lnet/lnet/api-ni.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  */
  31
  32 #define DEBUG_SUBSYSTEM S_LNET
  33
  34 #include <linux/ctype.h>
  35 #include <linux/log2.h>
  36 #include <linux/ktime.h>
  37 #include <linux/moduleparam.h>
  38 #include <linux/uaccess.h>
  39 #ifdef HAVE_SCHED_HEADERS
  40 #include <linux/sched/signal.h>
  41 #endif
  42 #include <lnet/udsp.h>
  43 #include <lnet/lib-lnet.h>
  44
  45 #define D_LNI D_CONSOLE
  46
  47 /*
  48  * initialize ln_api_mutex statically, since it needs to be used in
  49  * discovery_set callback. That module parameter callback can be called
  50  * before module init completes. The mutex needs to be ready for use then.
  51  */
  52 struct lnet the_lnet = {
  53         .ln_api_mutex = __MUTEX_INITIALIZER(the_lnet.ln_api_mutex),
  54 };              /* THE state of the network */
  55 EXPORT_SYMBOL(the_lnet);
  56
  57 static char *ip2nets = "";
  58 module_param(ip2nets, charp, 0444);
  59 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
  60
  61 static char *networks = "";
  62 module_param(networks, charp, 0444);
  63 MODULE_PARM_DESC(networks, "local networks");
  64
  65 static char *routes = "";
  66 module_param(routes, charp, 0444);
  67 MODULE_PARM_DESC(routes, "routes to non-local networks");
  68
  69 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
  70 module_param(rnet_htable_size, int, 0444);
  71 MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
  72
  73 static int use_tcp_bonding;
  74 module_param(use_tcp_bonding, int, 0444);
  75 MODULE_PARM_DESC(use_tcp_bonding,
  76                  "use_tcp_bonding parameter has been removed");
  77
  78 unsigned int lnet_numa_range = 0;
  79 module_param(lnet_numa_range, uint, 0444);
  80 MODULE_PARM_DESC(lnet_numa_range,
  81                 "NUMA range to consider during Multi-Rail selection");
  82
  83 /*
  84  * lnet_health_sensitivity determines by how much we decrement the health
  85  * value on sending error. The value defaults to 100, which means health
  86  * interface health is decremented by 100 points every failure.
  87  */
  88 unsigned int lnet_health_sensitivity = 100;
  89 static int sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp);
  90 #ifdef HAVE_KERNEL_PARAM_OPS
  91 static struct kernel_param_ops param_ops_health_sensitivity = {
  92         .set = sensitivity_set,
  93         .get = param_get_int,
  94 };
  95 #define param_check_health_sensitivity(name, p) \
  96                 __param_check(name, p, int)
  97 module_param(lnet_health_sensitivity, health_sensitivity, S_IRUGO|S_IWUSR);
  98 #else
  99 module_param_call(lnet_health_sensitivity, sensitivity_set, param_get_int,
 100                   &lnet_health_sensitivity, S_IRUGO|S_IWUSR);
 101 #endif
 102 MODULE_PARM_DESC(lnet_health_sensitivity,
 103                 "Value to decrement the health value by on error");
 104
 105 /*
 106  * lnet_recovery_interval determines how often we should perform recovery
 107  * on unhealthy interfaces.
 108  */
 109 unsigned int lnet_recovery_interval = 1;
 110 static int recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp);
 111 #ifdef HAVE_KERNEL_PARAM_OPS
 112 static struct kernel_param_ops param_ops_recovery_interval = {
 113         .set = recovery_interval_set,
 114         .get = param_get_int,
 115 };
 116 #define param_check_recovery_interval(name, p) \
 117                 __param_check(name, p, int)
 118 module_param(lnet_recovery_interval, recovery_interval, S_IRUGO|S_IWUSR);
 119 #else
 120 module_param_call(lnet_recovery_interval, recovery_interval_set, param_get_int,
 121                   &lnet_recovery_interval, S_IRUGO|S_IWUSR);
 122 #endif
 123 MODULE_PARM_DESC(lnet_recovery_interval,
 124                 "DEPRECATED - Interval to recover unhealthy interfaces in seconds");
 125
 126 unsigned int lnet_recovery_limit;
 127 module_param(lnet_recovery_limit, uint, 0644);
 128 MODULE_PARM_DESC(lnet_recovery_limit,
 129                  "How long to attempt recovery of unhealthy peer interfaces in seconds. Set to 0 to allow indefinite recovery");
 130
 131 static int lnet_interfaces_max = LNET_INTERFACES_MAX_DEFAULT;
 132 static int intf_max_set(const char *val, cfs_kernel_param_arg_t *kp);
 133
 134 static struct kernel_param_ops param_ops_interfaces_max = {
 135         .set = intf_max_set,
 136         .get = param_get_int,
 137 };
 138
 139 #define param_check_interfaces_max(name, p) \
 140                 __param_check(name, p, int)
 141
 142 #ifdef HAVE_KERNEL_PARAM_OPS
 143 module_param(lnet_interfaces_max, interfaces_max, 0644);
 144 #else
 145 module_param_call(lnet_interfaces_max, intf_max_set, param_get_int,
 146                   &param_ops_interfaces_max, 0644);
 147 #endif
 148 MODULE_PARM_DESC(lnet_interfaces_max,
 149                 "Maximum number of interfaces in a node.");
 150
 151 unsigned lnet_peer_discovery_disabled = 0;
 152 static int discovery_set(const char *val, cfs_kernel_param_arg_t *kp);
 153
 154 static struct kernel_param_ops param_ops_discovery_disabled = {
 155         .set = discovery_set,
 156         .get = param_get_int,
 157 };
 158
 159 #define param_check_discovery_disabled(name, p) \
 160                 __param_check(name, p, int)
 161 #ifdef HAVE_KERNEL_PARAM_OPS
 162 module_param(lnet_peer_discovery_disabled, discovery_disabled, 0644);
 163 #else
 164 module_param_call(lnet_peer_discovery_disabled, discovery_set, param_get_int,
 165                   &param_ops_discovery_disabled, 0644);
 166 #endif
 167 MODULE_PARM_DESC(lnet_peer_discovery_disabled,
 168                 "Set to 1 to disable peer discovery on this node.");
 169
 170 unsigned int lnet_drop_asym_route;
 171 static int drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp);
 172
 173 static struct kernel_param_ops param_ops_drop_asym_route = {
 174         .set = drop_asym_route_set,
 175         .get = param_get_int,
 176 };
 177
 178 #define param_check_drop_asym_route(name, p)    \
 179         __param_check(name, p, int)
 180 #ifdef HAVE_KERNEL_PARAM_OPS
 181 module_param(lnet_drop_asym_route, drop_asym_route, 0644);
 182 #else
 183 module_param_call(lnet_drop_asym_route, drop_asym_route_set, param_get_int,
 184                   &param_ops_drop_asym_route, 0644);
 185 #endif
 186 MODULE_PARM_DESC(lnet_drop_asym_route,
 187                  "Set to 1 to drop asymmetrical route messages.");
 188
 189 #define LNET_TRANSACTION_TIMEOUT_DEFAULT 50
 190 unsigned int lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_DEFAULT;
 191 static int transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp);
 192 #ifdef HAVE_KERNEL_PARAM_OPS
 193 static struct kernel_param_ops param_ops_transaction_timeout = {
 194         .set = transaction_to_set,
 195         .get = param_get_int,
 196 };
 197
 198 #define param_check_transaction_timeout(name, p) \
 199                 __param_check(name, p, int)
 200 module_param(lnet_transaction_timeout, transaction_timeout, S_IRUGO|S_IWUSR);
 201 #else
 202 module_param_call(lnet_transaction_timeout, transaction_to_set, param_get_int,
 203                   &lnet_transaction_timeout, S_IRUGO|S_IWUSR);
 204 #endif
 205 MODULE_PARM_DESC(lnet_transaction_timeout,
 206                 "Maximum number of seconds to wait for a peer response.");
 207
 208 #define LNET_RETRY_COUNT_DEFAULT 2
 209 unsigned int lnet_retry_count = LNET_RETRY_COUNT_DEFAULT;
 210 static int retry_count_set(const char *val, cfs_kernel_param_arg_t *kp);
 211 #ifdef HAVE_KERNEL_PARAM_OPS
 212 static struct kernel_param_ops param_ops_retry_count = {
 213         .set = retry_count_set,
 214         .get = param_get_int,
 215 };
 216
 217 #define param_check_retry_count(name, p) \
 218                 __param_check(name, p, int)
 219 module_param(lnet_retry_count, retry_count, S_IRUGO|S_IWUSR);
 220 #else
 221 module_param_call(lnet_retry_count, retry_count_set, param_get_int,
 222                   &lnet_retry_count, S_IRUGO|S_IWUSR);
 223 #endif
 224 MODULE_PARM_DESC(lnet_retry_count,
 225                  "Maximum number of times to retry transmitting a message");
 226
 227 unsigned int lnet_response_tracking = 3;
 228 static int response_tracking_set(const char *val, cfs_kernel_param_arg_t *kp);
 229
 230 #ifdef HAVE_KERNEL_PARAM_OPS
 231 static struct kernel_param_ops param_ops_response_tracking = {
 232         .set = response_tracking_set,
 233         .get = param_get_int,
 234 };
 235
 236 #define param_check_response_tracking(name, p)  \
 237         __param_check(name, p, int)
 238 module_param(lnet_response_tracking, response_tracking, 0644);
 239 #else
 240 module_param_call(lnet_response_tracking, response_tracking_set, param_get_int,
 241                   &lnet_response_tracking, 0644);
 242 #endif
 243 MODULE_PARM_DESC(lnet_response_tracking,
 244                  "(0|1|2|3) LNet Internal Only|GET Reply only|PUT ACK only|Full Tracking (default)");
 245
 246 #define LNET_LND_TIMEOUT_DEFAULT ((LNET_TRANSACTION_TIMEOUT_DEFAULT - 1) / \
 247                                   (LNET_RETRY_COUNT_DEFAULT + 1))
 248 unsigned int lnet_lnd_timeout = LNET_LND_TIMEOUT_DEFAULT;
 249 static void lnet_set_lnd_timeout(void)
 250 {
 251         lnet_lnd_timeout = (lnet_transaction_timeout - 1) /
 252                            (lnet_retry_count + 1);
 253 }
 254
 255 /*
 256  * This sequence number keeps track of how many times DLC was used to
 257  * update the local NIs. It is incremented when a NI is added or
 258  * removed and checked when sending a message to determine if there is
 259  * a need to re-run the selection algorithm. See lnet_select_pathway()
 260  * for more details on its usage.
 261  */
 262 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
 263
 264 static int lnet_ping(struct lnet_process_id id, struct lnet_nid *src_nid,
 265                      signed long timeout, struct lnet_process_id __user *ids,
 266                      int n_ids);
 267
 268 static int lnet_discover(struct lnet_process_id id, __u32 force,
 269                          struct lnet_process_id __user *ids, int n_ids);
 270
 271 static int
 272 sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
 273 {
 274         int rc;
 275         unsigned *sensitivity = (unsigned *)kp->arg;
 276         unsigned long value;
 277
 278         rc = kstrtoul(val, 0, &value);
 279         if (rc) {
 280                 CERROR("Invalid module parameter value for 'lnet_health_sensitivity'\n");
 281                 return rc;
 282         }
 283
 284         /*
 285          * The purpose of locking the api_mutex here is to ensure that
 286          * the correct value ends up stored properly.
 287          */
 288         mutex_lock(&the_lnet.ln_api_mutex);
 289
 290         if (value > LNET_MAX_HEALTH_VALUE) {
 291                 mutex_unlock(&the_lnet.ln_api_mutex);
 292                 CERROR("Invalid health value. Maximum: %d value = %lu\n",
 293                        LNET_MAX_HEALTH_VALUE, value);
 294                 return -EINVAL;
 295         }
 296
 297         if (*sensitivity != 0 && value == 0 && lnet_retry_count != 0) {
 298                 lnet_retry_count = 0;
 299                 lnet_set_lnd_timeout();
 300         }
 301
 302         *sensitivity = value;
 303
 304         mutex_unlock(&the_lnet.ln_api_mutex);
 305
 306         return 0;
 307 }
 308
 309 static int
 310 recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp)
 311 {
 312         CWARN("'lnet_recovery_interval' has been deprecated\n");
 313
 314         return 0;
 315 }
 316
 317 static int
 318 discovery_set(const char *val, cfs_kernel_param_arg_t *kp)
 319 {
 320         int rc;
 321         unsigned *discovery_off = (unsigned *)kp->arg;
 322         unsigned long value;
 323         struct lnet_ping_buffer *pbuf;
 324
 325         rc = kstrtoul(val, 0, &value);
 326         if (rc) {
 327                 CERROR("Invalid module parameter value for 'lnet_peer_discovery_disabled'\n");
 328                 return rc;
 329         }
 330
 331         value = (value) ? 1 : 0;
 332
 333         /*
 334          * The purpose of locking the api_mutex here is to ensure that
 335          * the correct value ends up stored properly.
 336          */
 337         mutex_lock(&the_lnet.ln_api_mutex);
 338
 339         if (value == *discovery_off) {
 340                 mutex_unlock(&the_lnet.ln_api_mutex);
 341                 return 0;
 342         }
 343
 344         /*
 345          * We still want to set the discovery value even when LNet is not
 346          * running. This is the case when LNet is being loaded and we want
 347          * the module parameters to take effect. Otherwise if we're
 348          * changing the value dynamically, we want to set it after
 349          * updating the peers
 350          */
 351         if (the_lnet.ln_state != LNET_STATE_RUNNING) {
 352                 *discovery_off = value;
 353                 mutex_unlock(&the_lnet.ln_api_mutex);
 354                 return 0;
 355         }
 356
 357         /* tell peers that discovery setting has changed */
 358         lnet_net_lock(LNET_LOCK_EX);
 359         pbuf = the_lnet.ln_ping_target;
 360         if (value)
 361                 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
 362         else
 363                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
 364         lnet_net_unlock(LNET_LOCK_EX);
 365
 366         /* only send a push when we're turning off discovery */
 367         if (*discovery_off <= 0 && value > 0)
 368                 lnet_push_update_to_peers(1);
 369         *discovery_off = value;
 370
 371         mutex_unlock(&the_lnet.ln_api_mutex);
 372
 373         return 0;
 374 }
 375
 376 static int
 377 drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp)
 378 {
 379         int rc;
 380         unsigned int *drop_asym_route = (unsigned int *)kp->arg;
 381         unsigned long value;
 382
 383         rc = kstrtoul(val, 0, &value);
 384         if (rc) {
 385                 CERROR("Invalid module parameter value for "
 386                        "'lnet_drop_asym_route'\n");
 387                 return rc;
 388         }
 389
 390         /*
 391          * The purpose of locking the api_mutex here is to ensure that
 392          * the correct value ends up stored properly.
 393          */
 394         mutex_lock(&the_lnet.ln_api_mutex);
 395
 396         if (value == *drop_asym_route) {
 397                 mutex_unlock(&the_lnet.ln_api_mutex);
 398                 return 0;
 399         }
 400
 401         *drop_asym_route = value;
 402
 403         mutex_unlock(&the_lnet.ln_api_mutex);
 404
 405         return 0;
 406 }
 407
 408 static int
 409 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
 410 {
 411         int rc;
 412         unsigned *transaction_to = (unsigned *)kp->arg;
 413         unsigned long value;
 414
 415         rc = kstrtoul(val, 0, &value);
 416         if (rc) {
 417                 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
 418                 return rc;
 419         }
 420
 421         /*
 422          * The purpose of locking the api_mutex here is to ensure that
 423          * the correct value ends up stored properly.
 424          */
 425         mutex_lock(&the_lnet.ln_api_mutex);
 426
 427         if (value <= lnet_retry_count || value == 0) {
 428                 mutex_unlock(&the_lnet.ln_api_mutex);
 429                 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
 430                        "Has to be greater than lnet_retry_count (%u)\n",
 431                        value, lnet_retry_count);
 432                 return -EINVAL;
 433         }
 434
 435         if (value == *transaction_to) {
 436                 mutex_unlock(&the_lnet.ln_api_mutex);
 437                 return 0;
 438         }
 439
 440         *transaction_to = value;
 441         /* Update the lnet_lnd_timeout now that we've modified the
 442          * transaction timeout
 443          */
 444         lnet_set_lnd_timeout();
 445
 446         mutex_unlock(&the_lnet.ln_api_mutex);
 447
 448         return 0;
 449 }
 450
 451 static int
 452 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
 453 {
 454         int rc;
 455         unsigned *retry_count = (unsigned *)kp->arg;
 456         unsigned long value;
 457
 458         rc = kstrtoul(val, 0, &value);
 459         if (rc) {
 460                 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
 461                 return rc;
 462         }
 463
 464         /*
 465          * The purpose of locking the api_mutex here is to ensure that
 466          * the correct value ends up stored properly.
 467          */
 468         mutex_lock(&the_lnet.ln_api_mutex);
 469
 470         if (lnet_health_sensitivity == 0 && value > 0) {
 471                 mutex_unlock(&the_lnet.ln_api_mutex);
 472                 CERROR("Can not set lnet_retry_count when health feature is turned off\n");
 473                 return -EINVAL;
 474         }
 475
 476         if (value > lnet_transaction_timeout) {
 477                 mutex_unlock(&the_lnet.ln_api_mutex);
 478                 CERROR("Invalid value for lnet_retry_count (%lu). "
 479                        "Has to be smaller than lnet_transaction_timeout (%u)\n",
 480                        value, lnet_transaction_timeout);
 481                 return -EINVAL;
 482         }
 483
 484         *retry_count = value;
 485
 486         /* Update the lnet_lnd_timeout now that we've modified the
 487          * retry count
 488          */
 489         lnet_set_lnd_timeout();
 490
 491         mutex_unlock(&the_lnet.ln_api_mutex);
 492
 493         return 0;
 494 }
 495
 496 static int
 497 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
 498 {
 499         int value, rc;
 500
 501         rc = kstrtoint(val, 0, &value);
 502         if (rc) {
 503                 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
 504                 return rc;
 505         }
 506
 507         if (value < LNET_INTERFACES_MIN) {
 508                 CWARN("max interfaces provided are too small, setting to %d\n",
 509                       LNET_INTERFACES_MAX_DEFAULT);
 510                 value = LNET_INTERFACES_MAX_DEFAULT;
 511         }
 512
 513         *(int *)kp->arg = value;
 514
 515         return 0;
 516 }
 517
 518 static int
 519 response_tracking_set(const char *val, cfs_kernel_param_arg_t *kp)
 520 {
 521         int rc;
 522         unsigned long new_value;
 523
 524         rc = kstrtoul(val, 0, &new_value);
 525         if (rc) {
 526                 CERROR("Invalid value for 'lnet_response_tracking'\n");
 527                 return -EINVAL;
 528         }
 529
 530         if (new_value < 0 || new_value > 3) {
 531                 CWARN("Invalid value (%lu) for 'lnet_response_tracking'\n",
 532                       new_value);
 533                 return -EINVAL;
 534         }
 535
 536         lnet_response_tracking = new_value;
 537
 538         return 0;
 539 }
 540
 541 static const char *
 542 lnet_get_routes(void)
 543 {
 544         return routes;
 545 }
 546
 547 static const char *
 548 lnet_get_networks(void)
 549 {
 550         const char *nets;
 551         int rc;
 552
 553         if (*networks != 0 && *ip2nets != 0) {
 554                 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
 555                                    "'ip2nets' but not both at once\n");
 556                 return NULL;
 557         }
 558
 559         if (*ip2nets != 0) {
 560                 rc = lnet_parse_ip2nets(&nets, ip2nets);
 561                 return (rc == 0) ? nets : NULL;
 562         }
 563
 564         if (*networks != 0)
 565                 return networks;
 566
 567         return "tcp";
 568 }
 569
 570 static void
 571 lnet_init_locks(void)
 572 {
 573         spin_lock_init(&the_lnet.ln_eq_wait_lock);
 574         spin_lock_init(&the_lnet.ln_msg_resend_lock);
 575         init_completion(&the_lnet.ln_mt_wait_complete);
 576         mutex_init(&the_lnet.ln_lnd_mutex);
 577 }
 578
 579 struct kmem_cache *lnet_mes_cachep;        /* MEs kmem_cache */
 580 struct kmem_cache *lnet_small_mds_cachep;  /* <= LNET_SMALL_MD_SIZE bytes
 581                                             *  MDs kmem_cache */
 582 struct kmem_cache *lnet_udsp_cachep;       /* udsp cache */
 583 struct kmem_cache *lnet_rspt_cachep;       /* response tracker cache */
 584 struct kmem_cache *lnet_msg_cachep;
 585
 586 static int
 587 lnet_slab_setup(void)
 588 {
 589         /* create specific kmem_cache for MEs and small MDs (i.e., originally
 590          * allocated in <size-xxx> kmem_cache).
