lnet/lnet/api-ni.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32
  33 #define DEBUG_SUBSYSTEM S_LNET
  34
  35 #include <linux/ctype.h>
  36 #include <linux/log2.h>
  37 #include <linux/ktime.h>
  38 #include <linux/moduleparam.h>
  39 #include <linux/uaccess.h>
  40 #ifdef HAVE_SCHED_HEADERS
  41 #include <linux/sched/signal.h>
  42 #endif
  43
  44 #include <lnet/lib-lnet.h>
  45
  46 #define D_LNI D_CONSOLE
  47
  48 /*
  49  * initialize ln_api_mutex statically, since it needs to be used in
  50  * discovery_set callback. That module parameter callback can be called
  51  * before module init completes. The mutex needs to be ready for use then.
  52  */
  53 struct lnet the_lnet = {
  54         .ln_api_mutex = __MUTEX_INITIALIZER(the_lnet.ln_api_mutex),
  55 };              /* THE state of the network */
  56 EXPORT_SYMBOL(the_lnet);
  57
  58 static char *ip2nets = "";
  59 module_param(ip2nets, charp, 0444);
  60 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
  61
  62 static char *networks = "";
  63 module_param(networks, charp, 0444);
  64 MODULE_PARM_DESC(networks, "local networks");
  65
  66 static char *routes = "";
  67 module_param(routes, charp, 0444);
  68 MODULE_PARM_DESC(routes, "routes to non-local networks");
  69
  70 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
  71 module_param(rnet_htable_size, int, 0444);
  72 MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
  73
  74 static int use_tcp_bonding = false;
  75 module_param(use_tcp_bonding, int, 0444);
  76 MODULE_PARM_DESC(use_tcp_bonding,
  77                  "Set to 1 to use socklnd bonding. 0 to use Multi-Rail");
  78
  79 unsigned int lnet_numa_range = 0;
  80 module_param(lnet_numa_range, uint, 0444);
  81 MODULE_PARM_DESC(lnet_numa_range,
  82                 "NUMA range to consider during Multi-Rail selection");
  83
  84 /*
  85  * lnet_health_sensitivity determines by how much we decrement the health
  86  * value on sending error. The value defaults to 100, which means health
  87  * interface health is decremented by 100 points every failure.
  88  */
  89 unsigned int lnet_health_sensitivity = 100;
  90 static int sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp);
  91 #ifdef HAVE_KERNEL_PARAM_OPS
  92 static struct kernel_param_ops param_ops_health_sensitivity = {
  93         .set = sensitivity_set,
  94         .get = param_get_int,
  95 };
  96 #define param_check_health_sensitivity(name, p) \
  97                 __param_check(name, p, int)
  98 module_param(lnet_health_sensitivity, health_sensitivity, S_IRUGO|S_IWUSR);
  99 #else
 100 module_param_call(lnet_health_sensitivity, sensitivity_set, param_get_int,
 101                   &lnet_health_sensitivity, S_IRUGO|S_IWUSR);
 102 #endif
 103 MODULE_PARM_DESC(lnet_health_sensitivity,
 104                 "Value to decrement the health value by on error");
 105
 106 /*
 107  * lnet_recovery_interval determines how often we should perform recovery
 108  * on unhealthy interfaces.
 109  */
 110 unsigned int lnet_recovery_interval = 1;
 111 static int recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp);
 112 #ifdef HAVE_KERNEL_PARAM_OPS
 113 static struct kernel_param_ops param_ops_recovery_interval = {
 114         .set = recovery_interval_set,
 115         .get = param_get_int,
 116 };
 117 #define param_check_recovery_interval(name, p) \
 118                 __param_check(name, p, int)
 119 module_param(lnet_recovery_interval, recovery_interval, S_IRUGO|S_IWUSR);
 120 #else
 121 module_param_call(lnet_recovery_interval, recovery_interval_set, param_get_int,
 122                   &lnet_recovery_interval, S_IRUGO|S_IWUSR);
 123 #endif
 124 MODULE_PARM_DESC(lnet_recovery_interval,
 125                 "Interval to recover unhealthy interfaces in seconds");
 126
 127 static int lnet_interfaces_max = LNET_INTERFACES_MAX_DEFAULT;
 128 static int intf_max_set(const char *val, cfs_kernel_param_arg_t *kp);
 129
 130 static struct kernel_param_ops param_ops_interfaces_max = {
 131         .set = intf_max_set,
 132         .get = param_get_int,
 133 };
 134
 135 #define param_check_interfaces_max(name, p) \
 136                 __param_check(name, p, int)
 137
 138 #ifdef HAVE_KERNEL_PARAM_OPS
 139 module_param(lnet_interfaces_max, interfaces_max, 0644);
 140 #else
 141 module_param_call(lnet_interfaces_max, intf_max_set, param_get_int,
 142                   &param_ops_interfaces_max, 0644);
 143 #endif
 144 MODULE_PARM_DESC(lnet_interfaces_max,
 145                 "Maximum number of interfaces in a node.");
 146
 147 unsigned lnet_peer_discovery_disabled = 0;
 148 static int discovery_set(const char *val, cfs_kernel_param_arg_t *kp);
 149
 150 static struct kernel_param_ops param_ops_discovery_disabled = {
 151         .set = discovery_set,
 152         .get = param_get_int,
 153 };
 154
 155 #define param_check_discovery_disabled(name, p) \
 156                 __param_check(name, p, int)
 157 #ifdef HAVE_KERNEL_PARAM_OPS
 158 module_param(lnet_peer_discovery_disabled, discovery_disabled, 0644);
 159 #else
 160 module_param_call(lnet_peer_discovery_disabled, discovery_set, param_get_int,
 161                   &param_ops_discovery_disabled, 0644);
 162 #endif
 163 MODULE_PARM_DESC(lnet_peer_discovery_disabled,
 164                 "Set to 1 to disable peer discovery on this node.");
 165
 166 unsigned int lnet_drop_asym_route;
 167 static int drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp);
 168
 169 static struct kernel_param_ops param_ops_drop_asym_route = {
 170         .set = drop_asym_route_set,
 171         .get = param_get_int,
 172 };
 173
