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  */
  31
  32 #define DEBUG_SUBSYSTEM S_LNET
  33
  34 #include <linux/ctype.h>
  35 #include <linux/log2.h>
  36 #include <linux/ktime.h>
  37 #include <linux/moduleparam.h>
  38 #include <linux/uaccess.h>
  39 #ifdef HAVE_SCHED_HEADERS
  40 #include <linux/sched/signal.h>
  41 #endif
  42 #include <lnet/udsp.h>
  43 #include <lnet/lib-lnet.h>
  44
  45 #define D_LNI D_CONSOLE
  46
  47 /*
  48  * initialize ln_api_mutex statically, since it needs to be used in
  49  * discovery_set callback. That module parameter callback can be called
  50  * before module init completes. The mutex needs to be ready for use then.
  51  */
  52 struct lnet the_lnet = {
  53         .ln_api_mutex = __MUTEX_INITIALIZER(the_lnet.ln_api_mutex),
  54 };              /* THE state of the network */
  55 EXPORT_SYMBOL(the_lnet);
  56
  57 static char *ip2nets = "";
  58 module_param(ip2nets, charp, 0444);
  59 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
  60
  61 static char *networks = "";
  62 module_param(networks, charp, 0444);
  63 MODULE_PARM_DESC(networks, "local networks");
  64
  65 static char *routes = "";
  66 module_param(routes, charp, 0444);
  67 MODULE_PARM_DESC(routes, "routes to non-local networks");
  68
  69 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
  70 module_param(rnet_htable_size, int, 0444);
  71 MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
  72
  73 static int use_tcp_bonding;
  74 module_param(use_tcp_bonding, int, 0444);
  75 MODULE_PARM_DESC(use_tcp_bonding,
  76                  "use_tcp_bonding parameter has been removed");
  77
  78 unsigned int lnet_numa_range = 0;
  79 module_param(lnet_numa_range, uint, 0444);
  80 MODULE_PARM_DESC(lnet_numa_range,
  81                 "NUMA range to consider during Multi-Rail selection");
  82
  83 /*
  84  * lnet_health_sensitivity determines by how much we decrement the health
  85  * value on sending error. The value defaults to 100, which means health
  86  * interface health is decremented by 100 points every failure.
  87  */
  88 unsigned int lnet_health_sensitivity = 100;
  89 static int sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp);
  90 #ifdef HAVE_KERNEL_PARAM_OPS
  91 static struct kernel_param_ops param_ops_health_sensitivity = {
  92         .set = sensitivity_set,
  93         .get = param_get_int,
  94 };
  95 #define param_check_health_sensitivity(name, p) \
  96                 __param_check(name, p, int)
  97 module_param(lnet_health_sensitivity, health_sensitivity, S_IRUGO|S_IWUSR);
  98 #else
  99 module_param_call(lnet_health_sensitivity, sensitivity_set, param_get_int,
 100                   &lnet_health_sensitivity, S_IRUGO|S_IWUSR);
 101 #endif
 102 MODULE_PARM_DESC(lnet_health_sensitivity,
 103                 "Value to decrement the health value by on error");
 104
 105 /*
 106  * lnet_recovery_interval determines how often we should perform recovery
 107  * on unhealthy interfaces.
 108  */
 109 unsigned int lnet_recovery_interval = 1;
 110 static int recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp);
 111 #ifdef HAVE_KERNEL_PARAM_OPS
 112 static struct kernel_param_ops param_ops_recovery_interval = {
 113         .set = recovery_interval_set,
 114         .get = param_get_int,
 115 };
 116 #define param_check_recovery_interval(name, p) \
 117                 __param_check(name, p, int)
 118 module_param(lnet_recovery_interval, recovery_interval, S_IRUGO|S_IWUSR);
 119 #else
 120 module_param_call(lnet_recovery_interval, recovery_interval_set, param_get_int,
 121                   &lnet_recovery_interval, S_IRUGO|S_IWUSR);
 122 #endif
 123 MODULE_PARM_DESC(lnet_recovery_interval,
 124                 "DEPRECATED - Interval to recover unhealthy interfaces in seconds");
 125
 126 unsigned int lnet_recovery_limit;
 127 module_param(lnet_recovery_limit, uint, 0644);
 128 MODULE_PARM_DESC(lnet_recovery_limit,
 129                  "How long to attempt recovery of unhealthy peer interfaces in seconds. Set to 0 to allow indefinite recovery");
 130
 131 static int lnet_interfaces_max = LNET_INTERFACES_MAX_DEFAULT;
 132 static int intf_max_set(const char *val, cfs_kernel_param_arg_t *kp);
 133
 134 static struct kernel_param_ops param_ops_interfaces_max = {
 135         .set = intf_max_set,
 136         .get = param_get_int,
 137 };
 138
 139 #define param_check_interfaces_max(name, p) \
 140                 __param_check(name, p, int)
 141
 142 #ifdef HAVE_KERNEL_PARAM_OPS
 143 module_param(lnet_interfaces_max, interfaces_max, 0644);
 144 #else
 145 module_param_call(lnet_interfaces_max, intf_max_set, param_get_int,
 146                   &param_ops_interfaces_max, 0644);
 147 #endif
 148 MODULE_PARM_DESC(lnet_interfaces_max,
 149                 "Maximum number of interfaces in a node.");
 150
 151 unsigned lnet_peer_discovery_disabled = 0;
 152 static int discovery_set(const char *val, cfs_kernel_param_arg_t *kp);
 153
 154 static struct kernel_param_ops param_ops_discovery_disabled = {
 155         .set = discovery_set,
 156         .get = param_get_int,
 157 };
 158
 159 #define param_check_discovery_disabled(name, p) \
 160                 __param_check(name, p, int)
 161 #ifdef HAVE_KERNEL_PARAM_OPS
 162 module_param(lnet_peer_discovery_disabled, discovery_disabled, 0644);
 163 #else
 164 module_param_call(lnet_peer_discovery_disabled, discovery_set, param_get_int,
 165                   &param_ops_discovery_disabled, 0644);
 166 #endif
 167 MODULE_PARM_DESC(lnet_peer_discovery_disabled,
 168                 "Set to 1 to disable peer discovery on this node.");
 169
 170 unsigned int lnet_drop_asym_route;
 171 static int drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp);
 172
 173 static struct kernel_param_ops param_ops_drop_asym_route = {
 174         .set = drop_asym_route_set,
 175         .get = param_get_int,
 176 };
 177
 178 #define param_check_drop_asym_route(name, p)    \
 179         __param_check(name, p, int)
 180 #ifdef HAVE_KERNEL_PARAM_OPS
 181 module_param(lnet_drop_asym_route, drop_asym_route, 0644);
 182 #else
 183 module_param_call(lnet_drop_asym_route, drop_asym_route_set, param_get_int,
 184                   &param_ops_drop_asym_route, 0644);
 185 #endif
 186 MODULE_PARM_DESC(lnet_drop_asym_route,
 187                  "Set to 1 to drop asymmetrical route messages.");
 188
 189 #define LNET_TRANSACTION_TIMEOUT_DEFAULT 50
 190 unsigned int lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_DEFAULT;
 191 static int transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp);
 192 #ifdef HAVE_KERNEL_PARAM_OPS
 193 static struct kernel_param_ops param_ops_transaction_timeout = {
 194         .set = transaction_to_set,
 195         .get = param_get_int,
 196 };
 197
 198 #define param_check_transaction_timeout(name, p) \
 199                 __param_check(name, p, int)
 200 module_param(lnet_transaction_timeout, transaction_timeout, S_IRUGO|S_IWUSR);
 201 #else
 202 module_param_call(lnet_transaction_timeout, transaction_to_set, param_get_int,
 203                   &lnet_transaction_timeout, S_IRUGO|S_IWUSR);
 204 #endif
 205 MODULE_PARM_DESC(lnet_transaction_timeout,
 206                 "Maximum number of seconds to wait for a peer response.");
 207
 208 #define LNET_RETRY_COUNT_DEFAULT 2
 209 unsigned int lnet_retry_count = LNET_RETRY_COUNT_DEFAULT;
 210 static int retry_count_set(const char *val, cfs_kernel_param_arg_t *kp);
 211 #ifdef HAVE_KERNEL_PARAM_OPS
 212 static struct kernel_param_ops param_ops_retry_count = {
 213         .set = retry_count_set,
 214         .get = param_get_int,
 215 };
 216
 217 #define param_check_retry_count(name, p) \
 218                 __param_check(name, p, int)
 219 module_param(lnet_retry_count, retry_count, S_IRUGO|S_IWUSR);
 220 #else
 221 module_param_call(lnet_retry_count, retry_count_set, param_get_int,
 222                   &lnet_retry_count, S_IRUGO|S_IWUSR);
 223 #endif
 224 MODULE_PARM_DESC(lnet_retry_count,
 225                  "Maximum number of times to retry transmitting a message");
 226
 227 unsigned int lnet_response_tracking = 3;
 228 static int response_tracking_set(const char *val, cfs_kernel_param_arg_t *kp);
 229
 230 #ifdef HAVE_KERNEL_PARAM_OPS
 231 static struct kernel_param_ops param_ops_response_tracking = {
 232         .set = response_tracking_set,
 233         .get = param_get_int,
 234 };
 235
 236 #define param_check_response_tracking(name, p)  \
 237         __param_check(name, p, int)
 238 module_param(lnet_response_tracking, response_tracking, 0644);
 239 #else
 240 module_param_call(lnet_response_tracking, response_tracking_set, param_get_int,
 241                   &lnet_response_tracking, 0644);
 242 #endif
 243 MODULE_PARM_DESC(lnet_response_tracking,
 244                  "(0|1|2|3) LNet Internal Only|GET Reply only|PUT ACK only|Full Tracking (default)");
 245
 246 #define LNET_LND_TIMEOUT_DEFAULT ((LNET_TRANSACTION_TIMEOUT_DEFAULT - 1) / \
 247                                   (LNET_RETRY_COUNT_DEFAULT + 1))
 248 unsigned int lnet_lnd_timeout = LNET_LND_TIMEOUT_DEFAULT;
 249 static void lnet_set_lnd_timeout(void)
 250 {
 251         lnet_lnd_timeout = (lnet_transaction_timeout - 1) /
 252                            (lnet_retry_count + 1);
 253 }
 254
 255 /*
 256  * This sequence number keeps track of how many times DLC was used to
 257  * update the local NIs. It is incremented when a NI is added or
 258  * removed and checked when sending a message to determine if there is
 259  * a need to re-run the selection algorithm. See lnet_select_pathway()
 260  * for more details on its usage.
 261  */
 262 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
 263
 264 static int lnet_ping(struct lnet_process_id id, struct lnet_nid *src_nid,
 265                      signed long timeout, struct lnet_process_id __user *ids,
 266                      int n_ids);
 267
 268 static int lnet_discover(struct lnet_process_id id, __u32 force,
 269                          struct lnet_process_id __user *ids, int n_ids);
 270
 271 static int
 272 sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
 273 {
 274         int rc;
 275         unsigned *sensitivity = (unsigned *)kp->arg;
 276         unsigned long value;
 277
 278         rc = kstrtoul(val, 0, &value);
 279         if (rc) {
 280                 CERROR("Invalid module parameter value for 'lnet_health_sensitivity'\n");
 281                 return rc;
 282         }
 283
 284         /*
 285          * The purpose of locking the api_mutex here is to ensure that
 286          * the correct value ends up stored properly.
 287          */
 288         mutex_lock(&the_lnet.ln_api_mutex);
 289
 290         if (value > LNET_MAX_HEALTH_VALUE) {
 291                 mutex_unlock(&the_lnet.ln_api_mutex);
 292                 CERROR("Invalid health value. Maximum: %d value = %lu\n",
 293                        LNET_MAX_HEALTH_VALUE, value);
 294                 return -EINVAL;
 295         }
 296
 297         if (*sensitivity != 0 && value == 0 && lnet_retry_count != 0) {
 298                 lnet_retry_count = 0;
 299                 lnet_set_lnd_timeout();
 300         }
 301
 302         *sensitivity = value;
 303
 304         mutex_unlock(&the_lnet.ln_api_mutex);
 305
 306         return 0;
 307 }
 308
 309 static int
 310 recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp)
 311 {
 312         CWARN("'lnet_recovery_interval' has been deprecated\n");
 313
 314         return 0;
 315 }
 316
 317 static int
 318 discovery_set(const char *val, cfs_kernel_param_arg_t *kp)
 319 {
 320         int rc;
 321         unsigned *discovery_off = (unsigned *)kp->arg;
 322         unsigned long value;
 323         struct lnet_ping_buffer *pbuf;
 324
 325         rc = kstrtoul(val, 0, &value);
 326         if (rc) {
 327                 CERROR("Invalid module parameter value for 'lnet_peer_discovery_disabled'\n");
 328                 return rc;
 329         }
 330
 331         value = (value) ? 1 : 0;
 332
 333         /*
 334          * The purpose of locking the api_mutex here is to ensure that
 335          * the correct value ends up stored properly.
 336          */
 337         mutex_lock(&the_lnet.ln_api_mutex);
 338
 339         if (value == *discovery_off) {
 340                 mutex_unlock(&the_lnet.ln_api_mutex);
 341                 return 0;
 342         }
 343
 344         /*
 345          * We still want to set the discovery value even when LNet is not
 346          * running. This is the case when LNet is being loaded and we want
 347          * the module parameters to take effect. Otherwise if we're
 348          * changing the value dynamically, we want to set it after
 349          * updating the peers
 350          */
 351         if (the_lnet.ln_state != LNET_STATE_RUNNING) {
 352                 *discovery_off = value;
 353                 mutex_unlock(&the_lnet.ln_api_mutex);
 354                 return 0;
 355         }
 356
 357         /* tell peers that discovery setting has changed */
 358         lnet_net_lock(LNET_LOCK_EX);
 359         pbuf = the_lnet.ln_ping_target;
 360         if (value)
 361                 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
 362         else
 363                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
 364         lnet_net_unlock(LNET_LOCK_EX);
 365
 366         /* only send a push when we're turning off discovery */
 367         if (*discovery_off <= 0 && value > 0)
 368                 lnet_push_update_to_peers(1);
 369         *discovery_off = value;
 370
 371         mutex_unlock(&the_lnet.ln_api_mutex);
 372
 373         return 0;
 374 }
 375
 376 static int
 377 drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp)
 378 {
 379         int rc;
 380         unsigned int *drop_asym_route = (unsigned int *)kp->arg;
 381         unsigned long value;
 382
 383         rc = kstrtoul(val, 0, &value);
 384         if (rc) {
 385                 CERROR("Invalid module parameter value for "
 386                        "'lnet_drop_asym_route'\n");
 387                 return rc;
 388         }
 389
 390         /*
 391          * The purpose of locking the api_mutex here is to ensure that
 392          * the correct value ends up stored properly.
 393          */
 394         mutex_lock(&the_lnet.ln_api_mutex);
 395
 396         if (value == *drop_asym_route) {
 397                 mutex_unlock(&the_lnet.ln_api_mutex);
 398                 return 0;
 399         }
 400
 401         *drop_asym_route = value;
 402
 403         mutex_unlock(&the_lnet.ln_api_mutex);
 404
 405         return 0;
 406 }
 407
 408 static int
 409 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
 410 {
 411         int rc;
 412         unsigned *transaction_to = (unsigned *)kp->arg;
 413         unsigned long value;
 414
 415         rc = kstrtoul(val, 0, &value);
 416         if (rc) {
 417                 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
 418                 return rc;
 419         }
 420
 421         /*
 422          * The purpose of locking the api_mutex here is to ensure that
 423          * the correct value ends up stored properly.
 424          */
 425         mutex_lock(&the_lnet.ln_api_mutex);
 426
 427         if (value <= lnet_retry_count || value == 0) {
 428                 mutex_unlock(&the_lnet.ln_api_mutex);
 429                 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
 430                        "Has to be greater than lnet_retry_count (%u)\n",
 431                        value, lnet_retry_count);
 432                 return -EINVAL;
 433         }
 434
 435         if (value == *transaction_to) {
 436                 mutex_unlock(&the_lnet.ln_api_mutex);
 437                 return 0;
 438         }
 439
 440         *transaction_to = value;
 441         /* Update the lnet_lnd_timeout now that we've modified the
 442          * transaction timeout
 443          */
 444         lnet_set_lnd_timeout();
 445
 446         mutex_unlock(&the_lnet.ln_api_mutex);
 447
 448         return 0;
 449 }
 450
 451 static int
 452 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
 453 {
 454         int rc;
 455         unsigned *retry_count = (unsigned *)kp->arg;
 456         unsigned long value;
 457
 458         rc = kstrtoul(val, 0, &value);
 459         if (rc) {
 460                 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
 461                 return rc;
 462         }
 463
 464         /*
 465          * The purpose of locking the api_mutex here is to ensure that
 466          * the correct value ends up stored properly.
 467          */
 468         mutex_lock(&the_lnet.ln_api_mutex);
 469
 470         if (lnet_health_sensitivity == 0 && value > 0) {
 471                 mutex_unlock(&the_lnet.ln_api_mutex);
 472                 CERROR("Can not set lnet_retry_count when health feature is turned off\n");
 473                 return -EINVAL;
 474         }
 475
 476         if (value > lnet_transaction_timeout) {
 477                 mutex_unlock(&the_lnet.ln_api_mutex);
 478                 CERROR("Invalid value for lnet_retry_count (%lu). "
 479                        "Has to be smaller than lnet_transaction_timeout (%u)\n",
 480                        value, lnet_transaction_timeout);
 481                 return -EINVAL;
 482         }
 483
 484         *retry_count = value;
 485
 486         /* Update the lnet_lnd_timeout now that we've modified the
 487          * retry count
 488          */
 489         lnet_set_lnd_timeout();
 490
 491         mutex_unlock(&the_lnet.ln_api_mutex);
 492
 493         return 0;
 494 }
 495
 496 static int
 497 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
 498 {
 499         int value, rc;
 500
 501         rc = kstrtoint(val, 0, &value);
 502         if (rc) {
 503                 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
 504                 return rc;
 505         }
 506
 507         if (value < LNET_INTERFACES_MIN) {
 508                 CWARN("max interfaces provided are too small, setting to %d\n",
 509                       LNET_INTERFACES_MAX_DEFAULT);
 510                 value = LNET_INTERFACES_MAX_DEFAULT;
 511         }
 512
 513         *(int *)kp->arg = value;
 514
 515         return 0;
 516 }
 517
 518 static int
 519 response_tracking_set(const char *val, cfs_kernel_param_arg_t *kp)
 520 {
 521         int rc;
 522         unsigned long new_value;
 523
 524         rc = kstrtoul(val, 0, &new_value);
 525         if (rc) {
 526                 CERROR("Invalid value for 'lnet_response_tracking'\n");
 527                 return -EINVAL;
 528         }
 529
 530         if (new_value < 0 || new_value > 3) {
 531                 CWARN("Invalid value (%lu) for 'lnet_response_tracking'\n",
 532                       new_value);
 533                 return -EINVAL;
 534         }
 535
 536         lnet_response_tracking = new_value;
 537
 538         return 0;
 539 }
 540
 541 static const char *
 542 lnet_get_routes(void)
 543 {
 544         return routes;
 545 }
 546
 547 static const char *
 548 lnet_get_networks(void)
 549 {
 550         const char *nets;
 551         int rc;
 552
 553         if (*networks != 0 && *ip2nets != 0) {
 554                 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
 555                                    "'ip2nets' but not both at once\n");
 556                 return NULL;
 557         }
 558
 559         if (*ip2nets != 0) {
 560                 rc = lnet_parse_ip2nets(&nets, ip2nets);
 561                 return (rc == 0) ? nets : NULL;
 562         }
 563
 564         if (*networks != 0)
 565                 return networks;
 566
 567         return "tcp";
 568 }
 569
 570 static void
 571 lnet_init_locks(void)
 572 {
 573         spin_lock_init(&the_lnet.ln_eq_wait_lock);
 574         spin_lock_init(&the_lnet.ln_msg_resend_lock);
 575         init_completion(&the_lnet.ln_mt_wait_complete);
 576         mutex_init(&the_lnet.ln_lnd_mutex);
 577 }
 578
 579 struct kmem_cache *lnet_mes_cachep;        /* MEs kmem_cache */
 580 struct kmem_cache *lnet_small_mds_cachep;  /* <= LNET_SMALL_MD_SIZE bytes
 581                                             *  MDs kmem_cache */
 582 struct kmem_cache *lnet_udsp_cachep;       /* udsp cache */
 583 struct kmem_cache *lnet_rspt_cachep;       /* response tracker cache */
 584 struct kmem_cache *lnet_msg_cachep;
 585
 586 static int
 587 lnet_slab_setup(void)
 588 {
 589         /* create specific kmem_cache for MEs and small MDs (i.e., originally
 590          * allocated in <size-xxx> kmem_cache).