 591          */
 592         lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
 593                                             0, 0, NULL);
 594         if (!lnet_mes_cachep)
 595                 return -ENOMEM;
 596
 597         lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
 598                                                   LNET_SMALL_MD_SIZE, 0, 0,
 599                                                   NULL);
 600         if (!lnet_small_mds_cachep)
 601                 return -ENOMEM;
 602
 603         lnet_udsp_cachep = kmem_cache_create("lnet_udsp",
 604                                              sizeof(struct lnet_udsp),
 605                                              0, 0, NULL);
 606         if (!lnet_udsp_cachep)
 607                 return -ENOMEM;
 608
 609         lnet_rspt_cachep = kmem_cache_create("lnet_rspt", sizeof(struct lnet_rsp_tracker),
 610                                             0, 0, NULL);
 611         if (!lnet_rspt_cachep)
 612                 return -ENOMEM;
 613
 614         lnet_msg_cachep = kmem_cache_create("lnet_msg", sizeof(struct lnet_msg),
 615                                             0, 0, NULL);
 616         if (!lnet_msg_cachep)
 617                 return -ENOMEM;
 618
 619         return 0;
 620 }
 621
 622 static void
 623 lnet_slab_cleanup(void)
 624 {
 625         if (lnet_msg_cachep) {
 626                 kmem_cache_destroy(lnet_msg_cachep);
 627                 lnet_msg_cachep = NULL;
 628         }
 629
 630         if (lnet_rspt_cachep) {
 631                 kmem_cache_destroy(lnet_rspt_cachep);
 632                 lnet_rspt_cachep = NULL;
 633         }
 634
 635         if (lnet_udsp_cachep) {
 636                 kmem_cache_destroy(lnet_udsp_cachep);
 637                 lnet_udsp_cachep = NULL;
 638         }
 639
 640         if (lnet_small_mds_cachep) {
 641                 kmem_cache_destroy(lnet_small_mds_cachep);
 642                 lnet_small_mds_cachep = NULL;
 643         }
 644
 645         if (lnet_mes_cachep) {
 646                 kmem_cache_destroy(lnet_mes_cachep);
 647                 lnet_mes_cachep = NULL;
 648         }
 649 }
 650
 651 static int
 652 lnet_create_remote_nets_table(void)
 653 {
 654         int               i;
 655         struct list_head *hash;
 656
 657         LASSERT(the_lnet.ln_remote_nets_hash == NULL);
 658         LASSERT(the_lnet.ln_remote_nets_hbits > 0);
 659         CFS_ALLOC_PTR_ARRAY(hash, LNET_REMOTE_NETS_HASH_SIZE);
 660         if (hash == NULL) {
 661                 CERROR("Failed to create remote nets hash table\n");
 662                 return -ENOMEM;
 663         }
 664
 665         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 666                 INIT_LIST_HEAD(&hash[i]);
 667         the_lnet.ln_remote_nets_hash = hash;
 668         return 0;
 669 }
 670
 671 static void
 672 lnet_destroy_remote_nets_table(void)
 673 {
 674         int i;
 675
 676         if (the_lnet.ln_remote_nets_hash == NULL)
 677                 return;
 678
 679         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 680                 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
 681
 682         CFS_FREE_PTR_ARRAY(the_lnet.ln_remote_nets_hash,
 683                            LNET_REMOTE_NETS_HASH_SIZE);
 684         the_lnet.ln_remote_nets_hash = NULL;
 685 }
 686
 687 static void
 688 lnet_destroy_locks(void)
 689 {
 690         if (the_lnet.ln_res_lock != NULL) {
 691                 cfs_percpt_lock_free(the_lnet.ln_res_lock);
 692                 the_lnet.ln_res_lock = NULL;
 693         }
 694
 695         if (the_lnet.ln_net_lock != NULL) {
 696                 cfs_percpt_lock_free(the_lnet.ln_net_lock);
 697                 the_lnet.ln_net_lock = NULL;
 698         }
 699 }
 700
 701 static int
 702 lnet_create_locks(void)
 703 {
 704         lnet_init_locks();
 705
 706         the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 707         if (the_lnet.ln_res_lock == NULL)
 708                 goto failed;
 709
 710         the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 711         if (the_lnet.ln_net_lock == NULL)
 712                 goto failed;
 713
 714         return 0;
 715
 716  failed:
 717         lnet_destroy_locks();
 718         return -ENOMEM;
 719 }
 720
 721 static void lnet_assert_wire_constants(void)
 722 {
 723         /* Wire protocol assertions generated by 'wirecheck'
 724          * running on Linux robert.bartonsoftware.com 2.6.8-1.521
 725          * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
 726          * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7)
 727          */
 728
 729         /* Constants... */
 730         BUILD_BUG_ON(LNET_PROTO_TCP_MAGIC != 0xeebc0ded);
 731         BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MAJOR != 1);
 732         BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MINOR != 0);
 733         BUILD_BUG_ON(LNET_MSG_ACK != 0);
 734         BUILD_BUG_ON(LNET_MSG_PUT != 1);
 735         BUILD_BUG_ON(LNET_MSG_GET != 2);
 736         BUILD_BUG_ON(LNET_MSG_REPLY != 3);
 737         BUILD_BUG_ON(LNET_MSG_HELLO != 4);
 738
 739         BUILD_BUG_ON((int)sizeof(lnet_nid_t) != 8);
 740         BUILD_BUG_ON((int)sizeof(lnet_pid_t) != 4);
 741
 742         /* Checks for struct lnet_nid */
 743         BUILD_BUG_ON((int)sizeof(struct lnet_nid) != 20);
 744         BUILD_BUG_ON((int)offsetof(struct lnet_nid, nid_size) != 0);
 745         BUILD_BUG_ON((int)sizeof(((struct lnet_nid *)0)->nid_size) != 1);
 746         BUILD_BUG_ON((int)offsetof(struct lnet_nid, nid_type) != 1);
 747         BUILD_BUG_ON((int)sizeof(((struct lnet_nid *)0)->nid_type) != 1);
 748         BUILD_BUG_ON((int)offsetof(struct lnet_nid, nid_num) != 2);
 749         BUILD_BUG_ON((int)sizeof(((struct lnet_nid *)0)->nid_num) != 2);
 750         BUILD_BUG_ON((int)offsetof(struct lnet_nid, nid_addr) != 4);
 751         BUILD_BUG_ON((int)sizeof(((struct lnet_nid *)0)->nid_addr) != 16);
 752
 753         /* Checks for struct lnet_process_id_packed */
 754         BUILD_BUG_ON((int)sizeof(struct lnet_process_id_packed) != 12);
 755         BUILD_BUG_ON((int)offsetof(struct lnet_process_id_packed, nid) != 0);
 756         BUILD_BUG_ON((int)sizeof(((struct lnet_process_id_packed *)0)->nid) != 8);
 757         BUILD_BUG_ON((int)offsetof(struct lnet_process_id_packed, pid) != 8);
 758         BUILD_BUG_ON((int)sizeof(((struct lnet_process_id_packed *)0)->pid) != 4);
 759
 760         /* Checks for struct lnet_handle_wire */
 761         BUILD_BUG_ON((int)sizeof(struct lnet_handle_wire) != 16);
 762         BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
 763                                    wh_interface_cookie) != 0);
 764         BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) != 8);
 765         BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
 766                                    wh_object_cookie) != 8);
 767         BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) != 8);
 768
 769         /* Checks for struct struct lnet_magicversion */
 770         BUILD_BUG_ON((int)sizeof(struct lnet_magicversion) != 8);
 771         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, magic) != 0);
 772         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->magic) != 4);
 773         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, version_major) != 4);
 774         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_major) != 2);
 775         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion,
 776                                    version_minor) != 6);
 777         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_minor) != 2);
 778
 779         /* Checks for struct _lnet_hdr_nid4 */
 780         BUILD_BUG_ON((int)sizeof(struct _lnet_hdr_nid4) != 72);
 781         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, dest_nid) != 0);
 782         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->dest_nid) != 8);
 783         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, src_nid) != 8);
 784         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->src_nid) != 8);
 785         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, dest_pid) != 16);
 786         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->dest_pid) != 4);
 787         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, src_pid) != 20);
 788         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->src_pid) != 4);
 789         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, type) != 24);
 790         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->type) != 4);
 791         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, payload_length) != 28);
 792         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->payload_length) != 4);
 793         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg) != 32);
 794         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg) != 40);
 795
 796         /* Ack */
 797         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.ack.dst_wmd) != 32);
 798         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.ack.dst_wmd) != 16);
 799         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.ack.match_bits) != 48);
 800         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.ack.match_bits) != 8);
 801         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.ack.mlength) != 56);
 802         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.ack.mlength) != 4);
 803
 804         /* Put */
 805         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.put.ack_wmd) != 32);
 806         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.put.ack_wmd) != 16);
 807         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.put.match_bits) != 48);
 808         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.put.match_bits) != 8);
 809         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.put.hdr_data) != 56);
 810         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.put.hdr_data) != 8);
 811         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.put.ptl_index) != 64);
 812         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.put.ptl_index) != 4);
 813         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.put.offset) != 68);
 814         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.put.offset) != 4);
 815
 816         /* Get */
 817         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.get.return_wmd) != 32);
 818         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.get.return_wmd) != 16);
 819         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.get.match_bits) != 48);
 820         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.get.match_bits) != 8);
 821         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.get.ptl_index) != 56);
 822         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.get.ptl_index) != 4);
 823         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.get.src_offset) != 60);
 824         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.get.src_offset) != 4);
 825         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.get.sink_length) != 64);
 826         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.get.sink_length) != 4);
 827
 828         /* Reply */
 829         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.reply.dst_wmd) != 32);
 830         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.reply.dst_wmd) != 16);
 831
 832         /* Hello */
 833         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.hello.incarnation) != 32);
 834         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.hello.incarnation) != 8);
 835         BUILD_BUG_ON((int)offsetof(struct _lnet_hdr_nid4, msg.hello.type) != 40);
 836         BUILD_BUG_ON((int)sizeof(((struct _lnet_hdr_nid4 *)0)->msg.hello.type) != 4);
 837
 838         /* Checks for struct lnet_ni_status and related constants */
 839         BUILD_BUG_ON(LNET_NI_STATUS_INVALID != 0x00000000);
 840         BUILD_BUG_ON(LNET_NI_STATUS_UP != 0x15aac0de);
 841         BUILD_BUG_ON(LNET_NI_STATUS_DOWN != 0xdeadface);
 842
 843         /* Checks for struct lnet_ni_status */
 844         BUILD_BUG_ON((int)sizeof(struct lnet_ni_status) != 16);
 845         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_nid) != 0);
 846         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) != 8);
 847         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_status) != 8);
 848         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_status) != 4);
 849         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_unused) != 12);
 850         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) != 4);
 851
 852         /* Checks for struct lnet_ping_info and related constants */
 853         BUILD_BUG_ON(LNET_PROTO_PING_MAGIC != 0x70696E67);
 854         BUILD_BUG_ON(LNET_PING_FEAT_INVAL != 0);
 855         BUILD_BUG_ON(LNET_PING_FEAT_BASE != 1);
 856         BUILD_BUG_ON(LNET_PING_FEAT_NI_STATUS != 2);
 857         BUILD_BUG_ON(LNET_PING_FEAT_RTE_DISABLED != 4);
 858         BUILD_BUG_ON(LNET_PING_FEAT_MULTI_RAIL != 8);
 859         BUILD_BUG_ON(LNET_PING_FEAT_DISCOVERY != 16);
 860         BUILD_BUG_ON(LNET_PING_FEAT_BITS != 31);
 861
 862         /* Checks for struct lnet_ping_info */
 863         BUILD_BUG_ON((int)sizeof(struct lnet_ping_info) != 16);
 864         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_magic) != 0);
 865         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) != 4);
 866         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_features) != 4);
 867         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_features) != 4);
 868         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_pid) != 8);
 869         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) != 4);
 870         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_nnis) != 12);
 871         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) != 4);
 872         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_ni) != 16);
 873         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) != 0);
 874
 875         /* Acceptor connection request */
 876         BUILD_BUG_ON(LNET_PROTO_ACCEPTOR_VERSION != 1);
 877
 878         /* Checks for struct lnet_acceptor_connreq */
 879         BUILD_BUG_ON((int)sizeof(struct lnet_acceptor_connreq) != 16);
 880         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_magic) != 0);
 881         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_magic) != 4);
 882         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_version) != 4);
 883         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_version) != 4);
 884         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_nid) != 8);
 885         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_nid) != 8);
 886
 887         /* Checks for struct lnet_acceptor_connreq_v2 */
 888         BUILD_BUG_ON((int)sizeof(struct lnet_acceptor_connreq_v2) != 28);
 889         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq_v2, acr_magic) != 0);
 890         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq_v2 *)0)->acr_magic) != 4);
 891         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq_v2, acr_version) != 4);
 892         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq_v2 *)0)->acr_version) != 4);
 893         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq_v2, acr_nid) != 8);
 894         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq_v2 *)0)->acr_nid) != 20);
 895
 896         /* Checks for struct lnet_counters_common */
 897         BUILD_BUG_ON((int)sizeof(struct lnet_counters_common) != 60);
 898         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_msgs_alloc) != 0);
 899         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_msgs_alloc) != 4);
 900         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_msgs_max) != 4);
 901         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_msgs_max) != 4);
 902         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_errors) != 8);
 903         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_errors) != 4);
 904         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_send_count) != 12);
 905         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_send_count) != 4);
 906         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_recv_count) != 16);
 907         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_recv_count) != 4);
 908         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_route_count) != 20);
 909         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_route_count) != 4);
 910         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_drop_count) != 24);
 911         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_drop_count) != 4);
 912         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_send_length) != 28);
 913         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_send_length) != 8);
 914         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_recv_length) != 36);
 915         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_recv_length) != 8);
 916         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_route_length) != 44);
 917         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_route_length) != 8);
 918         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_drop_length) != 52);
 919         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_drop_length) != 8);
 920 }
 921
 922 static const struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
 923 {
 924         const struct lnet_lnd *lnd;
 925
 926         /* holding lnd mutex */
 927         if (type >= NUM_LNDS)
 928                 return NULL;
 929         lnd = the_lnet.ln_lnds[type];
 930         LASSERT(!lnd || lnd->lnd_type == type);
 931
 932         return lnd;
 933 }
 934
 935 unsigned int
 936 lnet_get_lnd_timeout(void)
 937 {
 938         return lnet_lnd_timeout;
 939 }
 940 EXPORT_SYMBOL(lnet_get_lnd_timeout);
 941
 942 void
 943 lnet_register_lnd(const struct lnet_lnd *lnd)
 944 {
 945         mutex_lock(&the_lnet.ln_lnd_mutex);
 946
 947         LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
 948         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
 949
 950         the_lnet.ln_lnds[lnd->lnd_type] = lnd;
 951
 952         CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
 953
 954         mutex_unlock(&the_lnet.ln_lnd_mutex);
 955 }
 956 EXPORT_SYMBOL(lnet_register_lnd);
 957
 958 void
 959 lnet_unregister_lnd(const struct lnet_lnd *lnd)
 960 {
 961         mutex_lock(&the_lnet.ln_lnd_mutex);
 962
 963         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
 964
 965         the_lnet.ln_lnds[lnd->lnd_type] = NULL;
 966         CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
 967
 968         mutex_unlock(&the_lnet.ln_lnd_mutex);
 969 }
 970 EXPORT_SYMBOL(lnet_unregister_lnd);
 971
 972 static void
 973 lnet_counters_get_common_locked(struct lnet_counters_common *common)
 974 {
 975         struct lnet_counters *ctr;
 976         int i;
 977
 978         /* FIXME !!! Their is no assert_lnet_net_locked() to ensure this
 979          * actually called under the protection of the lnet_net_lock.