 174 #define param_check_drop_asym_route(name, p)    \
 175         __param_check(name, p, int)
 176 #ifdef HAVE_KERNEL_PARAM_OPS
 177 module_param(lnet_drop_asym_route, drop_asym_route, 0644);
 178 #else
 179 module_param_call(lnet_drop_asym_route, drop_asym_route_set, param_get_int,
 180                   &param_ops_drop_asym_route, 0644);
 181 #endif
 182 MODULE_PARM_DESC(lnet_drop_asym_route,
 183                  "Set to 1 to drop asymmetrical route messages.");
 184
 185 #define LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT 50
 186 #define LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT 50
 187
 188 unsigned lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
 189 static int transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp);
 190 #ifdef HAVE_KERNEL_PARAM_OPS
 191 static struct kernel_param_ops param_ops_transaction_timeout = {
 192         .set = transaction_to_set,
 193         .get = param_get_int,
 194 };
 195
 196 #define param_check_transaction_timeout(name, p) \
 197                 __param_check(name, p, int)
 198 module_param(lnet_transaction_timeout, transaction_timeout, S_IRUGO|S_IWUSR);
 199 #else
 200 module_param_call(lnet_transaction_timeout, transaction_to_set, param_get_int,
 201                   &lnet_transaction_timeout, S_IRUGO|S_IWUSR);
 202 #endif
 203 MODULE_PARM_DESC(lnet_transaction_timeout,
 204                 "Maximum number of seconds to wait for a peer response.");
 205
 206 #define LNET_RETRY_COUNT_HEALTH_DEFAULT 2
 207 unsigned lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
 208 static int retry_count_set(const char *val, cfs_kernel_param_arg_t *kp);
 209 #ifdef HAVE_KERNEL_PARAM_OPS
 210 static struct kernel_param_ops param_ops_retry_count = {
 211         .set = retry_count_set,
 212         .get = param_get_int,
 213 };
 214
 215 #define param_check_retry_count(name, p) \
 216                 __param_check(name, p, int)
 217 module_param(lnet_retry_count, retry_count, S_IRUGO|S_IWUSR);
 218 #else
 219 module_param_call(lnet_retry_count, retry_count_set, param_get_int,
 220                   &lnet_retry_count, S_IRUGO|S_IWUSR);
 221 #endif
 222 MODULE_PARM_DESC(lnet_retry_count,
 223                  "Maximum number of times to retry transmitting a message");
 224
 225
 226 unsigned lnet_lnd_timeout = LNET_LND_DEFAULT_TIMEOUT;
 227 unsigned int lnet_current_net_count;
 228
 229 /*
 230  * This sequence number keeps track of how many times DLC was used to
 231  * update the local NIs. It is incremented when a NI is added or
 232  * removed and checked when sending a message to determine if there is
 233  * a need to re-run the selection algorithm. See lnet_select_pathway()
 234  * for more details on its usage.
 235  */
 236 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
 237
 238 static int lnet_ping(struct lnet_process_id id, signed long timeout,
 239                      struct lnet_process_id __user *ids, int n_ids);
 240
 241 static int lnet_discover(struct lnet_process_id id, __u32 force,
 242                          struct lnet_process_id __user *ids, int n_ids);
 243
 244 static int
 245 sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
 246 {
 247         int rc;
 248         unsigned *sensitivity = (unsigned *)kp->arg;
 249         unsigned long value;
 250
 251         rc = kstrtoul(val, 0, &value);
 252         if (rc) {
 253                 CERROR("Invalid module parameter value for 'lnet_health_sensitivity'\n");
 254                 return rc;
 255         }
 256
 257         /*
 258          * The purpose of locking the api_mutex here is to ensure that
 259          * the correct value ends up stored properly.
 260          */
 261         mutex_lock(&the_lnet.ln_api_mutex);
 262
 263         if (value > LNET_MAX_HEALTH_VALUE) {
 264                 mutex_unlock(&the_lnet.ln_api_mutex);
 265                 CERROR("Invalid health value. Maximum: %d value = %lu\n",
 266                        LNET_MAX_HEALTH_VALUE, value);
 267                 return -EINVAL;
 268         }
 269
 270         /*
 271          * if we're turning on health then use the health timeout
 272          * defaults.
 273          */
 274         if (*sensitivity == 0 && value != 0) {
 275                 lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_HEALTH_DEFAULT;
 276                 lnet_retry_count = LNET_RETRY_COUNT_HEALTH_DEFAULT;
 277         /*
 278          * if we're turning off health then use the no health timeout
 279          * default.
 280          */
 281         } else if (*sensitivity != 0 && value == 0) {
 282                 lnet_transaction_timeout =
 283                         LNET_TRANSACTION_TIMEOUT_NO_HEALTH_DEFAULT;
 284                 lnet_retry_count = 0;
 285         }
 286
 287         *sensitivity = value;
 288
 289         mutex_unlock(&the_lnet.ln_api_mutex);
 290
 291         return 0;
 292 }
 293
 294 static int
 295 recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp)
 296 {
 297         int rc;
 298         unsigned *interval = (unsigned *)kp->arg;
 299         unsigned long value;
 300
 301         rc = kstrtoul(val, 0, &value);
 302         if (rc) {
 303                 CERROR("Invalid module parameter value for 'lnet_recovery_interval'\n");
 304                 return rc;
 305         }
 306
 307         if (value < 1) {
 308                 CERROR("lnet_recovery_interval must be at least 1 second\n");
 309                 return -EINVAL;
 310         }
 311
 312         /*
 313          * The purpose of locking the api_mutex here is to ensure that
 314          * the correct value ends up stored properly.
