         /*
4294          * Clearing the NIDS_UPTODATE flag ensures the peer will
4295          * be discovered, provided discovery has not been disabled.
4296          */
4297         lp = lpni->lpni_peer_net->lpn_peer;
4298         spin_lock(&lp->lp_lock);
4299         lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4300         /* If the force flag is set, force a PING and PUSH as well. */
4301         if (force)
4302                 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4303         spin_unlock(&lp->lp_lock);
4304         rc = lnet_discover_peer_locked(lpni, cpt, true);
4305         if (rc)
4306                 goto out_decref;
4307
4308         /* Peer may have changed. */
4309         lp = lpni->lpni_peer_net->lpn_peer;
4310         if (lp->lp_nnis < n_ids)
4311                 n_ids = lp->lp_nnis;
4312
4313         i = 0;
4314         p = NULL;
4315         while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4316                 buf[i].pid = id.pid;
4317                 buf[i].nid = p->lpni_nid;
4318                 if (++i >= n_ids)
4319                         break;
4320         }
4321
4322         lnet_net_unlock(cpt);
4323
4324         rc = -EFAULT;
4325         if (copy_to_user(ids, buf, n_ids * sizeof(*buf)))
4326                 goto out_relock;
4327         rc = n_ids;
4328 out_relock:
4329         lnet_net_lock(cpt);
4330 out_decref:
4331         lnet_peer_ni_decref_locked(lpni);
4332 out:
4333         lnet_net_unlock(cpt);
4334
4335         LIBCFS_FREE(buf, buf_size);
4336
4337         return rc;
4338 }
4339
4340 /**
4341  * Retrieve peer discovery status.
4342  *
4343  * \retval 1 if lnet_peer_discovery_disabled is 0
4344  * \retval 0 if lnet_peer_discovery_disabled is 1
4345  */
4346 int
4347 LNetGetPeerDiscoveryStatus(void)
4348 {
4349         return !lnet_peer_discovery_disabled;
4350 }
4351 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);