 980          */
 981         memset(common, 0, sizeof(*common));
 982
 983         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
 984                 common->lcc_msgs_max     += ctr->lct_common.lcc_msgs_max;
 985                 common->lcc_msgs_alloc   += ctr->lct_common.lcc_msgs_alloc;
 986                 common->lcc_errors       += ctr->lct_common.lcc_errors;
 987                 common->lcc_send_count   += ctr->lct_common.lcc_send_count;
 988                 common->lcc_recv_count   += ctr->lct_common.lcc_recv_count;
 989                 common->lcc_route_count  += ctr->lct_common.lcc_route_count;
 990                 common->lcc_drop_count   += ctr->lct_common.lcc_drop_count;
 991                 common->lcc_send_length  += ctr->lct_common.lcc_send_length;
 992                 common->lcc_recv_length  += ctr->lct_common.lcc_recv_length;
 993                 common->lcc_route_length += ctr->lct_common.lcc_route_length;
 994                 common->lcc_drop_length  += ctr->lct_common.lcc_drop_length;
 995         }
 996 }
 997
 998 void
 999 lnet_counters_get_common(struct lnet_counters_common *common)
1000 {
1001         lnet_net_lock(LNET_LOCK_EX);
1002         lnet_counters_get_common_locked(common);
1003         lnet_net_unlock(LNET_LOCK_EX);
1004 }
1005 EXPORT_SYMBOL(lnet_counters_get_common);
1006
1007 int
1008 lnet_counters_get(struct lnet_counters *counters)
1009 {
1010         struct lnet_counters *ctr;
1011         struct lnet_counters_health *health = &counters->lct_health;
1012         int i, rc = 0;
1013
1014         memset(counters, 0, sizeof(*counters));
1015
1016         lnet_net_lock(LNET_LOCK_EX);
1017
1018         if (the_lnet.ln_state != LNET_STATE_RUNNING)
1019                 GOTO(out_unlock, rc = -ENODEV);
1020
1021         lnet_counters_get_common_locked(&counters->lct_common);
1022
1023         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
1024                 health->lch_rst_alloc    += ctr->lct_health.lch_rst_alloc;
1025                 health->lch_resend_count += ctr->lct_health.lch_resend_count;
1026                 health->lch_response_timeout_count +=
1027                                 ctr->lct_health.lch_response_timeout_count;
1028                 health->lch_local_interrupt_count +=
1029                                 ctr->lct_health.lch_local_interrupt_count;
1030                 health->lch_local_dropped_count +=
1031                                 ctr->lct_health.lch_local_dropped_count;
1032                 health->lch_local_aborted_count +=
1033                                 ctr->lct_health.lch_local_aborted_count;
1034                 health->lch_local_no_route_count +=
1035                                 ctr->lct_health.lch_local_no_route_count;
1036                 health->lch_local_timeout_count +=
1037                                 ctr->lct_health.lch_local_timeout_count;
1038                 health->lch_local_error_count +=
1039                                 ctr->lct_health.lch_local_error_count;
1040                 health->lch_remote_dropped_count +=
1041                                 ctr->lct_health.lch_remote_dropped_count;
1042                 health->lch_remote_error_count +=
1043                                 ctr->lct_health.lch_remote_error_count;
1044                 health->lch_remote_timeout_count +=
1045                                 ctr->lct_health.lch_remote_timeout_count;
1046                 health->lch_network_timeout_count +=
1047                                 ctr->lct_health.lch_network_timeout_count;
1048         }
1049 out_unlock:
1050         lnet_net_unlock(LNET_LOCK_EX);
1051         return rc;
1052 }
1053 EXPORT_SYMBOL(lnet_counters_get);
1054
1055 void
1056 lnet_counters_reset(void)
1057 {
1058         struct lnet_counters *counters;
1059         int             i;
1060
1061         lnet_net_lock(LNET_LOCK_EX);
1062
1063         if (the_lnet.ln_state != LNET_STATE_RUNNING)
1064                 goto avoid_reset;
1065
1066         cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
1067                 memset(counters, 0, sizeof(struct lnet_counters));
1068 avoid_reset:
1069         lnet_net_unlock(LNET_LOCK_EX);
1070 }
1071
1072 static char *
1073 lnet_res_type2str(int type)
1074 {
1075         switch (type) {
1076         default:
1077                 LBUG();
1078         case LNET_COOKIE_TYPE_MD:
1079                 return "MD";
1080         case LNET_COOKIE_TYPE_ME:
1081                 return "ME";
1082         case LNET_COOKIE_TYPE_EQ:
1083                 return "EQ";
1084         }
1085 }
1086
1087 static void
1088 lnet_res_container_cleanup(struct lnet_res_container *rec)
1089 {
1090         int     count = 0;
1091
1092         if (rec->rec_type == 0) /* not set yet, it's uninitialized */
1093                 return;
1094
1095         while (!list_empty(&rec->rec_active)) {
1096                 struct list_head *e = rec->rec_active.next;
1097
1098                 list_del_init(e);
1099                 if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
1100                         lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
1101
1102                 } else { /* NB: Active MEs should be attached on portals */
1103                         LBUG();
1104                 }
1105                 count++;
1106         }
1107
1108         if (count > 0) {
1109                 /* Found alive MD/ME/EQ, user really should unlink/free
1110                  * all of them before finalize LNet, but if someone didn't,
1111                  * we have to recycle garbage for him */
1112                 CERROR("%d active elements on exit of %s container\n",
1113                        count, lnet_res_type2str(rec->rec_type));
1114         }
1115
1116         if (rec->rec_lh_hash != NULL) {
1117                 CFS_FREE_PTR_ARRAY(rec->rec_lh_hash, LNET_LH_HASH_SIZE);
1118                 rec->rec_lh_hash = NULL;
1119         }
1120
1121         rec->rec_type = 0; /* mark it as finalized */
1122 }
1123
1124 static int
1125 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
1126 {
1127         int     rc = 0;
1128         int     i;
1129
1130         LASSERT(rec->rec_type == 0);
1131
1132         rec->rec_type = type;
1133         INIT_LIST_HEAD(&rec->rec_active);
1134
1135         rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1136
1137         /* Arbitrary choice of hash table size */
1138         LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1139                          LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1140         if (rec->rec_lh_hash == NULL) {
1141                 rc = -ENOMEM;
1142                 goto out;
1143         }
1144
1145         for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1146                 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1147
1148         return 0;
1149
1150 out:
1151         CERROR("Failed to setup %s resource container\n",
1152                lnet_res_type2str(type));
1153         lnet_res_container_cleanup(rec);
1154         return rc;
1155 }
1156
1157 static void
1158 lnet_res_containers_destroy(struct lnet_res_container **recs)
1159 {
1160         struct lnet_res_container       *rec;
1161         int                             i;
1162
1163         cfs_percpt_for_each(rec, i, recs)
1164                 lnet_res_container_cleanup(rec);
1165
1166         cfs_percpt_free(recs);
1167 }
1168
1169 static struct lnet_res_container **
1170 lnet_res_containers_create(int type)
1171 {
1172         struct lnet_res_container       **recs;
1173         struct lnet_res_container       *rec;
1174         int                             rc;
1175         int                             i;
1176
1177         recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1178         if (recs == NULL) {
1179                 CERROR("Failed to allocate %s resource containers\n",
1180                        lnet_res_type2str(type));
1181                 return NULL;
1182         }
1183
1184         cfs_percpt_for_each(rec, i, recs) {
1185                 rc = lnet_res_container_setup(rec, i, type);
1186                 if (rc != 0) {
1187                         lnet_res_containers_destroy(recs);
1188                         return NULL;
1189                 }
1190         }
1191
1192         return recs;
1193 }
1194
1195 struct lnet_libhandle *
1196 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1197 {
1198         /* ALWAYS called with lnet_res_lock held */
1199         struct list_head        *head;
1200         struct lnet_libhandle   *lh;
1201         unsigned int            hash;
1202
1203         if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1204                 return NULL;
1205
1206         hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1207         head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1208
1209         list_for_each_entry(lh, head, lh_hash_chain) {
1210                 if (lh->lh_cookie == cookie)
1211                         return lh;
1212         }
1213
1214         return NULL;
1215 }
1216
1217 void
1218 lnet_res_lh_initialize(struct lnet_res_container *rec,
1219                        struct lnet_libhandle *lh)
1220 {
1221         /* ALWAYS called with lnet_res_lock held */
1222         unsigned int    ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1223         unsigned int    hash;
1224
1225         lh->lh_cookie = rec->rec_lh_cookie;
1226         rec->rec_lh_cookie += 1 << ibits;
1227
1228         hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1229
1230         list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1231 }
1232
1233 struct list_head **
1234 lnet_create_array_of_queues(void)
1235 {
1236         struct list_head **qs;
1237         struct list_head *q;
1238         int i;
1239
1240         qs = cfs_percpt_alloc(lnet_cpt_table(),
1241                               sizeof(struct list_head));
1242         if (!qs) {
1243                 CERROR("Failed to allocate queues\n");
1244                 return NULL;
1245         }
1246
1247         cfs_percpt_for_each(q, i, qs)
1248                 INIT_LIST_HEAD(q);
1249
1250         return qs;
1251 }
1252
1253 static int lnet_unprepare(void);
1254
1255 static int
1256 lnet_prepare(lnet_pid_t requested_pid)
1257 {
1258         /* Prepare to bring up the network */
1259         struct lnet_res_container **recs;
1260         int                       rc = 0;
1261
1262         if (requested_pid == LNET_PID_ANY) {
1263                 /* Don't instantiate LNET just for me */
1264                 return -ENETDOWN;
1265         }
1266
1267         LASSERT(the_lnet.ln_refcount == 0);
1268
1269         the_lnet.ln_routing = 0;
1270
1271         LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1272         the_lnet.ln_pid = requested_pid;
1273
1274         INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1275         INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1276         INIT_LIST_HEAD(&the_lnet.ln_nets);
1277         INIT_LIST_HEAD(&the_lnet.ln_routers);
1278         INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1279         INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1280         INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1281         INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1282         INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1283         INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1284         INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1285         INIT_LIST_HEAD(&the_lnet.ln_udsp_list);
1286         init_waitqueue_head(&the_lnet.ln_dc_waitq);
1287         the_lnet.ln_mt_handler = NULL;
1288         init_completion(&the_lnet.ln_started);
1289
1290         rc = lnet_slab_setup();
1291         if (rc != 0)
1292                 goto failed;
1293
1294         rc = lnet_create_remote_nets_table();
1295         if (rc != 0)
1296                 goto failed;
1297
1298         /*
1299          * NB the interface cookie in wire handles guards against delayed
1300          * replies and ACKs appearing valid after reboot.
1301          */
1302         the_lnet.ln_interface_cookie = ktime_get_real_ns();
1303
1304         the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1305                                                 sizeof(struct lnet_counters));
1306         if (the_lnet.ln_counters == NULL) {
1307                 CERROR("Failed to allocate counters for LNet\n");
1308                 rc = -ENOMEM;
1309                 goto failed;
1310         }
1311
1312         rc = lnet_peer_tables_create();
1313         if (rc != 0)
1314                 goto failed;
1315
1316         rc = lnet_msg_containers_create();
1317         if (rc != 0)
1318                 goto failed;
1319
1320         rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1321                                       LNET_COOKIE_TYPE_EQ);
1322         if (rc != 0)
1323                 goto failed;
1324
1325         recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1326         if (recs == NULL) {
1327                 rc = -ENOMEM;
1328                 goto failed;
1329         }
1330
1331         the_lnet.ln_md_containers = recs;
1332
1333         rc = lnet_portals_create();
1334         if (rc != 0) {
1335                 CERROR("Failed to create portals for LNet: %d\n", rc);
1336                 goto failed;
1337         }
1338
1339         the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1340         if (!the_lnet.ln_mt_zombie_rstqs) {
1341                 rc = -ENOMEM;
1342                 goto failed;
1343         }
1344
1345         return 0;
1346
1347  failed:
1348         lnet_unprepare();
1349         return rc;
1350 }
1351
1352 static int
1353 lnet_unprepare(void)
1354 {
1355         /* NB no LNET_LOCK since this is the last reference.  All LND instances
1356          * have shut down already, so it is safe to unlink and free all
1357          * descriptors, even those that appear committed to a network op (eg MD
1358          * with non-zero pending count) */
1359
1360         lnet_fail_nid(LNET_NID_ANY, 0);
1361
1362         LASSERT(the_lnet.ln_refcount == 0);
1363         LASSERT(list_empty(&the_lnet.ln_test_peers));
1364         LASSERT(list_empty(&the_lnet.ln_nets));
1365
1366         if (the_lnet.ln_mt_zombie_rstqs) {
1367                 lnet_clean_zombie_rstqs();
1368                 the_lnet.ln_mt_zombie_rstqs = NULL;
1369         }
1370
1371         lnet_assert_handler_unused(the_lnet.ln_mt_handler);
1372         the_lnet.ln_mt_handler = NULL;
1373
1374         lnet_portals_destroy();
1375
1376         if (the_lnet.ln_md_containers != NULL) {
1377                 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1378                 the_lnet.ln_md_containers = NULL;
1379         }
1380
1381         lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1382
1383         lnet_msg_containers_destroy();
1384         lnet_peer_uninit();
1385         lnet_rtrpools_free(0);
1386
1387         if (the_lnet.ln_counters != NULL) {
1388                 cfs_percpt_free(the_lnet.ln_counters);
1389                 the_lnet.ln_counters = NULL;
1390         }
1391         lnet_destroy_remote_nets_table();
1392         lnet_udsp_destroy(true);
1393         lnet_slab_cleanup();
1394
1395         return 0;
1396 }
1397
1398 struct lnet_ni  *
1399 lnet_net2ni_locked(__u32 net_id, int cpt)
1400 {
1401         struct lnet_ni   *ni;
1402         struct lnet_net  *net;
1403
1404         LASSERT(cpt != LNET_LOCK_EX);
1405
1406         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1407                 if (net->net_id == net_id) {
1408                         ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1409                                         ni_netlist);
1410                         return ni;
1411                 }
1412         }
1413
1414         return NULL;
1415 }
1416
1417 struct lnet_ni *
1418 lnet_net2ni_addref(__u32 net)
1419 {
1420         struct lnet_ni *ni;
1421
1422         lnet_net_lock(0);
1423         ni = lnet_net2ni_locked(net, 0);
1424         if (ni)
1425                 lnet_ni_addref_locked(ni, 0);
1426         lnet_net_unlock(0);
1427
1428         return ni;
1429 }
1430 EXPORT_SYMBOL(lnet_net2ni_addref);
1431
1432 struct lnet_net *
1433 lnet_get_net_locked(__u32 net_id)
1434 {
1435         struct lnet_net  *net;
1436
1437         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1438                 if (net->net_id == net_id)
1439                         return net;
1440         }
1441
1442         return NULL;
1443 }
1444
1445 void
1446 lnet_net_clr_pref_rtrs(struct lnet_net *net)
1447 {
1448         struct list_head zombies;
1449         struct lnet_nid_list *ne;
1450         struct lnet_nid_list *tmp;
1451
1452         INIT_LIST_HEAD(&zombies);
1453
1454         lnet_net_lock(LNET_LOCK_EX);
1455         list_splice_init(&net->net_rtr_pref_nids, &zombies);
1456         lnet_net_unlock(LNET_LOCK_EX);
1457
1458         list_for_each_entry_safe(ne, tmp, &zombies, nl_list) {
1459                 list_del_init(&ne->nl_list);
1460                 LIBCFS_FREE(ne, sizeof(*ne));
1461         }
1462 }
1463
1464 int
1465 lnet_net_add_pref_rtr(struct lnet_net *net,
1466                       struct lnet_nid *gw_nid)
1467 __must_hold(&the_lnet.ln_api_mutex)
1468 {
1469         struct lnet_nid_list *ne;
1470
1471         /* This function is called with api_mutex held. When the api_mutex
1472          * is held the list can not be modified, as it is only modified as
1473          * a result of applying a UDSP and that happens under api_mutex
1474          * lock.
1475          */
1476         list_for_each_entry(ne, &net->net_rtr_pref_nids, nl_list) {
1477                 if (nid_same(&ne->nl_nid, gw_nid))
1478                         return -EEXIST;
1479         }
1480
1481         LIBCFS_ALLOC(ne, sizeof(*ne));
1482         if (!ne)
1483                 return -ENOMEM;
1484
1485         ne->nl_nid = *gw_nid;
1486
1487         /* Lock the cpt to protect against addition and checks in the
1488          * selection algorithm
1489          */
1490         lnet_net_lock(LNET_LOCK_EX);
1491         list_add(&ne->nl_list, &net->net_rtr_pref_nids);
1492         lnet_net_unlock(LNET_LOCK_EX);
1493
1494         return 0;
1495 }
1496
1497 bool
1498 lnet_net_is_pref_rtr_locked(struct lnet_net *net, struct lnet_nid *rtr_nid)
1499 {
1500         struct lnet_nid_list *ne;
1501
1502         CDEBUG(D_NET, "%s: rtr pref empty: %d\n",
1503                libcfs_net2str(net->net_id),
1504                list_empty(&net->net_rtr_pref_nids));
1505
1506         if (list_empty(&net->net_rtr_pref_nids))
1507                 return false;
1508
1509         list_for_each_entry(ne, &net->net_rtr_pref_nids, nl_list) {
1510                 CDEBUG(D_NET, "Comparing pref %s with gw %s\n",
1511                        libcfs_nidstr(&ne->nl_nid),
1512                        libcfs_nidstr(rtr_nid));
1513                 if (nid_same(rtr_nid, &ne->nl_nid))
1514                         return true;
1515         }
1516
1517         return false;
1518 }
1519
1520 static unsigned int
1521 lnet_nid4_cpt_hash(lnet_nid_t nid, unsigned int number)
1522 {
1523         __u64 key = nid;
1524         __u64 pair_bits = 0x0001000100010001LLU;
1525         __u64 mask = pair_bits * 0xFF;
1526         __u64 pair_sum;
1527
1528         /* Use (sum-by-multiplication of nid bytes) mod (number of CPTs)
1529          * to match nid to a CPT.
1530          */
1531         pair_sum = (key & mask) + ((key >> 8) & mask);
1532         pair_sum = (pair_sum * pair_bits) >> 48;
1533
1534         CDEBUG(D_NET, "Match nid %s to cpt %u\n",
1535                libcfs_nid2str(nid), (unsigned int)(pair_sum) % number);
1536
1537         return (unsigned int)(pair_sum) % number;
1538 }
1539
1540 unsigned int
1541 lnet_nid_cpt_hash(struct lnet_nid *nid, unsigned int number)
1542 {
1543         unsigned int val;
1544         u32 h = 0;
1545         int i;
1546
1547         LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1548
1549         if (number == 1)
1550                 return 0;
1551
1552         if (nid_is_nid4(nid))
1553                 return lnet_nid4_cpt_hash(lnet_nid_to_nid4(nid), number);
1554
1555         for (i = 0; i < 4; i++)
1556                 h = hash_32(nid->nid_addr[i]^h, 32);
1557         val = hash_32(LNET_NID_NET(nid) ^ h, LNET_CPT_BITS);
1558         if (val < number)
1559                 return val;
1560         return (unsigned int)(h + val + (val >> 1)) % number;
1561 }
1562
1563 int
1564 lnet_cpt_of_nid_locked(struct lnet_nid *nid, struct lnet_ni *ni)
1565 {
1566         struct lnet_net *net;
1567
1568         /* must called with hold of lnet_net_lock */
1569         if (LNET_CPT_NUMBER == 1)
1570                 return 0; /* the only one */
1571
1572         /*
1573          * If NI is provided then use the CPT identified in the NI cpt
1574          * list if one exists. If one doesn't exist, then that NI is
1575          * associated with all CPTs and it follows that the net it belongs
1576          * to is implicitly associated with all CPTs, so just hash the nid
1577          * and return that.
1578          */
1579         if (ni != NULL) {
1580                 if (ni->ni_cpts != NULL)
1581                         return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1582                                                              ni->ni_ncpts)];
1583                 else
1584                         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1585         }
1586
1587         /* no NI provided so look at the net */
1588         net = lnet_get_net_locked(LNET_NID_NET(nid));
1589
1590         if (net != NULL && net->net_cpts != NULL) {
1591                 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1592         }
1593
1594         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1595 }
1596
1597 int
1598 lnet_nid2cpt(struct lnet_nid *nid, struct lnet_ni *ni)
1599 {
1600         int     cpt;
1601         int     cpt2;
1602
1603         if (LNET_CPT_NUMBER == 1)
1604                 return 0; /* the only one */
1605
1606         cpt = lnet_net_lock_current();
1607
1608         cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1609
1610         lnet_net_unlock(cpt);
1611
1612         return cpt2;
1613 }
1614 EXPORT_SYMBOL(lnet_nid2cpt);
1615
1616 int
1617 lnet_cpt_of_nid(lnet_nid_t nid4, struct lnet_ni *ni)
1618 {
1619         struct lnet_nid nid;
1620
1621         if (LNET_CPT_NUMBER == 1)
1622                 return 0; /* the only one */
1623
1624         lnet_nid4_to_nid(nid4, &nid);
1625         return lnet_nid2cpt(&nid, ni);
1626 }
1627 EXPORT_SYMBOL(lnet_cpt_of_nid);
1628
1629 int
1630 lnet_islocalnet_locked(__u32 net_id)
1631 {
1632         struct lnet_net *net;
1633         bool local;
1634
1635         net = lnet_get_net_locked(net_id);
1636
1637         local = net != NULL;
1638
1639         return local;
1640 }
1641
1642 int
1643 lnet_islocalnet(__u32 net_id)
1644 {
1645         int cpt;
1646         bool local;
1647
1648         cpt = lnet_net_lock_current();
1649
1650         local = lnet_islocalnet_locked(net_id);
1651
1652         lnet_net_unlock(cpt);
1653
1654         return local;
1655 }
1656
1657 struct lnet_ni  *