 591          */
 592         lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
 593                                             0, 0, NULL);
 594         if (!lnet_mes_cachep)
 595                 return -ENOMEM;
 596
 597         lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
 598                                                   LNET_SMALL_MD_SIZE, 0, 0,
 599                                                   NULL);
 600         if (!lnet_small_mds_cachep)
 601                 return -ENOMEM;
 602
 603         lnet_udsp_cachep = kmem_cache_create("lnet_udsp",
 604                                              sizeof(struct lnet_udsp),
 605                                              0, 0, NULL);
 606         if (!lnet_udsp_cachep)
 607                 return -ENOMEM;
 608
 609         lnet_rspt_cachep = kmem_cache_create("lnet_rspt", sizeof(struct lnet_rsp_tracker),
 610                                             0, 0, NULL);
 611         if (!lnet_rspt_cachep)
 612                 return -ENOMEM;
 613
 614         lnet_msg_cachep = kmem_cache_create("lnet_msg", sizeof(struct lnet_msg),
 615                                             0, 0, NULL);
 616         if (!lnet_msg_cachep)
 617                 return -ENOMEM;
 618
 619         return 0;
 620 }
 621
 622 static void
 623 lnet_slab_cleanup(void)
 624 {
 625         if (lnet_msg_cachep) {
 626                 kmem_cache_destroy(lnet_msg_cachep);
 627                 lnet_msg_cachep = NULL;
 628         }
 629
 630         if (lnet_rspt_cachep) {
 631                 kmem_cache_destroy(lnet_rspt_cachep);
 632                 lnet_rspt_cachep = NULL;
 633         }
 634
 635         if (lnet_udsp_cachep) {
 636                 kmem_cache_destroy(lnet_udsp_cachep);
 637                 lnet_udsp_cachep = NULL;
 638         }
 639
 640         if (lnet_small_mds_cachep) {
 641                 kmem_cache_destroy(lnet_small_mds_cachep);
 642                 lnet_small_mds_cachep = NULL;
 643         }
 644
 645         if (lnet_mes_cachep) {
 646                 kmem_cache_destroy(lnet_mes_cachep);
 647                 lnet_mes_cachep = NULL;
 648         }
 649 }
 650
 651 static int
 652 lnet_create_remote_nets_table(void)
 653 {
 654         int               i;
 655         struct list_head *hash;
 656
 657         LASSERT(the_lnet.ln_remote_nets_hash == NULL);
 658         LASSERT(the_lnet.ln_remote_nets_hbits > 0);
 659         CFS_ALLOC_PTR_ARRAY(hash, LNET_REMOTE_NETS_HASH_SIZE);
 660         if (hash == NULL) {
 661                 CERROR("Failed to create remote nets hash table\n");
 662                 return -ENOMEM;
 663         }
 664
 665         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 666                 INIT_LIST_HEAD(&hash[i]);
 667         the_lnet.ln_remote_nets_hash = hash;
 668         return 0;
 669 }
 670
 671 static void
 672 lnet_destroy_remote_nets_table(void)
 673 {
 674         int i;
 675
 676         if (the_lnet.ln_remote_nets_hash == NULL)
 677                 return;
 678
 679         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 680                 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
 681
 682         CFS_FREE_PTR_ARRAY(the_lnet.ln_remote_nets_hash,
 683                            LNET_REMOTE_NETS_HASH_SIZE);
 684         the_lnet.ln_remote_nets_hash = NULL;
 685 }
 686
 687 static void
 688 lnet_destroy_locks(void)
 689 {
 690         if (the_lnet.ln_res_lock != NULL) {
 691                 cfs_percpt_lock_free(the_lnet.ln_res_lock);
 692                 the_lnet.ln_res_lock = NULL;
 693         }
 694
 695         if (the_lnet.ln_net_lock != NULL) {
 696                 cfs_percpt_lock_free(the_lnet.ln_net_lock);
 697                 the_lnet.ln_net_lock = NULL;
 698         }
 699 }
 700
 701 static int
 702 lnet_create_locks(void)
 703 {
 704         lnet_init_locks();
 705
 706         the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 707         if (the_lnet.ln_res_lock == NULL)
 708                 goto failed;
 709
 710         the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 711         if (the_lnet.ln_net_lock == NULL)
 712                 goto failed;
 713
 714         return 0;
 715
 716  failed:
 717         lnet_destroy_locks();
 718         return -ENOMEM;
 719 }
 720
 721 static void lnet_assert_wire_constants(void)
 722 {
 723         /* Wire protocol assertions generated by 'wirecheck'
 724          * running on Linux robert.bartonsoftware.com 2.6.8-1.521
 725          * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
 726          * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7)
 727          */
 728
 729         /* Constants... */
 730         BUILD_BUG_ON(LNET_PROTO_TCP_MAGIC != 0xeebc0ded);
 731         BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MAJOR != 1);
 732         BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MINOR != 0);
 733         BUILD_BUG_ON(LNET_MSG_ACK != 0);
 734         BUILD_BUG_ON(LNET_MSG_PUT != 1);
 735         BUILD_BUG_ON(LNET_MSG_GET != 2);
 736         BUILD_BUG_ON(LNET_MSG_REPLY != 3);
 737         BUILD_BUG_ON(LNET_MSG_HELLO != 4);
 738
 739         BUILD_BUG_ON((int)sizeof(lnet_nid_t) != 8);
 740         BUILD_BUG_ON((int)sizeof(lnet_pid_t) != 4);
 741
 742         /* Checks for struct lnet_nid */
 743         BUILD_BUG_ON((int)sizeof(struct lnet_nid) != 20);
 744         BUILD_BUG_ON((int)offsetof(struct lnet_nid, nid_size) != 0);
 745         BUILD_BUG_ON((int)sizeof(((struct lnet_nid *)0)->nid_size) != 1);
 746         BUILD_BUG_ON((int)offsetof(struct lnet_nid, nid_type) != 1);
 747         BUILD_BUG_ON((int)sizeof(((struct lnet_nid *)0)->nid_type) != 1);
 748         BUILD_BUG_ON((int)offsetof(struct lnet_nid, nid_num) != 2);
 749         BUILD_BUG_ON((int)sizeof(((struct lnet_nid *)0)->nid_num) != 2);
 750         BUILD_BUG_ON((int)offsetof(struct lnet_nid, nid_addr) != 4);
 751         BUILD_BUG_ON((int)sizeof(((struct lnet_nid *)0)->nid_addr) != 16);
 752
 753         /* Checks for struct lnet_process_id_packed */
 754         BUILD_BUG_ON((int)sizeof(struct lnet_process_id_packed) != 12);
 755         BUILD_BUG_ON((int)offsetof(struct lnet_process_id_packed, nid) != 0);
 756         BUILD_BUG_ON((int)sizeof(((struct lnet_process_id_packed *)0)->nid) != 8);
 757         BUILD_BUG_ON((int)offsetof(struct lnet_process_id_packed, pid) != 8);
 758         BUILD_BUG_ON((int)sizeof(((struct lnet_process_id_packed *)0)->pid) != 4);
 759
 760         /* Checks for struct lnet_handle_wire */
 761         BUILD_BUG_ON((int)sizeof(struct lnet_handle_wire) != 16);
 762         BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
 763                                    wh_interface_cookie) != 0);
 764         BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) != 8);
 765         BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
 766                                    wh_object_cookie) != 8);
 767         BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) != 8);
 768
 769         /* Checks for struct struct lnet_magicversion */
 770         BUILD_BUG_ON((int)sizeof(struct lnet_magicversion) != 8);
 771         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, magic) != 0);
 772         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->magic) != 4);
 773         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, version_major) != 4);
 774         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_major) != 2);
 775         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion,
 776                                    version_minor) != 6);
 777         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_minor) != 2);
 778
 779         /* Checks for struct _lnet_hdr_nid4 */
 780         BUILD_BUG_ON((int)sizeof(struct _lnet_hdr_nid4) != 72);
 781         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, dest_nid) != 0);
 782         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->dest_nid) != 8);
 783         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, src_nid) != 8);
 784         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->src_nid) != 8);
 785         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, dest_pid) != 16);
 786         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->dest_pid) != 4);
 787         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, src_pid) != 20);
 788         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->src_pid) != 4);
 789         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, type) != 24);
 790         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->type) != 4);
 791         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, payload_length) != 28);
 792         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->payload_length) != 4);
 793         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg) != 32);
 794         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg) != 40);
 795
 796         /* Ack */
 797         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.ack.dst_wmd) != 32);
 798         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.ack.dst_wmd) != 16);
 799         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.ack.match_bits) != 48);
 800         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.ack.match_bits) != 8);
 801         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.ack.mlength) != 56);
 802         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.ack.mlength) != 4);
 803
 804         /* Put */
 805         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.put.ack_wmd) != 32);
 806         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.put.ack_wmd) != 16);
 807         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.put.match_bits) != 48);
 808         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.put.match_bits) != 8);
 809         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.put.hdr_data) != 56);
 810         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.put.hdr_data) != 8);
 811         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.put.ptl_index) != 64);
 812         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.put.ptl_index) != 4);
 813         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.put.offset) != 68);
 814         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.put.offset) != 4);
 815
 816         /* Get */
 817         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.get.return_wmd) != 32);
 818         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.get.return_wmd) != 16);
 819         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.get.match_bits) != 48);
 820         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.get.match_bits) != 8);
 821         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.get.ptl_index) != 56);
 822         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.get.ptl_index) != 4);
 823         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.get.src_offset) != 60);
 824         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.get.src_offset) != 4);
 825         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.get.sink_length) != 64);
 826         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.get.sink_length) != 4);
 827
 828         /* Reply */
 829         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.reply.dst_wmd) != 32);
 830         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.reply.dst_wmd) != 16);
 831
 832         /* Hello */
 833         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.hello.incarnation) != 32);
 834         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.hello.incarnation) != 8);
 835         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.hello.type) != 40);
 836         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.hello.type) != 4);
 837
 838         /* Checks for struct lnet_ni_status and related constants */
 839         BUILD_BUG_ON(LNET_NI_STATUS_INVALID != 0x00000000);
 840         BUILD_BUG_ON(LNET_NI_STATUS_UP != 0x15aac0de);
 841         BUILD_BUG_ON(LNET_NI_STATUS_DOWN != 0xdeadface);
 842
 843         /* Checks for struct lnet_ni_status */
 844         BUILD_BUG_ON((int)sizeof(struct lnet_ni_status) != 16);
 845         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_nid) != 0);
 846         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) != 8);
 847         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_status) != 8);
 848         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_status) != 4);
 849         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_unused) != 12);
 850         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) != 4);
 851
 852         /* Checks for struct lnet_ping_info and related constants */
 853         BUILD_BUG_ON(LNET_PROTO_PING_MAGIC != 0x70696E67);
 854         BUILD_BUG_ON(LNET_PING_FEAT_INVAL != 0);
 855         BUILD_BUG_ON(LNET_PING_FEAT_BASE != 1);
 856         BUILD_BUG_ON(LNET_PING_FEAT_NI_STATUS != 2);
 857         BUILD_BUG_ON(LNET_PING_FEAT_RTE_DISABLED != 4);
 858         BUILD_BUG_ON(LNET_PING_FEAT_MULTI_RAIL != 8);
 859         BUILD_BUG_ON(LNET_PING_FEAT_DISCOVERY != 16);
 860         BUILD_BUG_ON(LNET_PING_FEAT_BITS != 31);
 861
 862         /* Checks for struct lnet_ping_info */
 863         BUILD_BUG_ON((int)sizeof(struct lnet_ping_info) != 16);
 864         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_magic) != 0);
 865         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) != 4);
 866         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_features) != 4);
 867         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_features) != 4);
 868         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_pid) != 8);
 869         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) != 4);
 870         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_nnis) != 12);
 871         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) != 4);
 872         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_ni) != 16);
 873         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) != 0);
 874
 875         /* Acceptor connection request */
 876         BUILD_BUG_ON(LNET_PROTO_ACCEPTOR_VERSION != 1);
 877
 878         /* Checks for struct lnet_acceptor_connreq */
 879         BUILD_BUG_ON((int)sizeof(struct lnet_acceptor_connreq) != 16);
 880         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_magic) != 0);
 881         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_magic) != 4);
 882         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_version) != 4);
 883         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_version) != 4);
 884         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_nid) != 8);
 885         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_nid) != 8);
 886
 887         /* Checks for struct lnet_acceptor_connreq_v2 */
 888         BUILD_BUG_ON((int)sizeof(struct lnet_acceptor_connreq_v2) != 28);
 889         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq_v2, acr_magic) != 0);
 890         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq_v2 *)0)->acr_magic) != 4);
 891         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq_v2, acr_version) != 4);
 892         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq_v2 *)0)->acr_version) != 4);
 893         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq_v2, acr_nid) != 8);
 894         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq_v2 *)0)->acr_nid) != 20);
 895
 896         /* Checks for struct lnet_counters_common */
 897         BUILD_BUG_ON((int)sizeof(struct lnet_counters_common) != 60);
 898         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_msgs_alloc) != 0);
 899         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_msgs_alloc) != 4);
 900         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_msgs_max) != 4);
 901         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_msgs_max) != 4);
 902         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_errors) != 8);
 903         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_errors) != 4);
 904         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_send_count) != 12);
 905         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_send_count) != 4);
 906         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_recv_count) != 16);
 907         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_recv_count) != 4);
 908         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_route_count) != 20);
 909         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_route_count) != 4);
 910         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_drop_count) != 24);
 911         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_drop_count) != 4);
 912         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_send_length) != 28);
 913         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_send_length) != 8);
 914         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_recv_length) != 36);
 915         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_recv_length) != 8);
 916         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_route_length) != 44);
 917         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_route_length) != 8);
 918         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_drop_length) != 52);
 919         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_drop_length) != 8);
 920 }
 921
 922 static const struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
 923 {
 924         const struct lnet_lnd *lnd;
 925
 926         /* holding lnd mutex */
 927         if (type >= NUM_LNDS)
 928                 return NULL;
 929         lnd = the_lnet.ln_lnds[type];
 930         LASSERT(!lnd || lnd->lnd_type == type);
 931
 932         return lnd;
 933 }
 934
 935 unsigned int
 936 lnet_get_lnd_timeout(void)
 937 {
 938         return lnet_lnd_timeout;
 939 }
 940 EXPORT_SYMBOL(lnet_get_lnd_timeout);
 941
 942 void
 943 lnet_register_lnd(const struct lnet_lnd *lnd)
 944 {
 945         mutex_lock(&the_lnet.ln_lnd_mutex);
 946
 947         LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
 948         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
 949
 950         the_lnet.ln_lnds[lnd->lnd_type] = lnd;
 951
 952         CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
 953
 954         mutex_unlock(&the_lnet.ln_lnd_mutex);
 955 }
 956 EXPORT_SYMBOL(lnet_register_lnd);
 957
 958 void
 959 lnet_unregister_lnd(const struct lnet_lnd *lnd)
 960 {
 961         mutex_lock(&the_lnet.ln_lnd_mutex);
 962
 963         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
 964
 965         the_lnet.ln_lnds[lnd->lnd_type] = NULL;
 966         CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
 967
 968         mutex_unlock(&the_lnet.ln_lnd_mutex);
 969 }
 970 EXPORT_SYMBOL(lnet_unregister_lnd);
 971
 972 static void
 973 lnet_counters_get_common_locked(struct lnet_counters_common *common)
 974 {
 975         struct lnet_counters *ctr;
 976         int i;
 977
 978         /* FIXME !!! Their is no assert_lnet_net_locked() to ensure this
 979          * actually called under the protection of the lnet_net_lock.