 315          */
 316         mutex_lock(&the_lnet.ln_api_mutex);
 317
 318         *interval = value;
 319
 320         mutex_unlock(&the_lnet.ln_api_mutex);
 321
 322         return 0;
 323 }
 324
 325 static int
 326 discovery_set(const char *val, cfs_kernel_param_arg_t *kp)
 327 {
 328         int rc;
 329         unsigned *discovery = (unsigned *)kp->arg;
 330         unsigned long value;
 331         struct lnet_ping_buffer *pbuf;
 332
 333         rc = kstrtoul(val, 0, &value);
 334         if (rc) {
 335                 CERROR("Invalid module parameter value for 'lnet_peer_discovery_disabled'\n");
 336                 return rc;
 337         }
 338
 339         value = (value) ? 1 : 0;
 340
 341         /*
 342          * The purpose of locking the api_mutex here is to ensure that
 343          * the correct value ends up stored properly.
 344          */
 345         mutex_lock(&the_lnet.ln_api_mutex);
 346
 347         if (value == *discovery) {
 348                 mutex_unlock(&the_lnet.ln_api_mutex);
 349                 return 0;
 350         }
 351
 352         *discovery = value;
 353
 354         if (the_lnet.ln_state != LNET_STATE_RUNNING) {
 355                 mutex_unlock(&the_lnet.ln_api_mutex);
 356                 return 0;
 357         }
 358
 359         /* tell peers that discovery setting has changed */
 360         lnet_net_lock(LNET_LOCK_EX);
 361         pbuf = the_lnet.ln_ping_target;
 362         if (value)
 363                 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
 364         else
 365                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
 366         lnet_net_unlock(LNET_LOCK_EX);
 367
 368         lnet_push_update_to_peers(1);
 369
 370         mutex_unlock(&the_lnet.ln_api_mutex);
 371
 372         return 0;
 373 }
 374
 375 static int
 376 drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp)
 377 {
 378         int rc;
 379         unsigned int *drop_asym_route = (unsigned int *)kp->arg;
 380         unsigned long value;
 381
 382         rc = kstrtoul(val, 0, &value);
 383         if (rc) {
 384                 CERROR("Invalid module parameter value for "
 385                        "'lnet_drop_asym_route'\n");
 386                 return rc;
 387         }
 388
 389         /*
 390          * The purpose of locking the api_mutex here is to ensure that
 391          * the correct value ends up stored properly.
 392          */
 393         mutex_lock(&the_lnet.ln_api_mutex);
 394
 395         if (value == *drop_asym_route) {
 396                 mutex_unlock(&the_lnet.ln_api_mutex);
 397                 return 0;
 398         }
 399
 400         *drop_asym_route = value;
 401
 402         mutex_unlock(&the_lnet.ln_api_mutex);
 403
 404         return 0;
 405 }
 406
 407 static int
 408 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
 409 {
 410         int rc;
 411         unsigned *transaction_to = (unsigned *)kp->arg;
 412         unsigned long value;
 413
 414         rc = kstrtoul(val, 0, &value);
 415         if (rc) {
 416                 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
 417                 return rc;
 418         }
 419
 420         /*
 421          * The purpose of locking the api_mutex here is to ensure that
 422          * the correct value ends up stored properly.
 423          */
 424         mutex_lock(&the_lnet.ln_api_mutex);
 425
 426         if (value < lnet_retry_count || value == 0) {
 427                 mutex_unlock(&the_lnet.ln_api_mutex);
 428                 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
 429                        "Has to be greater than lnet_retry_count (%u)\n",
 430                        value, lnet_retry_count);
 431                 return -EINVAL;
 432         }
 433
 434         if (value == *transaction_to) {
 435                 mutex_unlock(&the_lnet.ln_api_mutex);
 436                 return 0;
 437         }
 438
 439         *transaction_to = value;
 440         if (lnet_retry_count == 0)
 441                 lnet_lnd_timeout = value;
 442         else
 443                 lnet_lnd_timeout = value / lnet_retry_count;
 444
 445         mutex_unlock(&the_lnet.ln_api_mutex);
 446
 447         return 0;
 448 }
 449
 450 static int
 451 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
 452 {
 453         int rc;
 454         unsigned *retry_count = (unsigned *)kp->arg;
 455         unsigned long value;
 456
 457         rc = kstrtoul(val, 0, &value);
 458         if (rc) {
 459                 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
 460                 return rc;
 461         }
 462
 463         /*
 464          * The purpose of locking the api_mutex here is to ensure that
 465          * the correct value ends up stored properly.