1658 lnet_nid_to_ni_locked(struct lnet_nid *nid, int cpt)
1659 {
1660         struct lnet_net  *net;
1661         struct lnet_ni *ni;
1662
1663         LASSERT(cpt != LNET_LOCK_EX);
1664
1665         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1666                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1667                         if (nid_same(&ni->ni_nid, nid))
1668                                 return ni;
1669                 }
1670         }
1671
1672         return NULL;
1673 }
1674
1675 struct lnet_ni  *
1676 lnet_nid2ni_locked(lnet_nid_t nid4, int cpt)
1677 {
1678         struct lnet_nid nid;
1679
1680         lnet_nid4_to_nid(nid4, &nid);
1681         return lnet_nid_to_ni_locked(&nid, cpt);
1682 }
1683
1684 struct lnet_ni *
1685 lnet_nid2ni_addref(lnet_nid_t nid4)
1686 {
1687         struct lnet_ni *ni;
1688         struct lnet_nid nid;
1689
1690         lnet_nid4_to_nid(nid4, &nid);
1691
1692         lnet_net_lock(0);
1693         ni = lnet_nid_to_ni_locked(&nid, 0);
1694         if (ni)
1695                 lnet_ni_addref_locked(ni, 0);
1696         lnet_net_unlock(0);
1697
1698         return ni;
1699 }
1700 EXPORT_SYMBOL(lnet_nid2ni_addref);
1701
1702 struct lnet_ni *
1703 lnet_nid_to_ni_addref(struct lnet_nid *nid)
1704 {
1705         struct lnet_ni *ni;
1706
1707         lnet_net_lock(0);
1708         ni = lnet_nid_to_ni_locked(nid, 0);
1709         if (ni)
1710                 lnet_ni_addref_locked(ni, 0);
1711         lnet_net_unlock(0);
1712
1713         return ni;
1714 }
1715 EXPORT_SYMBOL(lnet_nid_to_ni_addref);
1716
1717 int
1718 lnet_islocalnid(struct lnet_nid *nid)
1719 {
1720         struct lnet_ni  *ni;
1721         int             cpt;
1722
1723         cpt = lnet_net_lock_current();
1724         ni = lnet_nid_to_ni_locked(nid, cpt);
1725         lnet_net_unlock(cpt);
1726
1727         return ni != NULL;
1728 }
1729
1730 int
1731 lnet_count_acceptor_nets(void)
1732 {
1733         /* Return the # of NIs that need the acceptor. */
1734         int              count = 0;
1735         struct lnet_net  *net;
1736         int              cpt;
1737
1738         cpt = lnet_net_lock_current();
1739         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1740                 /* all socklnd type networks should have the acceptor
1741                  * thread started */
1742                 if (net->net_lnd->lnd_accept != NULL)
1743                         count++;
1744         }
1745
1746         lnet_net_unlock(cpt);
1747
1748         return count;
1749 }
1750
1751 struct lnet_ping_buffer *
1752 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1753 {
1754         struct lnet_ping_buffer *pbuf;
1755
1756         LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1757         if (pbuf) {
1758                 pbuf->pb_nnis = nnis;
1759                 pbuf->pb_needs_post = false;
1760                 atomic_set(&pbuf->pb_refcnt, 1);
1761         }
1762
1763         return pbuf;
1764 }
1765
1766 void
1767 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1768 {
1769         LASSERT(atomic_read(&pbuf->pb_refcnt) == 0);
1770         LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1771 }
1772
1773 static struct lnet_ping_buffer *
1774 lnet_ping_target_create(int nnis)
1775 {
1776         struct lnet_ping_buffer *pbuf;
1777
1778         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1779         if (pbuf == NULL) {
1780                 CERROR("Can't allocate ping source [%d]\n", nnis);
1781                 return NULL;
1782         }
1783
1784         pbuf->pb_info.pi_nnis = nnis;
1785         pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1786         pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1787         pbuf->pb_info.pi_features =
1788                 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1789
1790         return pbuf;
1791 }
1792
1793 static inline int
1794 lnet_get_net_ni_count_locked(struct lnet_net *net)
1795 {
1796         struct lnet_ni  *ni;
1797         int             count = 0;
1798
1799         list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1800                 count++;
1801
1802         return count;
1803 }
1804
1805 static inline int
1806 lnet_get_net_ni_count_pre(struct lnet_net *net)
1807 {
1808         struct lnet_ni  *ni;
1809         int             count = 0;
1810
1811         list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1812                 count++;
1813
1814         return count;
1815 }
1816
1817 static inline int
1818 lnet_get_ni_count(void)
1819 {
1820         struct lnet_ni  *ni;
1821         struct lnet_net *net;
1822         int             count = 0;
1823
1824         lnet_net_lock(0);
1825
1826         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1827                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1828                         count++;
1829         }
1830
1831         lnet_net_unlock(0);
1832
1833         return count;
1834 }
1835
1836 void
1837 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1838 {
1839         struct lnet_ni_status *stat;
1840         int nnis;
1841         int i;
1842
1843         __swab32s(&pbuf->pb_info.pi_magic);
1844         __swab32s(&pbuf->pb_info.pi_features);
1845         __swab32s(&pbuf->pb_info.pi_pid);
1846         __swab32s(&pbuf->pb_info.pi_nnis);
1847         nnis = pbuf->pb_info.pi_nnis;
1848         if (nnis > pbuf->pb_nnis)
1849                 nnis = pbuf->pb_nnis;
1850         for (i = 0; i < nnis; i++) {
1851                 stat = &pbuf->pb_info.pi_ni[i];
1852                 __swab64s(&stat->ns_nid);
1853                 __swab32s(&stat->ns_status);
1854         }
1855 }
1856
1857 int
1858 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1859 {
1860         if (!pinfo)
1861                 return -EINVAL;
1862         if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1863                 return -EPROTO;
1864         if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1865                 return -EPROTO;
1866         /* Loopback is guaranteed to be present */
1867         if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1868                 return -ERANGE;
1869         if (LNET_PING_INFO_LONI(pinfo) != LNET_NID_LO_0)
1870                 return -EPROTO;
1871         return 0;
1872 }
1873
1874 static void
1875 lnet_ping_target_destroy(void)
1876 {
1877         struct lnet_net *net;
1878         struct lnet_ni  *ni;
1879
1880         lnet_net_lock(LNET_LOCK_EX);
1881
1882         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1883                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1884                         lnet_ni_lock(ni);
1885                         ni->ni_status = NULL;
1886                         lnet_ni_unlock(ni);
1887                 }
1888         }
1889
1890         lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1891         the_lnet.ln_ping_target = NULL;
1892
1893         lnet_net_unlock(LNET_LOCK_EX);
1894 }
1895
1896 static void
1897 lnet_ping_target_event_handler(struct lnet_event *event)
1898 {
1899         struct lnet_ping_buffer *pbuf = event->md_user_ptr;
1900
1901         if (event->unlinked)
1902                 lnet_ping_buffer_decref(pbuf);
1903 }
1904
1905 static int
1906 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1907                        struct lnet_handle_md *ping_mdh,
1908                        int ni_count, bool set_eq)
1909 {
1910         struct lnet_processid id = {
1911                 .nid = LNET_ANY_NID,
1912                 .pid = LNET_PID_ANY
1913         };
1914         struct lnet_me *me;
1915         struct lnet_md md = { NULL };
1916         int rc;
1917
1918         if (set_eq)
1919                 the_lnet.ln_ping_target_handler =
1920                         lnet_ping_target_event_handler;
1921
1922         *ppbuf = lnet_ping_target_create(ni_count);
1923         if (*ppbuf == NULL) {
1924                 rc = -ENOMEM;
1925                 goto fail_free_eq;
1926         }
1927
1928         /* Ping target ME/MD */
1929         me = LNetMEAttach(LNET_RESERVED_PORTAL, &id,
1930                           LNET_PROTO_PING_MATCHBITS, 0,
1931                           LNET_UNLINK, LNET_INS_AFTER);
1932         if (IS_ERR(me)) {
1933                 rc = PTR_ERR(me);
1934                 CERROR("Can't create ping target ME: %d\n", rc);
1935                 goto fail_decref_ping_buffer;
1936         }
1937
1938         /* initialize md content */
1939         md.start     = &(*ppbuf)->pb_info;
1940         md.length    = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1941         md.threshold = LNET_MD_THRESH_INF;
1942         md.max_size  = 0;
1943         md.options   = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1944                        LNET_MD_MANAGE_REMOTE;
1945         md.handler   = the_lnet.ln_ping_target_handler;
1946         md.user_ptr  = *ppbuf;
1947
1948         rc = LNetMDAttach(me, &md, LNET_RETAIN, ping_mdh);
1949         if (rc != 0) {
1950                 CERROR("Can't attach ping target MD: %d\n", rc);
1951                 goto fail_decref_ping_buffer;
1952         }
1953         lnet_ping_buffer_addref(*ppbuf);
1954
1955         return 0;
1956
1957 fail_decref_ping_buffer:
1958         LASSERT(atomic_read(&(*ppbuf)->pb_refcnt) == 1);
1959         lnet_ping_buffer_decref(*ppbuf);
1960         *ppbuf = NULL;
1961 fail_free_eq:
1962         return rc;
1963 }
1964
1965 static void
1966 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1967                     struct lnet_handle_md *ping_mdh)
1968 {
1969         LNetMDUnlink(*ping_mdh);
1970         LNetInvalidateMDHandle(ping_mdh);
1971
1972         /* NB the MD could be busy; this just starts the unlink */
1973         wait_var_event_warning(&pbuf->pb_refcnt,
1974                                atomic_read(&pbuf->pb_refcnt) <= 1,
1975                                "Still waiting for ping data MD to unlink\n");
1976 }
1977
1978 static void
1979 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1980 {
1981         struct lnet_ni          *ni;
1982         struct lnet_net         *net;
1983         struct lnet_ni_status *ns;
1984         int                     i;
1985         int                     rc;
1986
1987         i = 0;
1988         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1989                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1990                         LASSERT(i < pbuf->pb_nnis);
1991
1992                         ns = &pbuf->pb_info.pi_ni[i];
1993
1994                         if (!nid_is_nid4(&ni->ni_nid))
1995                                 continue;
1996                         ns->ns_nid = lnet_nid_to_nid4(&ni->ni_nid);
1997
1998                         lnet_ni_lock(ni);
1999                         ns->ns_status = lnet_ni_get_status_locked(ni);
2000                         ni->ni_status = ns;
2001                         lnet_ni_unlock(ni);
2002
2003                         i++;
2004                 }
2005         }
2006         /*
2007          * We (ab)use the ns_status of the loopback interface to
2008          * transmit the sequence number. The first interface listed
2009          * must be the loopback interface.
2010          */
2011         rc = lnet_ping_info_validate(&pbuf->pb_info);
2012         if (rc) {
2013                 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
2014                 LBUG();
2015         }
2016         LNET_PING_BUFFER_SEQNO(pbuf) =
2017                 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
2018 }
2019
2020 static void
2021 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
2022                         struct lnet_handle_md ping_mdh)
2023 {
2024         struct lnet_ping_buffer *old_pbuf = NULL;
2025         struct lnet_handle_md old_ping_md;
2026
2027         /* switch the NIs to point to the new ping info created */
2028         lnet_net_lock(LNET_LOCK_EX);
2029
2030         if (!the_lnet.ln_routing)
2031                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
2032         if (!lnet_peer_discovery_disabled)
2033                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
2034
2035         /* Ensure only known feature bits have been set. */
2036         LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
2037         LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
2038
2039         lnet_ping_target_install_locked(pbuf);
2040
2041         if (the_lnet.ln_ping_target) {
2042                 old_pbuf = the_lnet.ln_ping_target;
2043                 old_ping_md = the_lnet.ln_ping_target_md;
2044         }
2045         the_lnet.ln_ping_target_md = ping_mdh;
2046         the_lnet.ln_ping_target = pbuf;
2047
2048         lnet_net_unlock(LNET_LOCK_EX);
2049
2050         if (old_pbuf) {
2051                 /* unlink and free the old ping info */
2052                 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
2053                 lnet_ping_buffer_decref(old_pbuf);
2054         }
2055
2056         lnet_push_update_to_peers(0);
2057 }
2058
2059 static void
2060 lnet_ping_target_fini(void)
2061 {
2062         lnet_ping_md_unlink(the_lnet.ln_ping_target,
2063                             &the_lnet.ln_ping_target_md);
2064
2065         lnet_assert_handler_unused(the_lnet.ln_ping_target_handler);
2066         lnet_ping_target_destroy();
2067 }
2068
2069 /* Resize the push target. */
2070 int lnet_push_target_resize(void)
2071 {
2072         struct lnet_handle_md mdh;
2073         struct lnet_handle_md old_mdh;
2074         struct lnet_ping_buffer *pbuf;
2075         struct lnet_ping_buffer *old_pbuf;
2076         int nnis;
2077         int rc;
2078
2079 again:
2080         nnis = the_lnet.ln_push_target_nnis;
2081         if (nnis <= 0) {
2082                 CDEBUG(D_NET, "Invalid nnis %d\n", nnis);
2083                 return -EINVAL;
2084         }
2085
2086         /* NB: lnet_ping_buffer_alloc() sets pbuf refcount to 1. That ref is
2087          * dropped when we need to resize again (see "old_pbuf" below) or when
2088          * LNet is shutdown (see lnet_push_target_fini())
2089          */
2090         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
2091         if (!pbuf) {
2092                 CDEBUG(D_NET, "Can't allocate pbuf for nnis %d\n", nnis);
2093                 return -ENOMEM;
2094         }
2095
2096         rc = lnet_push_target_post(pbuf, &mdh);
2097         if (rc) {
2098                 CDEBUG(D_NET, "Failed to post push target: %d\n", rc);
2099                 lnet_ping_buffer_decref(pbuf);
2100                 return rc;
2101         }
2102
2103         lnet_net_lock(LNET_LOCK_EX);
2104         old_pbuf = the_lnet.ln_push_target;
2105         old_mdh = the_lnet.ln_push_target_md;
2106         the_lnet.ln_push_target = pbuf;
2107         the_lnet.ln_push_target_md = mdh;
2108         lnet_net_unlock(LNET_LOCK_EX);
2109
2110         if (old_pbuf) {
2111                 LNetMDUnlink(old_mdh);
2112                 /* Drop ref set by lnet_ping_buffer_alloc() */
2113                 lnet_ping_buffer_decref(old_pbuf);
2114         }
2115
2116         /* Received another push or reply that requires a larger buffer */
2117         if (nnis < the_lnet.ln_push_target_nnis)
2118                 goto again;
2119
2120         CDEBUG(D_NET, "nnis %d success\n", nnis);
2121         return 0;
2122 }
2123
2124 int lnet_push_target_post(struct lnet_ping_buffer *pbuf,
2125                           struct lnet_handle_md *mdhp)
2126 {
2127         struct lnet_processid id = { LNET_ANY_NID, LNET_PID_ANY };
2128         struct lnet_md md = { NULL };
2129         struct lnet_me *me;
2130         int rc;
2131
2132         me = LNetMEAttach(LNET_RESERVED_PORTAL, &id,
2133                           LNET_PROTO_PING_MATCHBITS, 0,
2134                           LNET_UNLINK, LNET_INS_AFTER);
2135         if (IS_ERR(me)) {
2136                 rc = PTR_ERR(me);
2137                 CERROR("Can't create push target ME: %d\n", rc);
2138                 return rc;
2139         }
2140
2141         pbuf->pb_needs_post = false;
2142
2143         /* This reference is dropped by lnet_push_target_event_handler() */
2144         lnet_ping_buffer_addref(pbuf);
2145
2146         /* initialize md content */
2147         md.start     = &pbuf->pb_info;
2148         md.length    = LNET_PING_INFO_SIZE(pbuf->pb_nnis);
2149         md.threshold = 1;
2150         md.max_size  = 0;
2151         md.options   = LNET_MD_OP_PUT | LNET_MD_TRUNCATE;
2152         md.user_ptr  = pbuf;
2153         md.handler   = the_lnet.ln_push_target_handler;
2154
2155         rc = LNetMDAttach(me, &md, LNET_UNLINK, mdhp);
2156         if (rc) {
2157                 CERROR("Can't attach push MD: %d\n", rc);
2158                 lnet_ping_buffer_decref(pbuf);
2159                 pbuf->pb_needs_post = true;
2160                 return rc;
2161         }
2162
2163         CDEBUG(D_NET, "posted push target %p\n", pbuf);
2164
2165         return 0;
2166 }
2167
2168 static void lnet_push_target_event_handler(struct lnet_event *ev)
2169 {
2170         struct lnet_ping_buffer *pbuf = ev->md_user_ptr;
2171
2172         CDEBUG(D_NET, "type %d status %d unlinked %d\n", ev->type, ev->status,
2173                ev->unlinked);
2174
2175         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
2176                 lnet_swap_pinginfo(pbuf);
2177
2178         if (ev->type == LNET_EVENT_UNLINK) {
2179                 /* Drop ref added by lnet_push_target_post() */
2180                 lnet_ping_buffer_decref(pbuf);
2181                 return;
2182         }
2183
2184         lnet_peer_push_event(ev);
2185         if (ev->unlinked)
2186                 /* Drop ref added by lnet_push_target_post */
2187                 lnet_ping_buffer_decref(pbuf);
2188 }
2189
2190 /* Initialize the push target. */
2191 static int lnet_push_target_init(void)
2192 {
2193         int rc;
2194
2195         if (the_lnet.ln_push_target)
2196                 return -EALREADY;
2197
2198         the_lnet.ln_push_target_handler =
2199                 lnet_push_target_event_handler;
2200
2201         rc = LNetSetLazyPortal(LNET_RESERVED_PORTAL);
2202         LASSERT(rc == 0);
2203
2204         /* Start at the required minimum, we'll enlarge if required. */
2205         the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
2206
2207         rc = lnet_push_target_resize();
2208
2209         if (rc) {
2210                 LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2211                 the_lnet.ln_push_target_handler = NULL;
2212         }
2213
2214         return rc;
2215 }
2216
2217 /* Clean up the push target. */
2218 static void lnet_push_target_fini(void)
2219 {
2220         if (!the_lnet.ln_push_target)
2221                 return;
2222
2223         /* Unlink and invalidate to prevent new references. */
2224         LNetMDUnlink(the_lnet.ln_push_target_md);
2225         LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
2226
2227         /* Wait for the unlink to complete. */
2228         wait_var_event_warning(&the_lnet.ln_push_target->pb_refcnt,
2229                                atomic_read(&the_lnet.ln_push_target->pb_refcnt) <= 1,
2230                                "Still waiting for ping data MD to unlink\n");
2231
2232         /* Drop ref set by lnet_ping_buffer_alloc() */
2233         lnet_ping_buffer_decref(the_lnet.ln_push_target);
2234         the_lnet.ln_push_target = NULL;
2235         the_lnet.ln_push_target_nnis = 0;
2236
2237         LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2238         lnet_assert_handler_unused(the_lnet.ln_push_target_handler);
2239         the_lnet.ln_push_target_handler = NULL;
2240 }
2241
2242 static int
2243 lnet_ni_tq_credits(struct lnet_ni *ni)
2244 {
2245         int     credits;
2246
2247         LASSERT(ni->ni_ncpts >= 1);
2248
2249         if (ni->ni_ncpts == 1)
2250                 return ni->ni_net->net_tunables.lct_max_tx_credits;
2251
2252         credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2253         credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2254         credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2255
2256         return credits;
2257 }
2258
2259 static void
2260 lnet_ni_unlink_locked(struct lnet_ni *ni)
2261 {
2262         /* move it to zombie list and nobody can find it anymore */
2263         LASSERT(!list_empty(&ni->ni_netlist));
2264         list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2265         lnet_ni_decref_locked(ni, 0);
2266 }
2267
2268 static void
2269 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2270 {
2271         int             i;
2272         int             islo;
2273         struct lnet_ni  *ni;
2274         struct list_head *zombie_list = &net->net_ni_zombie;
2275
2276         /*
2277          * Now wait for the NIs I just nuked to show up on the zombie
2278          * list and shut them down in guaranteed thread context
2279          */
2280         i = 2;
2281         while (!list_empty(zombie_list)) {
2282                 int     *ref;
2283                 int     j;
2284
2285                 ni = list_entry(zombie_list->next,
2286                                 struct lnet_ni, ni_netlist);
2287                 list_del_init(&ni->ni_netlist);
2288                 /* the ni should be in deleting state. If it's not it's
2289                  * a bug */
2290                 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2291                 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2292                         if (*ref == 0)
2293                                 continue;
2294                         /* still busy, add it back to zombie list */
2295                         list_add(&ni->ni_netlist, zombie_list);
2296                         break;
2297                 }
2298
2299                 if (!list_empty(&ni->ni_netlist)) {
2300                         /* Unlock mutex while waiting to allow other
2301                          * threads to read the LNet state and fall through
2302                          * to avoid deadlock
2303                          */
2304                         lnet_net_unlock(LNET_LOCK_EX);
2305                         mutex_unlock(&the_lnet.ln_api_mutex);
2306
2307                         ++i;
2308                         if ((i & (-i)) == i) {
2309                                 CDEBUG(D_WARNING,
2310                                        "Waiting for zombie LNI %s\n",
2311                                        libcfs_nidstr(&ni->ni_nid));
2312                         }
2313                         schedule_timeout_uninterruptible(cfs_time_seconds(1));
2314
2315                         mutex_lock(&the_lnet.ln_api_mutex);
2316                         lnet_net_lock(LNET_LOCK_EX);
2317                         continue;
2318                 }
2319
2320                 lnet_net_unlock(LNET_LOCK_EX);
2321
2322                 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2323
2324                 LASSERT(!in_interrupt());
2325                 /* Holding the LND mutex makes it safe for lnd_shutdown
2326                  * to call module_put(). Module unload cannot finish
2327                  * until lnet_unregister_lnd() completes, and that
2328                  * requires the LND mutex.