 980          */
 981         memset(common, 0, sizeof(*common));
 982
 983         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
 984                 common->lcc_msgs_max     += ctr->lct_common.lcc_msgs_max;
 985                 common->lcc_msgs_alloc   += ctr->lct_common.lcc_msgs_alloc;
 986                 common->lcc_errors       += ctr->lct_common.lcc_errors;
 987                 common->lcc_send_count   += ctr->lct_common.lcc_send_count;
 988                 common->lcc_recv_count   += ctr->lct_common.lcc_recv_count;
 989                 common->lcc_route_count  += ctr->lct_common.lcc_route_count;
 990                 common->lcc_drop_count   += ctr->lct_common.lcc_drop_count;
 991                 common->lcc_send_length  += ctr->lct_common.lcc_send_length;
 992                 common->lcc_recv_length  += ctr->lct_common.lcc_recv_length;
 993                 common->lcc_route_length += ctr->lct_common.lcc_route_length;
 994                 common->lcc_drop_length  += ctr->lct_common.lcc_drop_length;
 995         }
 996 }
 997
 998 void
 999 lnet_counters_get_common(struct lnet_counters_common *common)
1000 {
1001         lnet_net_lock(LNET_LOCK_EX);
1002         lnet_counters_get_common_locked(common);
1003         lnet_net_unlock(LNET_LOCK_EX);
1004 }
1005 EXPORT_SYMBOL(lnet_counters_get_common);
1006
1007 int
1008 lnet_counters_get(struct lnet_counters *counters)
1009 {
1010         struct lnet_counters *ctr;
1011         struct lnet_counters_health *health = &counters->lct_health;
1012         int i, rc = 0;
1013
1014         memset(counters, 0, sizeof(*counters));
1015
1016         lnet_net_lock(LNET_LOCK_EX);
1017
1018         if (the_lnet.ln_state != LNET_STATE_RUNNING)
1019                 GOTO(out_unlock, rc = -ENODEV);
1020
1021         lnet_counters_get_common_locked(&counters->lct_common);
1022
1023         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
1024                 health->lch_rst_alloc    += ctr->lct_health.lch_rst_alloc;
1025                 health->lch_resend_count += ctr->lct_health.lch_resend_count;
1026                 health->lch_response_timeout_count +=
1027                                 ctr->lct_health.lch_response_timeout_count;
1028                 health->lch_local_interrupt_count +=
1029                                 ctr->lct_health.lch_local_interrupt_count;
1030                 health->lch_local_dropped_count +=
1031                                 ctr->lct_health.lch_local_dropped_count;
1032                 health->lch_local_aborted_count +=
1033                                 ctr->lct_health.lch_local_aborted_count;
1034                 health->lch_local_no_route_count +=
1035                                 ctr->lct_health.lch_local_no_route_count;
1036                 health->lch_local_timeout_count +=
1037                                 ctr->lct_health.lch_local_timeout_count;
1038                 health->lch_local_error_count +=
1039                                 ctr->lct_health.lch_local_error_count;
1040                 health->lch_remote_dropped_count +=
1041                                 ctr->lct_health.lch_remote_dropped_count;
1042                 health->lch_remote_error_count +=
1043                                 ctr->lct_health.lch_remote_error_count;
1044                 health->lch_remote_timeout_count +=
1045                                 ctr->lct_health.lch_remote_timeout_count;
1046                 health->lch_network_timeout_count +=
1047                                 ctr->lct_health.lch_network_timeout_count;
1048         }
1049 out_unlock:
1050         lnet_net_unlock(LNET_LOCK_EX);
1051         return rc;
1052 }
1053 EXPORT_SYMBOL(lnet_counters_get);
1054
1055 void
1056 lnet_counters_reset(void)
1057 {
1058         struct lnet_counters *counters;
1059         int             i;
1060
1061         lnet_net_lock(LNET_LOCK_EX);
1062
1063         if (the_lnet.ln_state != LNET_STATE_RUNNING)
1064                 goto avoid_reset;
1065
1066         cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
1067                 memset(counters, 0, sizeof(struct lnet_counters));
1068 avoid_reset:
1069         lnet_net_unlock(LNET_LOCK_EX);
1070 }
1071
1072 static char *
1073 lnet_res_type2str(int type)
1074 {
1075         switch (type) {
1076         default:
1077                 LBUG();
1078         case LNET_COOKIE_TYPE_MD:
1079                 return "MD";
1080         case LNET_COOKIE_TYPE_ME:
1081                 return "ME";
1082         case LNET_COOKIE_TYPE_EQ:
1083                 return "EQ";
1084         }
1085 }
1086
1087 static void
1088 lnet_res_container_cleanup(struct lnet_res_container *rec)
1089 {
1090         int     count = 0;
1091
1092         if (rec->rec_type == 0) /* not set yet, it's uninitialized */
1093                 return;
1094
1095         while (!list_empty(&rec->rec_active)) {
1096                 struct list_head *e = rec->rec_active.next;
1097
1098                 list_del_init(e);
1099                 if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
1100                         lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
1101
1102                 } else { /* NB: Active MEs should be attached on portals */
1103                         LBUG();
1104                 }
1105                 count++;
1106         }
1107
1108         if (count > 0) {
1109                 /* Found alive MD/ME/EQ, user really should unlink/free
1110                  * all of them before finalize LNet, but if someone didn't,
1111                  * we have to recycle garbage for him */
1112                 CERROR("%d active elements on exit of %s container\n",
1113                        count, lnet_res_type2str(rec->rec_type));
1114         }
1115
1116         if (rec->rec_lh_hash != NULL) {
1117                 CFS_FREE_PTR_ARRAY(rec->rec_lh_hash, LNET_LH_HASH_SIZE);
1118                 rec->rec_lh_hash = NULL;
1119         }
1120
1121         rec->rec_type = 0; /* mark it as finalized */
1122 }
1123
1124 static int
1125 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
1126 {
1127         int     rc = 0;
1128         int     i;
1129
1130         LASSERT(rec->rec_type == 0);
1131
1132         rec->rec_type = type;
1133         INIT_LIST_HEAD(&rec->rec_active);
1134
1135         rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1136
1137         /* Arbitrary choice of hash table size */
1138         LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1139                          LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1140         if (rec->rec_lh_hash == NULL) {
1141                 rc = -ENOMEM;
1142                 goto out;
1143         }
1144
1145         for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1146                 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1147
1148         return 0;
1149
1150 out:
1151         CERROR("Failed to setup %s resource container\n",
1152                lnet_res_type2str(type));
1153         lnet_res_container_cleanup(rec);
1154         return rc;
1155 }
1156
1157 static void
1158 lnet_res_containers_destroy(struct lnet_res_container **recs)
1159 {
1160         struct lnet_res_container       *rec;
1161         int                             i;
1162
1163         cfs_percpt_for_each(rec, i, recs)
1164                 lnet_res_container_cleanup(rec);
1165
1166         cfs_percpt_free(recs);
1167 }
1168
1169 static struct lnet_res_container **
1170 lnet_res_containers_create(int type)
1171 {
1172         struct lnet_res_container       **recs;
1173         struct lnet_res_container       *rec;
1174         int                             rc;
1175         int                             i;
1176
1177         recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1178         if (recs == NULL) {
1179                 CERROR("Failed to allocate %s resource containers\n",
1180                        lnet_res_type2str(type));
1181                 return NULL;
1182         }
1183
1184         cfs_percpt_for_each(rec, i, recs) {
1185                 rc = lnet_res_container_setup(rec, i, type);
1186                 if (rc != 0) {
1187                         lnet_res_containers_destroy(recs);
1188                         return NULL;
1189                 }
1190         }
1191
1192         return recs;
1193 }
1194
1195 struct lnet_libhandle *
1196 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1197 {
1198         /* ALWAYS called with lnet_res_lock held */
1199         struct list_head        *head;
1200         struct lnet_libhandle   *lh;
1201         unsigned int            hash;
1202
1203         if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1204                 return NULL;
1205
1206         hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1207         head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1208
1209         list_for_each_entry(lh, head, lh_hash_chain) {
1210                 if (lh->lh_cookie == cookie)
1211                         return lh;
1212         }
1213
1214         return NULL;
1215 }
1216
1217 void
1218 lnet_res_lh_initialize(struct lnet_res_container *rec,
1219                        struct lnet_libhandle *lh)
1220 {
1221         /* ALWAYS called with lnet_res_lock held */
1222         unsigned int    ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1223         unsigned int    hash;
1224
1225         lh->lh_cookie = rec->rec_lh_cookie;
1226         rec->rec_lh_cookie += 1 << ibits;
1227
1228         hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1229
1230         list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1231 }
1232
1233 struct list_head **
1234 lnet_create_array_of_queues(void)
1235 {
1236         struct list_head **qs;
1237         struct list_head *q;
1238         int i;
1239
1240         qs = cfs_percpt_alloc(lnet_cpt_table(),
1241                               sizeof(struct list_head));
1242         if (!qs) {
1243                 CERROR("Failed to allocate queues\n");
1244                 return NULL;
1245         }
1246
1247         cfs_percpt_for_each(q, i, qs)
1248                 INIT_LIST_HEAD(q);
1249
1250         return qs;
1251 }
1252
1253 static int lnet_unprepare(void);
1254
1255 static int
1256 lnet_prepare(lnet_pid_t requested_pid)
1257 {
1258         /* Prepare to bring up the network */
1259         struct lnet_res_container **recs;
1260         int                       rc = 0;
1261
1262         if (requested_pid == LNET_PID_ANY) {
1263                 /* Don't instantiate LNET just for me */
1264                 return -ENETDOWN;
1265         }
1266
1267         LASSERT(the_lnet.ln_refcount == 0);
1268
1269         the_lnet.ln_routing = 0;
1270
1271         LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1272         the_lnet.ln_pid = requested_pid;
1273
1274         INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1275         INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1276         INIT_LIST_HEAD(&the_lnet.ln_nets);
1277         INIT_LIST_HEAD(&the_lnet.ln_routers);
1278         INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1279         INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1280         INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1281         INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1282         INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1283         INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1284         INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1285         INIT_LIST_HEAD(&the_lnet.ln_udsp_list);
1286         init_waitqueue_head(&the_lnet.ln_dc_waitq);
1287         the_lnet.ln_mt_handler = NULL;
1288         init_completion(&the_lnet.ln_started);
1289
1290         rc = lnet_slab_setup();
1291         if (rc != 0)
1292                 goto failed;
1293
1294         rc = lnet_create_remote_nets_table();
1295         if (rc != 0)
1296                 goto failed;
1297
1298         /*
1299          * NB the interface cookie in wire handles guards against delayed
1300          * replies and ACKs appearing valid after reboot.
1301          */
1302         the_lnet.ln_interface_cookie = ktime_get_real_ns();
1303
1304         the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1305                                                 sizeof(struct lnet_counters));
1306         if (the_lnet.ln_counters == NULL) {
1307                 CERROR("Failed to allocate counters for LNet\n");
1308                 rc = -ENOMEM;
1309                 goto failed;
1310         }
1311
1312         rc = lnet_peer_tables_create();
1313         if (rc != 0)
1314                 goto failed;
1315
1316         rc = lnet_msg_containers_create();
1317         if (rc != 0)
1318                 goto failed;
1319
1320         rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1321                                       LNET_COOKIE_TYPE_EQ);
1322         if (rc != 0)
1323                 goto failed;
1324
1325         recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1326         if (recs == NULL) {
1327                 rc = -ENOMEM;
1328                 goto failed;
1329         }
1330
1331         the_lnet.ln_md_containers = recs;
1332
1333         rc = lnet_portals_create();
1334         if (rc != 0) {
1335                 CERROR("Failed to create portals for LNet: %d\n", rc);
1336                 goto failed;
1337         }
1338
1339         the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1340         if (!the_lnet.ln_mt_zombie_rstqs) {
1341                 rc = -ENOMEM;
1342                 goto failed;
1343         }
1344
1345         return 0;
1346
1347  failed:
1348         lnet_unprepare();
1349         return rc;
1350 }
1351
1352 static int
1353 lnet_unprepare (void)
1354 {
1355         /* NB no LNET_LOCK since this is the last reference.  All LND instances
1356          * have shut down already, so it is safe to unlink and free all
1357          * descriptors, even those that appear committed to a network op (eg MD
1358          * with non-zero pending count) */
1359
1360         lnet_fail_nid(LNET_NID_ANY, 0);
1361
1362         LASSERT(the_lnet.ln_refcount == 0);
1363         LASSERT(list_empty(&the_lnet.ln_test_peers));
1364         LASSERT(list_empty(&the_lnet.ln_nets));
1365
1366         if (the_lnet.ln_mt_zombie_rstqs) {
1367                 lnet_clean_zombie_rstqs();
1368                 the_lnet.ln_mt_zombie_rstqs = NULL;
1369         }
1370
1371         lnet_assert_handler_unused(the_lnet.ln_mt_handler);
1372         the_lnet.ln_mt_handler = NULL;
1373
1374         lnet_portals_destroy();
1375
1376         if (the_lnet.ln_md_containers != NULL) {
1377                 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1378                 the_lnet.ln_md_containers = NULL;
1379         }
1380
1381         lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1382
1383         lnet_msg_containers_destroy();
1384         lnet_peer_uninit();
1385         lnet_rtrpools_free(0);
1386
1387         if (the_lnet.ln_counters != NULL) {
1388                 cfs_percpt_free(the_lnet.ln_counters);
1389                 the_lnet.ln_counters = NULL;
1390         }
1391         lnet_destroy_remote_nets_table();
1392         lnet_udsp_destroy(true);
1393         lnet_slab_cleanup();
1394
1395         return 0;
1396 }
1397
1398 struct lnet_ni  *
1399 lnet_net2ni_locked(__u32 net_id, int cpt)
1400 {
1401         struct lnet_ni   *ni;
1402         struct lnet_net  *net;
1403
1404         LASSERT(cpt != LNET_LOCK_EX);
1405
1406         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1407                 if (net->net_id == net_id) {
1408                         ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1409                                         ni_netlist);
1410                         return ni;
1411                 }
1412         }
1413
1414         return NULL;
1415 }
1416
1417 struct lnet_ni *
1418 lnet_net2ni_addref(__u32 net)
1419 {
1420         struct lnet_ni *ni;
1421
1422         lnet_net_lock(0);
1423         ni = lnet_net2ni_locked(net, 0);
1424         if (ni)
1425                 lnet_ni_addref_locked(ni, 0);
1426         lnet_net_unlock(0);
1427
1428         return ni;
1429 }
1430 EXPORT_SYMBOL(lnet_net2ni_addref);
1431
1432 struct lnet_net *
1433 lnet_get_net_locked(__u32 net_id)
1434 {
1435         struct lnet_net  *net;
1436
1437         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1438                 if (net->net_id == net_id)
1439                         return net;
1440         }
1441
1442         return NULL;
1443 }
1444
1445 void
1446 lnet_net_clr_pref_rtrs(struct lnet_net *net)
1447 {
1448         struct list_head zombies;
1449         struct lnet_nid_list *ne;
1450         struct lnet_nid_list *tmp;
1451
1452         INIT_LIST_HEAD(&zombies);
1453
1454         lnet_net_lock(LNET_LOCK_EX);
1455         list_splice_init(&net->net_rtr_pref_nids, &zombies);
1456         lnet_net_unlock(LNET_LOCK_EX);
1457
1458         list_for_each_entry_safe(ne, tmp, &zombies, nl_list) {
1459                 list_del_init(&ne->nl_list);
1460                 LIBCFS_FREE(ne, sizeof(*ne));
1461         }
1462 }
1463
1464 int
1465 lnet_net_add_pref_rtr(struct lnet_net *net,
1466                       struct lnet_nid *gw_nid)
1467 __must_hold(&the_lnet.ln_api_mutex)
1468 {
1469         struct lnet_nid_list *ne;
1470
1471         /* This function is called with api_mutex held. When the api_mutex
1472          * is held the list can not be modified, as it is only modified as
1473          * a result of applying a UDSP and that happens under api_mutex
1474          * lock.
1475          */
1476         list_for_each_entry(ne, &net->net_rtr_pref_nids, nl_list) {
1477                 if (nid_same(&ne->nl_nid, gw_nid))
1478                         return -EEXIST;
1479         }
1480
1481         LIBCFS_ALLOC(ne, sizeof(*ne));
1482         if (!ne)
1483                 return -ENOMEM;
1484
1485         ne->nl_nid = *gw_nid;
1486
1487         /* Lock the cpt to protect against addition and checks in the
1488          * selection algorithm
1489          */
1490         lnet_net_lock(LNET_LOCK_EX);
1491         list_add(&ne->nl_list, &net->net_rtr_pref_nids);
1492         lnet_net_unlock(LNET_LOCK_EX);
1493
1494         return 0;
1495 }
1496
1497 bool
1498 lnet_net_is_pref_rtr_locked(struct lnet_net *net, struct lnet_nid *rtr_nid)
1499 {
1500         struct lnet_nid_list *ne;
1501
1502         CDEBUG(D_NET, "%s: rtr pref empty: %d\n",
1503                libcfs_net2str(net->net_id),
1504                list_empty(&net->net_rtr_pref_nids));
1505
1506         if (list_empty(&net->net_rtr_pref_nids))
1507                 return false;
1508
1509         list_for_each_entry(ne, &net->net_rtr_pref_nids, nl_list) {
1510                 CDEBUG(D_NET, "Comparing pref %s with gw %s\n",
1511                        libcfs_nidstr(&ne->nl_nid),
1512                        libcfs_nidstr(rtr_nid));
1513                 if (nid_same(rtr_nid, &ne->nl_nid))
1514                         return true;
1515         }
1516
1517         return false;
1518 }
1519
1520 static unsigned int
1521 lnet_nid4_cpt_hash(lnet_nid_t nid, unsigned int number)
1522 {
1523         __u64           key = nid;
1524         unsigned int    val;
1525
1526         LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1527
1528         if (number == 1)
1529                 return 0;
1530
1531         val = hash_long(key, LNET_CPT_BITS);
1532         /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1533         if (val < number)
1534                 return val;
1535
1536         return (unsigned int)(key + val + (val >> 1)) % number;
1537 }
1538
1539 unsigned int
1540 lnet_nid_cpt_hash(struct lnet_nid *nid, unsigned int number)
1541 {
1542         unsigned int val;
1543         u32 h = 0;
1544         int i;
1545
1546         LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1547
1548         if (number == 1)
1549                 return 0;
1550
1551         if (nid_is_nid4(nid))
1552                 return lnet_nid4_cpt_hash(lnet_nid_to_nid4(nid), number);
1553
1554         for (i = 0; i < 4; i++)
1555                 h = hash_32(nid->nid_addr[i]^h, 32);
1556         val = hash_32(LNET_NID_NET(nid) ^ h, LNET_CPT_BITS);
1557         if (val < number)
1558                 return val;
1559         return (unsigned int)(h + val + (val >> 1)) % number;
1560 }
1561
1562 int
1563 lnet_cpt_of_nid_locked(struct lnet_nid *nid, struct lnet_ni *ni)
1564 {
1565         struct lnet_net *net;
1566
1567         /* must called with hold of lnet_net_lock */
1568         if (LNET_CPT_NUMBER == 1)
1569                 return 0; /* the only one */
1570
1571         /*
1572          * If NI is provided then use the CPT identified in the NI cpt
1573          * list if one exists. If one doesn't exist, then that NI is
1574          * associated with all CPTs and it follows that the net it belongs
1575          * to is implicitly associated with all CPTs, so just hash the nid
1576          * and return that.