 466          */
 467         mutex_lock(&the_lnet.ln_api_mutex);
 468
 469         if (lnet_health_sensitivity == 0) {
 470                 mutex_unlock(&the_lnet.ln_api_mutex);
 471                 CERROR("Can not set retry_count when health feature is turned off\n");
 472                 return -EINVAL;
 473         }
 474
 475         if (value > lnet_transaction_timeout) {
 476                 mutex_unlock(&the_lnet.ln_api_mutex);
 477                 CERROR("Invalid value for lnet_retry_count (%lu). "
 478                        "Has to be smaller than lnet_transaction_timeout (%u)\n",
 479                        value, lnet_transaction_timeout);
 480                 return -EINVAL;
 481         }
 482
 483         *retry_count = value;
 484
 485         if (value == 0)
 486                 lnet_lnd_timeout = lnet_transaction_timeout;
 487         else
 488                 lnet_lnd_timeout = lnet_transaction_timeout / value;
 489
 490         mutex_unlock(&the_lnet.ln_api_mutex);
 491
 492         return 0;
 493 }
 494
 495 static int
 496 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
 497 {
 498         int value, rc;
 499
 500         rc = kstrtoint(val, 0, &value);
 501         if (rc) {
 502                 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
 503                 return rc;
 504         }
 505
 506         if (value < LNET_INTERFACES_MIN) {
 507                 CWARN("max interfaces provided are too small, setting to %d\n",
 508                       LNET_INTERFACES_MAX_DEFAULT);
 509                 value = LNET_INTERFACES_MAX_DEFAULT;
 510         }
 511
 512         *(int *)kp->arg = value;
 513
 514         return 0;
 515 }
 516
 517 static char *
 518 lnet_get_routes(void)
 519 {
 520         return routes;
 521 }
 522
 523 static char *
 524 lnet_get_networks(void)
 525 {
 526         char   *nets;
 527         int     rc;
 528
 529         if (*networks != 0 && *ip2nets != 0) {
 530                 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
 531                                    "'ip2nets' but not both at once\n");
 532                 return NULL;
 533         }
 534
 535         if (*ip2nets != 0) {
 536                 rc = lnet_parse_ip2nets(&nets, ip2nets);
 537                 return (rc == 0) ? nets : NULL;
 538         }
 539
 540         if (*networks != 0)
 541                 return networks;
 542
 543         return "tcp";
 544 }
 545
 546 static void
 547 lnet_init_locks(void)
 548 {
 549         spin_lock_init(&the_lnet.ln_eq_wait_lock);
 550         spin_lock_init(&the_lnet.ln_msg_resend_lock);
 551         init_waitqueue_head(&the_lnet.ln_eq_waitq);
 552         init_completion(&the_lnet.ln_mt_wait_complete);
 553         mutex_init(&the_lnet.ln_lnd_mutex);
 554 }
 555
 556 struct kmem_cache *lnet_mes_cachep;        /* MEs kmem_cache */
 557 struct kmem_cache *lnet_small_mds_cachep;  /* <= LNET_SMALL_MD_SIZE bytes
 558                                             *  MDs kmem_cache */
 559 struct kmem_cache *lnet_rspt_cachep;       /* response tracker cache */
 560 struct kmem_cache *lnet_msg_cachep;
 561
 562 static int
 563 lnet_slab_setup(void)
 564 {
 565         /* create specific kmem_cache for MEs and small MDs (i.e., originally
 566          * allocated in <size-xxx> kmem_cache).
 567          */
 568         lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
 569                                             0, 0, NULL);
 570         if (!lnet_mes_cachep)
 571                 return -ENOMEM;
 572
 573         lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
 574                                                   LNET_SMALL_MD_SIZE, 0, 0,
 575                                                   NULL);
 576         if (!lnet_small_mds_cachep)
 577                 return -ENOMEM;
 578
 579         lnet_rspt_cachep = kmem_cache_create("lnet_rspt", sizeof(struct lnet_rsp_tracker),
 580                                             0, 0, NULL);
 581         if (!lnet_rspt_cachep)
 582                 return -ENOMEM;
 583
 584         lnet_msg_cachep = kmem_cache_create("lnet_msg", sizeof(struct lnet_msg),
 585                                             0, 0, NULL);
 586         if (!lnet_msg_cachep)
 587                 return -ENOMEM;
 588
 589         return 0;
 590 }
 591
 592 static void
 593 lnet_slab_cleanup(void)
 594 {
 595         if (lnet_msg_cachep) {
 596                 kmem_cache_destroy(lnet_msg_cachep);
 597                 lnet_msg_cachep = NULL;
 598         }
 599
 600
 601         if (lnet_rspt_cachep) {
 602                 kmem_cache_destroy(lnet_rspt_cachep);
 603                 lnet_rspt_cachep = NULL;
 604         }
 605
 606         if (lnet_small_mds_cachep) {
 607                 kmem_cache_destroy(lnet_small_mds_cachep);
 608                 lnet_small_mds_cachep = NULL;
 609         }
 610
 611         if (lnet_mes_cachep) {
 612                 kmem_cache_destroy(lnet_mes_cachep);
 613                 lnet_mes_cachep = NULL;
 614         }
 615 }
 616
 617 static int
 618 lnet_create_remote_nets_table(void)
 619 {
 620         int               i;
 621         struct list_head *hash;
 622
 623         LASSERT(the_lnet.ln_remote_nets_hash == NULL);
 624         LASSERT(the_lnet.ln_remote_nets_hbits > 0);
 625         LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
 626         if (hash == NULL) {
 627                 CERROR("Failed to create remote nets hash table\n");
 628                 return -ENOMEM;
 629         }
 630
 631         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 632                 INIT_LIST_HEAD(&hash[i]);
 633         the_lnet.ln_remote_nets_hash = hash;
 634         return 0;
 635 }
 636
 637 static void
 638 lnet_destroy_remote_nets_table(void)
 639 {
 640         int i;
 641
 642         if (the_lnet.ln_remote_nets_hash == NULL)
 643                 return;
 644
 645         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 646                 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
 647
 648         LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
 649                     LNET_REMOTE_NETS_HASH_SIZE *
 650                     sizeof(the_lnet.ln_remote_nets_hash[0]));
 651         the_lnet.ln_remote_nets_hash = NULL;
 652 }
 653
 654 static void
 655 lnet_destroy_locks(void)
 656 {
 657         if (the_lnet.ln_res_lock != NULL) {
 658                 cfs_percpt_lock_free(the_lnet.ln_res_lock);
 659                 the_lnet.ln_res_lock = NULL;
 660         }
 661
 662         if (the_lnet.ln_net_lock != NULL) {
 663                 cfs_percpt_lock_free(the_lnet.ln_net_lock);
 664                 the_lnet.ln_net_lock = NULL;
 665         }
 666 }
 667
 668 static int
 669 lnet_create_locks(void)
 670 {
 671         lnet_init_locks();
 672
 673         the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 674         if (the_lnet.ln_res_lock == NULL)
 675                 goto failed;
 676
 677         the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 678         if (the_lnet.ln_net_lock == NULL)
 679                 goto failed;
 680
 681         return 0;
 682
 683  failed:
 684         lnet_destroy_locks();