2329                  */
2330                 mutex_unlock(&the_lnet.ln_api_mutex);
2331                 mutex_lock(&the_lnet.ln_lnd_mutex);
2332                 (net->net_lnd->lnd_shutdown)(ni);
2333                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2334                 mutex_lock(&the_lnet.ln_api_mutex);
2335
2336                 if (!islo)
2337                         CDEBUG(D_LNI, "Removed LNI %s\n",
2338                               libcfs_nidstr(&ni->ni_nid));
2339
2340                 lnet_ni_free(ni);
2341                 i = 2;
2342                 lnet_net_lock(LNET_LOCK_EX);
2343         }
2344 }
2345
2346 /* shutdown down the NI and release refcount */
2347 static void
2348 lnet_shutdown_lndni(struct lnet_ni *ni)
2349 {
2350         int i;
2351         struct lnet_net *net = ni->ni_net;
2352
2353         lnet_net_lock(LNET_LOCK_EX);
2354         lnet_ni_lock(ni);
2355         ni->ni_state = LNET_NI_STATE_DELETING;
2356         lnet_ni_unlock(ni);
2357         lnet_ni_unlink_locked(ni);
2358         lnet_incr_dlc_seq();
2359         lnet_net_unlock(LNET_LOCK_EX);
2360
2361         /* clear messages for this NI on the lazy portal */
2362         for (i = 0; i < the_lnet.ln_nportals; i++)
2363                 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2364
2365         lnet_net_lock(LNET_LOCK_EX);
2366         lnet_clear_zombies_nis_locked(net);
2367         lnet_net_unlock(LNET_LOCK_EX);
2368 }
2369
2370 static void
2371 lnet_shutdown_lndnet(struct lnet_net *net)
2372 {
2373         struct lnet_ni *ni;
2374
2375         lnet_net_lock(LNET_LOCK_EX);
2376
2377         list_del_init(&net->net_list);
2378
2379         while (!list_empty(&net->net_ni_list)) {
2380                 ni = list_entry(net->net_ni_list.next,
2381                                 struct lnet_ni, ni_netlist);
2382                 lnet_net_unlock(LNET_LOCK_EX);
2383                 lnet_shutdown_lndni(ni);
2384                 lnet_net_lock(LNET_LOCK_EX);
2385         }
2386
2387         lnet_net_unlock(LNET_LOCK_EX);
2388
2389         /* Do peer table cleanup for this net */
2390         lnet_peer_tables_cleanup(net);
2391
2392         lnet_net_free(net);
2393 }
2394
2395 static void
2396 lnet_shutdown_lndnets(void)
2397 {
2398         struct lnet_net *net;
2399         LIST_HEAD(resend);
2400         struct lnet_msg *msg, *tmp;
2401
2402         /* NB called holding the global mutex */
2403
2404         /* All quiet on the API front */
2405         LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING ||
2406                 the_lnet.ln_state == LNET_STATE_STOPPING);
2407         LASSERT(the_lnet.ln_refcount == 0);
2408
2409         lnet_net_lock(LNET_LOCK_EX);
2410         the_lnet.ln_state = LNET_STATE_STOPPING;
2411
2412         /*
2413          * move the nets to the zombie list to avoid them being
2414          * picked up for new work. LONET is also included in the
2415          * Nets that will be moved to the zombie list
2416          */
2417         list_splice_init(&the_lnet.ln_nets, &the_lnet.ln_net_zombie);
2418
2419         /* Drop the cached loopback Net. */
2420         if (the_lnet.ln_loni != NULL) {
2421                 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2422                 the_lnet.ln_loni = NULL;
2423         }
2424         lnet_net_unlock(LNET_LOCK_EX);
2425
2426         /* iterate through the net zombie list and delete each net */
2427         while (!list_empty(&the_lnet.ln_net_zombie)) {
2428                 net = list_entry(the_lnet.ln_net_zombie.next,
2429                                  struct lnet_net, net_list);
2430                 lnet_shutdown_lndnet(net);
2431         }
2432
2433         spin_lock(&the_lnet.ln_msg_resend_lock);
2434         list_splice(&the_lnet.ln_msg_resend, &resend);
2435         spin_unlock(&the_lnet.ln_msg_resend_lock);
2436
2437         list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2438                 list_del_init(&msg->msg_list);
2439                 msg->msg_no_resend = true;
2440                 lnet_finalize(msg, -ECANCELED);
2441         }
2442
2443         lnet_net_lock(LNET_LOCK_EX);
2444         the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2445         lnet_net_unlock(LNET_LOCK_EX);
2446 }
2447
2448 static int
2449 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2450 {
2451         int                     rc = -EINVAL;
2452         struct lnet_tx_queue    *tq;
2453         int                     i;
2454         struct lnet_net         *net = ni->ni_net;
2455
2456         mutex_lock(&the_lnet.ln_lnd_mutex);
2457
2458         if (tun) {
2459                 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2460                 ni->ni_lnd_tunables_set = true;
2461         }
2462
2463         rc = (net->net_lnd->lnd_startup)(ni);
2464
2465         mutex_unlock(&the_lnet.ln_lnd_mutex);
2466
2467         if (rc != 0) {
2468                 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2469                                    rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2470                 goto failed0;
2471         }
2472
2473         lnet_ni_lock(ni);
2474         ni->ni_state = LNET_NI_STATE_ACTIVE;
2475         lnet_ni_unlock(ni);
2476
2477         /* We keep a reference on the loopback net through the loopback NI */
2478         if (net->net_lnd->lnd_type == LOLND) {
2479                 lnet_ni_addref(ni);
2480                 LASSERT(the_lnet.ln_loni == NULL);
2481                 the_lnet.ln_loni = ni;
2482                 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2483                 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2484                 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2485                 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2486                 return 0;
2487         }
2488
2489         if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2490             ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2491                 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2492                                    libcfs_lnd2str(net->net_lnd->lnd_type),
2493                                    ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2494                                         "" : "per-peer ");
2495                 /* shutdown the NI since if we get here then it must've already
2496                  * been started
2497                  */
2498                 lnet_shutdown_lndni(ni);
2499                 return -EINVAL;
2500         }
2501
2502         cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2503                 tq->tq_credits_min =
2504                 tq->tq_credits_max =
2505                 tq->tq_credits = lnet_ni_tq_credits(ni);
2506         }
2507
2508         atomic_set(&ni->ni_tx_credits,
2509                    lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2510         atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2511
2512         CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2513                 libcfs_nidstr(&ni->ni_nid),
2514                 ni->ni_net->net_tunables.lct_peer_tx_credits,
2515                 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2516                 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2517                 ni->ni_net->net_tunables.lct_peer_timeout);
2518
2519         return 0;
2520 failed0:
2521         lnet_ni_free(ni);
2522         return rc;
2523 }
2524
2525 static int
2526 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2527 {
2528         struct lnet_ni *ni;
2529         struct lnet_net *net_l = NULL;
2530         LIST_HEAD(local_ni_list);
2531         int rc;
2532         int ni_count = 0;
2533         __u32 lnd_type;
2534         const struct lnet_lnd  *lnd;
2535         int peer_timeout =
2536                 net->net_tunables.lct_peer_timeout;
2537         int maxtxcredits =
2538                 net->net_tunables.lct_max_tx_credits;
2539         int peerrtrcredits =
2540                 net->net_tunables.lct_peer_rtr_credits;
2541
2542         /*
2543          * make sure that this net is unique. If it isn't then
2544          * we are adding interfaces to an already existing network, and
2545          * 'net' is just a convenient way to pass in the list.
2546          * if it is unique we need to find the LND and load it if
2547          * necessary.
2548          */
2549         if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2550                 lnd_type = LNET_NETTYP(net->net_id);
2551
2552                 mutex_lock(&the_lnet.ln_lnd_mutex);
2553                 lnd = lnet_find_lnd_by_type(lnd_type);
2554
2555                 if (lnd == NULL) {
2556                         mutex_unlock(&the_lnet.ln_lnd_mutex);
2557                         rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2558                         mutex_lock(&the_lnet.ln_lnd_mutex);
2559
2560                         lnd = lnet_find_lnd_by_type(lnd_type);
2561                         if (lnd == NULL) {
2562                                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2563                                 CERROR("Can't load LND %s, module %s, rc=%d\n",
2564                                 libcfs_lnd2str(lnd_type),
2565                                 libcfs_lnd2modname(lnd_type), rc);
2566 #ifndef HAVE_MODULE_LOADING_SUPPORT
2567                                 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2568                                                 "compiled with kernel module "
2569                                                 "loading support.");
2570 #endif
2571                                 rc = -EINVAL;
2572                                 goto failed0;
2573                         }
2574                 }
2575
2576                 net->net_lnd = lnd;
2577
2578                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2579
2580                 net_l = net;
2581         }
2582
2583         /*
2584          * net_l: if the network being added is unique then net_l
2585          *        will point to that network
2586          *        if the network being added is not unique then
2587          *        net_l points to the existing network.
2588          *
2589          * When we enter the loop below, we'll pick NIs off he
2590          * network beign added and start them up, then add them to
2591          * a local ni list. Once we've successfully started all
2592          * the NIs then we join the local NI list (of started up
2593          * networks) with the net_l->net_ni_list, which should
2594          * point to the correct network to add the new ni list to
2595          *
2596          * If any of the new NIs fail to start up, then we want to
2597          * iterate through the local ni list, which should include
2598          * any NIs which were successfully started up, and shut
2599          * them down.
2600          *
2601          * After than we want to delete the network being added,
2602          * to avoid a memory leak.
2603          */
2604         while (!list_empty(&net->net_ni_added)) {
2605                 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2606                                 ni_netlist);
2607                 list_del_init(&ni->ni_netlist);
2608
2609                 /* make sure that the the NI we're about to start
2610                  * up is actually unique. if it's not fail. */
2611                 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2612                                         ni->ni_interface)) {
2613                         rc = -EEXIST;
2614                         goto failed1;
2615                 }
2616
2617                 /* adjust the pointer the parent network, just in case it
2618                  * the net is a duplicate */
2619                 ni->ni_net = net_l;
2620
2621                 rc = lnet_startup_lndni(ni, tun);
2622
2623                 if (rc < 0)
2624                         goto failed1;
2625
2626                 lnet_ni_addref(ni);
2627                 list_add_tail(&ni->ni_netlist, &local_ni_list);
2628
2629                 ni_count++;
2630         }
2631
2632         lnet_net_lock(LNET_LOCK_EX);
2633         list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2634         lnet_incr_dlc_seq();
2635         lnet_net_unlock(LNET_LOCK_EX);
2636
2637         /* if the network is not unique then we don't want to keep
2638          * it around after we're done. Free it. Otherwise add that
2639          * net to the global the_lnet.ln_nets */
2640         if (net_l != net && net_l != NULL) {
2641                 /*
2642                  * TODO - note. currently the tunables can not be updated
2643                  * once added
2644                  */
2645                 lnet_net_free(net);
2646         } else {
2647                 /*
2648                  * restore tunables after it has been overwitten by the
2649                  * lnd
2650                  */
2651                 if (peer_timeout != -1)
2652                         net->net_tunables.lct_peer_timeout = peer_timeout;
2653                 if (maxtxcredits != -1)
2654                         net->net_tunables.lct_max_tx_credits = maxtxcredits;
2655                 if (peerrtrcredits != -1)
2656                         net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2657
2658                 lnet_net_lock(LNET_LOCK_EX);
2659                 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2660                 lnet_net_unlock(LNET_LOCK_EX);
2661         }
2662
2663         return ni_count;
2664
2665 failed1:
2666         /*
2667          * shutdown the new NIs that are being started up
2668          * free the NET being started
2669          */
2670         while (!list_empty(&local_ni_list)) {
2671                 ni = list_entry(local_ni_list.next, struct lnet_ni,
2672                                 ni_netlist);
2673
2674                 lnet_shutdown_lndni(ni);
2675         }
2676
2677 failed0:
2678         lnet_net_free(net);
2679
2680         return rc;
2681 }
2682
2683 static int
2684 lnet_startup_lndnets(struct list_head *netlist)
2685 {
2686         struct lnet_net         *net;
2687         int                     rc;
2688         int                     ni_count = 0;
2689
2690         /*
2691          * Change to running state before bringing up the LNDs. This
2692          * allows lnet_shutdown_lndnets() to assert that we've passed
2693          * through here.
2694          */
2695         lnet_net_lock(LNET_LOCK_EX);
2696         the_lnet.ln_state = LNET_STATE_RUNNING;
2697         lnet_net_unlock(LNET_LOCK_EX);
2698
2699         while (!list_empty(netlist)) {
2700                 net = list_entry(netlist->next, struct lnet_net, net_list);
2701                 list_del_init(&net->net_list);
2702
2703                 rc = lnet_startup_lndnet(net, NULL);
2704
2705                 if (rc < 0)
2706                         goto failed;
2707
2708                 ni_count += rc;
2709         }
2710
2711         return ni_count;
2712 failed:
2713         lnet_shutdown_lndnets();
2714
2715         return rc;
2716 }
2717
2718 static int lnet_genl_parse_list(struct sk_buff *msg,
2719                                 const struct ln_key_list *data[], u16 idx)
2720 {
2721         const struct ln_key_list *list = data[idx];
2722         const struct ln_key_props *props;
2723         struct nlattr *node;
2724         u16 count;
2725
2726         if (!list)
2727                 return 0;
2728
2729         if (!list->lkl_maxattr)
2730                 return -ERANGE;
2731
2732         props = list->lkl_list;
2733         if (!props)
2734                 return -EINVAL;
2735
2736         node = nla_nest_start(msg, LN_SCALAR_ATTR_LIST);
2737         if (!node)
2738                 return -ENOBUFS;
2739
2740         for (count = 1; count <= list->lkl_maxattr; count++) {
2741                 struct nlattr *key = nla_nest_start(msg, count);
2742
2743                 if (count == 1)
2744                         nla_put_u16(msg, LN_SCALAR_ATTR_LIST_SIZE,
2745                                     list->lkl_maxattr);
2746
2747                 nla_put_u16(msg, LN_SCALAR_ATTR_INDEX, count);
2748                 if (props[count].lkp_value)
2749                         nla_put_string(msg, LN_SCALAR_ATTR_VALUE,
2750                                        props[count].lkp_value);
2751                 if (props[count].lkp_key_format)
2752                         nla_put_u16(msg, LN_SCALAR_ATTR_KEY_FORMAT,
2753                                     props[count].lkp_key_format);
2754                 nla_put_u16(msg, LN_SCALAR_ATTR_NLA_TYPE,
2755                             props[count].lkp_data_type);
2756                 if (props[count].lkp_data_type == NLA_NESTED) {
2757                         int rc;
2758
2759                         rc = lnet_genl_parse_list(msg, data, ++idx);
2760                         if (rc < 0)
2761                                 return rc;
2762                         idx = rc;
2763                 }
2764
2765                 nla_nest_end(msg, key);
2766         }
2767
2768         nla_nest_end(msg, node);
2769         return idx;
2770 }
2771
2772 int lnet_genl_send_scalar_list(struct sk_buff *msg, u32 portid, u32 seq,
2773                                const struct genl_family *family, int flags,
2774                                u8 cmd, const struct ln_key_list *data[])
2775 {
2776         int rc = 0;
2777         void *hdr;
2778
2779         if (!data[0])
2780                 return -EINVAL;
2781
2782         hdr = genlmsg_put(msg, portid, seq, family, flags, cmd);
2783         if (!hdr)
2784                 GOTO(canceled, rc = -EMSGSIZE);
2785
2786         rc = lnet_genl_parse_list(msg, data, 0);
2787         if (rc < 0)
2788                 GOTO(canceled, rc);
2789
2790         genlmsg_end(msg, hdr);
2791 canceled:
2792         if (rc < 0)
2793                 genlmsg_cancel(msg, hdr);
2794         return rc > 0 ? 0 : rc;
2795 }
2796 EXPORT_SYMBOL(lnet_genl_send_scalar_list);
2797
2798 /**
2799  * Initialize LNet library.
2800  *
2801  * Automatically called at module loading time. Caller has to call
2802  * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2803  * latter returned 0. It must be called exactly once.
2804  *
2805  * \retval 0 on success
2806  * \retval -ve on failures.
2807  */
2808 int lnet_lib_init(void)
2809 {
2810         int rc;
2811
2812         lnet_assert_wire_constants();
2813
2814         /* refer to global cfs_cpt_table for now */
2815         the_lnet.ln_cpt_table = cfs_cpt_tab;
2816         the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_tab);
2817
2818         LASSERT(the_lnet.ln_cpt_number > 0);
2819         if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2820                 /* we are under risk of consuming all lh_cookie */
2821                 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2822                        "please change setting of CPT-table and retry\n",
2823                        the_lnet.ln_cpt_number, LNET_CPT_MAX);
2824                 return -E2BIG;
2825         }
2826
2827         while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2828                 the_lnet.ln_cpt_bits++;
2829
2830         rc = lnet_create_locks();
2831         if (rc != 0) {
2832                 CERROR("Can't create LNet global locks: %d\n", rc);
2833                 return rc;
2834         }
2835
2836         the_lnet.ln_refcount = 0;
2837         INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2838         INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2839
2840         /* The hash table size is the number of bits it takes to express the set
2841          * ln_num_routes, minus 1 (better to under estimate than over so we
2842          * don't waste memory). */
2843         if (rnet_htable_size <= 0)
2844                 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2845         else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2846                 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2847         the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2848                                            order_base_2(rnet_htable_size) - 1);
2849
2850         /* All LNDs apart from the LOLND are in separate modules.  They
2851          * register themselves when their module loads, and unregister
2852          * themselves when their module is unloaded. */
2853         lnet_register_lnd(&the_lolnd);
2854         return 0;
2855 }
2856
2857 /**
2858  * Finalize LNet library.
2859  *
2860  * \pre lnet_lib_init() called with success.
2861  * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2862  *
2863  * As this happens at module-unload, all lnds must already be unloaded,
2864  * so they must already be unregistered.
2865  */
2866 void lnet_lib_exit(void)
2867 {
2868         int i;
2869
2870         LASSERT(the_lnet.ln_refcount == 0);
2871         lnet_unregister_lnd(&the_lolnd);
2872         for (i = 0; i < NUM_LNDS; i++)
2873                 LASSERT(!the_lnet.ln_lnds[i]);
2874         lnet_destroy_locks();
2875 }
2876
2877 /**
2878  * Set LNet PID and start LNet interfaces, routing, and forwarding.
2879  *
2880  * Users must call this function at least once before any other functions.
2881  * For each successful call there must be a corresponding call to
2882  * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2883  * ignored.
2884  *
2885  * The PID used by LNet may be different from the one requested.
2886  * See LNetGetId().
2887  *
2888  * \param requested_pid PID requested by the caller.
2889  *
2890  * \return >= 0 on success, and < 0 error code on failures.
2891  */
2892 int
2893 LNetNIInit(lnet_pid_t requested_pid)
2894 {
2895         int                     im_a_router = 0;
2896         int                     rc;
2897         int                     ni_count;
2898         struct lnet_ping_buffer *pbuf;
2899         struct lnet_handle_md   ping_mdh;
2900         LIST_HEAD(net_head);
2901         struct lnet_net         *net;
2902
2903         mutex_lock(&the_lnet.ln_api_mutex);
2904
2905         CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2906
2907         if (the_lnet.ln_state == LNET_STATE_STOPPING) {
2908                 mutex_unlock(&the_lnet.ln_api_mutex);
2909                 return -ESHUTDOWN;
2910         }
2911
2912         if (the_lnet.ln_refcount > 0) {
2913                 rc = the_lnet.ln_refcount++;
2914                 mutex_unlock(&the_lnet.ln_api_mutex);
2915                 return rc;
2916         }
2917
2918         rc = lnet_prepare(requested_pid);
2919         if (rc != 0) {
2920                 mutex_unlock(&the_lnet.ln_api_mutex);
2921                 return rc;
2922         }
2923
2924         /* create a network for Loopback network */
2925         net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2926         if (net == NULL) {
2927                 rc = -ENOMEM;
2928                 goto err_empty_list;
2929         }
2930
2931         /* Add in the loopback NI */
2932         if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2933                 rc = -ENOMEM;
2934                 goto err_empty_list;
2935         }
2936
2937         if (use_tcp_bonding)
2938                 CWARN("use_tcp_bonding has been removed. Use Multi-Rail and Dynamic Discovery instead, see LU-13641\n");
2939
2940         /* If LNet is being initialized via DLC it is possible
2941          * that the user requests not to load module parameters (ones which
2942          * are supported by DLC) on initialization.  Therefore, make sure not
2943          * to load networks, routes and forwarding from module parameters
2944          * in this case.  On cleanup in case of failure only clean up
2945          * routes if it has been loaded */
2946         if (!the_lnet.ln_nis_from_mod_params) {
2947                 rc = lnet_parse_networks(&net_head, lnet_get_networks());
2948                 if (rc < 0)
2949                         goto err_empty_list;
2950         }
2951
2952         ni_count = lnet_startup_lndnets(&net_head);
2953         if (ni_count < 0) {
2954                 rc = ni_count;
2955                 goto err_empty_list;
2956         }
2957
2958         if (!the_lnet.ln_nis_from_mod_params) {
2959                 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2960                 if (rc != 0)
2961                         goto err_shutdown_lndnis;
2962
2963                 rc = lnet_rtrpools_alloc(im_a_router);
2964                 if (rc != 0)
2965                         goto err_destroy_routes;
2966         }
2967
2968         rc = lnet_acceptor_start();
2969         if (rc != 0)
2970                 goto err_destroy_routes;
2971
2972         the_lnet.ln_refcount = 1;
2973         /* Now I may use my own API functions... */
2974
2975         rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2976         if (rc != 0)
2977                 goto err_acceptor_stop;
2978
2979         lnet_ping_target_update(pbuf, ping_mdh);
2980
2981         the_lnet.ln_mt_handler = lnet_mt_event_handler;
2982
2983         rc = lnet_push_target_init();
2984         if (rc != 0)
2985                 goto err_stop_ping;
2986
2987         rc = lnet_peer_discovery_start();
2988         if (rc != 0)
2989                 goto err_destroy_push_target;
2990
2991         rc = lnet_monitor_thr_start();
2992         if (rc != 0)
2993                 goto err_stop_discovery_thr;
2994
2995         lnet_fault_init();
2996         lnet_router_debugfs_init();
2997
2998         mutex_unlock(&the_lnet.ln_api_mutex);
2999
3000         complete_all(&the_lnet.ln_started);
3001
3002         /* wait for all routers to start */
3003         lnet_wait_router_start();
3004
3005         return 0;
3006
3007 err_stop_discovery_thr:
3008         lnet_peer_discovery_stop();
3009 err_destroy_push_target:
3010         lnet_push_target_fini();
3011 err_stop_ping:
3012         lnet_ping_target_fini();
3013 err_acceptor_stop:
3014         the_lnet.ln_refcount = 0;
3015         lnet_acceptor_stop();
3016 err_destroy_routes:
3017         if (!the_lnet.ln_nis_from_mod_params)
3018                 lnet_destroy_routes();
3019 err_shutdown_lndnis:
3020         lnet_shutdown_lndnets();
3021 err_empty_list:
3022         lnet_unprepare();
3023         LASSERT(rc < 0);
3024         mutex_unlock(&the_lnet.ln_api_mutex);
3025         while (!list_empty(&net_head)) {
3026                 struct lnet_net *net;
3027
3028                 net = list_entry(net_head.next, struct lnet_net, net_list);
3029                 list_del_init(&net->net_list);
3030                 lnet_net_free(net);
3031         }
3032         return rc;
3033 }
3034 EXPORT_SYMBOL(LNetNIInit);
3035
3036 /**
3037  * Stop LNet interfaces, routing, and forwarding.
3038  *
3039  * Users must call this function once for each successful call to LNetNIInit().
3040  * Once the LNetNIFini() operation has been started, the results of pending
3041  * API operations are undefined.
3042  *
3043  * \return always 0 for current implementation.