1577          */
1578         if (ni != NULL) {
1579                 if (ni->ni_cpts != NULL)
1580                         return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1581                                                              ni->ni_ncpts)];
1582                 else
1583                         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1584         }
1585
1586         /* no NI provided so look at the net */
1587         net = lnet_get_net_locked(LNET_NID_NET(nid));
1588
1589         if (net != NULL && net->net_cpts != NULL) {
1590                 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1591         }
1592
1593         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1594 }
1595
1596 int
1597 lnet_nid2cpt(struct lnet_nid *nid, struct lnet_ni *ni)
1598 {
1599         int     cpt;
1600         int     cpt2;
1601
1602         if (LNET_CPT_NUMBER == 1)
1603                 return 0; /* the only one */
1604
1605         cpt = lnet_net_lock_current();
1606
1607         cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1608
1609         lnet_net_unlock(cpt);
1610
1611         return cpt2;
1612 }
1613 EXPORT_SYMBOL(lnet_nid2cpt);
1614
1615 int
1616 lnet_cpt_of_nid(lnet_nid_t nid4, struct lnet_ni *ni)
1617 {
1618         struct lnet_nid nid;
1619
1620         if (LNET_CPT_NUMBER == 1)
1621                 return 0; /* the only one */
1622
1623         lnet_nid4_to_nid(nid4, &nid);
1624         return lnet_nid2cpt(&nid, ni);
1625 }
1626 EXPORT_SYMBOL(lnet_cpt_of_nid);
1627
1628 int
1629 lnet_islocalnet_locked(__u32 net_id)
1630 {
1631         struct lnet_net *net;
1632         bool local;
1633
1634         net = lnet_get_net_locked(net_id);
1635
1636         local = net != NULL;
1637
1638         return local;
1639 }
1640
1641 int
1642 lnet_islocalnet(__u32 net_id)
1643 {
1644         int cpt;
1645         bool local;
1646
1647         cpt = lnet_net_lock_current();
1648
1649         local = lnet_islocalnet_locked(net_id);
1650
1651         lnet_net_unlock(cpt);
1652
1653         return local;
1654 }
1655
1656 struct lnet_ni  *
1657 lnet_nid_to_ni_locked(struct lnet_nid *nid, int cpt)
1658 {
1659         struct lnet_net  *net;
1660         struct lnet_ni *ni;
1661
1662         LASSERT(cpt != LNET_LOCK_EX);
1663
1664         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1665                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1666                         if (nid_same(&ni->ni_nid, nid))
1667                                 return ni;
1668                 }
1669         }
1670
1671         return NULL;
1672 }
1673
1674 struct lnet_ni  *
1675 lnet_nid2ni_locked(lnet_nid_t nid4, int cpt)
1676 {
1677         struct lnet_nid nid;
1678
1679         lnet_nid4_to_nid(nid4, &nid);
1680         return lnet_nid_to_ni_locked(&nid, cpt);
1681 }
1682
1683 struct lnet_ni *
1684 lnet_nid2ni_addref(lnet_nid_t nid4)
1685 {
1686         struct lnet_ni *ni;
1687         struct lnet_nid nid;
1688
1689         lnet_nid4_to_nid(nid4, &nid);
1690
1691         lnet_net_lock(0);
1692         ni = lnet_nid_to_ni_locked(&nid, 0);
1693         if (ni)
1694                 lnet_ni_addref_locked(ni, 0);
1695         lnet_net_unlock(0);
1696
1697         return ni;
1698 }
1699 EXPORT_SYMBOL(lnet_nid2ni_addref);
1700
1701 struct lnet_ni *
1702 lnet_nid_to_ni_addref(struct lnet_nid *nid)
1703 {
1704         struct lnet_ni *ni;
1705
1706         lnet_net_lock(0);
1707         ni = lnet_nid_to_ni_locked(nid, 0);
1708         if (ni)
1709                 lnet_ni_addref_locked(ni, 0);
1710         lnet_net_unlock(0);
1711
1712         return ni;
1713 }
1714 EXPORT_SYMBOL(lnet_nid_to_ni_addref);
1715
1716 int
1717 lnet_islocalnid(struct lnet_nid *nid)
1718 {
1719         struct lnet_ni  *ni;
1720         int             cpt;
1721
1722         cpt = lnet_net_lock_current();
1723         ni = lnet_nid_to_ni_locked(nid, cpt);
1724         lnet_net_unlock(cpt);
1725
1726         return ni != NULL;
1727 }
1728
1729 int
1730 lnet_count_acceptor_nets(void)
1731 {
1732         /* Return the # of NIs that need the acceptor. */
1733         int              count = 0;
1734         struct lnet_net  *net;
1735         int              cpt;
1736
1737         cpt = lnet_net_lock_current();
1738         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1739                 /* all socklnd type networks should have the acceptor
1740                  * thread started */
1741                 if (net->net_lnd->lnd_accept != NULL)
1742                         count++;
1743         }
1744
1745         lnet_net_unlock(cpt);
1746
1747         return count;
1748 }
1749
1750 struct lnet_ping_buffer *
1751 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1752 {
1753         struct lnet_ping_buffer *pbuf;
1754
1755         LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1756         if (pbuf) {
1757                 pbuf->pb_nnis = nnis;
1758                 pbuf->pb_needs_post = false;
1759                 atomic_set(&pbuf->pb_refcnt, 1);
1760         }
1761
1762         return pbuf;
1763 }
1764
1765 void
1766 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1767 {
1768         LASSERT(atomic_read(&pbuf->pb_refcnt) == 0);
1769         LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1770 }
1771
1772 static struct lnet_ping_buffer *
1773 lnet_ping_target_create(int nnis)
1774 {
1775         struct lnet_ping_buffer *pbuf;
1776
1777         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1778         if (pbuf == NULL) {
1779                 CERROR("Can't allocate ping source [%d]\n", nnis);
1780                 return NULL;
1781         }
1782
1783         pbuf->pb_info.pi_nnis = nnis;
1784         pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1785         pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1786         pbuf->pb_info.pi_features =
1787                 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1788
1789         return pbuf;
1790 }
1791
1792 static inline int
1793 lnet_get_net_ni_count_locked(struct lnet_net *net)
1794 {
1795         struct lnet_ni  *ni;
1796         int             count = 0;
1797
1798         list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1799                 count++;
1800
1801         return count;
1802 }
1803
1804 static inline int
1805 lnet_get_net_ni_count_pre(struct lnet_net *net)
1806 {
1807         struct lnet_ni  *ni;
1808         int             count = 0;
1809
1810         list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1811                 count++;
1812
1813         return count;
1814 }
1815
1816 static inline int
1817 lnet_get_ni_count(void)
1818 {
1819         struct lnet_ni  *ni;
1820         struct lnet_net *net;
1821         int             count = 0;
1822
1823         lnet_net_lock(0);
1824
1825         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1826                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1827                         count++;
1828         }
1829
1830         lnet_net_unlock(0);
1831
1832         return count;
1833 }
1834
1835 void
1836 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1837 {
1838         struct lnet_ni_status *stat;
1839         int nnis;
1840         int i;
1841
1842         __swab32s(&pbuf->pb_info.pi_magic);
1843         __swab32s(&pbuf->pb_info.pi_features);
1844         __swab32s(&pbuf->pb_info.pi_pid);
1845         __swab32s(&pbuf->pb_info.pi_nnis);
1846         nnis = pbuf->pb_info.pi_nnis;
1847         if (nnis > pbuf->pb_nnis)
1848                 nnis = pbuf->pb_nnis;
1849         for (i = 0; i < nnis; i++) {
1850                 stat = &pbuf->pb_info.pi_ni[i];
1851                 __swab64s(&stat->ns_nid);
1852                 __swab32s(&stat->ns_status);
1853         }
1854 }
1855
1856 int
1857 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1858 {
1859         if (!pinfo)
1860                 return -EINVAL;
1861         if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1862                 return -EPROTO;
1863         if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1864                 return -EPROTO;
1865         /* Loopback is guaranteed to be present */
1866         if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1867                 return -ERANGE;
1868         if (LNET_PING_INFO_LONI(pinfo) != LNET_NID_LO_0)
1869                 return -EPROTO;
1870         return 0;
1871 }
1872
1873 static void
1874 lnet_ping_target_destroy(void)
1875 {
1876         struct lnet_net *net;
1877         struct lnet_ni  *ni;
1878
1879         lnet_net_lock(LNET_LOCK_EX);
1880
1881         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1882                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1883                         lnet_ni_lock(ni);
1884                         ni->ni_status = NULL;
1885                         lnet_ni_unlock(ni);
1886                 }
1887         }
1888
1889         lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1890         the_lnet.ln_ping_target = NULL;
1891
1892         lnet_net_unlock(LNET_LOCK_EX);
1893 }
1894
1895 static void
1896 lnet_ping_target_event_handler(struct lnet_event *event)
1897 {
1898         struct lnet_ping_buffer *pbuf = event->md_user_ptr;
1899
1900         if (event->unlinked)
1901                 lnet_ping_buffer_decref(pbuf);
1902 }
1903
1904 static int
1905 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1906                        struct lnet_handle_md *ping_mdh,
1907                        int ni_count, bool set_eq)
1908 {
1909         struct lnet_processid id = {
1910                 .nid = LNET_ANY_NID,
1911                 .pid = LNET_PID_ANY
1912         };
1913         struct lnet_me *me;
1914         struct lnet_md md = { NULL };
1915         int rc;
1916
1917         if (set_eq)
1918                 the_lnet.ln_ping_target_handler =
1919                         lnet_ping_target_event_handler;
1920
1921         *ppbuf = lnet_ping_target_create(ni_count);
1922         if (*ppbuf == NULL) {
1923                 rc = -ENOMEM;
1924                 goto fail_free_eq;
1925         }
1926
1927         /* Ping target ME/MD */
1928         me = LNetMEAttach(LNET_RESERVED_PORTAL, &id,
1929                           LNET_PROTO_PING_MATCHBITS, 0,
1930                           LNET_UNLINK, LNET_INS_AFTER);
1931         if (IS_ERR(me)) {
1932                 rc = PTR_ERR(me);
1933                 CERROR("Can't create ping target ME: %d\n", rc);
1934                 goto fail_decref_ping_buffer;
1935         }
1936
1937         /* initialize md content */
1938         md.start     = &(*ppbuf)->pb_info;
1939         md.length    = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1940         md.threshold = LNET_MD_THRESH_INF;
1941         md.max_size  = 0;
1942         md.options   = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1943                        LNET_MD_MANAGE_REMOTE;
1944         md.handler   = the_lnet.ln_ping_target_handler;
1945         md.user_ptr  = *ppbuf;
1946
1947         rc = LNetMDAttach(me, &md, LNET_RETAIN, ping_mdh);
1948         if (rc != 0) {
1949                 CERROR("Can't attach ping target MD: %d\n", rc);
1950                 goto fail_decref_ping_buffer;
1951         }
1952         lnet_ping_buffer_addref(*ppbuf);
1953
1954         return 0;
1955
1956 fail_decref_ping_buffer:
1957         LASSERT(atomic_read(&(*ppbuf)->pb_refcnt) == 1);
1958         lnet_ping_buffer_decref(*ppbuf);
1959         *ppbuf = NULL;
1960 fail_free_eq:
1961         return rc;
1962 }
1963
1964 static void
1965 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1966                     struct lnet_handle_md *ping_mdh)
1967 {
1968         LNetMDUnlink(*ping_mdh);
1969         LNetInvalidateMDHandle(ping_mdh);
1970
1971         /* NB the MD could be busy; this just starts the unlink */
1972         wait_var_event_warning(&pbuf->pb_refcnt,
1973                                atomic_read(&pbuf->pb_refcnt) <= 1,
1974                                "Still waiting for ping data MD to unlink\n");
1975 }
1976
1977 static void
1978 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1979 {
1980         struct lnet_ni          *ni;
1981         struct lnet_net         *net;
1982         struct lnet_ni_status *ns;
1983         int                     i;
1984         int                     rc;
1985
1986         i = 0;
1987         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1988                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1989                         LASSERT(i < pbuf->pb_nnis);
1990
1991                         ns = &pbuf->pb_info.pi_ni[i];
1992
1993                         if (!nid_is_nid4(&ni->ni_nid))
1994                                 continue;
1995                         ns->ns_nid = lnet_nid_to_nid4(&ni->ni_nid);
1996
1997                         lnet_ni_lock(ni);
1998                         ns->ns_status = lnet_ni_get_status_locked(ni);
1999                         ni->ni_status = ns;
2000                         lnet_ni_unlock(ni);
2001
2002                         i++;
2003                 }
2004         }
2005         /*
2006          * We (ab)use the ns_status of the loopback interface to
2007          * transmit the sequence number. The first interface listed
2008          * must be the loopback interface.
2009          */
2010         rc = lnet_ping_info_validate(&pbuf->pb_info);
2011         if (rc) {
2012                 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
2013                 LBUG();
2014         }
2015         LNET_PING_BUFFER_SEQNO(pbuf) =
2016                 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
2017 }
2018
2019 static void
2020 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
2021                         struct lnet_handle_md ping_mdh)
2022 {
2023         struct lnet_ping_buffer *old_pbuf = NULL;
2024         struct lnet_handle_md old_ping_md;
2025
2026         /* switch the NIs to point to the new ping info created */
2027         lnet_net_lock(LNET_LOCK_EX);
2028
2029         if (!the_lnet.ln_routing)
2030                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
2031         if (!lnet_peer_discovery_disabled)
2032                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
2033
2034         /* Ensure only known feature bits have been set. */
2035         LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
2036         LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
2037
2038         lnet_ping_target_install_locked(pbuf);
2039
2040         if (the_lnet.ln_ping_target) {
2041                 old_pbuf = the_lnet.ln_ping_target;
2042                 old_ping_md = the_lnet.ln_ping_target_md;
2043         }
2044         the_lnet.ln_ping_target_md = ping_mdh;
2045         the_lnet.ln_ping_target = pbuf;
2046
2047         lnet_net_unlock(LNET_LOCK_EX);
2048
2049         if (old_pbuf) {
2050                 /* unlink and free the old ping info */
2051                 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
2052                 lnet_ping_buffer_decref(old_pbuf);
2053         }
2054
2055         lnet_push_update_to_peers(0);
2056 }
2057
2058 static void
2059 lnet_ping_target_fini(void)
2060 {
2061         lnet_ping_md_unlink(the_lnet.ln_ping_target,
2062                             &the_lnet.ln_ping_target_md);
2063
2064         lnet_assert_handler_unused(the_lnet.ln_ping_target_handler);
2065         lnet_ping_target_destroy();
2066 }
2067
2068 /* Resize the push target. */
2069 int lnet_push_target_resize(void)
2070 {
2071         struct lnet_handle_md mdh;
2072         struct lnet_handle_md old_mdh;
2073         struct lnet_ping_buffer *pbuf;
2074         struct lnet_ping_buffer *old_pbuf;
2075         int nnis;
2076         int rc;
2077
2078 again:
2079         nnis = the_lnet.ln_push_target_nnis;
2080         if (nnis <= 0) {
2081                 CDEBUG(D_NET, "Invalid nnis %d\n", nnis);
2082                 return -EINVAL;
2083         }
2084
2085         /* NB: lnet_ping_buffer_alloc() sets pbuf refcount to 1. That ref is
2086          * dropped when we need to resize again (see "old_pbuf" below) or when
2087          * LNet is shutdown (see lnet_push_target_fini())
2088          */
2089         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
2090         if (!pbuf) {
2091                 CDEBUG(D_NET, "Can't allocate pbuf for nnis %d\n", nnis);
2092                 return -ENOMEM;
2093         }
2094
2095         rc = lnet_push_target_post(pbuf, &mdh);
2096         if (rc) {
2097                 CDEBUG(D_NET, "Failed to post push target: %d\n", rc);
2098                 lnet_ping_buffer_decref(pbuf);
2099                 return rc;
2100         }
2101
2102         lnet_net_lock(LNET_LOCK_EX);
2103         old_pbuf = the_lnet.ln_push_target;
2104         old_mdh = the_lnet.ln_push_target_md;
2105         the_lnet.ln_push_target = pbuf;
2106         the_lnet.ln_push_target_md = mdh;
2107         lnet_net_unlock(LNET_LOCK_EX);
2108
2109         if (old_pbuf) {
2110                 LNetMDUnlink(old_mdh);
2111                 /* Drop ref set by lnet_ping_buffer_alloc() */
2112                 lnet_ping_buffer_decref(old_pbuf);
2113         }
2114
2115         /* Received another push or reply that requires a larger buffer */
2116         if (nnis < the_lnet.ln_push_target_nnis)
2117                 goto again;
2118
2119         CDEBUG(D_NET, "nnis %d success\n", nnis);
2120         return 0;
2121 }
2122
2123 int lnet_push_target_post(struct lnet_ping_buffer *pbuf,
2124                           struct lnet_handle_md *mdhp)
2125 {
2126         struct lnet_processid id = { LNET_ANY_NID, LNET_PID_ANY };
2127         struct lnet_md md = { NULL };
2128         struct lnet_me *me;
2129         int rc;
2130
2131         me = LNetMEAttach(LNET_RESERVED_PORTAL, &id,
2132                           LNET_PROTO_PING_MATCHBITS, 0,
2133                           LNET_UNLINK, LNET_INS_AFTER);
2134         if (IS_ERR(me)) {
2135                 rc = PTR_ERR(me);
2136                 CERROR("Can't create push target ME: %d\n", rc);
2137                 return rc;
2138         }
2139
2140         pbuf->pb_needs_post = false;
2141
2142         /* This reference is dropped by lnet_push_target_event_handler() */
2143         lnet_ping_buffer_addref(pbuf);
2144
2145         /* initialize md content */
2146         md.start     = &pbuf->pb_info;
2147         md.length    = LNET_PING_INFO_SIZE(pbuf->pb_nnis);
2148         md.threshold = 1;
2149         md.max_size  = 0;
2150         md.options   = LNET_MD_OP_PUT | LNET_MD_TRUNCATE;
2151         md.user_ptr  = pbuf;
2152         md.handler   = the_lnet.ln_push_target_handler;
2153
2154         rc = LNetMDAttach(me, &md, LNET_UNLINK, mdhp);
2155         if (rc) {
2156                 CERROR("Can't attach push MD: %d\n", rc);
2157                 lnet_ping_buffer_decref(pbuf);
2158                 pbuf->pb_needs_post = true;
2159                 return rc;
2160         }
2161
2162         CDEBUG(D_NET, "posted push target %p\n", pbuf);
2163
2164         return 0;
2165 }
2166
2167 static void lnet_push_target_event_handler(struct lnet_event *ev)
2168 {
2169         struct lnet_ping_buffer *pbuf = ev->md_user_ptr;
2170
2171         CDEBUG(D_NET, "type %d status %d unlinked %d\n", ev->type, ev->status,
2172                ev->unlinked);
2173
2174         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
2175                 lnet_swap_pinginfo(pbuf);
2176
2177         if (ev->type == LNET_EVENT_UNLINK) {
2178                 /* Drop ref added by lnet_push_target_post() */
2179                 lnet_ping_buffer_decref(pbuf);
2180                 return;
2181         }
2182
2183         lnet_peer_push_event(ev);
2184         if (ev->unlinked)
2185                 /* Drop ref added by lnet_push_target_post */
2186                 lnet_ping_buffer_decref(pbuf);
2187 }
2188
2189 /* Initialize the push target. */
2190 static int lnet_push_target_init(void)
2191 {
2192         int rc;
2193
2194         if (the_lnet.ln_push_target)
2195                 return -EALREADY;
2196
2197         the_lnet.ln_push_target_handler =
2198                 lnet_push_target_event_handler;
2199
2200         rc = LNetSetLazyPortal(LNET_RESERVED_PORTAL);
2201         LASSERT(rc == 0);
2202
2203         /* Start at the required minimum, we'll enlarge if required. */
2204         the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
2205
2206         rc = lnet_push_target_resize();
2207
2208         if (rc) {
2209                 LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2210                 the_lnet.ln_push_target_handler = NULL;
2211         }
2212
2213         return rc;
2214 }
2215
2216 /* Clean up the push target. */
2217 static void lnet_push_target_fini(void)
2218 {
2219         if (!the_lnet.ln_push_target)
2220                 return;
2221
2222         /* Unlink and invalidate to prevent new references. */
2223         LNetMDUnlink(the_lnet.ln_push_target_md);
2224         LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
2225
2226         /* Wait for the unlink to complete. */
2227         wait_var_event_warning(&the_lnet.ln_push_target->pb_refcnt,
2228                                atomic_read(&the_lnet.ln_push_target->pb_refcnt) <= 1,
2229                                "Still waiting for ping data MD to unlink\n");
2230
2231         /* Drop ref set by lnet_ping_buffer_alloc() */
2232         lnet_ping_buffer_decref(the_lnet.ln_push_target);
2233         the_lnet.ln_push_target = NULL;
2234         the_lnet.ln_push_target_nnis = 0;
2235
2236         LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2237         lnet_assert_handler_unused(the_lnet.ln_push_target_handler);
2238         the_lnet.ln_push_target_handler = NULL;
2239 }
2240
2241 static int
2242 lnet_ni_tq_credits(struct lnet_ni *ni)
2243 {
2244         int     credits;
2245
2246         LASSERT(ni->ni_ncpts >= 1);
2247
2248         if (ni->ni_ncpts == 1)
2249                 return ni->ni_net->net_tunables.lct_max_tx_credits;
2250
2251         credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2252         credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2253         credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2254
2255         return credits;
2256 }
2257
2258 static void
2259 lnet_ni_unlink_locked(struct lnet_ni *ni)
2260 {
2261         /* move it to zombie list and nobody can find it anymore */
2262         LASSERT(!list_empty(&ni->ni_netlist));
2263         list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2264         lnet_ni_decref_locked(ni, 0);
2265 }
2266
2267 static void
2268 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2269 {
2270         int             i;
2271         int             islo;
2272         struct lnet_ni  *ni;
2273         struct list_head *zombie_list = &net->net_ni_zombie;
2274
2275         /*
2276          * Now wait for the NIs I just nuked to show up on the zombie
2277          * list and shut them down in guaranteed thread context
2278          */
2279         i = 2;
2280         while (!list_empty(zombie_list)) {
2281                 int     *ref;
2282                 int     j;
2283
2284                 ni = list_entry(zombie_list->next,
2285                                 struct lnet_ni, ni_netlist);
2286                 list_del_init(&ni->ni_netlist);
2287                 /* the ni should be in deleting state. If it's not it's
2288                  * a bug */
2289                 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2290                 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2291                         if (*ref == 0)
2292                                 continue;
2293                         /* still busy, add it back to zombie list */
2294                         list_add(&ni->ni_netlist, zombie_list);
2295                         break;
2296                 }
2297
2298                 if (!list_empty(&ni->ni_netlist)) {
2299                         /* Unlock mutex while waiting to allow other
2300                          * threads to read the LNet state and fall through
2301                          * to avoid deadlock
2302                          */
2303                         lnet_net_unlock(LNET_LOCK_EX);
2304                         mutex_unlock(&the_lnet.ln_api_mutex);
2305
2306                         ++i;
2307                         if ((i & (-i)) == i) {
2308                                 CDEBUG(D_WARNING,
2309                                        "Waiting for zombie LNI %s\n",
2310                                        libcfs_nidstr(&ni->ni_nid));
2311                         }
2312                         schedule_timeout_uninterruptible(cfs_time_seconds(1));
2313
2314                         mutex_lock(&the_lnet.ln_api_mutex);
2315                         lnet_net_lock(LNET_LOCK_EX);
2316                         continue;
2317                 }
2318
2319                 lnet_net_unlock(LNET_LOCK_EX);
2320
2321                 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2322
2323                 LASSERT(!in_interrupt());
2324                 /* Holding the mutex makes it safe for lnd_shutdown
2325                  * to call module_put(). Module unload cannot finish
2326                  * until lnet_unregister_lnd() completes, and that
2327                  * requires the mutex.