 685         return -ENOMEM;
 686 }
 687
 688 static void lnet_assert_wire_constants(void)
 689 {
 690         /* Wire protocol assertions generated by 'wirecheck'
 691          * running on Linux robert.bartonsoftware.com 2.6.8-1.521
 692          * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
 693          * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
 694
 695         /* Constants... */
 696         BUILD_BUG_ON(LNET_PROTO_TCP_MAGIC != 0xeebc0ded);
 697         BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MAJOR != 1);
 698         BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MINOR != 0);
 699         BUILD_BUG_ON(LNET_MSG_ACK != 0);
 700         BUILD_BUG_ON(LNET_MSG_PUT != 1);
 701         BUILD_BUG_ON(LNET_MSG_GET != 2);
 702         BUILD_BUG_ON(LNET_MSG_REPLY != 3);
 703         BUILD_BUG_ON(LNET_MSG_HELLO != 4);
 704
 705         /* Checks for struct lnet_handle_wire */
 706         BUILD_BUG_ON((int)sizeof(struct lnet_handle_wire) != 16);
 707         BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
 708                                    wh_interface_cookie) != 0);
 709         BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) != 8);
 710         BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
 711                                    wh_object_cookie) != 8);
 712         BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) != 8);
 713
 714         /* Checks for struct struct lnet_magicversion */
 715         BUILD_BUG_ON((int)sizeof(struct lnet_magicversion) != 8);
 716         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, magic) != 0);
 717         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->magic) != 4);
 718         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, version_major) != 4);
 719         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_major) != 2);
 720         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion,
 721                                    version_minor) != 6);
 722         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_minor) != 2);
 723
 724         /* Checks for struct struct lnet_hdr */
 725         BUILD_BUG_ON((int)sizeof(struct lnet_hdr) != 72);
 726         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_nid) != 0);
 727         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_nid) != 8);
 728         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_nid) != 8);
 729         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_nid) != 8);
 730         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_pid) != 16);
 731         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_pid) != 4);
 732         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_pid) != 20);
 733         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_pid) != 4);
 734         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, type) != 24);
 735         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->type) != 4);
 736         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, payload_length) != 28);
 737         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->payload_length) != 4);
 738         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg) != 32);
 739         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg) != 40);
 740
 741         /* Ack */
 742         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) != 32);
 743         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) != 16);
 744         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.match_bits) != 48);
 745         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) != 8);
 746         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.mlength) != 56);
 747         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) != 4);
 748
 749         /* Put */
 750         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) != 32);
 751         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) != 16);
 752         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.match_bits) != 48);
 753         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) != 8);
 754         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.hdr_data) != 56);
 755         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) != 8);
 756         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ptl_index) != 64);
 757         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) != 4);
 758         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.offset) != 68);
 759         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) != 4);
 760
 761         /* Get */
 762         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.return_wmd) != 32);
 763         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) != 16);
 764         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.match_bits) != 48);
 765         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) != 8);
 766         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.ptl_index) != 56);
 767         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) != 4);
 768         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.src_offset) != 60);
 769         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) != 4);
 770         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.sink_length) != 64);
 771         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) != 4);
 772
 773         /* Reply */
 774         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) != 32);
 775         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) != 16);
 776
 777         /* Hello */
 778         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.incarnation) != 32);
 779         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) != 8);
 780         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.type) != 40);
 781         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) != 4);
 782
 783         /* Checks for struct lnet_ni_status and related constants */
 784         BUILD_BUG_ON(LNET_NI_STATUS_INVALID != 0x00000000);
 785         BUILD_BUG_ON(LNET_NI_STATUS_UP != 0x15aac0de);
 786         BUILD_BUG_ON(LNET_NI_STATUS_DOWN != 0xdeadface);
 787
 788         /* Checks for struct lnet_ni_status */
 789         BUILD_BUG_ON((int)sizeof(struct lnet_ni_status) != 16);
 790         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_nid) != 0);
 791         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) != 8);
 792         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_status) != 8);
 793         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_status) != 4);
 794         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_unused) != 12);
 795         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) != 4);
 796
 797         /* Checks for struct lnet_ping_info and related constants */
 798         BUILD_BUG_ON(LNET_PROTO_PING_MAGIC != 0x70696E67);
 799         BUILD_BUG_ON(LNET_PING_FEAT_INVAL != 0);
 800         BUILD_BUG_ON(LNET_PING_FEAT_BASE != 1);
 801         BUILD_BUG_ON(LNET_PING_FEAT_NI_STATUS != 2);