3044  */
3045 int
3046 LNetNIFini(void)
3047 {
3048         mutex_lock(&the_lnet.ln_api_mutex);
3049
3050         LASSERT(the_lnet.ln_refcount > 0);
3051
3052         if (the_lnet.ln_refcount != 1) {
3053                 the_lnet.ln_refcount--;
3054         } else {
3055                 LASSERT(!the_lnet.ln_niinit_self);
3056
3057                 lnet_net_lock(LNET_LOCK_EX);
3058                 the_lnet.ln_state = LNET_STATE_STOPPING;
3059                 lnet_net_unlock(LNET_LOCK_EX);
3060
3061                 lnet_fault_fini();
3062
3063                 lnet_router_debugfs_fini();
3064                 lnet_monitor_thr_stop();
3065                 lnet_peer_discovery_stop();
3066                 lnet_push_target_fini();
3067                 lnet_ping_target_fini();
3068
3069                 /* Teardown fns that use my own API functions BEFORE here */
3070                 the_lnet.ln_refcount = 0;
3071
3072                 lnet_acceptor_stop();
3073                 lnet_destroy_routes();
3074                 lnet_shutdown_lndnets();
3075                 lnet_unprepare();
3076         }
3077
3078         mutex_unlock(&the_lnet.ln_api_mutex);
3079         return 0;
3080 }
3081 EXPORT_SYMBOL(LNetNIFini);
3082
3083 /**
3084  * Grabs the ni data from the ni structure and fills the out
3085  * parameters
3086  *
3087  * \param[in] ni network        interface structure
3088  * \param[out] cfg_ni           NI config information
3089  * \param[out] tun              network and LND tunables
3090  */
3091 static void
3092 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
3093                    struct lnet_ioctl_config_lnd_tunables *tun,
3094                    struct lnet_ioctl_element_stats *stats,
3095                    __u32 tun_size)
3096 {
3097         size_t min_size = 0;
3098         int i;
3099
3100         if (!ni || !cfg_ni || !tun || !nid_is_nid4(&ni->ni_nid))
3101                 return;
3102
3103         if (ni->ni_interface != NULL) {
3104                 strncpy(cfg_ni->lic_ni_intf,
3105                         ni->ni_interface,
3106                         sizeof(cfg_ni->lic_ni_intf));
3107         }
3108
3109         cfg_ni->lic_nid = lnet_nid_to_nid4(&ni->ni_nid);
3110         cfg_ni->lic_status = lnet_ni_get_status_locked(ni);
3111         cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
3112
3113         memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
3114
3115         if (stats) {
3116                 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
3117                                                        LNET_STATS_TYPE_SEND);
3118                 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
3119                                                        LNET_STATS_TYPE_RECV);
3120                 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
3121                                                        LNET_STATS_TYPE_DROP);
3122         }
3123
3124         /*
3125          * tun->lt_tun will always be present, but in order to be
3126          * backwards compatible, we need to deal with the cases when
3127          * tun->lt_tun is smaller than what the kernel has, because it
3128          * comes from an older version of a userspace program, then we'll
3129          * need to copy as much information as we have available space.
3130          */
3131         min_size = tun_size - sizeof(tun->lt_cmn);
3132         memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
3133
3134         /* copy over the cpts */
3135         if (ni->ni_ncpts == LNET_CPT_NUMBER &&
3136             ni->ni_cpts == NULL)  {
3137                 for (i = 0; i < ni->ni_ncpts; i++)
3138                         cfg_ni->lic_cpts[i] = i;
3139         } else {
3140                 for (i = 0;
3141                      ni->ni_cpts != NULL && i < ni->ni_ncpts &&
3142                      i < LNET_MAX_SHOW_NUM_CPT;
3143                      i++)
3144                         cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
3145         }
3146         cfg_ni->lic_ncpts = ni->ni_ncpts;
3147 }
3148
3149 /**
3150  * NOTE: This is a legacy function left in the code to be backwards
3151  * compatible with older userspace programs. It should eventually be
3152  * removed.
3153  *
3154  * Grabs the ni data from the ni structure and fills the out
3155  * parameters
3156  *
3157  * \param[in] ni network        interface structure
3158  * \param[out] config           config information
3159  */
3160 static void
3161 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
3162                          struct lnet_ioctl_config_data *config)
3163 {
3164         struct lnet_ioctl_net_config *net_config;
3165         struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
3166         size_t min_size, tunable_size = 0;
3167         int i;
3168
3169         if (!ni || !config || !nid_is_nid4(&ni->ni_nid))
3170                 return;
3171
3172         net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
3173         if (!net_config)
3174                 return;
3175
3176         if (!ni->ni_interface)
3177                 return;
3178
3179         strncpy(net_config->ni_interface,
3180                 ni->ni_interface,
3181                 sizeof(net_config->ni_interface));
3182
3183         config->cfg_nid = lnet_nid_to_nid4(&ni->ni_nid);
3184         config->cfg_config_u.cfg_net.net_peer_timeout =
3185                 ni->ni_net->net_tunables.lct_peer_timeout;
3186         config->cfg_config_u.cfg_net.net_max_tx_credits =
3187                 ni->ni_net->net_tunables.lct_max_tx_credits;
3188         config->cfg_config_u.cfg_net.net_peer_tx_credits =
3189                 ni->ni_net->net_tunables.lct_peer_tx_credits;
3190         config->cfg_config_u.cfg_net.net_peer_rtr_credits =
3191                 ni->ni_net->net_tunables.lct_peer_rtr_credits;
3192
3193         net_config->ni_status = lnet_ni_get_status_locked(ni);
3194
3195         if (ni->ni_cpts) {
3196                 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
3197
3198                 for (i = 0; i < num_cpts; i++)
3199                         net_config->ni_cpts[i] = ni->ni_cpts[i];
3200
3201                 config->cfg_ncpts = num_cpts;
3202         }
3203
3204         /*
3205          * See if user land tools sent in a newer and larger version
3206          * of struct lnet_tunables than what the kernel uses.
3207          */
3208         min_size = sizeof(*config) + sizeof(*net_config);
3209
3210         if (config->cfg_hdr.ioc_len > min_size)
3211                 tunable_size = config->cfg_hdr.ioc_len - min_size;
3212
3213         /* Don't copy too much data to user space */
3214         min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
3215         lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
3216
3217         if (lnd_cfg && min_size) {
3218                 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
3219                 config->cfg_config_u.cfg_net.net_interface_count = 1;
3220
3221                 /* Tell user land that kernel side has less data */
3222                 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
3223                         min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
3224                         config->cfg_hdr.ioc_len -= min_size;
3225                 }
3226         }
3227 }
3228
3229 struct lnet_ni *
3230 lnet_get_ni_idx_locked(int idx)
3231 {
3232         struct lnet_ni          *ni;
3233         struct lnet_net         *net;
3234
3235         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3236                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3237                         if (idx-- == 0)
3238                                 return ni;
3239                 }
3240         }
3241
3242         return NULL;
3243 }
3244
3245 int lnet_get_net_healthv_locked(struct lnet_net *net)
3246 {
3247         struct lnet_ni *ni;
3248         int best_healthv = 0;
3249         int healthv, ni_fatal;
3250
3251         list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3252                 healthv = atomic_read(&ni->ni_healthv);
3253                 ni_fatal = atomic_read(&ni->ni_fatal_error_on);
3254                 if (!ni_fatal && healthv > best_healthv)
3255                         best_healthv = healthv;
3256         }
3257
3258         return best_healthv;
3259 }
3260
3261 struct lnet_ni *
3262 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
3263 {
3264         struct lnet_ni          *ni;
3265         struct lnet_net         *net = mynet;
3266
3267         /*
3268          * It is possible that the net has been cleaned out while there is
3269          * a message being sent. This function accessed the net without
3270          * checking if the list is empty
3271          */
3272         if (prev == NULL) {
3273                 if (net == NULL)
3274                         net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
3275                                         net_list);
3276                 if (list_empty(&net->net_ni_list))
3277                         return NULL;
3278                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3279                                 ni_netlist);
3280
3281                 return ni;
3282         }
3283
3284         if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
3285                 /* if you reached the end of the ni list and the net is
3286                  * specified, then there are no more nis in that net */
3287                 if (net != NULL)
3288                         return NULL;
3289
3290                 /* we reached the end of this net ni list. move to the
3291                  * next net */
3292                 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
3293                         /* no more nets and no more NIs. */
3294                         return NULL;
3295
3296                 /* get the next net */
3297                 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
3298                                  net_list);
3299                 if (list_empty(&net->net_ni_list))
3300                         return NULL;
3301                 /* get the ni on it */
3302                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3303                                 ni_netlist);
3304
3305                 return ni;
3306         }
3307
3308         if (list_empty(&prev->ni_netlist))
3309                 return NULL;
3310
3311         /* there are more nis left */
3312         ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
3313
3314         return ni;
3315 }
3316
3317 int
3318 lnet_get_net_config(struct lnet_ioctl_config_data *config)
3319 {
3320         struct lnet_ni *ni;
3321         int cpt;
3322         int rc = -ENOENT;
3323         int idx = config->cfg_count;
3324
3325         cpt = lnet_net_lock_current();
3326
3327         ni = lnet_get_ni_idx_locked(idx);
3328
3329         if (ni != NULL) {
3330                 rc = 0;
3331                 lnet_ni_lock(ni);
3332                 lnet_fill_ni_info_legacy(ni, config);
3333                 lnet_ni_unlock(ni);
3334         }
3335
3336         lnet_net_unlock(cpt);
3337         return rc;
3338 }
3339
3340 int
3341 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3342                    struct lnet_ioctl_config_lnd_tunables *tun,
3343                    struct lnet_ioctl_element_stats *stats,
3344                    __u32 tun_size)
3345 {
3346         struct lnet_ni          *ni;
3347         int                     cpt;
3348         int                     rc = -ENOENT;
3349
3350         if (!cfg_ni || !tun || !stats)
3351                 return -EINVAL;
3352
3353         cpt = lnet_net_lock_current();
3354
3355         ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3356
3357         if (ni) {
3358                 rc = 0;
3359                 lnet_ni_lock(ni);
3360                 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3361                 lnet_ni_unlock(ni);
3362         }
3363
3364         lnet_net_unlock(cpt);
3365         return rc;
3366 }
3367
3368 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3369 {
3370         struct lnet_ni *ni;
3371         int cpt;
3372         int rc = -ENOENT;
3373
3374         if (!msg_stats)
3375                 return -EINVAL;
3376
3377         cpt = lnet_net_lock_current();
3378
3379         ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3380
3381         if (ni) {
3382                 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3383                 rc = 0;
3384         }
3385
3386         lnet_net_unlock(cpt);
3387
3388         return rc;
3389 }
3390
3391 static int lnet_add_net_common(struct lnet_net *net,
3392                                struct lnet_ioctl_config_lnd_tunables *tun)
3393 {
3394         struct lnet_handle_md ping_mdh;
3395         struct lnet_ping_buffer *pbuf;
3396         struct lnet_remotenet *rnet;
3397         struct lnet_ni *ni;
3398         int net_ni_count;
3399         __u32 net_id;
3400         int rc;
3401
3402         lnet_net_lock(LNET_LOCK_EX);
3403         rnet = lnet_find_rnet_locked(net->net_id);
3404         lnet_net_unlock(LNET_LOCK_EX);
3405         /*
3406          * make sure that the net added doesn't invalidate the current
3407          * configuration LNet is keeping
3408          */
3409         if (rnet) {
3410                 CERROR("Adding net %s will invalidate routing configuration\n",
3411                        libcfs_net2str(net->net_id));
3412                 lnet_net_free(net);
3413                 return -EUSERS;
3414         }
3415
3416         /*
3417          * make sure you calculate the correct number of slots in the ping
3418          * buffer. Since the ping info is a flattened list of all the NIs,
3419          * we should allocate enough slots to accomodate the number of NIs
3420          * which will be added.
3421          *
3422          * since ni hasn't been configured yet, use
3423          * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3424          */
3425         net_ni_count = lnet_get_net_ni_count_pre(net);
3426
3427         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3428                                     net_ni_count + lnet_get_ni_count(),
3429                                     false);
3430         if (rc < 0) {
3431                 lnet_net_free(net);
3432                 return rc;
3433         }
3434
3435         if (tun)
3436                 memcpy(&net->net_tunables,
3437                        &tun->lt_cmn, sizeof(net->net_tunables));
3438         else
3439                 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3440
3441         net_id = net->net_id;
3442
3443         rc = lnet_startup_lndnet(net,
3444                                  (tun) ? &tun->lt_tun : NULL);
3445         if (rc < 0)
3446                 goto failed;
3447
3448         lnet_net_lock(LNET_LOCK_EX);
3449         net = lnet_get_net_locked(net_id);
3450         LASSERT(net);
3451
3452         /* apply the UDSPs */
3453         rc = lnet_udsp_apply_policies_on_net(net);
3454         if (rc)
3455                 CERROR("Failed to apply UDSPs on local net %s\n",
3456                        libcfs_net2str(net->net_id));
3457
3458         /* At this point we lost track of which NI was just added, so we
3459          * just re-apply the policies on all of the NIs on this net
3460          */
3461         list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3462                 rc = lnet_udsp_apply_policies_on_ni(ni);
3463                 if (rc)
3464                         CERROR("Failed to apply UDSPs on ni %s\n",
3465                                libcfs_nidstr(&ni->ni_nid));
3466         }
3467         lnet_net_unlock(LNET_LOCK_EX);
3468
3469         /*
3470          * Start the acceptor thread if this is the first network
3471          * being added that requires the thread.
3472          */
3473         if (net->net_lnd->lnd_accept) {
3474                 rc = lnet_acceptor_start();
3475                 if (rc < 0) {
3476                         /* shutdown the net that we just started */
3477                         CERROR("Failed to start up acceptor thread\n");
3478                         lnet_shutdown_lndnet(net);
3479                         goto failed;
3480                 }
3481         }
3482
3483         lnet_net_lock(LNET_LOCK_EX);
3484         lnet_peer_net_added(net);
3485         lnet_net_unlock(LNET_LOCK_EX);
3486
3487         lnet_ping_target_update(pbuf, ping_mdh);
3488
3489         return 0;
3490
3491 failed:
3492         lnet_ping_md_unlink(pbuf, &ping_mdh);
3493         lnet_ping_buffer_decref(pbuf);
3494         return rc;
3495 }
3496
3497 static void
3498 lnet_set_tune_defaults(struct lnet_ioctl_config_lnd_tunables *tun)
3499 {
3500         if (tun) {
3501                 if (!tun->lt_cmn.lct_peer_timeout)
3502                         tun->lt_cmn.lct_peer_timeout = DEFAULT_PEER_TIMEOUT;
3503                 if (!tun->lt_cmn.lct_peer_tx_credits)
3504                         tun->lt_cmn.lct_peer_tx_credits = DEFAULT_PEER_CREDITS;
3505                 if (!tun->lt_cmn.lct_max_tx_credits)
3506                         tun->lt_cmn.lct_max_tx_credits = DEFAULT_CREDITS;
3507         }
3508 }
3509
3510 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3511                                       struct lnet_ioctl_config_lnd_tunables *tun)
3512 {
3513         struct lnet_net *net;
3514         const char *nets;
3515         int rc;
3516         LIST_HEAD(net_head);
3517
3518         rc = lnet_parse_ip2nets(&nets, ip2nets);
3519         if (rc < 0)
3520                 return rc;
3521
3522         rc = lnet_parse_networks(&net_head, nets);
3523         if (rc < 0)
3524                 return rc;
3525
3526         lnet_set_tune_defaults(tun);
3527
3528         mutex_lock(&the_lnet.ln_api_mutex);
3529         if (the_lnet.ln_state != LNET_STATE_RUNNING) {
3530                 rc = -ESHUTDOWN;
3531                 goto out;
3532         }
3533         while (!list_empty(&net_head)) {
3534                 net = list_entry(net_head.next, struct lnet_net, net_list);
3535                 list_del_init(&net->net_list);
3536                 rc = lnet_add_net_common(net, tun);
3537                 if (rc < 0)
3538                         goto out;
3539         }
3540
3541 out:
3542         mutex_unlock(&the_lnet.ln_api_mutex);
3543
3544         while (!list_empty(&net_head)) {
3545                 net = list_entry(net_head.next, struct lnet_net, net_list);
3546                 list_del_init(&net->net_list);
3547                 lnet_net_free(net);
3548         }
3549         return rc;
3550 }
3551
3552 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3553 {
3554         struct lnet_net *net;
3555         struct lnet_ni *ni;
3556         struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3557         int rc, i;
3558         __u32 net_id, lnd_type;
3559
3560         /* get the tunables if they are available */
3561         if (conf->lic_cfg_hdr.ioc_len >=
3562             sizeof(*conf) + sizeof(*tun))
3563                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3564                         conf->lic_bulk;
3565
3566         /* handle legacy ip2nets from DLC */
3567         if (conf->lic_legacy_ip2nets[0] != '\0')
3568                 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3569                                                   tun);
3570
3571         net_id = LNET_NIDNET(conf->lic_nid);
3572         lnd_type = LNET_NETTYP(net_id);
3573
3574         if (!libcfs_isknown_lnd(lnd_type)) {
3575                 CERROR("No valid net and lnd information provided\n");
3576                 return -EINVAL;
3577         }
3578
3579         net = lnet_net_alloc(net_id, NULL);
3580         if (!net)
3581                 return -ENOMEM;
3582
3583         for (i = 0; i < conf->lic_ncpts; i++) {
3584                 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3585                         return -EINVAL;
3586         }
3587
3588         ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3589                                        conf->lic_ni_intf);
3590         if (!ni)
3591                 return -ENOMEM;
3592
3593         lnet_set_tune_defaults(tun);
3594
3595         mutex_lock(&the_lnet.ln_api_mutex);
3596         if (the_lnet.ln_state != LNET_STATE_RUNNING)
3597                 rc = -ESHUTDOWN;
3598         else
3599                 rc = lnet_add_net_common(net, tun);
3600
3601         mutex_unlock(&the_lnet.ln_api_mutex);
3602
3603         return rc;
3604 }
3605
3606 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3607 {
3608         struct lnet_net  *net;
3609         struct lnet_ni *ni;
3610         __u32 net_id = LNET_NIDNET(conf->lic_nid);
3611         struct lnet_ping_buffer *pbuf;
3612         struct lnet_handle_md  ping_mdh;
3613         int               rc;
3614         int               net_count;
3615         __u32             addr;
3616
3617         /* don't allow userspace to shutdown the LOLND */
3618         if (LNET_NETTYP(net_id) == LOLND)
3619                 return -EINVAL;
3620
3621         mutex_lock(&the_lnet.ln_api_mutex);
3622         if (the_lnet.ln_state != LNET_STATE_RUNNING) {
3623                 rc = -ESHUTDOWN;
3624                 goto unlock_api_mutex;
3625         }
3626
3627         lnet_net_lock(0);
3628
3629         net = lnet_get_net_locked(net_id);
3630         if (!net) {
3631                 CERROR("net %s not found\n",
3632                        libcfs_net2str(net_id));
3633                 rc = -ENOENT;
3634                 goto unlock_net;
3635         }
3636
3637         addr = LNET_NIDADDR(conf->lic_nid);
3638         if (addr == 0) {
3639                 /* remove the entire net */
3640                 net_count = lnet_get_net_ni_count_locked(net);
3641
3642                 lnet_net_unlock(0);
3643
3644                 /* create and link a new ping info, before removing the old one */
3645                 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3646                                         lnet_get_ni_count() - net_count,
3647                                         false);
3648                 if (rc != 0)
3649                         goto unlock_api_mutex;
3650
3651                 lnet_shutdown_lndnet(net);
3652
3653                 lnet_acceptor_stop();
3654
3655                 lnet_ping_target_update(pbuf, ping_mdh);
3656
3657                 goto unlock_api_mutex;
3658         }
3659
3660         ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3661         if (!ni) {
3662                 CERROR("nid %s not found\n",
3663                        libcfs_nid2str(conf->lic_nid));
3664                 rc = -ENOENT;
3665                 goto unlock_net;
3666         }
3667
3668         net_count = lnet_get_net_ni_count_locked(net);
3669
3670         lnet_net_unlock(0);
3671
3672         /* create and link a new ping info, before removing the old one */
3673         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3674                                   lnet_get_ni_count() - 1, false);
3675         if (rc != 0)
3676                 goto unlock_api_mutex;
3677
3678         lnet_shutdown_lndni(ni);
3679
3680         lnet_acceptor_stop();
3681
3682         lnet_ping_target_update(pbuf, ping_mdh);
3683
3684         /* check if the net is empty and remove it if it is */
3685         if (net_count == 1)
3686                 lnet_shutdown_lndnet(net);
3687
3688         goto unlock_api_mutex;
3689
3690 unlock_net:
3691         lnet_net_unlock(0);
3692 unlock_api_mutex:
3693         mutex_unlock(&the_lnet.ln_api_mutex);
3694
3695         return rc;
3696 }
3697
3698 /*
3699  * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3700  * They are only expected to be called for unique networks.