2328                  */
2329                 mutex_lock(&the_lnet.ln_lnd_mutex);
2330                 (net->net_lnd->lnd_shutdown)(ni);
2331                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2332
2333                 if (!islo)
2334                         CDEBUG(D_LNI, "Removed LNI %s\n",
2335                               libcfs_nidstr(&ni->ni_nid));
2336
2337                 lnet_ni_free(ni);
2338                 i = 2;
2339                 lnet_net_lock(LNET_LOCK_EX);
2340         }
2341 }
2342
2343 /* shutdown down the NI and release refcount */
2344 static void
2345 lnet_shutdown_lndni(struct lnet_ni *ni)
2346 {
2347         int i;
2348         struct lnet_net *net = ni->ni_net;
2349
2350         lnet_net_lock(LNET_LOCK_EX);
2351         lnet_ni_lock(ni);
2352         ni->ni_state = LNET_NI_STATE_DELETING;
2353         lnet_ni_unlock(ni);
2354         lnet_ni_unlink_locked(ni);
2355         lnet_incr_dlc_seq();
2356         lnet_net_unlock(LNET_LOCK_EX);
2357
2358         /* clear messages for this NI on the lazy portal */
2359         for (i = 0; i < the_lnet.ln_nportals; i++)
2360                 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2361
2362         lnet_net_lock(LNET_LOCK_EX);
2363         lnet_clear_zombies_nis_locked(net);
2364         lnet_net_unlock(LNET_LOCK_EX);
2365 }
2366
2367 static void
2368 lnet_shutdown_lndnet(struct lnet_net *net)
2369 {
2370         struct lnet_ni *ni;
2371
2372         lnet_net_lock(LNET_LOCK_EX);
2373
2374         list_del_init(&net->net_list);
2375
2376         while (!list_empty(&net->net_ni_list)) {
2377                 ni = list_entry(net->net_ni_list.next,
2378                                 struct lnet_ni, ni_netlist);
2379                 lnet_net_unlock(LNET_LOCK_EX);
2380                 lnet_shutdown_lndni(ni);
2381                 lnet_net_lock(LNET_LOCK_EX);
2382         }
2383
2384         lnet_net_unlock(LNET_LOCK_EX);
2385
2386         /* Do peer table cleanup for this net */
2387         lnet_peer_tables_cleanup(net);
2388
2389         lnet_net_free(net);
2390 }
2391
2392 static void
2393 lnet_shutdown_lndnets(void)
2394 {
2395         struct lnet_net *net;
2396         LIST_HEAD(resend);
2397         struct lnet_msg *msg, *tmp;
2398
2399         /* NB called holding the global mutex */
2400
2401         /* All quiet on the API front */
2402         LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2403         LASSERT(the_lnet.ln_refcount == 0);
2404
2405         lnet_net_lock(LNET_LOCK_EX);
2406         the_lnet.ln_state = LNET_STATE_STOPPING;
2407
2408         /*
2409          * move the nets to the zombie list to avoid them being
2410          * picked up for new work. LONET is also included in the
2411          * Nets that will be moved to the zombie list
2412          */
2413         list_splice_init(&the_lnet.ln_nets, &the_lnet.ln_net_zombie);
2414
2415         /* Drop the cached loopback Net. */
2416         if (the_lnet.ln_loni != NULL) {
2417                 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2418                 the_lnet.ln_loni = NULL;
2419         }
2420         lnet_net_unlock(LNET_LOCK_EX);
2421
2422         /* iterate through the net zombie list and delete each net */
2423         while (!list_empty(&the_lnet.ln_net_zombie)) {
2424                 net = list_entry(the_lnet.ln_net_zombie.next,
2425                                  struct lnet_net, net_list);
2426                 lnet_shutdown_lndnet(net);
2427         }
2428
2429         spin_lock(&the_lnet.ln_msg_resend_lock);
2430         list_splice(&the_lnet.ln_msg_resend, &resend);
2431         spin_unlock(&the_lnet.ln_msg_resend_lock);
2432
2433         list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2434                 list_del_init(&msg->msg_list);
2435                 msg->msg_no_resend = true;
2436                 lnet_finalize(msg, -ECANCELED);
2437         }
2438
2439         lnet_net_lock(LNET_LOCK_EX);
2440         the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2441         lnet_net_unlock(LNET_LOCK_EX);
2442 }
2443
2444 static int
2445 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2446 {
2447         int                     rc = -EINVAL;
2448         struct lnet_tx_queue    *tq;
2449         int                     i;
2450         struct lnet_net         *net = ni->ni_net;
2451
2452         mutex_lock(&the_lnet.ln_lnd_mutex);
2453
2454         if (tun) {
2455                 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2456                 ni->ni_lnd_tunables_set = true;
2457         }
2458
2459         rc = (net->net_lnd->lnd_startup)(ni);
2460
2461         mutex_unlock(&the_lnet.ln_lnd_mutex);
2462
2463         if (rc != 0) {
2464                 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2465                                    rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2466                 goto failed0;
2467         }
2468
2469         lnet_ni_lock(ni);
2470         ni->ni_state = LNET_NI_STATE_ACTIVE;
2471         lnet_ni_unlock(ni);
2472
2473         /* We keep a reference on the loopback net through the loopback NI */
2474         if (net->net_lnd->lnd_type == LOLND) {
2475                 lnet_ni_addref(ni);
2476                 LASSERT(the_lnet.ln_loni == NULL);
2477                 the_lnet.ln_loni = ni;
2478                 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2479                 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2480                 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2481                 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2482                 return 0;
2483         }
2484
2485         if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2486             ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2487                 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2488                                    libcfs_lnd2str(net->net_lnd->lnd_type),
2489                                    ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2490                                         "" : "per-peer ");
2491                 /* shutdown the NI since if we get here then it must've already
2492                  * been started
2493                  */
2494                 lnet_shutdown_lndni(ni);
2495                 return -EINVAL;
2496         }
2497
2498         cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2499                 tq->tq_credits_min =
2500                 tq->tq_credits_max =
2501                 tq->tq_credits = lnet_ni_tq_credits(ni);
2502         }
2503
2504         atomic_set(&ni->ni_tx_credits,
2505                    lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2506         atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2507
2508         CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2509                 libcfs_nidstr(&ni->ni_nid),
2510                 ni->ni_net->net_tunables.lct_peer_tx_credits,
2511                 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2512                 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2513                 ni->ni_net->net_tunables.lct_peer_timeout);
2514
2515         return 0;
2516 failed0:
2517         lnet_ni_free(ni);
2518         return rc;
2519 }
2520
2521 static int
2522 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2523 {
2524         struct lnet_ni *ni;
2525         struct lnet_net *net_l = NULL;
2526         LIST_HEAD(local_ni_list);
2527         int rc;
2528         int ni_count = 0;
2529         __u32 lnd_type;
2530         const struct lnet_lnd  *lnd;
2531         int peer_timeout =
2532                 net->net_tunables.lct_peer_timeout;
2533         int maxtxcredits =
2534                 net->net_tunables.lct_max_tx_credits;
2535         int peerrtrcredits =
2536                 net->net_tunables.lct_peer_rtr_credits;
2537
2538         /*
2539          * make sure that this net is unique. If it isn't then
2540          * we are adding interfaces to an already existing network, and
2541          * 'net' is just a convenient way to pass in the list.
2542          * if it is unique we need to find the LND and load it if
2543          * necessary.
2544          */
2545         if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2546                 lnd_type = LNET_NETTYP(net->net_id);
2547
2548                 mutex_lock(&the_lnet.ln_lnd_mutex);
2549                 lnd = lnet_find_lnd_by_type(lnd_type);
2550
2551                 if (lnd == NULL) {
2552                         mutex_unlock(&the_lnet.ln_lnd_mutex);
2553                         rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2554                         mutex_lock(&the_lnet.ln_lnd_mutex);
2555
2556                         lnd = lnet_find_lnd_by_type(lnd_type);
2557                         if (lnd == NULL) {
2558                                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2559                                 CERROR("Can't load LND %s, module %s, rc=%d\n",
2560                                 libcfs_lnd2str(lnd_type),
2561                                 libcfs_lnd2modname(lnd_type), rc);
2562 #ifndef HAVE_MODULE_LOADING_SUPPORT
2563                                 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2564                                                 "compiled with kernel module "
2565                                                 "loading support.");
2566 #endif
2567                                 rc = -EINVAL;
2568                                 goto failed0;
2569                         }
2570                 }
2571
2572                 net->net_lnd = lnd;
2573
2574                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2575
2576                 net_l = net;
2577         }
2578
2579         /*
2580          * net_l: if the network being added is unique then net_l
2581          *        will point to that network
2582          *        if the network being added is not unique then
2583          *        net_l points to the existing network.
2584          *
2585          * When we enter the loop below, we'll pick NIs off he
2586          * network beign added and start them up, then add them to
2587          * a local ni list. Once we've successfully started all
2588          * the NIs then we join the local NI list (of started up
2589          * networks) with the net_l->net_ni_list, which should
2590          * point to the correct network to add the new ni list to
2591          *
2592          * If any of the new NIs fail to start up, then we want to
2593          * iterate through the local ni list, which should include
2594          * any NIs which were successfully started up, and shut
2595          * them down.
2596          *
2597          * After than we want to delete the network being added,
2598          * to avoid a memory leak.
2599          */
2600         while (!list_empty(&net->net_ni_added)) {
2601                 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2602                                 ni_netlist);
2603                 list_del_init(&ni->ni_netlist);
2604
2605                 /* make sure that the the NI we're about to start
2606                  * up is actually unique. if it's not fail. */
2607                 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2608                                         ni->ni_interface)) {
2609                         rc = -EEXIST;
2610                         goto failed1;
2611                 }
2612
2613                 /* adjust the pointer the parent network, just in case it
2614                  * the net is a duplicate */
2615                 ni->ni_net = net_l;
2616
2617                 rc = lnet_startup_lndni(ni, tun);
2618
2619                 if (rc < 0)
2620                         goto failed1;
2621
2622                 lnet_ni_addref(ni);
2623                 list_add_tail(&ni->ni_netlist, &local_ni_list);
2624
2625                 ni_count++;
2626         }
2627
2628         lnet_net_lock(LNET_LOCK_EX);
2629         list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2630         lnet_incr_dlc_seq();
2631         lnet_net_unlock(LNET_LOCK_EX);
2632
2633         /* if the network is not unique then we don't want to keep
2634          * it around after we're done. Free it. Otherwise add that
2635          * net to the global the_lnet.ln_nets */
2636         if (net_l != net && net_l != NULL) {
2637                 /*
2638                  * TODO - note. currently the tunables can not be updated
2639                  * once added
2640                  */
2641                 lnet_net_free(net);
2642         } else {
2643                 /*
2644                  * restore tunables after it has been overwitten by the
2645                  * lnd
2646                  */
2647                 if (peer_timeout != -1)
2648                         net->net_tunables.lct_peer_timeout = peer_timeout;
2649                 if (maxtxcredits != -1)
2650                         net->net_tunables.lct_max_tx_credits = maxtxcredits;
2651                 if (peerrtrcredits != -1)
2652                         net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2653
2654                 lnet_net_lock(LNET_LOCK_EX);
2655                 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2656                 lnet_net_unlock(LNET_LOCK_EX);
2657         }
2658
2659         return ni_count;
2660
2661 failed1:
2662         /*
2663          * shutdown the new NIs that are being started up
2664          * free the NET being started
2665          */
2666         while (!list_empty(&local_ni_list)) {
2667                 ni = list_entry(local_ni_list.next, struct lnet_ni,
2668                                 ni_netlist);
2669
2670                 lnet_shutdown_lndni(ni);
2671         }
2672
2673 failed0:
2674         lnet_net_free(net);
2675
2676         return rc;
2677 }
2678
2679 static int
2680 lnet_startup_lndnets(struct list_head *netlist)
2681 {
2682         struct lnet_net         *net;
2683         int                     rc;
2684         int                     ni_count = 0;
2685
2686         /*
2687          * Change to running state before bringing up the LNDs. This
2688          * allows lnet_shutdown_lndnets() to assert that we've passed
2689          * through here.
2690          */
2691         lnet_net_lock(LNET_LOCK_EX);
2692         the_lnet.ln_state = LNET_STATE_RUNNING;
2693         lnet_net_unlock(LNET_LOCK_EX);
2694
2695         while (!list_empty(netlist)) {
2696                 net = list_entry(netlist->next, struct lnet_net, net_list);
2697                 list_del_init(&net->net_list);
2698
2699                 rc = lnet_startup_lndnet(net, NULL);
2700
2701                 if (rc < 0)
2702                         goto failed;
2703
2704                 ni_count += rc;
2705         }
2706
2707         return ni_count;
2708 failed:
2709         lnet_shutdown_lndnets();
2710
2711         return rc;
2712 }
2713
2714 static int lnet_genl_parse_list(struct sk_buff *msg,
2715                                 const struct ln_key_list *data[], u16 idx)
2716 {
2717         const struct ln_key_list *list = data[idx];
2718         const struct ln_key_props *props;
2719         struct nlattr *node;
2720         u16 count;
2721
2722         if (!list)
2723                 return 0;
2724
2725         if (!list->lkl_maxattr)
2726                 return -ERANGE;
2727
2728         props = list->lkl_list;
2729         if (!props)
2730                 return -EINVAL;
2731
2732         node = nla_nest_start(msg, LN_SCALAR_ATTR_LIST);
2733         if (!node)
2734                 return -ENOBUFS;
2735
2736         for (count = 1; count <= list->lkl_maxattr; count++) {
2737                 struct nlattr *key = nla_nest_start(msg, count);
2738
2739                 if (count == 1)
2740                         nla_put_u16(msg, LN_SCALAR_ATTR_LIST_SIZE,
2741                                     list->lkl_maxattr);
2742
2743                 nla_put_u16(msg, LN_SCALAR_ATTR_INDEX, count);
2744                 if (props[count].lkp_value)
2745                         nla_put_string(msg, LN_SCALAR_ATTR_VALUE,
2746                                        props[count].lkp_value);
2747                 if (props[count].lkp_key_format)
2748                         nla_put_u16(msg, LN_SCALAR_ATTR_KEY_FORMAT,
2749                                     props[count].lkp_key_format);
2750                 nla_put_u16(msg, LN_SCALAR_ATTR_NLA_TYPE,
2751                             props[count].lkp_data_type);
2752                 if (props[count].lkp_data_type == NLA_NESTED) {
2753                         int rc;
2754
2755                         rc = lnet_genl_parse_list(msg, data, ++idx);
2756                         if (rc < 0)
2757                                 return rc;
2758                         idx = rc;
2759                 }
2760
2761                 nla_nest_end(msg, key);
2762         }
2763
2764         nla_nest_end(msg, node);
2765         return idx;
2766 }
2767
2768 int lnet_genl_send_scalar_list(struct sk_buff *msg, u32 portid, u32 seq,
2769                                const struct genl_family *family, int flags,
2770                                u8 cmd, const struct ln_key_list *data[])
2771 {
2772         int rc = 0;
2773         void *hdr;
2774
2775         if (!data[0])
2776                 return -EINVAL;
2777
2778         hdr = genlmsg_put(msg, portid, seq, family, flags, cmd);
2779         if (!hdr)
2780                 GOTO(canceled, rc = -EMSGSIZE);
2781
2782         rc = lnet_genl_parse_list(msg, data, 0);
2783         if (rc < 0)
2784                 GOTO(canceled, rc);
2785
2786         genlmsg_end(msg, hdr);
2787 canceled:
2788         if (rc < 0)
2789                 genlmsg_cancel(msg, hdr);
2790         return rc > 0 ? 0 : rc;
2791 }
2792 EXPORT_SYMBOL(lnet_genl_send_scalar_list);
2793
2794 /**
2795  * Initialize LNet library.
2796  *
2797  * Automatically called at module loading time. Caller has to call
2798  * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2799  * latter returned 0. It must be called exactly once.
2800  *
2801  * \retval 0 on success
2802  * \retval -ve on failures.
2803  */
2804 int lnet_lib_init(void)
2805 {
2806         int rc;
2807
2808         lnet_assert_wire_constants();
2809
2810         /* refer to global cfs_cpt_table for now */
2811         the_lnet.ln_cpt_table = cfs_cpt_tab;
2812         the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_tab);
2813
2814         LASSERT(the_lnet.ln_cpt_number > 0);
2815         if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2816                 /* we are under risk of consuming all lh_cookie */
2817                 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2818                        "please change setting of CPT-table and retry\n",
2819                        the_lnet.ln_cpt_number, LNET_CPT_MAX);
2820                 return -E2BIG;
2821         }
2822
2823         while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2824                 the_lnet.ln_cpt_bits++;
2825
2826         rc = lnet_create_locks();
2827         if (rc != 0) {
2828                 CERROR("Can't create LNet global locks: %d\n", rc);
2829                 return rc;
2830         }
2831
2832         the_lnet.ln_refcount = 0;
2833         INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2834         INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2835
2836         /* The hash table size is the number of bits it takes to express the set
2837          * ln_num_routes, minus 1 (better to under estimate than over so we
2838          * don't waste memory). */
2839         if (rnet_htable_size <= 0)
2840                 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2841         else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2842                 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2843         the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2844                                            order_base_2(rnet_htable_size) - 1);
2845
2846         /* All LNDs apart from the LOLND are in separate modules.  They
2847          * register themselves when their module loads, and unregister
2848          * themselves when their module is unloaded. */
2849         lnet_register_lnd(&the_lolnd);
2850         return 0;
2851 }
2852
2853 /**
2854  * Finalize LNet library.
2855  *
2856  * \pre lnet_lib_init() called with success.
2857  * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2858  *
2859  * As this happens at module-unload, all lnds must already be unloaded,
2860  * so they must already be unregistered.
2861  */
2862 void lnet_lib_exit(void)
2863 {
2864         int i;
2865
2866         LASSERT(the_lnet.ln_refcount == 0);
2867         lnet_unregister_lnd(&the_lolnd);
2868         for (i = 0; i < NUM_LNDS; i++)
2869                 LASSERT(!the_lnet.ln_lnds[i]);
2870         lnet_destroy_locks();
2871 }
2872
2873 /**
2874  * Set LNet PID and start LNet interfaces, routing, and forwarding.
2875  *
2876  * Users must call this function at least once before any other functions.
2877  * For each successful call there must be a corresponding call to
2878  * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2879  * ignored.
2880  *
2881  * The PID used by LNet may be different from the one requested.
2882  * See LNetGetId().
2883  *
2884  * \param requested_pid PID requested by the caller.
2885  *
2886  * \return >= 0 on success, and < 0 error code on failures.
2887  */
2888 int
2889 LNetNIInit(lnet_pid_t requested_pid)
2890 {
2891         int                     im_a_router = 0;
2892         int                     rc;
2893         int                     ni_count;
2894         struct lnet_ping_buffer *pbuf;
2895         struct lnet_handle_md   ping_mdh;
2896         LIST_HEAD(net_head);
2897         struct lnet_net         *net;
2898
2899         mutex_lock(&the_lnet.ln_api_mutex);
2900
2901         CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2902
2903         if (the_lnet.ln_refcount > 0) {
2904                 rc = the_lnet.ln_refcount++;
2905                 mutex_unlock(&the_lnet.ln_api_mutex);
2906                 return rc;
2907         }
2908
2909         rc = lnet_prepare(requested_pid);
2910         if (rc != 0) {
2911                 mutex_unlock(&the_lnet.ln_api_mutex);
2912                 return rc;
2913         }
2914
2915         /* create a network for Loopback network */
2916         net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2917         if (net == NULL) {
2918                 rc = -ENOMEM;
2919                 goto err_empty_list;
2920         }
2921
2922         /* Add in the loopback NI */
2923         if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2924                 rc = -ENOMEM;
2925                 goto err_empty_list;
2926         }
2927
2928         if (use_tcp_bonding)
2929                 CWARN("use_tcp_bonding has been removed. Use Multi-Rail and Dynamic Discovery instead, see LU-13641\n");
2930
2931         /* If LNet is being initialized via DLC it is possible
2932          * that the user requests not to load module parameters (ones which
2933          * are supported by DLC) on initialization.  Therefore, make sure not
2934          * to load networks, routes and forwarding from module parameters
2935          * in this case.  On cleanup in case of failure only clean up
2936          * routes if it has been loaded */
2937         if (!the_lnet.ln_nis_from_mod_params) {
2938                 rc = lnet_parse_networks(&net_head, lnet_get_networks());
2939                 if (rc < 0)
2940                         goto err_empty_list;
2941         }
2942
2943         ni_count = lnet_startup_lndnets(&net_head);
2944         if (ni_count < 0) {
2945                 rc = ni_count;
2946                 goto err_empty_list;
2947         }
2948
2949         if (!the_lnet.ln_nis_from_mod_params) {
2950                 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2951                 if (rc != 0)
2952                         goto err_shutdown_lndnis;
2953
2954                 rc = lnet_rtrpools_alloc(im_a_router);
2955                 if (rc != 0)
2956                         goto err_destroy_routes;
2957         }
2958
2959         rc = lnet_acceptor_start();
2960         if (rc != 0)
2961                 goto err_destroy_routes;
2962
2963         the_lnet.ln_refcount = 1;
2964         /* Now I may use my own API functions... */
2965
2966         rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2967         if (rc != 0)
2968                 goto err_acceptor_stop;
2969
2970         lnet_ping_target_update(pbuf, ping_mdh);
2971
2972         the_lnet.ln_mt_handler = lnet_mt_event_handler;
2973
2974         rc = lnet_push_target_init();
2975         if (rc != 0)
2976                 goto err_stop_ping;
2977
2978         rc = lnet_peer_discovery_start();
2979         if (rc != 0)
2980                 goto err_destroy_push_target;
2981
2982         rc = lnet_monitor_thr_start();
2983         if (rc != 0)
2984                 goto err_stop_discovery_thr;
2985
2986         lnet_fault_init();
2987         lnet_router_debugfs_init();
2988
2989         mutex_unlock(&the_lnet.ln_api_mutex);
2990
2991         complete_all(&the_lnet.ln_started);
2992
2993         /* wait for all routers to start */
2994         lnet_wait_router_start();
2995
2996         return 0;
2997
2998 err_stop_discovery_thr:
2999         lnet_peer_discovery_stop();
3000 err_destroy_push_target:
3001         lnet_push_target_fini();
3002 err_stop_ping:
3003         lnet_ping_target_fini();
3004 err_acceptor_stop:
3005         the_lnet.ln_refcount = 0;
3006         lnet_acceptor_stop();
3007 err_destroy_routes:
3008         if (!the_lnet.ln_nis_from_mod_params)
3009                 lnet_destroy_routes();
3010 err_shutdown_lndnis:
3011         lnet_shutdown_lndnets();
3012 err_empty_list:
3013         lnet_unprepare();
3014         LASSERT(rc < 0);
3015         mutex_unlock(&the_lnet.ln_api_mutex);
3016         while (!list_empty(&net_head)) {
3017                 struct lnet_net *net;
3018
3019                 net = list_entry(net_head.next, struct lnet_net, net_list);
3020                 list_del_init(&net->net_list);
3021                 lnet_net_free(net);
3022         }
3023         return rc;
3024 }
3025 EXPORT_SYMBOL(LNetNIInit);
3026
3027 /**
3028  * Stop LNet interfaces, routing, and forwarding.