 802         BUILD_BUG_ON(LNET_PING_FEAT_RTE_DISABLED != 4);
 803         BUILD_BUG_ON(LNET_PING_FEAT_MULTI_RAIL != 8);
 804         BUILD_BUG_ON(LNET_PING_FEAT_DISCOVERY != 16);
 805         BUILD_BUG_ON(LNET_PING_FEAT_BITS != 31);
 806
 807         /* Checks for struct lnet_ping_info */
 808         BUILD_BUG_ON((int)sizeof(struct lnet_ping_info) != 16);
 809         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_magic) != 0);
 810         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) != 4);
 811         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_features) != 4);
 812         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_features) != 4);
 813         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_pid) != 8);
 814         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) != 4);
 815         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_nnis) != 12);
 816         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) != 4);
 817         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_ni) != 16);
 818         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) != 0);
 819 }
 820
 821 static const struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
 822 {
 823         const struct lnet_lnd *lnd;
 824
 825         /* holding lnd mutex */
 826         if (type >= NUM_LNDS)
 827                 return NULL;
 828         lnd = the_lnet.ln_lnds[type];
 829         LASSERT(!lnd || lnd->lnd_type == type);
 830
 831         return lnd;
 832 }
 833
 834 unsigned int
 835 lnet_get_lnd_timeout(void)
 836 {
 837         return lnet_lnd_timeout;
 838 }
 839 EXPORT_SYMBOL(lnet_get_lnd_timeout);
 840
 841 void
 842 lnet_register_lnd(const struct lnet_lnd *lnd)
 843 {
 844         mutex_lock(&the_lnet.ln_lnd_mutex);
 845
 846         LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
 847         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
 848
 849         the_lnet.ln_lnds[lnd->lnd_type] = lnd;
 850
 851         CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
 852
 853         mutex_unlock(&the_lnet.ln_lnd_mutex);
 854 }
 855 EXPORT_SYMBOL(lnet_register_lnd);
 856
 857 void
 858 lnet_unregister_lnd(const struct lnet_lnd *lnd)
 859 {
 860         mutex_lock(&the_lnet.ln_lnd_mutex);
 861
 862         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
 863
 864         the_lnet.ln_lnds[lnd->lnd_type] = NULL;
 865         CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
 866
 867         mutex_unlock(&the_lnet.ln_lnd_mutex);
 868 }
 869 EXPORT_SYMBOL(lnet_unregister_lnd);
 870
 871 void
 872 lnet_counters_get_common(struct lnet_counters_common *common)
 873 {
 874         struct lnet_counters *ctr;
 875         int i;
 876
 877         memset(common, 0, sizeof(*common));
 878
 879         lnet_net_lock(LNET_LOCK_EX);
 880
 881         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
 882                 common->lcc_msgs_max     += ctr->lct_common.lcc_msgs_max;
 883                 common->lcc_msgs_alloc   += ctr->lct_common.lcc_msgs_alloc;
 884                 common->lcc_errors       += ctr->lct_common.lcc_errors;
 885                 common->lcc_send_count   += ctr->lct_common.lcc_send_count;
 886                 common->lcc_recv_count   += ctr->lct_common.lcc_recv_count;
 887                 common->lcc_route_count  += ctr->lct_common.lcc_route_count;
 888                 common->lcc_drop_count   += ctr->lct_common.lcc_drop_count;
 889                 common->lcc_send_length  += ctr->lct_common.lcc_send_length;
 890                 common->lcc_recv_length  += ctr->lct_common.lcc_recv_length;
 891                 common->lcc_route_length += ctr->lct_common.lcc_route_length;
 892                 common->lcc_drop_length  += ctr->lct_common.lcc_drop_length;
 893         }
 894         lnet_net_unlock(LNET_LOCK_EX);
 895 }
 896 EXPORT_SYMBOL(lnet_counters_get_common);
 897
 898 void
 899 lnet_counters_get(struct lnet_counters *counters)
 900 {
 901         struct lnet_counters *ctr;
 902         struct lnet_counters_health *health = &counters->lct_health;
 903         int             i;
 904
 905         memset(counters, 0, sizeof(*counters));
 906
 907         lnet_counters_get_common(&counters->lct_common);
 908
 909         lnet_net_lock(LNET_LOCK_EX);
 910
 911         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
 912                 health->lch_rst_alloc    += ctr->lct_health.lch_rst_alloc;
 913                 health->lch_resend_count += ctr->lct_health.lch_resend_count;
 914                 health->lch_response_timeout_count +=
 915                                 ctr->lct_health.lch_response_timeout_count;
 916                 health->lch_local_interrupt_count +=
 917                                 ctr->lct_health.lch_local_interrupt_count;
 918                 health->lch_local_dropped_count +=
 919                                 ctr->lct_health.lch_local_dropped_count;
 920                 health->lch_local_aborted_count +=
 921                                 ctr->lct_health.lch_local_aborted_count;
 922                 health->lch_local_no_route_count +=
 923                                 ctr->lct_health.lch_local_no_route_count;
 924                 health->lch_local_timeout_count +=
 925                                 ctr->lct_health.lch_local_timeout_count;
 926                 health->lch_local_error_count +=
 927                                 ctr->lct_health.lch_local_error_count;
 928                 health->lch_remote_dropped_count +=
 929                                 ctr->lct_health.lch_remote_dropped_count;
 930                 health->lch_remote_error_count +=
 931                                 ctr->lct_health.lch_remote_error_count;
 932                 health->lch_remote_timeout_count +=
 933                                 ctr->lct_health.lch_remote_timeout_count;
 934                 health->lch_network_timeout_count +=
 935                                 ctr->lct_health.lch_network_timeout_count;
 936         }
 937         lnet_net_unlock(LNET_LOCK_EX);
 938 }
 939 EXPORT_SYMBOL(lnet_counters_get);
 940
 941 void
 942 lnet_counters_reset(void)
 943 {
 944         struct lnet_counters *counters;
 945         int             i;
 946
 947         lnet_net_lock(LNET_LOCK_EX);
 948
 949         cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
 950                 memset(counters, 0, sizeof(struct lnet_counters));
 951
 952         lnet_net_unlock(LNET_LOCK_EX);
 953 }
 954
 955 static char *
 956 lnet_res_type2str(int type)
 957 {
 958         switch (type) {
 959         default:
 960                 LBUG();
 961         case LNET_COOKIE_TYPE_MD:
 962                 return "MD";
 963         case LNET_COOKIE_TYPE_ME:
 964                 return "ME";
 965         case LNET_COOKIE_TYPE_EQ:
 966                 return "EQ";
 967         }
 968 }
 969
 970 static void
 971 lnet_res_container_cleanup(struct lnet_res_container *rec)
 972 {
 973         int     count = 0;
 974
 975         if (rec->rec_type == 0) /* not set yet, it's uninitialized */
 976                 return;
 977
 978         while (!list_empty(&rec->rec_active)) {
 979                 struct list_head *e = rec->rec_active.next;
