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