3029  *
3030  * Users must call this function once for each successful call to LNetNIInit().
3031  * Once the LNetNIFini() operation has been started, the results of pending
3032  * API operations are undefined.
3033  *
3034  * \return always 0 for current implementation.
3035  */
3036 int
3037 LNetNIFini(void)
3038 {
3039         mutex_lock(&the_lnet.ln_api_mutex);
3040
3041         LASSERT(the_lnet.ln_refcount > 0);
3042
3043         if (the_lnet.ln_refcount != 1) {
3044                 the_lnet.ln_refcount--;
3045         } else {
3046                 LASSERT(!the_lnet.ln_niinit_self);
3047
3048                 lnet_fault_fini();
3049
3050                 lnet_router_debugfs_fini();
3051                 lnet_monitor_thr_stop();
3052                 lnet_peer_discovery_stop();
3053                 lnet_push_target_fini();
3054                 lnet_ping_target_fini();
3055
3056                 /* Teardown fns that use my own API functions BEFORE here */
3057                 the_lnet.ln_refcount = 0;
3058
3059                 lnet_acceptor_stop();
3060                 lnet_destroy_routes();
3061                 lnet_shutdown_lndnets();
3062                 lnet_unprepare();
3063         }
3064
3065         mutex_unlock(&the_lnet.ln_api_mutex);
3066         return 0;
3067 }
3068 EXPORT_SYMBOL(LNetNIFini);
3069
3070 /**
3071  * Grabs the ni data from the ni structure and fills the out
3072  * parameters
3073  *
3074  * \param[in] ni network        interface structure
3075  * \param[out] cfg_ni           NI config information
3076  * \param[out] tun              network and LND tunables
3077  */
3078 static void
3079 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
3080                    struct lnet_ioctl_config_lnd_tunables *tun,
3081                    struct lnet_ioctl_element_stats *stats,
3082                    __u32 tun_size)
3083 {
3084         size_t min_size = 0;
3085         int i;
3086
3087         if (!ni || !cfg_ni || !tun || !nid_is_nid4(&ni->ni_nid))
3088                 return;
3089
3090         if (ni->ni_interface != NULL) {
3091                 strncpy(cfg_ni->lic_ni_intf,
3092                         ni->ni_interface,
3093                         sizeof(cfg_ni->lic_ni_intf));
3094         }
3095
3096         cfg_ni->lic_nid = lnet_nid_to_nid4(&ni->ni_nid);
3097         cfg_ni->lic_status = lnet_ni_get_status_locked(ni);
3098         cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
3099
3100         memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
3101
3102         if (stats) {
3103                 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
3104                                                        LNET_STATS_TYPE_SEND);
3105                 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
3106                                                        LNET_STATS_TYPE_RECV);
3107                 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
3108                                                        LNET_STATS_TYPE_DROP);
3109         }
3110
3111         /*
3112          * tun->lt_tun will always be present, but in order to be
3113          * backwards compatible, we need to deal with the cases when
3114          * tun->lt_tun is smaller than what the kernel has, because it
3115          * comes from an older version of a userspace program, then we'll
3116          * need to copy as much information as we have available space.
3117          */
3118         min_size = tun_size - sizeof(tun->lt_cmn);
3119         memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
3120
3121         /* copy over the cpts */
3122         if (ni->ni_ncpts == LNET_CPT_NUMBER &&
3123             ni->ni_cpts == NULL)  {
3124                 for (i = 0; i < ni->ni_ncpts; i++)
3125                         cfg_ni->lic_cpts[i] = i;
3126         } else {
3127                 for (i = 0;
3128                      ni->ni_cpts != NULL && i < ni->ni_ncpts &&
3129                      i < LNET_MAX_SHOW_NUM_CPT;
3130                      i++)
3131                         cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
3132         }
3133         cfg_ni->lic_ncpts = ni->ni_ncpts;
3134 }
3135
3136 /**
3137  * NOTE: This is a legacy function left in the code to be backwards
3138  * compatible with older userspace programs. It should eventually be
3139  * removed.
3140  *
3141  * Grabs the ni data from the ni structure and fills the out
3142  * parameters
3143  *
3144  * \param[in] ni network        interface structure
3145  * \param[out] config           config information
3146  */
3147 static void
3148 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
3149                          struct lnet_ioctl_config_data *config)
3150 {
3151         struct lnet_ioctl_net_config *net_config;
3152         struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
3153         size_t min_size, tunable_size = 0;
3154         int i;
3155
3156         if (!ni || !config || !nid_is_nid4(&ni->ni_nid))
3157                 return;
3158
3159         net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
3160         if (!net_config)
3161                 return;
3162
3163         if (!ni->ni_interface)
3164                 return;
3165
3166         strncpy(net_config->ni_interface,
3167                 ni->ni_interface,
3168                 sizeof(net_config->ni_interface));
3169
3170         config->cfg_nid = lnet_nid_to_nid4(&ni->ni_nid);
3171         config->cfg_config_u.cfg_net.net_peer_timeout =
3172                 ni->ni_net->net_tunables.lct_peer_timeout;
3173         config->cfg_config_u.cfg_net.net_max_tx_credits =
3174                 ni->ni_net->net_tunables.lct_max_tx_credits;
3175         config->cfg_config_u.cfg_net.net_peer_tx_credits =
3176                 ni->ni_net->net_tunables.lct_peer_tx_credits;
3177         config->cfg_config_u.cfg_net.net_peer_rtr_credits =
3178                 ni->ni_net->net_tunables.lct_peer_rtr_credits;
3179
3180         net_config->ni_status = lnet_ni_get_status_locked(ni);
3181
3182         if (ni->ni_cpts) {
3183                 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
3184
3185                 for (i = 0; i < num_cpts; i++)
3186                         net_config->ni_cpts[i] = ni->ni_cpts[i];
3187
3188                 config->cfg_ncpts = num_cpts;
3189         }
3190
3191         /*
3192          * See if user land tools sent in a newer and larger version
3193          * of struct lnet_tunables than what the kernel uses.
3194          */
3195         min_size = sizeof(*config) + sizeof(*net_config);
3196
3197         if (config->cfg_hdr.ioc_len > min_size)
3198                 tunable_size = config->cfg_hdr.ioc_len - min_size;
3199
3200         /* Don't copy too much data to user space */
3201         min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
3202         lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
3203
3204         if (lnd_cfg && min_size) {
3205                 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
3206                 config->cfg_config_u.cfg_net.net_interface_count = 1;
3207
3208                 /* Tell user land that kernel side has less data */
3209                 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
3210                         min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
3211                         config->cfg_hdr.ioc_len -= min_size;
3212                 }
3213         }
3214 }
3215
3216 struct lnet_ni *
3217 lnet_get_ni_idx_locked(int idx)
3218 {
3219         struct lnet_ni          *ni;
3220         struct lnet_net         *net;
3221
3222         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3223                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3224                         if (idx-- == 0)
3225                                 return ni;
3226                 }
3227         }
3228
3229         return NULL;
3230 }
3231
3232 int lnet_get_net_healthv_locked(struct lnet_net *net)
3233 {
3234         struct lnet_ni *ni;
3235         int best_healthv = 0;
3236         int healthv, ni_fatal;
3237
3238         list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3239                 healthv = atomic_read(&ni->ni_healthv);
3240                 ni_fatal = atomic_read(&ni->ni_fatal_error_on);
3241                 if (!ni_fatal && healthv > best_healthv)
3242                         best_healthv = healthv;
3243         }
3244
3245         return best_healthv;
3246 }
3247
3248 struct lnet_ni *
3249 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
3250 {
3251         struct lnet_ni          *ni;
3252         struct lnet_net         *net = mynet;
3253
3254         /*
3255          * It is possible that the net has been cleaned out while there is
3256          * a message being sent. This function accessed the net without
3257          * checking if the list is empty
3258          */
3259         if (prev == NULL) {
3260                 if (net == NULL)
3261                         net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
3262                                         net_list);
3263                 if (list_empty(&net->net_ni_list))
3264                         return NULL;
3265                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3266                                 ni_netlist);
3267
3268                 return ni;
3269         }
3270
3271         if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
3272                 /* if you reached the end of the ni list and the net is
3273                  * specified, then there are no more nis in that net */
3274                 if (net != NULL)
3275                         return NULL;
3276
3277                 /* we reached the end of this net ni list. move to the
3278                  * next net */
3279                 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
3280                         /* no more nets and no more NIs. */
3281                         return NULL;
3282
3283                 /* get the next net */
3284                 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
3285                                  net_list);
3286                 if (list_empty(&net->net_ni_list))
3287                         return NULL;
3288                 /* get the ni on it */
3289                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3290                                 ni_netlist);
3291
3292                 return ni;
3293         }
3294
3295         if (list_empty(&prev->ni_netlist))
3296                 return NULL;
3297
3298         /* there are more nis left */
3299         ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
3300
3301         return ni;
3302 }
3303
3304 int
3305 lnet_get_net_config(struct lnet_ioctl_config_data *config)
3306 {
3307         struct lnet_ni *ni;
3308         int cpt;
3309         int rc = -ENOENT;
3310         int idx = config->cfg_count;
3311
3312         cpt = lnet_net_lock_current();
3313
3314         ni = lnet_get_ni_idx_locked(idx);
3315
3316         if (ni != NULL) {
3317                 rc = 0;
3318                 lnet_ni_lock(ni);
3319                 lnet_fill_ni_info_legacy(ni, config);
3320                 lnet_ni_unlock(ni);
3321         }
3322
3323         lnet_net_unlock(cpt);
3324         return rc;
3325 }
3326
3327 int
3328 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3329                    struct lnet_ioctl_config_lnd_tunables *tun,
3330                    struct lnet_ioctl_element_stats *stats,
3331                    __u32 tun_size)
3332 {
3333         struct lnet_ni          *ni;
3334         int                     cpt;
3335         int                     rc = -ENOENT;
3336
3337         if (!cfg_ni || !tun || !stats)
3338                 return -EINVAL;
3339
3340         cpt = lnet_net_lock_current();
3341
3342         ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3343
3344         if (ni) {
3345                 rc = 0;
3346                 lnet_ni_lock(ni);
3347                 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3348                 lnet_ni_unlock(ni);
3349         }
3350
3351         lnet_net_unlock(cpt);
3352         return rc;
3353 }
3354
3355 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3356 {
3357         struct lnet_ni *ni;
3358         int cpt;
3359         int rc = -ENOENT;
3360
3361         if (!msg_stats)
3362                 return -EINVAL;
3363
3364         cpt = lnet_net_lock_current();
3365
3366         ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3367
3368         if (ni) {
3369                 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3370                 rc = 0;
3371         }
3372
3373         lnet_net_unlock(cpt);
3374
3375         return rc;
3376 }
3377
3378 static int lnet_add_net_common(struct lnet_net *net,
3379                                struct lnet_ioctl_config_lnd_tunables *tun)
3380 {
3381         struct lnet_handle_md ping_mdh;
3382         struct lnet_ping_buffer *pbuf;
3383         struct lnet_remotenet *rnet;
3384         struct lnet_ni *ni;
3385         int net_ni_count;
3386         __u32 net_id;
3387         int rc;
3388
3389         lnet_net_lock(LNET_LOCK_EX);
3390         rnet = lnet_find_rnet_locked(net->net_id);
3391         lnet_net_unlock(LNET_LOCK_EX);
3392         /*
3393          * make sure that the net added doesn't invalidate the current
3394          * configuration LNet is keeping
3395          */
3396         if (rnet) {
3397                 CERROR("Adding net %s will invalidate routing configuration\n",
3398                        libcfs_net2str(net->net_id));
3399                 lnet_net_free(net);
3400                 return -EUSERS;
3401         }
3402
3403         /*
3404          * make sure you calculate the correct number of slots in the ping
3405          * buffer. Since the ping info is a flattened list of all the NIs,
3406          * we should allocate enough slots to accomodate the number of NIs
3407          * which will be added.
3408          *
3409          * since ni hasn't been configured yet, use
3410          * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3411          */
3412         net_ni_count = lnet_get_net_ni_count_pre(net);
3413
3414         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3415                                     net_ni_count + lnet_get_ni_count(),
3416                                     false);
3417         if (rc < 0) {
3418                 lnet_net_free(net);
3419                 return rc;
3420         }
3421
3422         if (tun)
3423                 memcpy(&net->net_tunables,
3424                        &tun->lt_cmn, sizeof(net->net_tunables));
3425         else
3426                 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3427
3428         net_id = net->net_id;
3429
3430         rc = lnet_startup_lndnet(net,
3431                                  (tun) ? &tun->lt_tun : NULL);
3432         if (rc < 0)
3433                 goto failed;
3434
3435         lnet_net_lock(LNET_LOCK_EX);
3436         net = lnet_get_net_locked(net_id);
3437         LASSERT(net);
3438
3439         /* apply the UDSPs */
3440         rc = lnet_udsp_apply_policies_on_net(net);
3441         if (rc)
3442                 CERROR("Failed to apply UDSPs on local net %s\n",
3443                        libcfs_net2str(net->net_id));
3444
3445         /* At this point we lost track of which NI was just added, so we
3446          * just re-apply the policies on all of the NIs on this net
3447          */
3448         list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3449                 rc = lnet_udsp_apply_policies_on_ni(ni);
3450                 if (rc)
3451                         CERROR("Failed to apply UDSPs on ni %s\n",
3452                                libcfs_nidstr(&ni->ni_nid));
3453         }
3454         lnet_net_unlock(LNET_LOCK_EX);
3455
3456         /*
3457          * Start the acceptor thread if this is the first network
3458          * being added that requires the thread.
3459          */
3460         if (net->net_lnd->lnd_accept) {
3461                 rc = lnet_acceptor_start();
3462                 if (rc < 0) {
3463                         /* shutdown the net that we just started */
3464                         CERROR("Failed to start up acceptor thread\n");
3465                         lnet_shutdown_lndnet(net);
3466                         goto failed;
3467                 }
3468         }
3469
3470         lnet_net_lock(LNET_LOCK_EX);
3471         lnet_peer_net_added(net);
3472         lnet_net_unlock(LNET_LOCK_EX);
3473
3474         lnet_ping_target_update(pbuf, ping_mdh);
3475
3476         return 0;
3477
3478 failed:
3479         lnet_ping_md_unlink(pbuf, &ping_mdh);
3480         lnet_ping_buffer_decref(pbuf);
3481         return rc;
3482 }
3483
3484 static void
3485 lnet_set_tune_defaults(struct lnet_ioctl_config_lnd_tunables *tun)
3486 {
3487         if (tun) {
3488                 if (!tun->lt_cmn.lct_peer_timeout)
3489                         tun->lt_cmn.lct_peer_timeout = DEFAULT_PEER_TIMEOUT;
3490                 if (!tun->lt_cmn.lct_peer_tx_credits)
3491                         tun->lt_cmn.lct_peer_tx_credits = DEFAULT_PEER_CREDITS;
3492                 if (!tun->lt_cmn.lct_max_tx_credits)
3493                         tun->lt_cmn.lct_max_tx_credits = DEFAULT_CREDITS;
3494         }
3495 }
3496
3497 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3498                                       struct lnet_ioctl_config_lnd_tunables *tun)
3499 {
3500         struct lnet_net *net;
3501         const char *nets;
3502         int rc;
3503         LIST_HEAD(net_head);
3504
3505         rc = lnet_parse_ip2nets(&nets, ip2nets);
3506         if (rc < 0)
3507                 return rc;
3508
3509         rc = lnet_parse_networks(&net_head, nets);
3510         if (rc < 0)
3511                 return rc;
3512
3513         lnet_set_tune_defaults(tun);
3514
3515         mutex_lock(&the_lnet.ln_api_mutex);
3516         while (!list_empty(&net_head)) {
3517                 net = list_entry(net_head.next, struct lnet_net, net_list);
3518                 list_del_init(&net->net_list);
3519                 rc = lnet_add_net_common(net, tun);
3520                 if (rc < 0)
3521                         goto out;
3522         }
3523
3524 out:
3525         mutex_unlock(&the_lnet.ln_api_mutex);
3526
3527         while (!list_empty(&net_head)) {
3528                 net = list_entry(net_head.next, struct lnet_net, net_list);
3529                 list_del_init(&net->net_list);
3530                 lnet_net_free(net);
3531         }
3532         return rc;
3533 }
3534
3535 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3536 {
3537         struct lnet_net *net;
3538         struct lnet_ni *ni;
3539         struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3540         int rc, i;
3541         __u32 net_id, lnd_type;
3542
3543         /* get the tunables if they are available */
3544         if (conf->lic_cfg_hdr.ioc_len >=
3545             sizeof(*conf) + sizeof(*tun))
3546                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3547                         conf->lic_bulk;
3548
3549         /* handle legacy ip2nets from DLC */
3550         if (conf->lic_legacy_ip2nets[0] != '\0')
3551                 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3552                                                   tun);
3553
3554         net_id = LNET_NIDNET(conf->lic_nid);
3555         lnd_type = LNET_NETTYP(net_id);
3556
3557         if (!libcfs_isknown_lnd(lnd_type)) {
3558                 CERROR("No valid net and lnd information provided\n");
3559                 return -EINVAL;
3560         }
3561
3562         net = lnet_net_alloc(net_id, NULL);
3563         if (!net)
3564                 return -ENOMEM;
3565
3566         for (i = 0; i < conf->lic_ncpts; i++) {
3567                 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3568                         return -EINVAL;
3569         }
3570
3571         ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3572                                        conf->lic_ni_intf);
3573         if (!ni)
3574                 return -ENOMEM;
3575
3576         lnet_set_tune_defaults(tun);
3577
3578         mutex_lock(&the_lnet.ln_api_mutex);
3579
3580         rc = lnet_add_net_common(net, tun);
3581
3582         mutex_unlock(&the_lnet.ln_api_mutex);
3583
3584         return rc;
3585 }
3586
3587 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3588 {
3589         struct lnet_net  *net;
3590         struct lnet_ni *ni;
3591         __u32 net_id = LNET_NIDNET(conf->lic_nid);
3592         struct lnet_ping_buffer *pbuf;
3593         struct lnet_handle_md  ping_mdh;
3594         int               rc;
3595         int               net_count;
3596         __u32             addr;
3597
3598         /* don't allow userspace to shutdown the LOLND */
3599         if (LNET_NETTYP(net_id) == LOLND)
3600                 return -EINVAL;
3601
3602         mutex_lock(&the_lnet.ln_api_mutex);
3603
3604         lnet_net_lock(0);
3605
3606         net = lnet_get_net_locked(net_id);
3607         if (!net) {
3608                 CERROR("net %s not found\n",
3609                        libcfs_net2str(net_id));
3610                 rc = -ENOENT;
3611                 goto unlock_net;
3612         }
3613
3614         addr = LNET_NIDADDR(conf->lic_nid);
3615         if (addr == 0) {
3616                 /* remove the entire net */
3617                 net_count = lnet_get_net_ni_count_locked(net);
3618
3619                 lnet_net_unlock(0);
3620
3621                 /* create and link a new ping info, before removing the old one */
3622                 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3623                                         lnet_get_ni_count() - net_count,
3624                                         false);
3625                 if (rc != 0)
3626                         goto unlock_api_mutex;
3627
3628                 lnet_shutdown_lndnet(net);
3629
3630                 lnet_acceptor_stop();
3631
3632                 lnet_ping_target_update(pbuf, ping_mdh);
3633
3634                 goto unlock_api_mutex;
3635         }
3636
3637         ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3638         if (!ni) {
3639                 CERROR("nid %s not found\n",
3640                        libcfs_nid2str(conf->lic_nid));
3641                 rc = -ENOENT;
3642                 goto unlock_net;
3643         }
3644
3645         net_count = lnet_get_net_ni_count_locked(net);
3646
3647         lnet_net_unlock(0);
3648
3649         /* create and link a new ping info, before removing the old one */
3650         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3651                                   lnet_get_ni_count() - 1, false);
3652         if (rc != 0)
3653                 goto unlock_api_mutex;
3654
3655         lnet_shutdown_lndni(ni);
3656
3657         lnet_acceptor_stop();
3658
3659         lnet_ping_target_update(pbuf, ping_mdh);
3660
3661         /* check if the net is empty and remove it if it is */
3662         if (net_count == 1)
3663                 lnet_shutdown_lndnet(net);
3664
3665         goto unlock_api_mutex;
3666
3667 unlock_net:
3668         lnet_net_unlock(0);
3669 unlock_api_mutex:
3670         mutex_unlock(&the_lnet.ln_api_mutex);
3671
3672         return rc;
3673 }
3674
3675 /*
3676  * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3677  * They are only expected to be called for unique networks.