 980
 981                 list_del_init(e);
 982                 if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
 983                         lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
 984
 985                 } else { /* NB: Active MEs should be attached on portals */
 986                         LBUG();
 987                 }
 988                 count++;
 989         }
 990
 991         if (count > 0) {
 992                 /* Found alive MD/ME/EQ, user really should unlink/free
 993                  * all of them before finalize LNet, but if someone didn't,
 994                  * we have to recycle garbage for him */
 995                 CERROR("%d active elements on exit of %s container\n",
 996                        count, lnet_res_type2str(rec->rec_type));
 997         }
 998
 999         if (rec->rec_lh_hash != NULL) {
1000                 LIBCFS_FREE(rec->rec_lh_hash,
1001                             LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1002                 rec->rec_lh_hash = NULL;
1003         }
1004
1005         rec->rec_type = 0; /* mark it as finalized */
1006 }
1007
1008 static int
1009 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
1010 {
1011         int     rc = 0;
1012         int     i;
1013
1014         LASSERT(rec->rec_type == 0);
1015
1016         rec->rec_type = type;
1017         INIT_LIST_HEAD(&rec->rec_active);
1018
1019         rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1020
1021         /* Arbitrary choice of hash table size */
1022         LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1023                          LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1024         if (rec->rec_lh_hash == NULL) {
1025                 rc = -ENOMEM;
1026                 goto out;
1027         }
1028
1029         for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1030                 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1031
1032         return 0;
1033
1034 out:
1035         CERROR("Failed to setup %s resource container\n",
1036                lnet_res_type2str(type));
1037         lnet_res_container_cleanup(rec);
1038         return rc;
1039 }
1040
1041 static void
1042 lnet_res_containers_destroy(struct lnet_res_container **recs)
1043 {
1044         struct lnet_res_container       *rec;
1045         int                             i;
1046
1047         cfs_percpt_for_each(rec, i, recs)
1048                 lnet_res_container_cleanup(rec);
1049
1050         cfs_percpt_free(recs);
1051 }
1052
1053 static struct lnet_res_container **
1054 lnet_res_containers_create(int type)
1055 {
1056         struct lnet_res_container       **recs;
1057         struct lnet_res_container       *rec;
1058         int                             rc;
1059         int                             i;
1060
1061         recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1062         if (recs == NULL) {
1063                 CERROR("Failed to allocate %s resource containers\n",
1064                        lnet_res_type2str(type));
1065                 return NULL;
1066         }
1067
1068         cfs_percpt_for_each(rec, i, recs) {
1069                 rc = lnet_res_container_setup(rec, i, type);
1070                 if (rc != 0) {
1071                         lnet_res_containers_destroy(recs);
1072                         return NULL;
1073                 }
1074         }
1075
1076         return recs;
1077 }
1078
1079 struct lnet_libhandle *
1080 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1081 {
1082         /* ALWAYS called with lnet_res_lock held */
1083         struct list_head        *head;
1084         struct lnet_libhandle   *lh;
1085         unsigned int            hash;
1086
1087         if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1088                 return NULL;
1089
1090         hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1091         head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1092
1093         list_for_each_entry(lh, head, lh_hash_chain) {
1094                 if (lh->lh_cookie == cookie)
1095                         return lh;
1096         }
1097
1098         return NULL;
1099 }
1100
1101 void
1102 lnet_res_lh_initialize(struct lnet_res_container *rec,
1103                        struct lnet_libhandle *lh)
1104 {
1105         /* ALWAYS called with lnet_res_lock held */
1106         unsigned int    ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1107         unsigned int    hash;
1108
1109         lh->lh_cookie = rec->rec_lh_cookie;
1110         rec->rec_lh_cookie += 1 << ibits;
1111
1112         hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1113
1114         list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1115 }
1116
1117 struct list_head **
1118 lnet_create_array_of_queues(void)
1119 {
1120         struct list_head **qs;
1121         struct list_head *q;
1122         int i;
1123
1124         qs = cfs_percpt_alloc(lnet_cpt_table(),
1125                               sizeof(struct list_head));
1126         if (!qs) {
1127                 CERROR("Failed to allocate queues\n");
1128                 return NULL;
1129         }
1130
1131         cfs_percpt_for_each(q, i, qs)
1132                 INIT_LIST_HEAD(q);
1133
1134         return qs;
1135 }
1136
1137 static int lnet_unprepare(void);
1138
1139 static int
1140 lnet_prepare(lnet_pid_t requested_pid)
1141 {
1142         /* Prepare to bring up the network */
1143         struct lnet_res_container **recs;
1144         int                       rc = 0;
1145
1146         if (requested_pid == LNET_PID_ANY) {
1147                 /* Don't instantiate LNET just for me */
1148                 return -ENETDOWN;
1149         }
1150
1151         LASSERT(the_lnet.ln_refcount == 0);
1152
1153         the_lnet.ln_routing = 0;
1154
1155         LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1156         the_lnet.ln_pid = requested_pid;
1157
1158         INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1159         INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1160         INIT_LIST_HEAD(&the_lnet.ln_nets);
1161         INIT_LIST_HEAD(&the_lnet.ln_routers);
1162         INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1163         INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1164         INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1165         INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1166         INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1167         INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1168         INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1169         init_waitqueue_head(&the_lnet.ln_dc_waitq);
1170         the_lnet.ln_mt_eq = NULL;
1171         init_completion(&the_lnet.ln_started);
1172
1173         rc = lnet_slab_setup();
1174         if (rc != 0)
1175                 goto failed;
1176
1177         rc = lnet_create_remote_nets_table();
1178         if (rc != 0)
1179                 goto failed;
1180
1181         /*
1182          * NB the interface cookie in wire handles guards against delayed
1183          * replies and ACKs appearing valid after reboot.