3701  * That can be as a result of older DLC library
3702  * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3703  */
3704 int
3705 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3706 {
3707         struct lnet_net *net;
3708         LIST_HEAD(net_head);
3709         int rc;
3710         struct lnet_ioctl_config_lnd_tunables tun;
3711         const char *nets = conf->cfg_config_u.cfg_net.net_intf;
3712
3713         /* Create a net/ni structures for the network string */
3714         rc = lnet_parse_networks(&net_head, nets);
3715         if (rc <= 0)
3716                 return rc == 0 ? -EINVAL : rc;
3717
3718         mutex_lock(&the_lnet.ln_api_mutex);
3719         if (the_lnet.ln_state != LNET_STATE_RUNNING) {
3720                 rc = -ESHUTDOWN;
3721                 goto out_unlock_clean;
3722         }
3723
3724         if (rc > 1) {
3725                 rc = -EINVAL; /* only add one network per call */
3726                 goto out_unlock_clean;
3727         }
3728
3729         net = list_entry(net_head.next, struct lnet_net, net_list);
3730         list_del_init(&net->net_list);
3731
3732         LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3733
3734         memset(&tun, 0, sizeof(tun));
3735
3736         tun.lt_cmn.lct_peer_timeout =
3737           (!conf->cfg_config_u.cfg_net.net_peer_timeout) ? DEFAULT_PEER_TIMEOUT :
3738                 conf->cfg_config_u.cfg_net.net_peer_timeout;
3739         tun.lt_cmn.lct_peer_tx_credits =
3740           (!conf->cfg_config_u.cfg_net.net_peer_tx_credits) ? DEFAULT_PEER_CREDITS :
3741                 conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3742         tun.lt_cmn.lct_peer_rtr_credits =
3743           conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3744         tun.lt_cmn.lct_max_tx_credits =
3745           (!conf->cfg_config_u.cfg_net.net_max_tx_credits) ? DEFAULT_CREDITS :
3746                 conf->cfg_config_u.cfg_net.net_max_tx_credits;
3747
3748         rc = lnet_add_net_common(net, &tun);
3749
3750 out_unlock_clean:
3751         mutex_unlock(&the_lnet.ln_api_mutex);
3752         while (!list_empty(&net_head)) {
3753                 /* net_head list is empty in success case */
3754                 net = list_entry(net_head.next, struct lnet_net, net_list);
3755                 list_del_init(&net->net_list);
3756                 lnet_net_free(net);
3757         }
3758         return rc;
3759 }
3760
3761 int
3762 lnet_dyn_del_net(__u32 net_id)
3763 {
3764         struct lnet_net  *net;
3765         struct lnet_ping_buffer *pbuf;
3766         struct lnet_handle_md ping_mdh;
3767         int               rc;
3768         int               net_ni_count;
3769
3770         /* don't allow userspace to shutdown the LOLND */
3771         if (LNET_NETTYP(net_id) == LOLND)
3772                 return -EINVAL;
3773
3774         mutex_lock(&the_lnet.ln_api_mutex);
3775         if (the_lnet.ln_state != LNET_STATE_RUNNING) {
3776                 rc = -ESHUTDOWN;
3777                 goto out;
3778         }
3779
3780         lnet_net_lock(0);
3781
3782         net = lnet_get_net_locked(net_id);
3783         if (net == NULL) {
3784                 lnet_net_unlock(0);
3785                 rc = -EINVAL;
3786                 goto out;
3787         }
3788
3789         net_ni_count = lnet_get_net_ni_count_locked(net);
3790
3791         lnet_net_unlock(0);
3792
3793         /* create and link a new ping info, before removing the old one */
3794         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3795                                     lnet_get_ni_count() - net_ni_count, false);
3796         if (rc != 0)
3797                 goto out;
3798
3799         lnet_shutdown_lndnet(net);
3800
3801         lnet_acceptor_stop();
3802
3803         lnet_ping_target_update(pbuf, ping_mdh);
3804
3805 out:
3806         mutex_unlock(&the_lnet.ln_api_mutex);
3807
3808         return rc;
3809 }
3810
3811 void lnet_incr_dlc_seq(void)
3812 {
3813         atomic_inc(&lnet_dlc_seq_no);
3814 }
3815
3816 __u32 lnet_get_dlc_seq_locked(void)
3817 {
3818         return atomic_read(&lnet_dlc_seq_no);
3819 }
3820
3821 static void
3822 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3823 {
3824         struct lnet_net *net;
3825         struct lnet_ni *ni;
3826
3827         lnet_net_lock(LNET_LOCK_EX);
3828         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3829                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3830                         if (all || (nid_is_nid4(&ni->ni_nid) &&
3831                                     lnet_nid_to_nid4(&ni->ni_nid) == nid)) {
3832                                 atomic_set(&ni->ni_healthv, value);
3833                                 if (list_empty(&ni->ni_recovery) &&
3834                                     value < LNET_MAX_HEALTH_VALUE) {
3835                                         CERROR("manually adding local NI %s to recovery\n",
3836                                                libcfs_nidstr(&ni->ni_nid));
3837                                         list_add_tail(&ni->ni_recovery,
3838                                                       &the_lnet.ln_mt_localNIRecovq);
3839                                         lnet_ni_addref_locked(ni, 0);
3840                                 }
3841                                 if (!all) {
3842                                         lnet_net_unlock(LNET_LOCK_EX);
3843                                         return;
3844                                 }
3845                         }
3846                 }
3847         }
3848         lnet_net_unlock(LNET_LOCK_EX);
3849 }
3850
3851 static void
3852 lnet_ni_set_conns_per_peer(lnet_nid_t nid, int value, bool all)
3853 {
3854         struct lnet_net *net;
3855         struct lnet_ni *ni;
3856
3857         lnet_net_lock(LNET_LOCK_EX);
3858         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3859                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3860                         if (lnet_nid_to_nid4(&ni->ni_nid) != nid && !all)
3861                                 continue;
3862                         if (LNET_NETTYP(net->net_id) == SOCKLND)
3863                                 ni->ni_lnd_tunables.lnd_tun_u.lnd_sock.lnd_conns_per_peer = value;
3864                         else if (LNET_NETTYP(net->net_id) == O2IBLND)
3865                                 ni->ni_lnd_tunables.lnd_tun_u.lnd_o2ib.lnd_conns_per_peer = value;
3866                         if (!all) {
3867                                 lnet_net_unlock(LNET_LOCK_EX);
3868                                 return;
3869                         }
3870                 }
3871         }
3872         lnet_net_unlock(LNET_LOCK_EX);
3873 }
3874
3875 static int
3876 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3877 {
3878         int cpt, rc = 0;
3879         struct lnet_ni *ni;
3880         lnet_nid_t nid = stats->hlni_nid;
3881
3882         cpt = lnet_net_lock_current();
3883         ni = lnet_nid2ni_locked(nid, cpt);
3884
3885         if (!ni) {
3886                 rc = -ENOENT;
3887                 goto unlock;
3888         }
3889
3890         stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3891         stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3892         stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3893         stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3894         stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3895         stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3896         stats->hlni_fatal_error = atomic_read(&ni->ni_fatal_error_on);
3897         stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3898         stats->hlni_ping_count = ni->ni_ping_count;
3899         stats->hlni_next_ping = ni->ni_next_ping;
3900
3901 unlock:
3902         lnet_net_unlock(cpt);
3903
3904         return rc;
3905 }
3906
3907 static int
3908 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3909 {
3910         struct lnet_ni *ni;
3911         int i = 0;
3912
3913         lnet_net_lock(LNET_LOCK_EX);
3914         list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3915                 if (!nid_is_nid4(&ni->ni_nid))
3916                         continue;
3917                 list->rlst_nid_array[i] = lnet_nid_to_nid4(&ni->ni_nid);
3918                 i++;
3919                 if (i >= LNET_MAX_SHOW_NUM_NID)
3920                         break;
3921         }
3922         lnet_net_unlock(LNET_LOCK_EX);
3923         list->rlst_num_nids = i;
3924
3925         return 0;
3926 }
3927
3928 static int
3929 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3930 {
3931         struct lnet_peer_ni *lpni;
3932         int i = 0;
3933
3934         lnet_net_lock(LNET_LOCK_EX);
3935         list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3936                 list->rlst_nid_array[i] = lnet_nid_to_nid4(&lpni->lpni_nid);
3937                 i++;
3938                 if (i >= LNET_MAX_SHOW_NUM_NID)
3939                         break;
3940         }
3941         lnet_net_unlock(LNET_LOCK_EX);
3942         list->rlst_num_nids = i;
3943
3944         return 0;
3945 }
3946
3947 /**
3948  * LNet ioctl handler.
3949  *
3950  */
3951 int
3952 LNetCtl(unsigned int cmd, void *arg)
3953 {
3954         struct libcfs_ioctl_data *data = arg;
3955         struct lnet_ioctl_config_data *config;
3956         struct lnet_process_id    id4 = {};
3957         struct lnet_processid     id = {};
3958         struct lnet_ni           *ni;
3959         struct lnet_nid           nid;
3960         int                       rc;
3961
3962         BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3963                      sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3964
3965         switch (cmd) {
3966         case IOC_LIBCFS_GET_NI:
3967                 rc = LNetGetId(data->ioc_count, &id);
3968                 data->ioc_nid = lnet_nid_to_nid4(&id.nid);
3969                 return rc;
3970
3971         case IOC_LIBCFS_FAIL_NID:
3972                 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3973
3974         case IOC_LIBCFS_ADD_ROUTE: {
3975                 /* default router sensitivity to 1 */
3976                 unsigned int sensitivity = 1;
3977                 config = arg;
3978
3979                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3980                         return -EINVAL;
3981
3982                 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3983                         sensitivity =
3984                           config->cfg_config_u.cfg_route.rtr_sensitivity;
3985                 }
3986
3987                 lnet_nid4_to_nid(config->cfg_nid, &nid);
3988                 mutex_lock(&the_lnet.ln_api_mutex);
3989                 rc = lnet_add_route(config->cfg_net,
3990                                     config->cfg_config_u.cfg_route.rtr_hop,
3991                                     &nid,
3992                                     config->cfg_config_u.cfg_route.
3993                                         rtr_priority, sensitivity);
3994                 mutex_unlock(&the_lnet.ln_api_mutex);
3995                 return rc;
3996         }
3997
3998         case IOC_LIBCFS_DEL_ROUTE:
3999                 config = arg;
4000
4001                 if (config->cfg_hdr.ioc_len < sizeof(*config))
4002                         return -EINVAL;
4003
4004                 lnet_nid4_to_nid(config->cfg_nid, &nid);
4005                 mutex_lock(&the_lnet.ln_api_mutex);
4006                 rc = lnet_del_route(config->cfg_net, &nid);
4007                 mutex_unlock(&the_lnet.ln_api_mutex);
4008                 return rc;
4009
4010         case IOC_LIBCFS_GET_ROUTE:
4011                 config = arg;
4012
4013                 if (config->cfg_hdr.ioc_len < sizeof(*config))
4014                         return -EINVAL;
4015
4016                 mutex_lock(&the_lnet.ln_api_mutex);
4017                 rc = lnet_get_route(config->cfg_count,
4018                                     &config->cfg_net,
4019                                     &config->cfg_config_u.cfg_route.rtr_hop,
4020                                     &config->cfg_nid,
4021                                     &config->cfg_config_u.cfg_route.rtr_flags,
4022                                     &config->cfg_config_u.cfg_route.
4023                                         rtr_priority,
4024                                     &config->cfg_config_u.cfg_route.
4025                                         rtr_sensitivity);
4026                 mutex_unlock(&the_lnet.ln_api_mutex);
4027                 return rc;
4028
4029         case IOC_LIBCFS_GET_LOCAL_NI: {
4030                 struct lnet_ioctl_config_ni *cfg_ni;
4031                 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
4032                 struct lnet_ioctl_element_stats *stats;
4033                 __u32 tun_size;
4034
4035                 cfg_ni = arg;
4036
4037                 /* get the tunables if they are available */
4038                 if (cfg_ni->lic_cfg_hdr.ioc_len <
4039                     sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
4040                         return -EINVAL;
4041
4042                 stats = (struct lnet_ioctl_element_stats *)
4043                         cfg_ni->lic_bulk;
4044                 tun = (struct lnet_ioctl_config_lnd_tunables *)
4045                                 (cfg_ni->lic_bulk + sizeof(*stats));
4046
4047                 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
4048                         sizeof(*stats);
4049
4050                 mutex_lock(&the_lnet.ln_api_mutex);
4051                 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
4052                 mutex_unlock(&the_lnet.ln_api_mutex);
4053                 return rc;
4054         }
4055
4056         case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
4057                 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
4058
4059                 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
4060                         return -EINVAL;
4061
4062                 mutex_lock(&the_lnet.ln_api_mutex);
4063                 rc = lnet_get_ni_stats(msg_stats);
4064                 mutex_unlock(&the_lnet.ln_api_mutex);
4065
4066                 return rc;
4067         }
4068
4069         case IOC_LIBCFS_GET_NET: {
4070                 size_t total = sizeof(*config) +
4071                                sizeof(struct lnet_ioctl_net_config);
4072                 config = arg;
4073
4074                 if (config->cfg_hdr.ioc_len < total)
4075                         return -EINVAL;
4076
4077                 mutex_lock(&the_lnet.ln_api_mutex);
4078                 rc = lnet_get_net_config(config);
4079                 mutex_unlock(&the_lnet.ln_api_mutex);
4080                 return rc;
4081         }
4082
4083         case IOC_LIBCFS_GET_LNET_STATS:
4084         {
4085                 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
4086
4087                 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
4088                         return -EINVAL;
4089
4090                 mutex_lock(&the_lnet.ln_api_mutex);
4091                 rc = lnet_counters_get(&lnet_stats->st_cntrs);
4092                 mutex_unlock(&the_lnet.ln_api_mutex);
4093                 return rc;
4094         }
4095
4096         case IOC_LIBCFS_RESET_LNET_STATS:
4097         {
4098                 mutex_lock(&the_lnet.ln_api_mutex);
4099                 lnet_counters_reset();
4100                 mutex_unlock(&the_lnet.ln_api_mutex);
4101                 return 0;
4102         }
4103
4104         case IOC_LIBCFS_CONFIG_RTR:
4105                 config = arg;
4106
4107                 if (config->cfg_hdr.ioc_len < sizeof(*config))
4108                         return -EINVAL;
4109
4110                 mutex_lock(&the_lnet.ln_api_mutex);
4111                 if (config->cfg_config_u.cfg_buffers.buf_enable) {
4112                         rc = lnet_rtrpools_enable();
4113                         mutex_unlock(&the_lnet.ln_api_mutex);
4114                         return rc;
4115                 }
4116                 lnet_rtrpools_disable();
4117                 mutex_unlock(&the_lnet.ln_api_mutex);
4118                 return 0;
4119
4120         case IOC_LIBCFS_ADD_BUF:
4121                 config = arg;
4122
4123                 if (config->cfg_hdr.ioc_len < sizeof(*config))
4124                         return -EINVAL;
4125
4126                 mutex_lock(&the_lnet.ln_api_mutex);
4127                 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
4128                                                 buf_tiny,
4129                                           config->cfg_config_u.cfg_buffers.
4130                                                 buf_small,
4131                                           config->cfg_config_u.cfg_buffers.
4132                                                 buf_large);
4133                 mutex_unlock(&the_lnet.ln_api_mutex);
4134                 return rc;
4135
4136         case IOC_LIBCFS_SET_NUMA_RANGE: {
4137                 struct lnet_ioctl_set_value *numa;
4138                 numa = arg;
4139                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
4140                         return -EINVAL;
4141                 lnet_net_lock(LNET_LOCK_EX);
4142                 lnet_numa_range = numa->sv_value;
4143                 lnet_net_unlock(LNET_LOCK_EX);
4144                 return 0;
4145         }
4146
4147         case IOC_LIBCFS_GET_NUMA_RANGE: {
4148                 struct lnet_ioctl_set_value *numa;
4149                 numa = arg;
4150                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
4151                         return -EINVAL;
4152                 numa->sv_value = lnet_numa_range;
4153                 return 0;
4154         }
4155
4156         case IOC_LIBCFS_GET_BUF: {
4157                 struct lnet_ioctl_pool_cfg *pool_cfg;
4158                 size_t total = sizeof(*config) + sizeof(*pool_cfg);
4159
4160                 config = arg;
4161
4162                 if (config->cfg_hdr.ioc_len < total)
4163                         return -EINVAL;
4164
4165                 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
4166
4167                 mutex_lock(&the_lnet.ln_api_mutex);
4168                 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
4169                 mutex_unlock(&the_lnet.ln_api_mutex);
4170                 return rc;
4171         }
4172
4173         case IOC_LIBCFS_GET_LOCAL_HSTATS: {
4174                 struct lnet_ioctl_local_ni_hstats *stats = arg;
4175
4176                 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
4177                         return -EINVAL;
4178
4179                 mutex_lock(&the_lnet.ln_api_mutex);
4180                 rc = lnet_get_local_ni_hstats(stats);
4181                 mutex_unlock(&the_lnet.ln_api_mutex);
4182
4183                 return rc;
4184         }
4185
4186         case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
4187                 struct lnet_ioctl_recovery_list *list = arg;
4188                 if (list->rlst_hdr.ioc_len < sizeof(*list))
4189                         return -EINVAL;
4190
4191                 mutex_lock(&the_lnet.ln_api_mutex);
4192                 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
4193                         rc = lnet_get_local_ni_recovery_list(list);
4194                 else
4195                         rc = lnet_get_peer_ni_recovery_list(list);
4196                 mutex_unlock(&the_lnet.ln_api_mutex);
4197                 return rc;
4198         }
4199
4200         case IOC_LIBCFS_ADD_PEER_NI: {
4201                 struct lnet_ioctl_peer_cfg *cfg = arg;
4202
4203                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4204                         return -EINVAL;
4205
4206                 mutex_lock(&the_lnet.ln_api_mutex);
4207                 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
4208                                       cfg->prcfg_cfg_nid,
4209                                       cfg->prcfg_mr, false);
4210                 mutex_unlock(&the_lnet.ln_api_mutex);
4211                 return rc;
4212         }
4213
4214         case IOC_LIBCFS_DEL_PEER_NI: {
4215                 struct lnet_ioctl_peer_cfg *cfg = arg;
4216
4217                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4218                         return -EINVAL;
4219
4220                 mutex_lock(&the_lnet.ln_api_mutex);
4221                 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
4222                                       cfg->prcfg_cfg_nid);
4223                 mutex_unlock(&the_lnet.ln_api_mutex);
4224                 return rc;
4225         }
4226
4227         case IOC_LIBCFS_GET_PEER_INFO: {
4228                 struct lnet_ioctl_peer *peer_info = arg;
4229
4230                 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
4231                         return -EINVAL;
4232
4233                 mutex_lock(&the_lnet.ln_api_mutex);
4234                 rc = lnet_get_peer_ni_info(
4235                    peer_info->pr_count,
4236                    &peer_info->pr_nid,
4237                    peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
4238                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
4239                    &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
4240                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
4241                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
4242                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
4243                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
4244                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
4245                 mutex_unlock(&the_lnet.ln_api_mutex);
4246                 return rc;
4247         }
4248
4249         case IOC_LIBCFS_GET_PEER_NI: {
4250                 struct lnet_ioctl_peer_cfg *cfg = arg;
4251
4252                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4253                         return -EINVAL;
4254
4255                 mutex_lock(&the_lnet.ln_api_mutex);
4256                 rc = lnet_get_peer_info(cfg,
4257                                         (void __user *)cfg->prcfg_bulk);
4258                 mutex_unlock(&the_lnet.ln_api_mutex);
4259                 return rc;
4260         }
4261
4262         case IOC_LIBCFS_GET_PEER_LIST: {
4263                 struct lnet_ioctl_peer_cfg *cfg = arg;
4264
4265                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4266                         return -EINVAL;
4267
4268                 mutex_lock(&the_lnet.ln_api_mutex);
4269                 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
4270                                 (struct lnet_process_id __user *)cfg->prcfg_bulk);
4271                 mutex_unlock(&the_lnet.ln_api_mutex);
4272                 return rc;
4273         }
4274
4275         case IOC_LIBCFS_SET_HEALHV: {
4276                 struct lnet_ioctl_reset_health_cfg *cfg = arg;
4277                 int value;
4278                 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
4279                         return -EINVAL;
4280                 if (cfg->rh_value < 0 ||
4281                     cfg->rh_value > LNET_MAX_HEALTH_VALUE)
4282                         value = LNET_MAX_HEALTH_VALUE;
4283                 else
4284                         value = cfg->rh_value;
4285                 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
4286                        value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
4287                        "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
4288                 mutex_lock(&the_lnet.ln_api_mutex);
4289                 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
4290                         lnet_ni_set_healthv(cfg->rh_nid, value,
4291                                              cfg->rh_all);
4292                 else
4293                         lnet_peer_ni_set_healthv(cfg->rh_nid, value,
4294                                                   cfg->rh_all);
4295                 mutex_unlock(&the_lnet.ln_api_mutex);
4296                 return 0;
4297         }
4298
4299         case IOC_LIBCFS_SET_CONNS_PER_PEER: {
4300                 struct lnet_ioctl_reset_conns_per_peer_cfg *cfg = arg;
4301                 int value;
4302
4303                 if (cfg->rcpp_hdr.ioc_len < sizeof(*cfg))
4304                         return -EINVAL;
4305                 if (cfg->rcpp_value < 0)
4306                         value = 1;
4307                 else
4308                         value = cfg->rcpp_value;
4309                 CDEBUG(D_NET,
4310                        "Setting conns_per_peer to %d for %s. all = %d\n",
4311                        value, libcfs_nid2str(cfg->rcpp_nid), cfg->rcpp_all);
4312                 mutex_lock(&the_lnet.ln_api_mutex);
4313                 lnet_ni_set_conns_per_peer(cfg->rcpp_nid, value, cfg->rcpp_all);
4314                 mutex_unlock(&the_lnet.ln_api_mutex);
4315                 return 0;
4316         }
4317
4318         case IOC_LIBCFS_NOTIFY_ROUTER: {
4319                 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
4320
4321                 /* The deadline passed in by the user should be some time in
4322                  * seconds in the future since the UNIX epoch. We have to map
4323                  * that deadline to the wall clock.