3678  * That can be as a result of older DLC library
3679  * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3680  */
3681 int
3682 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3683 {
3684         struct lnet_net *net;
3685         LIST_HEAD(net_head);
3686         int rc;
3687         struct lnet_ioctl_config_lnd_tunables tun;
3688         const char *nets = conf->cfg_config_u.cfg_net.net_intf;
3689
3690         /* Create a net/ni structures for the network string */
3691         rc = lnet_parse_networks(&net_head, nets);
3692         if (rc <= 0)
3693                 return rc == 0 ? -EINVAL : rc;
3694
3695         mutex_lock(&the_lnet.ln_api_mutex);
3696
3697         if (rc > 1) {
3698                 rc = -EINVAL; /* only add one network per call */
3699                 goto out_unlock_clean;
3700         }
3701
3702         net = list_entry(net_head.next, struct lnet_net, net_list);
3703         list_del_init(&net->net_list);
3704
3705         LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3706
3707         memset(&tun, 0, sizeof(tun));
3708
3709         tun.lt_cmn.lct_peer_timeout =
3710           (!conf->cfg_config_u.cfg_net.net_peer_timeout) ? DEFAULT_PEER_TIMEOUT :
3711                 conf->cfg_config_u.cfg_net.net_peer_timeout;
3712         tun.lt_cmn.lct_peer_tx_credits =
3713           (!conf->cfg_config_u.cfg_net.net_peer_tx_credits) ? DEFAULT_PEER_CREDITS :
3714                 conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3715         tun.lt_cmn.lct_peer_rtr_credits =
3716           conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3717         tun.lt_cmn.lct_max_tx_credits =
3718           (!conf->cfg_config_u.cfg_net.net_max_tx_credits) ? DEFAULT_CREDITS :
3719                 conf->cfg_config_u.cfg_net.net_max_tx_credits;
3720
3721         rc = lnet_add_net_common(net, &tun);
3722
3723 out_unlock_clean:
3724         mutex_unlock(&the_lnet.ln_api_mutex);
3725         while (!list_empty(&net_head)) {
3726                 /* net_head list is empty in success case */
3727                 net = list_entry(net_head.next, struct lnet_net, net_list);
3728                 list_del_init(&net->net_list);
3729                 lnet_net_free(net);
3730         }
3731         return rc;
3732 }
3733
3734 int
3735 lnet_dyn_del_net(__u32 net_id)
3736 {
3737         struct lnet_net  *net;
3738         struct lnet_ping_buffer *pbuf;
3739         struct lnet_handle_md ping_mdh;
3740         int               rc;
3741         int               net_ni_count;
3742
3743         /* don't allow userspace to shutdown the LOLND */
3744         if (LNET_NETTYP(net_id) == LOLND)
3745                 return -EINVAL;
3746
3747         mutex_lock(&the_lnet.ln_api_mutex);
3748
3749         lnet_net_lock(0);
3750
3751         net = lnet_get_net_locked(net_id);
3752         if (net == NULL) {
3753                 lnet_net_unlock(0);
3754                 rc = -EINVAL;
3755                 goto out;
3756         }
3757
3758         net_ni_count = lnet_get_net_ni_count_locked(net);
3759
3760         lnet_net_unlock(0);
3761
3762         /* create and link a new ping info, before removing the old one */
3763         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3764                                     lnet_get_ni_count() - net_ni_count, false);
3765         if (rc != 0)
3766                 goto out;
3767
3768         lnet_shutdown_lndnet(net);
3769
3770         lnet_acceptor_stop();
3771
3772         lnet_ping_target_update(pbuf, ping_mdh);
3773
3774 out:
3775         mutex_unlock(&the_lnet.ln_api_mutex);
3776
3777         return rc;
3778 }
3779
3780 void lnet_incr_dlc_seq(void)
3781 {
3782         atomic_inc(&lnet_dlc_seq_no);
3783 }
3784
3785 __u32 lnet_get_dlc_seq_locked(void)
3786 {
3787         return atomic_read(&lnet_dlc_seq_no);
3788 }
3789
3790 static void
3791 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3792 {
3793         struct lnet_net *net;
3794         struct lnet_ni *ni;
3795
3796         lnet_net_lock(LNET_LOCK_EX);
3797         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3798                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3799                         if (all || (nid_is_nid4(&ni->ni_nid) &&
3800                                     lnet_nid_to_nid4(&ni->ni_nid) == nid)) {
3801                                 atomic_set(&ni->ni_healthv, value);
3802                                 if (list_empty(&ni->ni_recovery) &&
3803                                     value < LNET_MAX_HEALTH_VALUE) {
3804                                         CERROR("manually adding local NI %s to recovery\n",
3805                                                libcfs_nidstr(&ni->ni_nid));
3806                                         list_add_tail(&ni->ni_recovery,
3807                                                       &the_lnet.ln_mt_localNIRecovq);
3808                                         lnet_ni_addref_locked(ni, 0);
3809                                 }
3810                                 if (!all) {
3811                                         lnet_net_unlock(LNET_LOCK_EX);
3812                                         return;
3813                                 }
3814                         }
3815                 }
3816         }
3817         lnet_net_unlock(LNET_LOCK_EX);
3818 }
3819
3820 static void
3821 lnet_ni_set_conns_per_peer(lnet_nid_t nid, int value, bool all)
3822 {
3823         struct lnet_net *net;
3824         struct lnet_ni *ni;
3825
3826         lnet_net_lock(LNET_LOCK_EX);
3827         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3828                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3829                         if (lnet_nid_to_nid4(&ni->ni_nid) != nid && !all)
3830                                 continue;
3831                         if (LNET_NETTYP(net->net_id) == SOCKLND)
3832                                 ni->ni_lnd_tunables.lnd_tun_u.lnd_sock.lnd_conns_per_peer = value;
3833                         else if (LNET_NETTYP(net->net_id) == O2IBLND)
3834                                 ni->ni_lnd_tunables.lnd_tun_u.lnd_o2ib.lnd_conns_per_peer = value;
3835                         if (!all) {
3836                                 lnet_net_unlock(LNET_LOCK_EX);
3837                                 return;
3838                         }
3839                 }
3840         }
3841         lnet_net_unlock(LNET_LOCK_EX);
3842 }
3843
3844 static int
3845 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3846 {
3847         int cpt, rc = 0;
3848         struct lnet_ni *ni;
3849         lnet_nid_t nid = stats->hlni_nid;
3850
3851         cpt = lnet_net_lock_current();
3852         ni = lnet_nid2ni_locked(nid, cpt);
3853
3854         if (!ni) {
3855                 rc = -ENOENT;
3856                 goto unlock;
3857         }
3858
3859         stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3860         stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3861         stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3862         stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3863         stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3864         stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3865         stats->hlni_fatal_error = atomic_read(&ni->ni_fatal_error_on);
3866         stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3867         stats->hlni_ping_count = ni->ni_ping_count;
3868         stats->hlni_next_ping = ni->ni_next_ping;
3869
3870 unlock:
3871         lnet_net_unlock(cpt);
3872
3873         return rc;
3874 }
3875
3876 static int
3877 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3878 {
3879         struct lnet_ni *ni;
3880         int i = 0;
3881
3882         lnet_net_lock(LNET_LOCK_EX);
3883         list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3884                 if (!nid_is_nid4(&ni->ni_nid))
3885                         continue;
3886                 list->rlst_nid_array[i] = lnet_nid_to_nid4(&ni->ni_nid);
3887                 i++;
3888                 if (i >= LNET_MAX_SHOW_NUM_NID)
3889                         break;
3890         }
3891         lnet_net_unlock(LNET_LOCK_EX);
3892         list->rlst_num_nids = i;
3893
3894         return 0;
3895 }
3896
3897 static int
3898 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3899 {
3900         struct lnet_peer_ni *lpni;
3901         int i = 0;
3902
3903         lnet_net_lock(LNET_LOCK_EX);
3904         list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3905                 list->rlst_nid_array[i] = lnet_nid_to_nid4(&lpni->lpni_nid);
3906                 i++;
3907                 if (i >= LNET_MAX_SHOW_NUM_NID)
3908                         break;
3909         }
3910         lnet_net_unlock(LNET_LOCK_EX);
3911         list->rlst_num_nids = i;
3912
3913         return 0;
3914 }
3915
3916 /**
3917  * LNet ioctl handler.
3918  *
3919  */
3920 int
3921 LNetCtl(unsigned int cmd, void *arg)
3922 {
3923         struct libcfs_ioctl_data *data = arg;
3924         struct lnet_ioctl_config_data *config;
3925         struct lnet_process_id    id4 = {};
3926         struct lnet_processid     id = {};
3927         struct lnet_ni           *ni;
3928         struct lnet_nid           nid;
3929         int                       rc;
3930
3931         BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3932                      sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3933
3934         switch (cmd) {
3935         case IOC_LIBCFS_GET_NI:
3936                 rc = LNetGetId(data->ioc_count, &id);
3937                 data->ioc_nid = lnet_nid_to_nid4(&id.nid);
3938                 return rc;
3939
3940         case IOC_LIBCFS_FAIL_NID:
3941                 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3942
3943         case IOC_LIBCFS_ADD_ROUTE: {
3944                 /* default router sensitivity to 1 */
3945                 unsigned int sensitivity = 1;
3946                 config = arg;
3947
3948                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3949                         return -EINVAL;
3950
3951                 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3952                         sensitivity =
3953                           config->cfg_config_u.cfg_route.rtr_sensitivity;
3954                 }
3955
3956                 lnet_nid4_to_nid(config->cfg_nid, &nid);
3957                 mutex_lock(&the_lnet.ln_api_mutex);
3958                 rc = lnet_add_route(config->cfg_net,
3959                                     config->cfg_config_u.cfg_route.rtr_hop,
3960                                     &nid,
3961                                     config->cfg_config_u.cfg_route.
3962                                         rtr_priority, sensitivity);
3963                 mutex_unlock(&the_lnet.ln_api_mutex);
3964                 return rc;
3965         }
3966
3967         case IOC_LIBCFS_DEL_ROUTE:
3968                 config = arg;
3969
3970                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3971                         return -EINVAL;
3972
3973                 lnet_nid4_to_nid(config->cfg_nid, &nid);
3974                 mutex_lock(&the_lnet.ln_api_mutex);
3975                 rc = lnet_del_route(config->cfg_net, &nid);
3976                 mutex_unlock(&the_lnet.ln_api_mutex);
3977                 return rc;
3978
3979         case IOC_LIBCFS_GET_ROUTE:
3980                 config = arg;
3981
3982                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3983                         return -EINVAL;
3984
3985                 mutex_lock(&the_lnet.ln_api_mutex);
3986                 rc = lnet_get_route(config->cfg_count,
3987                                     &config->cfg_net,
3988                                     &config->cfg_config_u.cfg_route.rtr_hop,
3989                                     &config->cfg_nid,
3990                                     &config->cfg_config_u.cfg_route.rtr_flags,
3991                                     &config->cfg_config_u.cfg_route.
3992                                         rtr_priority,
3993                                     &config->cfg_config_u.cfg_route.
3994                                         rtr_sensitivity);
3995                 mutex_unlock(&the_lnet.ln_api_mutex);
3996                 return rc;
3997
3998         case IOC_LIBCFS_GET_LOCAL_NI: {
3999                 struct lnet_ioctl_config_ni *cfg_ni;
4000                 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
4001                 struct lnet_ioctl_element_stats *stats;
4002                 __u32 tun_size;
4003
4004                 cfg_ni = arg;
4005
4006                 /* get the tunables if they are available */
4007                 if (cfg_ni->lic_cfg_hdr.ioc_len <
4008                     sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
4009                         return -EINVAL;
4010
4011                 stats = (struct lnet_ioctl_element_stats *)
4012                         cfg_ni->lic_bulk;
4013                 tun = (struct lnet_ioctl_config_lnd_tunables *)
4014                                 (cfg_ni->lic_bulk + sizeof(*stats));
4015
4016                 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
4017                         sizeof(*stats);
4018
4019                 mutex_lock(&the_lnet.ln_api_mutex);
4020                 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
4021                 mutex_unlock(&the_lnet.ln_api_mutex);
4022                 return rc;
4023         }
4024
4025         case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
4026                 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
4027
4028                 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
4029                         return -EINVAL;
4030
4031                 mutex_lock(&the_lnet.ln_api_mutex);
4032                 rc = lnet_get_ni_stats(msg_stats);
4033                 mutex_unlock(&the_lnet.ln_api_mutex);
4034
4035                 return rc;
4036         }
4037
4038         case IOC_LIBCFS_GET_NET: {
4039                 size_t total = sizeof(*config) +
4040                                sizeof(struct lnet_ioctl_net_config);
4041                 config = arg;
4042
4043                 if (config->cfg_hdr.ioc_len < total)
4044                         return -EINVAL;
4045
4046                 mutex_lock(&the_lnet.ln_api_mutex);
4047                 rc = lnet_get_net_config(config);
4048                 mutex_unlock(&the_lnet.ln_api_mutex);
4049                 return rc;
4050         }
4051
4052         case IOC_LIBCFS_GET_LNET_STATS:
4053         {
4054                 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
4055
4056                 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
4057                         return -EINVAL;
4058
4059                 mutex_lock(&the_lnet.ln_api_mutex);
4060                 rc = lnet_counters_get(&lnet_stats->st_cntrs);
4061                 mutex_unlock(&the_lnet.ln_api_mutex);
4062                 return rc;
4063         }
4064
4065         case IOC_LIBCFS_RESET_LNET_STATS:
4066         {
4067                 mutex_lock(&the_lnet.ln_api_mutex);
4068                 lnet_counters_reset();
4069                 mutex_unlock(&the_lnet.ln_api_mutex);
4070                 return 0;
4071         }
4072
4073         case IOC_LIBCFS_CONFIG_RTR:
4074                 config = arg;
4075
4076                 if (config->cfg_hdr.ioc_len < sizeof(*config))
4077                         return -EINVAL;
4078
4079                 mutex_lock(&the_lnet.ln_api_mutex);
4080                 if (config->cfg_config_u.cfg_buffers.buf_enable) {
4081                         rc = lnet_rtrpools_enable();
4082                         mutex_unlock(&the_lnet.ln_api_mutex);
4083                         return rc;
4084                 }
4085                 lnet_rtrpools_disable();
4086                 mutex_unlock(&the_lnet.ln_api_mutex);
4087                 return 0;
4088
4089         case IOC_LIBCFS_ADD_BUF:
4090                 config = arg;
4091
4092                 if (config->cfg_hdr.ioc_len < sizeof(*config))
4093                         return -EINVAL;
4094
4095                 mutex_lock(&the_lnet.ln_api_mutex);
4096                 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
4097                                                 buf_tiny,
4098                                           config->cfg_config_u.cfg_buffers.
4099                                                 buf_small,
4100                                           config->cfg_config_u.cfg_buffers.
4101                                                 buf_large);
4102                 mutex_unlock(&the_lnet.ln_api_mutex);
4103                 return rc;
4104
4105         case IOC_LIBCFS_SET_NUMA_RANGE: {
4106                 struct lnet_ioctl_set_value *numa;
4107                 numa = arg;
4108                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
4109                         return -EINVAL;
4110                 lnet_net_lock(LNET_LOCK_EX);
4111                 lnet_numa_range = numa->sv_value;
4112                 lnet_net_unlock(LNET_LOCK_EX);
4113                 return 0;
4114         }
4115
4116         case IOC_LIBCFS_GET_NUMA_RANGE: {
4117                 struct lnet_ioctl_set_value *numa;
4118                 numa = arg;
4119                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
4120                         return -EINVAL;
4121                 numa->sv_value = lnet_numa_range;
4122                 return 0;
4123         }
4124
4125         case IOC_LIBCFS_GET_BUF: {
4126                 struct lnet_ioctl_pool_cfg *pool_cfg;
4127                 size_t total = sizeof(*config) + sizeof(*pool_cfg);
4128
4129                 config = arg;
4130
4131                 if (config->cfg_hdr.ioc_len < total)
4132                         return -EINVAL;
4133
4134                 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
4135
4136                 mutex_lock(&the_lnet.ln_api_mutex);
4137                 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
4138                 mutex_unlock(&the_lnet.ln_api_mutex);
4139                 return rc;
4140         }
4141
4142         case IOC_LIBCFS_GET_LOCAL_HSTATS: {
4143                 struct lnet_ioctl_local_ni_hstats *stats = arg;
4144
4145                 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
4146                         return -EINVAL;
4147
4148                 mutex_lock(&the_lnet.ln_api_mutex);
4149                 rc = lnet_get_local_ni_hstats(stats);
4150                 mutex_unlock(&the_lnet.ln_api_mutex);
4151
4152                 return rc;
4153         }
4154
4155         case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
4156                 struct lnet_ioctl_recovery_list *list = arg;
4157                 if (list->rlst_hdr.ioc_len < sizeof(*list))
4158                         return -EINVAL;
4159
4160                 mutex_lock(&the_lnet.ln_api_mutex);
4161                 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
4162                         rc = lnet_get_local_ni_recovery_list(list);
4163                 else
4164                         rc = lnet_get_peer_ni_recovery_list(list);
4165                 mutex_unlock(&the_lnet.ln_api_mutex);
4166                 return rc;
4167         }
4168
4169         case IOC_LIBCFS_ADD_PEER_NI: {
4170                 struct lnet_ioctl_peer_cfg *cfg = arg;
4171
4172                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4173                         return -EINVAL;
4174
4175                 mutex_lock(&the_lnet.ln_api_mutex);
4176                 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
4177                                       cfg->prcfg_cfg_nid,
4178                                       cfg->prcfg_mr, false);
4179                 mutex_unlock(&the_lnet.ln_api_mutex);
4180                 return rc;
4181         }
4182
4183         case IOC_LIBCFS_DEL_PEER_NI: {
4184                 struct lnet_ioctl_peer_cfg *cfg = arg;
4185
4186                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4187                         return -EINVAL;
4188
4189                 mutex_lock(&the_lnet.ln_api_mutex);
4190                 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
4191                                       cfg->prcfg_cfg_nid);
4192                 mutex_unlock(&the_lnet.ln_api_mutex);
4193                 return rc;
4194         }
4195
4196         case IOC_LIBCFS_GET_PEER_INFO: {
4197                 struct lnet_ioctl_peer *peer_info = arg;
4198
4199                 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
4200                         return -EINVAL;
4201
4202                 mutex_lock(&the_lnet.ln_api_mutex);
4203                 rc = lnet_get_peer_ni_info(
4204                    peer_info->pr_count,
4205                    &peer_info->pr_nid,
4206                    peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
4207                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
4208                    &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
4209                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
4210                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
4211                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
4212                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
4213                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
4214                 mutex_unlock(&the_lnet.ln_api_mutex);
4215                 return rc;
4216         }
4217
4218         case IOC_LIBCFS_GET_PEER_NI: {
4219                 struct lnet_ioctl_peer_cfg *cfg = arg;
4220
4221                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4222                         return -EINVAL;
4223
4224                 mutex_lock(&the_lnet.ln_api_mutex);
4225                 rc = lnet_get_peer_info(cfg,
4226                                         (void __user *)cfg->prcfg_bulk);
4227                 mutex_unlock(&the_lnet.ln_api_mutex);
4228                 return rc;
4229         }
4230
4231         case IOC_LIBCFS_GET_PEER_LIST: {
4232                 struct lnet_ioctl_peer_cfg *cfg = arg;
4233
4234                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4235                         return -EINVAL;
4236
4237                 mutex_lock(&the_lnet.ln_api_mutex);
4238                 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
4239                                 (struct lnet_process_id __user *)cfg->prcfg_bulk);
4240                 mutex_unlock(&the_lnet.ln_api_mutex);
4241                 return rc;
4242         }
4243
4244         case IOC_LIBCFS_SET_HEALHV: {
4245                 struct lnet_ioctl_reset_health_cfg *cfg = arg;
4246                 int value;
4247                 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
4248                         return -EINVAL;
4249                 if (cfg->rh_value < 0 ||
4250                     cfg->rh_value > LNET_MAX_HEALTH_VALUE)
4251                         value = LNET_MAX_HEALTH_VALUE;
4252                 else
4253                         value = cfg->rh_value;
4254                 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
4255                        value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
4256                        "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
4257                 mutex_lock(&the_lnet.ln_api_mutex);
4258                 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
4259                         lnet_ni_set_healthv(cfg->rh_nid, value,
4260                                              cfg->rh_all);
4261                 else
4262                         lnet_peer_ni_set_healthv(cfg->rh_nid, value,
4263                                                   cfg->rh_all);
4264                 mutex_unlock(&the_lnet.ln_api_mutex);
4265                 return 0;
4266         }
4267
4268         case IOC_LIBCFS_SET_CONNS_PER_PEER: {
4269                 struct lnet_ioctl_reset_conns_per_peer_cfg *cfg = arg;
4270                 int value;
4271
4272                 if (cfg->rcpp_hdr.ioc_len < sizeof(*cfg))
4273                         return -EINVAL;
4274                 if (cfg->rcpp_value < 0)
4275                         value = 1;
4276                 else
4277                         value = cfg->rcpp_value;
4278                 CDEBUG(D_NET,
4279                        "Setting conns_per_peer to %d for %s. all = %d\n",
4280                        value, libcfs_nid2str(cfg->rcpp_nid), cfg->rcpp_all);
4281                 mutex_lock(&the_lnet.ln_api_mutex);
4282                 lnet_ni_set_conns_per_peer(cfg->rcpp_nid, value, cfg->rcpp_all);
4283                 mutex_unlock(&the_lnet.ln_api_mutex);
4284                 return 0;
4285         }
4286
4287         case IOC_LIBCFS_NOTIFY_ROUTER: {
4288                 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
4289
4290                 /* The deadline passed in by the user should be some time in
4291                  * seconds in the future since the UNIX epoch. We have to map
4292                  * that deadline to the wall clock.
4293                  */
4294                 deadline += ktime_get_seconds();
4295                 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
4296                                    deadline);
4297         }
4298
4299         case IOC_LIBCFS_LNET_DIST:
4300                 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
4301                 if (rc < 0 && rc != -EHOSTUNREACH)
4302                         return rc;
4303
4304                 data->ioc_u32[0] = rc;
4305                 return 0;
4306
4307         case IOC_LIBCFS_TESTPROTOCOMPAT:
4308                 the_lnet.ln_testprotocompat = data->ioc_flags;
4309                 return 0;
4310
4311         case IOC_LIBCFS_LNET_FAULT:
4312                 return lnet_fault_ctl(data->ioc_flags, data);
4313
4314         case IOC_LIBCFS_PING: {
4315                 signed long timeout;
4316
4317                 id4.nid = data->ioc_nid;
4318                 id4.pid = data->ioc_u32[0];
4319
4320                 /* If timeout is negative then set default of 3 minutes */
4321                 if (((s32)data->ioc_u32[1] <= 0) ||
4322                     data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
4323                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
4324                 else
4325                         timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
4326
4327                 rc = lnet_ping(id4, &LNET_ANY_NID, timeout, data->ioc_pbuf1,
4328                                data->ioc_plen1 / sizeof(struct lnet_process_id));
4329
4330                 if (rc < 0)
4331                         return rc;
4332
4333                 data->ioc_count = rc;
4334                 return 0;
4335         }
4336
4337         case IOC_LIBCFS_PING_PEER: {
4338                 struct lnet_ioctl_ping_data *ping = arg;
4339                 struct lnet_nid src_nid = LNET_ANY_NID;
4340                 struct lnet_peer *lp;
4341                 signed long timeout;
4342
4343                 /* Check if the supplied ping data supports source nid
4344                  * NB: This check is sufficient if lnet_ioctl_ping_data has
4345                  * additional fields added, but if they are re-ordered or
4346                  * fields removed then this will break. It is expected that
4347                  * these ioctls will be replaced with netlink implementation, so
4348                  * it is probably not worth coming up with a more robust version
4349                  * compatibility scheme.
4350                  */
4351                 if (ping->ping_hdr.ioc_len >= sizeof(struct lnet_ioctl_ping_data))
4352                         lnet_nid4_to_nid(ping->ping_src, &src_nid);
4353
4354                 /* If timeout is negative then set default of 3 minutes */
4355                 if (((s32)ping->op_param) <= 0 ||
4356                     ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
4357                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
4358                 else
4359                         timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
4360
4361                 rc = lnet_ping(ping->ping_id, &src_nid, timeout,
4362                                ping->ping_buf,
4363                                ping->ping_count);
4364                 if (rc < 0)
4365                         return rc;
4366
4367                 mutex_lock(&the_lnet.ln_api_mutex);
4368                 lp = lnet_find_peer4(ping->ping_id.nid);
4369                 if (lp) {
4370                         ping->ping_id.nid =
4371                                 lnet_nid_to_nid4(&lp->lp_primary_nid);
4372                         ping->mr_info = lnet_peer_is_multi_rail(lp);
4373                         lnet_peer_decref_locked(lp);
4374                 }
4375                 mutex_unlock(&the_lnet.ln_api_mutex);
4376
4377                 ping->ping_count = rc;
4378                 return 0;
4379         }
4380
4381         case IOC_LIBCFS_DISCOVER: {
4382                 struct lnet_ioctl_ping_data *discover = arg;
4383                 struct lnet_peer *lp;
4384
4385                 rc = lnet_discover(discover->ping_id, discover->op_param,
4386                                    discover->ping_buf,
4387                                    discover->ping_count);
4388                 if (rc < 0)
4389                         return rc;
4390
4391                 mutex_lock(&the_lnet.ln_api_mutex);
4392                 lp = lnet_find_peer4(discover->ping_id.nid);
4393                 if (lp) {
4394                         discover->ping_id.nid =
4395                                 lnet_nid_to_nid4(&lp->lp_primary_nid);
4396                         discover->mr_info = lnet_peer_is_multi_rail(lp);
4397                         lnet_peer_decref_locked(lp);
4398                 }
4399                 mutex_unlock(&the_lnet.ln_api_mutex);
4400
4401                 discover->ping_count = rc;
4402                 return 0;
4403         }
4404
4405         case IOC_LIBCFS_ADD_UDSP: {
4406                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4407                 __u32 bulk_size = ioc_udsp->iou_hdr.ioc_len;
4408
4409                 mutex_lock(&the_lnet.ln_api_mutex);
4410                 rc = lnet_udsp_demarshal_add(arg, bulk_size);
4411                 if (!rc) {
4412                         rc = lnet_udsp_apply_policies(NULL, false);
4413                         CDEBUG(D_NET, "policy application returned %d\n", rc);
4414                         rc = 0;
4415                 }
4416                 mutex_unlock(&the_lnet.ln_api_mutex);
4417
4418                 return rc;
4419         }
4420
4421         case IOC_LIBCFS_DEL_UDSP: {
4422                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4423                 int idx = ioc_udsp->iou_idx;
4424
4425                 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4426                         return -EINVAL;
4427
4428                 mutex_lock(&the_lnet.ln_api_mutex);
4429                 rc = lnet_udsp_del_policy(idx);
4430                 if (!rc) {
4431                         rc = lnet_udsp_apply_policies(NULL, false);
4432                         CDEBUG(D_NET, "policy re-application returned %d\n",
4433                                rc);
4434                         rc = 0;
4435                 }
4436                 mutex_unlock(&the_lnet.ln_api_mutex);
4437
4438                 return rc;
4439         }
4440
4441         case IOC_LIBCFS_GET_UDSP_SIZE: {
4442                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4443                 struct lnet_udsp *udsp;
4444
4445                 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4446                         return -EINVAL;
4447
4448                 rc = 0;
4449
4450                 mutex_lock(&the_lnet.ln_api_mutex);
4451                 udsp = lnet_udsp_get_policy(ioc_udsp->iou_idx);
4452                 if (!udsp) {
4453                         rc = -ENOENT;
4454                 } else {
4455                         /* coming in iou_idx will hold the idx of the udsp
4456                          * to get the size of. going out the iou_idx will
4457                          * hold the size of the UDSP found at the passed
4458                          * in index.
4459                          */
4460                         ioc_udsp->iou_idx = lnet_get_udsp_size(udsp);
4461                         if (ioc_udsp->iou_idx < 0)
4462                                 rc = -EINVAL;
4463                 }
4464                 mutex_unlock(&the_lnet.ln_api_mutex);
4465
4466                 return rc;
4467         }
4468
4469         case IOC_LIBCFS_GET_UDSP: {
4470                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4471                 struct lnet_udsp *udsp;
4472
4473                 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4474                         return -EINVAL;
4475
4476                 rc = 0;
4477
4478                 mutex_lock(&the_lnet.ln_api_mutex);
4479                 udsp = lnet_udsp_get_policy(ioc_udsp->iou_idx);
4480                 if (!udsp)
4481                         rc = -ENOENT;
4482                 else
4483                         rc = lnet_udsp_marshal(udsp, ioc_udsp);
4484                 mutex_unlock(&the_lnet.ln_api_mutex);
4485
4486                 return rc;
4487         }
4488
4489         case IOC_LIBCFS_GET_CONST_UDSP_INFO: {
4490                 struct lnet_ioctl_construct_udsp_info *info = arg;
4491
4492                 if (info->cud_hdr.ioc_len < sizeof(*info))
4493                         return -EINVAL;
4494
4495                 CDEBUG(D_NET, "GET_UDSP_INFO for %s\n",
4496                        libcfs_nid2str(info->cud_nid));
4497
4498                 mutex_lock(&the_lnet.ln_api_mutex);
4499                 lnet_udsp_get_construct_info(info);
4500                 mutex_unlock(&the_lnet.ln_api_mutex);
4501
4502                 return 0;
4503         }
4504
4505         default:
4506                 ni = lnet_net2ni_addref(data->ioc_net);
4507                 if (ni == NULL)
4508                         return -EINVAL;
4509
4510                 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
4511                         rc = -EINVAL;
4512                 else
4513                         rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
4514
4515                 lnet_ni_decref(ni);
4516                 return rc;
4517         }
4518         /* not reached */
4519 }
4520 EXPORT_SYMBOL(LNetCtl);
4521
4522 void LNetDebugPeer(struct lnet_processid *id)
4523 {
4524         lnet_debug_peer(lnet_nid_to_nid4(&id->nid));
4525 }
4526 EXPORT_SYMBOL(LNetDebugPeer);
4527
4528 /**
4529  * Determine if the specified peer \a nid is on the local node.
4530  *
4531  * \param nid   peer nid to check
4532  *
4533  * \retval true         If peer NID is on the local node.
4534  * \retval false        If peer NID is not on the local node.
4535  */
4536 bool LNetIsPeerLocal(lnet_nid_t nid)
4537 {
4538         struct lnet_net *net;
4539         struct lnet_ni *ni;
4540         int cpt;
4541
4542         cpt = lnet_net_lock_current();
4543         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4544                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4545                         if (lnet_nid_to_nid4(&ni->ni_nid) == nid) {
4546                                 lnet_net_unlock(cpt);
4547                                 return true;
4548                         }
4549                 }
4550         }
4551         lnet_net_unlock(cpt);
4552
4553         return false;
4554 }
4555 EXPORT_SYMBOL(LNetIsPeerLocal);
4556
4557 /**
4558  * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4559  * Note that all interfaces share a same PID, as requested by LNetNIInit().
4560  *
4561  * \param index Index of the interface to look up.
4562  * \param id On successful return, this location will hold the
4563  * struct lnet_process_id ID of the interface.
4564  *
4565  * \retval 0 If an interface exists at \a index.
4566  * \retval -ENOENT If no interface has been found.
4567  */
4568 int
4569 LNetGetId(unsigned int index, struct lnet_processid *id)
4570 {
4571         struct lnet_ni   *ni;
4572         struct lnet_net  *net;
4573         int               cpt;
4574         int               rc = -ENOENT;
4575
4576         LASSERT(the_lnet.ln_refcount > 0);
4577
4578         cpt = lnet_net_lock_current();
4579
4580         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4581                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4582                         if (!nid_is_nid4(&ni->ni_nid))
4583                                 /* FIXME this needs to be handled */
4584                                 continue;
4585                         if (index-- != 0)
4586                                 continue;
4587
4588                         id->nid = ni->ni_nid;
4589                         id->pid = the_lnet.ln_pid;
4590                         rc = 0;
4591                         break;
4592                 }
4593         }
4594
4595         lnet_net_unlock(cpt);
4596         return rc;
4597 }
4598 EXPORT_SYMBOL(LNetGetId);
4599
4600 struct ping_data {
4601         int rc;
4602         int replied;
4603         struct lnet_handle_md mdh;
4604         struct completion completion;
4605 };
4606
4607 static void
4608 lnet_ping_event_handler(struct lnet_event *event)
4609 {
4610         struct ping_data *pd = event->md_user_ptr;
4611
4612         CDEBUG(D_NET, "ping event (%d %d)%s\n",
4613                event->type, event->status,
4614                event->unlinked ? " unlinked" : "");
4615
4616         if (event->status) {
4617                 if (!pd->rc)
4618                         pd->rc = event->status;
4619         } else if (event->type == LNET_EVENT_REPLY) {
4620                 pd->replied = 1;
4621                 pd->rc = event->mlength;
4622         }
4623         if (event->unlinked)
4624                 complete(&pd->completion);
4625 }
4626
4627 static int lnet_ping(struct lnet_process_id id, struct lnet_nid *src_nid,
4628                      signed long timeout, struct lnet_process_id __user *ids,
4629                      int n_ids)
4630 {
4631         struct lnet_md md = { NULL };
4632         struct ping_data pd = { 0 };
4633         struct lnet_ping_buffer *pbuf;
4634         struct lnet_process_id tmpid;
4635         int i;
4636         int nob;
4637         int rc;
4638         int rc2;
4639
4640         /* n_ids limit is arbitrary */
4641         if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4642                 return -EINVAL;
4643
4644         /*
4645          * if the user buffer has more space than the lnet_interfaces_max
4646          * then only fill it up to lnet_interfaces_max
4647          */
4648         if (n_ids > lnet_interfaces_max)
4649                 n_ids = lnet_interfaces_max;
4650
4651         if (id.pid == LNET_PID_ANY)
4652                 id.pid = LNET_PID_LUSTRE;
4653
4654         pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4655         if (!pbuf)
4656                 return -ENOMEM;
4657
4658         /* initialize md content */
4659         md.start     = &pbuf->pb_info;
4660         md.length    = LNET_PING_INFO_SIZE(n_ids);
4661         md.threshold = 2; /* GET/REPLY */
4662         md.max_size  = 0;
4663         md.options   = LNET_MD_TRUNCATE;
4664         md.user_ptr  = &pd;
4665         md.handler   = lnet_ping_event_handler;
4666
4667         init_completion(&pd.completion);
4668
4669         rc = LNetMDBind(&md, LNET_UNLINK, &pd.mdh);
4670         if (rc != 0) {
4671                 CERROR("Can't bind MD: %d\n", rc);
4672                 goto fail_ping_buffer_decref;
4673         }
4674
4675         rc = LNetGet(lnet_nid_to_nid4(src_nid), pd.mdh, id,
4676                      LNET_RESERVED_PORTAL,
4677                      LNET_PROTO_PING_MATCHBITS, 0, false);
4678
4679         if (rc != 0) {
4680                 /* Don't CERROR; this could be deliberate! */
4681                 rc2 = LNetMDUnlink(pd.mdh);
4682                 LASSERT(rc2 == 0);
4683
4684                 /* NB must wait for the UNLINK event below... */
4685         }
4686
4687         if (wait_for_completion_timeout(&pd.completion, timeout) == 0) {
4688                 /* Ensure completion in finite time... */
4689                 LNetMDUnlink(pd.mdh);
4690                 wait_for_completion(&pd.completion);
4691         }
4692         if (!pd.replied) {
4693                 rc = -EIO;
4694                 goto fail_ping_buffer_decref;
4695         }
4696
4697         nob = pd.rc;
4698         LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4699
4700         rc = -EPROTO;           /* if I can't parse... */
4701
4702         if (nob < 8) {
4703                 CERROR("%s: ping info too short %d\n",
4704                        libcfs_id2str(id), nob);
4705                 goto fail_ping_buffer_decref;
4706         }
4707
4708         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4709                 lnet_swap_pinginfo(pbuf);
4710         } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4711                 CERROR("%s: Unexpected magic %08x\n",
4712                        libcfs_id2str(id), pbuf->pb_info.pi_magic);
4713                 goto fail_ping_buffer_decref;
4714         }
4715
4716         if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4717                 CERROR("%s: ping w/o NI status: 0x%x\n",
4718                        libcfs_id2str(id), pbuf->pb_info.pi_features);
4719                 goto fail_ping_buffer_decref;
4720         }
4721
4722         if (nob < LNET_PING_INFO_SIZE(0)) {
4723                 CERROR("%s: Short reply %d(%d min)\n",
4724                        libcfs_id2str(id),
4725                        nob, (int)LNET_PING_INFO_SIZE(0));
4726                 goto fail_ping_buffer_decref;
4727         }
4728
4729         if (pbuf->pb_info.pi_nnis < n_ids)
4730                 n_ids = pbuf->pb_info.pi_nnis;
4731
4732         if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4733                 CERROR("%s: Short reply %d(%d expected)\n",
4734                        libcfs_id2str(id),
4735                        nob, (int)LNET_PING_INFO_SIZE(n_ids));
4736                 goto fail_ping_buffer_decref;
4737         }
4738
4739         rc = -EFAULT;           /* if I segv in copy_to_user()... */
4740
4741         memset(&tmpid, 0, sizeof(tmpid));
4742         for (i = 0; i < n_ids; i++) {
4743                 tmpid.pid = pbuf->pb_info.pi_pid;
4744                 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4745                 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4746                         goto fail_ping_buffer_decref;
4747         }
4748         rc = pbuf->pb_info.pi_nnis;
4749
4750  fail_ping_buffer_decref:
4751         lnet_ping_buffer_decref(pbuf);
4752         return rc;
4753 }
4754
4755 static int
4756 lnet_discover(struct lnet_process_id id, __u32 force,
4757               struct lnet_process_id __user *ids, int n_ids)
4758 {
4759         struct lnet_peer_ni *lpni;
4760         struct lnet_peer_ni *p;
4761         struct lnet_peer *lp;
4762         struct lnet_process_id *buf;
4763         int cpt;
4764         int i;
4765         int rc;
4766
4767         if (n_ids <= 0 ||
4768             id.nid == LNET_NID_ANY)
4769                 return -EINVAL;
4770
4771         if (id.pid == LNET_PID_ANY)
4772                 id.pid = LNET_PID_LUSTRE;
4773
4774         /*
4775          * If the user buffer has more space than the lnet_interfaces_max,
4776          * then only fill it up to lnet_interfaces_max.
4777          */
4778         if (n_ids > lnet_interfaces_max)
4779                 n_ids = lnet_interfaces_max;
4780
4781         CFS_ALLOC_PTR_ARRAY(buf, n_ids);
4782         if (!buf)
4783                 return -ENOMEM;
4784
4785         cpt = lnet_net_lock_current();
4786         lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4787         if (IS_ERR(lpni)) {
4788                 rc = PTR_ERR(lpni);
4789                 goto out;
4790         }
4791
4792         /*
4793          * Clearing the NIDS_UPTODATE flag ensures the peer will
4794          * be discovered, provided discovery has not been disabled.
4795          */
4796         lp = lpni->lpni_peer_net->lpn_peer;
4797         spin_lock(&lp->lp_lock);
4798         lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4799         /* If the force flag is set, force a PING and PUSH as well. */
4800         if (force)
4801                 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4802         spin_unlock(&lp->lp_lock);
4803         rc = lnet_discover_peer_locked(lpni, cpt, true);
4804         if (rc)
4805                 goto out_decref;
4806
4807         /* The lpni (or lp) for this NID may have changed and our ref is
4808          * the only thing keeping the old one around. Release the ref
4809          * and lookup the lpni again
4810          */
4811         lnet_peer_ni_decref_locked(lpni);
4812         lpni = lnet_find_peer_ni_locked(id.nid);
4813         if (!lpni) {
4814                 rc = -ENOENT;
4815                 goto out;
4816         }
4817         lp = lpni->lpni_peer_net->lpn_peer;
4818
4819         i = 0;
4820         p = NULL;
4821         while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4822                 buf[i].pid = id.pid;
4823                 buf[i].nid = lnet_nid_to_nid4(&p->lpni_nid);
4824                 if (++i >= n_ids)
4825                         break;
4826         }
4827         rc = i;
4828
4829 out_decref:
4830         lnet_peer_ni_decref_locked(lpni);
4831 out:
4832         lnet_net_unlock(cpt);
4833
4834         if (rc >= 0)
4835                 if (copy_to_user(ids, buf, rc * sizeof(*buf)))
4836                         rc = -EFAULT;
4837         CFS_FREE_PTR_ARRAY(buf, n_ids);
4838
4839         return rc;
4840 }
4841
4842 /**
4843  * Retrieve peer discovery status.
4844  *
4845  * \retval 1 if lnet_peer_discovery_disabled is 0
4846  * \retval 0 if lnet_peer_discovery_disabled is 1
4847  */
4848 int
4849 LNetGetPeerDiscoveryStatus(void)
4850 {
4851         return !lnet_peer_discovery_disabled;
4852 }
4853 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);