4324                  */
4325                 deadline += ktime_get_seconds();
4326                 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
4327                                    deadline);
4328         }
4329
4330         case IOC_LIBCFS_LNET_DIST:
4331                 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
4332                 if (rc < 0 && rc != -EHOSTUNREACH)
4333                         return rc;
4334
4335                 data->ioc_u32[0] = rc;
4336                 return 0;
4337
4338         case IOC_LIBCFS_TESTPROTOCOMPAT:
4339                 the_lnet.ln_testprotocompat = data->ioc_flags;
4340                 return 0;
4341
4342         case IOC_LIBCFS_LNET_FAULT:
4343                 return lnet_fault_ctl(data->ioc_flags, data);
4344
4345         case IOC_LIBCFS_PING: {
4346                 signed long timeout;
4347
4348                 id4.nid = data->ioc_nid;
4349                 id4.pid = data->ioc_u32[0];
4350
4351                 /* If timeout is negative then set default of 3 minutes */
4352                 if (((s32)data->ioc_u32[1] <= 0) ||
4353                     data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
4354                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
4355                 else
4356                         timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
4357
4358                 rc = lnet_ping(id4, &LNET_ANY_NID, timeout, data->ioc_pbuf1,
4359                                data->ioc_plen1 / sizeof(struct lnet_process_id));
4360
4361                 if (rc < 0)
4362                         return rc;
4363
4364                 data->ioc_count = rc;
4365                 return 0;
4366         }
4367
4368         case IOC_LIBCFS_PING_PEER: {
4369                 struct lnet_ioctl_ping_data *ping = arg;
4370                 struct lnet_nid src_nid = LNET_ANY_NID;
4371                 struct lnet_peer *lp;
4372                 signed long timeout;
4373
4374                 /* Check if the supplied ping data supports source nid
4375                  * NB: This check is sufficient if lnet_ioctl_ping_data has
4376                  * additional fields added, but if they are re-ordered or
4377                  * fields removed then this will break. It is expected that
4378                  * these ioctls will be replaced with netlink implementation, so
4379                  * it is probably not worth coming up with a more robust version
4380                  * compatibility scheme.
4381                  */
4382                 if (ping->ping_hdr.ioc_len >= sizeof(struct lnet_ioctl_ping_data))
4383                         lnet_nid4_to_nid(ping->ping_src, &src_nid);
4384
4385                 /* If timeout is negative then set default of 3 minutes */
4386                 if (((s32)ping->op_param) <= 0 ||
4387                     ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
4388                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
4389                 else
4390                         timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
4391
4392                 rc = lnet_ping(ping->ping_id, &src_nid, timeout,
4393                                ping->ping_buf,
4394                                ping->ping_count);
4395                 if (rc < 0)
4396                         return rc;
4397
4398                 mutex_lock(&the_lnet.ln_api_mutex);
4399                 lp = lnet_find_peer4(ping->ping_id.nid);
4400                 if (lp) {
4401                         ping->ping_id.nid =
4402                                 lnet_nid_to_nid4(&lp->lp_primary_nid);
4403                         ping->mr_info = lnet_peer_is_multi_rail(lp);
4404                         lnet_peer_decref_locked(lp);
4405                 }
4406                 mutex_unlock(&the_lnet.ln_api_mutex);
4407
4408                 ping->ping_count = rc;
4409                 return 0;
4410         }
4411
4412         case IOC_LIBCFS_DISCOVER: {
4413                 struct lnet_ioctl_ping_data *discover = arg;
4414                 struct lnet_peer *lp;
4415
4416                 rc = lnet_discover(discover->ping_id, discover->op_param,
4417                                    discover->ping_buf,
4418                                    discover->ping_count);
4419                 if (rc < 0)
4420                         return rc;
4421
4422                 mutex_lock(&the_lnet.ln_api_mutex);
4423                 lp = lnet_find_peer4(discover->ping_id.nid);
4424                 if (lp) {
4425                         discover->ping_id.nid =
4426                                 lnet_nid_to_nid4(&lp->lp_primary_nid);
4427                         discover->mr_info = lnet_peer_is_multi_rail(lp);
4428                         lnet_peer_decref_locked(lp);
4429                 }
4430                 mutex_unlock(&the_lnet.ln_api_mutex);
4431
4432                 discover->ping_count = rc;
4433                 return 0;
4434         }
4435
4436         case IOC_LIBCFS_ADD_UDSP: {
4437                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4438                 __u32 bulk_size = ioc_udsp->iou_hdr.ioc_len;
4439
4440                 mutex_lock(&the_lnet.ln_api_mutex);
4441                 rc = lnet_udsp_demarshal_add(arg, bulk_size);
4442                 if (!rc) {
4443                         rc = lnet_udsp_apply_policies(NULL, false);
4444                         CDEBUG(D_NET, "policy application returned %d\n", rc);
4445                         rc = 0;
4446                 }
4447                 mutex_unlock(&the_lnet.ln_api_mutex);
4448
4449                 return rc;
4450         }
4451
4452         case IOC_LIBCFS_DEL_UDSP: {
4453                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4454                 int idx = ioc_udsp->iou_idx;
4455
4456                 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4457                         return -EINVAL;
4458
4459                 mutex_lock(&the_lnet.ln_api_mutex);
4460                 rc = lnet_udsp_del_policy(idx);
4461                 if (!rc) {
4462                         rc = lnet_udsp_apply_policies(NULL, false);
4463                         CDEBUG(D_NET, "policy re-application returned %d\n",
4464                                rc);
4465                         rc = 0;
4466                 }
4467                 mutex_unlock(&the_lnet.ln_api_mutex);
4468
4469                 return rc;
4470         }
4471
4472         case IOC_LIBCFS_GET_UDSP_SIZE: {
4473                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4474                 struct lnet_udsp *udsp;
4475
4476                 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4477                         return -EINVAL;
4478
4479                 rc = 0;
4480
4481                 mutex_lock(&the_lnet.ln_api_mutex);
4482                 udsp = lnet_udsp_get_policy(ioc_udsp->iou_idx);
4483                 if (!udsp) {
4484                         rc = -ENOENT;
4485                 } else {
4486                         /* coming in iou_idx will hold the idx of the udsp
4487                          * to get the size of. going out the iou_idx will
4488                          * hold the size of the UDSP found at the passed
4489                          * in index.
4490                          */
4491                         ioc_udsp->iou_idx = lnet_get_udsp_size(udsp);
4492                         if (ioc_udsp->iou_idx < 0)
4493                                 rc = -EINVAL;
4494                 }
4495                 mutex_unlock(&the_lnet.ln_api_mutex);
4496
4497                 return rc;
4498         }
4499
4500         case IOC_LIBCFS_GET_UDSP: {
4501                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4502                 struct lnet_udsp *udsp;
4503
4504                 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4505                         return -EINVAL;
4506
4507                 rc = 0;
4508
4509                 mutex_lock(&the_lnet.ln_api_mutex);
4510                 udsp = lnet_udsp_get_policy(ioc_udsp->iou_idx);
4511                 if (!udsp)
4512                         rc = -ENOENT;
4513                 else
4514                         rc = lnet_udsp_marshal(udsp, ioc_udsp);
4515                 mutex_unlock(&the_lnet.ln_api_mutex);
4516
4517                 return rc;
4518         }
4519
4520         case IOC_LIBCFS_GET_CONST_UDSP_INFO: {
4521                 struct lnet_ioctl_construct_udsp_info *info = arg;
4522
4523                 if (info->cud_hdr.ioc_len < sizeof(*info))
4524                         return -EINVAL;
4525
4526                 CDEBUG(D_NET, "GET_UDSP_INFO for %s\n",
4527                        libcfs_nid2str(info->cud_nid));
4528
4529                 mutex_lock(&the_lnet.ln_api_mutex);
4530                 lnet_udsp_get_construct_info(info);
4531                 mutex_unlock(&the_lnet.ln_api_mutex);
4532
4533                 return 0;
4534         }
4535
4536         default:
4537                 ni = lnet_net2ni_addref(data->ioc_net);
4538                 if (ni == NULL)
4539                         return -EINVAL;
4540
4541                 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
4542                         rc = -EINVAL;
4543                 else
4544                         rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
4545
4546                 lnet_ni_decref(ni);
4547                 return rc;
4548         }
4549         /* not reached */
4550 }
4551 EXPORT_SYMBOL(LNetCtl);
4552
4553 void LNetDebugPeer(struct lnet_processid *id)
4554 {
4555         lnet_debug_peer(lnet_nid_to_nid4(&id->nid));
4556 }
4557 EXPORT_SYMBOL(LNetDebugPeer);
4558
4559 /**
4560  * Determine if the specified peer \a nid is on the local node.
4561  *
4562  * \param nid   peer nid to check
4563  *
4564  * \retval true         If peer NID is on the local node.
4565  * \retval false        If peer NID is not on the local node.
4566  */
4567 bool LNetIsPeerLocal(lnet_nid_t nid)
4568 {
4569         struct lnet_net *net;
4570         struct lnet_ni *ni;
4571         int cpt;
4572
4573         cpt = lnet_net_lock_current();
4574         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4575                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4576                         if (lnet_nid_to_nid4(&ni->ni_nid) == nid) {
4577                                 lnet_net_unlock(cpt);
4578                                 return true;
4579                         }
4580                 }
4581         }
4582         lnet_net_unlock(cpt);
4583
4584         return false;
4585 }
4586 EXPORT_SYMBOL(LNetIsPeerLocal);
4587
4588 /**
4589  * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4590  * Note that all interfaces share a same PID, as requested by LNetNIInit().
4591  *
4592  * \param index Index of the interface to look up.
4593  * \param id On successful return, this location will hold the
4594  * struct lnet_process_id ID of the interface.
4595  *
4596  * \retval 0 If an interface exists at \a index.
4597  * \retval -ENOENT If no interface has been found.
4598  */
4599 int
4600 LNetGetId(unsigned int index, struct lnet_processid *id)
4601 {
4602         struct lnet_ni   *ni;
4603         struct lnet_net  *net;
4604         int               cpt;
4605         int               rc = -ENOENT;
4606
4607         LASSERT(the_lnet.ln_refcount > 0);
4608
4609         cpt = lnet_net_lock_current();
4610
4611         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4612                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4613                         if (!nid_is_nid4(&ni->ni_nid))
4614                                 /* FIXME this needs to be handled */
4615                                 continue;
4616                         if (index-- != 0)
4617                                 continue;
4618
4619                         id->nid = ni->ni_nid;
4620                         id->pid = the_lnet.ln_pid;
4621                         rc = 0;
4622                         break;
4623                 }
4624         }
4625
4626         lnet_net_unlock(cpt);
4627         return rc;
4628 }
4629 EXPORT_SYMBOL(LNetGetId);
4630
4631 struct ping_data {
4632         int rc;
4633         int replied;
4634         struct lnet_handle_md mdh;
4635         struct completion completion;
4636 };
4637
4638 static void
4639 lnet_ping_event_handler(struct lnet_event *event)
4640 {
4641         struct ping_data *pd = event->md_user_ptr;
4642
4643         CDEBUG(D_NET, "ping event (%d %d)%s\n",
4644                event->type, event->status,
4645                event->unlinked ? " unlinked" : "");
4646
4647         if (event->status) {
4648                 if (!pd->rc)
4649                         pd->rc = event->status;
4650         } else if (event->type == LNET_EVENT_REPLY) {
4651                 pd->replied = 1;
4652                 pd->rc = event->mlength;
4653         }
4654         if (event->unlinked)
4655                 complete(&pd->completion);
4656 }
4657
4658 static int lnet_ping(struct lnet_process_id id, struct lnet_nid *src_nid,
4659                      signed long timeout, struct lnet_process_id __user *ids,
4660                      int n_ids)
4661 {
4662         struct lnet_md md = { NULL };
4663         struct ping_data pd = { 0 };
4664         struct lnet_ping_buffer *pbuf;
4665         struct lnet_process_id tmpid;
4666         int i;
4667         int nob;
4668         int rc;
4669         int rc2;
4670
4671         /* n_ids limit is arbitrary */
4672         if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4673                 return -EINVAL;
4674
4675         /*
4676          * if the user buffer has more space than the lnet_interfaces_max
4677          * then only fill it up to lnet_interfaces_max
4678          */
4679         if (n_ids > lnet_interfaces_max)
4680                 n_ids = lnet_interfaces_max;
4681
4682         if (id.pid == LNET_PID_ANY)
4683                 id.pid = LNET_PID_LUSTRE;
4684
4685         pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4686         if (!pbuf)
4687                 return -ENOMEM;
4688
4689         /* initialize md content */
4690         md.start     = &pbuf->pb_info;
4691         md.length    = LNET_PING_INFO_SIZE(n_ids);
4692         md.threshold = 2; /* GET/REPLY */
4693         md.max_size  = 0;
4694         md.options   = LNET_MD_TRUNCATE;
4695         md.user_ptr  = &pd;
4696         md.handler   = lnet_ping_event_handler;
4697
4698         init_completion(&pd.completion);
4699
4700         rc = LNetMDBind(&md, LNET_UNLINK, &pd.mdh);
4701         if (rc != 0) {
4702                 CERROR("Can't bind MD: %d\n", rc);
4703                 goto fail_ping_buffer_decref;
4704         }
4705
4706         rc = LNetGet(lnet_nid_to_nid4(src_nid), pd.mdh, id,
4707                      LNET_RESERVED_PORTAL,
4708                      LNET_PROTO_PING_MATCHBITS, 0, false);
4709
4710         if (rc != 0) {
4711                 /* Don't CERROR; this could be deliberate! */
4712                 rc2 = LNetMDUnlink(pd.mdh);
4713                 LASSERT(rc2 == 0);
4714
4715                 /* NB must wait for the UNLINK event below... */
4716         }
4717
4718         if (wait_for_completion_timeout(&pd.completion, timeout) == 0) {
4719                 /* Ensure completion in finite time... */
4720                 LNetMDUnlink(pd.mdh);
4721                 wait_for_completion(&pd.completion);
4722         }
4723         if (!pd.replied) {
4724                 rc = -EIO;
4725                 goto fail_ping_buffer_decref;
4726         }
4727
4728         nob = pd.rc;
4729         LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4730
4731         rc = -EPROTO;           /* if I can't parse... */
4732
4733         if (nob < 8) {
4734                 CERROR("%s: ping info too short %d\n",
4735                        libcfs_id2str(id), nob);
4736                 goto fail_ping_buffer_decref;
4737         }
4738
4739         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4740                 lnet_swap_pinginfo(pbuf);
4741         } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4742                 CERROR("%s: Unexpected magic %08x\n",
4743                        libcfs_id2str(id), pbuf->pb_info.pi_magic);
4744                 goto fail_ping_buffer_decref;
4745         }
4746
4747         if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4748                 CERROR("%s: ping w/o NI status: 0x%x\n",
4749                        libcfs_id2str(id), pbuf->pb_info.pi_features);
4750                 goto fail_ping_buffer_decref;
4751         }
4752
4753         if (nob < LNET_PING_INFO_SIZE(0)) {
4754                 CERROR("%s: Short reply %d(%d min)\n",
4755                        libcfs_id2str(id),
4756                        nob, (int)LNET_PING_INFO_SIZE(0));
4757                 goto fail_ping_buffer_decref;
4758         }
4759
4760         if (pbuf->pb_info.pi_nnis < n_ids)
4761                 n_ids = pbuf->pb_info.pi_nnis;
4762
4763         if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4764                 CERROR("%s: Short reply %d(%d expected)\n",
4765                        libcfs_id2str(id),
4766                        nob, (int)LNET_PING_INFO_SIZE(n_ids));
4767                 goto fail_ping_buffer_decref;
4768         }
4769
4770         rc = -EFAULT;           /* if I segv in copy_to_user()... */
4771
4772         memset(&tmpid, 0, sizeof(tmpid));
4773         for (i = 0; i < n_ids; i++) {
4774                 tmpid.pid = pbuf->pb_info.pi_pid;
4775                 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4776                 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4777                         goto fail_ping_buffer_decref;
4778         }
4779         rc = pbuf->pb_info.pi_nnis;
4780
4781  fail_ping_buffer_decref:
4782         lnet_ping_buffer_decref(pbuf);
4783         return rc;
4784 }
4785
4786 static int
4787 lnet_discover(struct lnet_process_id id, __u32 force,
4788               struct lnet_process_id __user *ids, int n_ids)
4789 {
4790         struct lnet_peer_ni *lpni;
4791         struct lnet_peer_ni *p;
4792         struct lnet_peer *lp;
4793         struct lnet_process_id *buf;
4794         int cpt;
4795         int i;
4796         int rc;
4797
4798         if (n_ids <= 0 ||
4799             id.nid == LNET_NID_ANY)
4800                 return -EINVAL;
4801
4802         if (id.pid == LNET_PID_ANY)
4803                 id.pid = LNET_PID_LUSTRE;
4804
4805         /*
4806          * If the user buffer has more space than the lnet_interfaces_max,
4807          * then only fill it up to lnet_interfaces_max.
4808          */
4809         if (n_ids > lnet_interfaces_max)
4810                 n_ids = lnet_interfaces_max;
4811
4812         CFS_ALLOC_PTR_ARRAY(buf, n_ids);
4813         if (!buf)
4814                 return -ENOMEM;
4815
4816         cpt = lnet_net_lock_current();
4817         lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4818         if (IS_ERR(lpni)) {
4819                 rc = PTR_ERR(lpni);
4820                 goto out;
4821         }
4822
4823         /*
4824          * Clearing the NIDS_UPTODATE flag ensures the peer will
4825          * be discovered, provided discovery has not been disabled.
4826          */
4827         lp = lpni->lpni_peer_net->lpn_peer;
4828         spin_lock(&lp->lp_lock);
4829         lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4830         /* If the force flag is set, force a PING and PUSH as well. */
4831         if (force)
4832                 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4833         spin_unlock(&lp->lp_lock);
4834         rc = lnet_discover_peer_locked(lpni, cpt, true);
4835         if (rc)
4836                 goto out_decref;
4837
4838         /* The lpni (or lp) for this NID may have changed and our ref is
4839          * the only thing keeping the old one around. Release the ref
4840          * and lookup the lpni again
4841          */
4842         lnet_peer_ni_decref_locked(lpni);
4843         lpni = lnet_find_peer_ni_locked(id.nid);
4844         if (!lpni) {
4845                 rc = -ENOENT;
4846                 goto out;
4847         }
4848         lp = lpni->lpni_peer_net->lpn_peer;
4849
4850         i = 0;
4851         p = NULL;
4852         while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4853                 buf[i].pid = id.pid;
4854                 buf[i].nid = lnet_nid_to_nid4(&p->lpni_nid);
4855                 if (++i >= n_ids)
4856                         break;
4857         }
4858         rc = i;
4859
4860 out_decref:
4861         lnet_peer_ni_decref_locked(lpni);
4862 out:
4863         lnet_net_unlock(cpt);
4864
4865         if (rc >= 0)
4866                 if (copy_to_user(ids, buf, rc * sizeof(*buf)))
4867                         rc = -EFAULT;
4868         CFS_FREE_PTR_ARRAY(buf, n_ids);
4869
4870         return rc;
4871 }
4872
4873 /**
4874  * Retrieve peer discovery status.
4875  *
4876  * \retval 1 if lnet_peer_discovery_disabled is 0
4877  * \retval 0 if lnet_peer_discovery_disabled is 1
4878  */
4879 int
4880 LNetGetPeerDiscoveryStatus(void)
4881 {
4882         return !lnet_peer_discovery_disabled;
4883 }
4884 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);