1184          */
1185         the_lnet.ln_interface_cookie = ktime_get_real_ns();
1186
1187         the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1188                                                 sizeof(struct lnet_counters));
1189         if (the_lnet.ln_counters == NULL) {
1190                 CERROR("Failed to allocate counters for LNet\n");
1191                 rc = -ENOMEM;
1192                 goto failed;
1193         }
1194
1195         rc = lnet_peer_tables_create();
1196         if (rc != 0)
1197                 goto failed;
1198
1199         rc = lnet_msg_containers_create();
1200         if (rc != 0)
1201                 goto failed;
1202
1203         rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1204                                       LNET_COOKIE_TYPE_EQ);
1205         if (rc != 0)
1206                 goto failed;
1207
1208         recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1209         if (recs == NULL) {
1210                 rc = -ENOMEM;
1211                 goto failed;
1212         }
1213
1214         the_lnet.ln_md_containers = recs;
1215
1216         rc = lnet_portals_create();
1217         if (rc != 0) {
1218                 CERROR("Failed to create portals for LNet: %d\n", rc);
1219                 goto failed;
1220         }
1221
1222         the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1223         if (!the_lnet.ln_mt_zombie_rstqs) {
1224                 rc = -ENOMEM;
1225                 goto failed;
1226         }
1227
1228         return 0;
1229
1230  failed:
1231         lnet_unprepare();
1232         return rc;
1233 }
1234
1235 static int
1236 lnet_unprepare (void)
1237 {
1238         int rc;
1239
1240         /* NB no LNET_LOCK since this is the last reference.  All LND instances
1241          * have shut down already, so it is safe to unlink and free all
1242          * descriptors, even those that appear committed to a network op (eg MD
1243          * with non-zero pending count) */
1244
1245         lnet_fail_nid(LNET_NID_ANY, 0);
1246
1247         LASSERT(the_lnet.ln_refcount == 0);
1248         LASSERT(list_empty(&the_lnet.ln_test_peers));
1249         LASSERT(list_empty(&the_lnet.ln_nets));
1250
1251         if (the_lnet.ln_mt_zombie_rstqs) {
1252                 lnet_clean_zombie_rstqs();
1253                 the_lnet.ln_mt_zombie_rstqs = NULL;
1254         }
1255
1256         if (the_lnet.ln_mt_eq) {
1257                 rc = LNetEQFree(the_lnet.ln_mt_eq);
1258                 the_lnet.ln_mt_eq = NULL;
1259                 LASSERT(rc == 0);
1260         }
1261
1262         lnet_portals_destroy();
1263
1264         if (the_lnet.ln_md_containers != NULL) {
1265                 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1266                 the_lnet.ln_md_containers = NULL;
1267         }
1268
1269         lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1270
1271         lnet_msg_containers_destroy();
1272         lnet_peer_uninit();
1273         lnet_rtrpools_free(0);
1274
1275         if (the_lnet.ln_counters != NULL) {
1276                 cfs_percpt_free(the_lnet.ln_counters);
1277                 the_lnet.ln_counters = NULL;
1278         }
1279         lnet_destroy_remote_nets_table();
1280         lnet_slab_cleanup();
1281
1282         return 0;
1283 }
1284
1285 struct lnet_ni  *
1286 lnet_net2ni_locked(__u32 net_id, int cpt)
1287 {
1288         struct lnet_ni   *ni;
1289         struct lnet_net  *net;
1290
1291         LASSERT(cpt != LNET_LOCK_EX);
1292
1293         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1294                 if (net->net_id == net_id) {
1295                         ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1296                                         ni_netlist);
1297                         return ni;
1298                 }
1299         }
1300
1301         return NULL;
1302 }
1303
1304 struct lnet_ni *
1305 lnet_net2ni_addref(__u32 net)
1306 {
1307         struct lnet_ni *ni;
1308
1309         lnet_net_lock(0);
1310         ni = lnet_net2ni_locked(net, 0);
1311         if (ni)
1312                 lnet_ni_addref_locked(ni, 0);
1313         lnet_net_unlock(0);
1314
1315         return ni;
1316 }
1317 EXPORT_SYMBOL(lnet_net2ni_addref);
1318
1319 struct lnet_net *
1320 lnet_get_net_locked(__u32 net_id)
1321 {
1322         struct lnet_net  *net;
1323
1324         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1325                 if (net->net_id == net_id)
1326                         return net;
1327         }
1328
1329         return NULL;
1330 }
1331
1332 unsigned int
1333 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1334 {
1335         __u64           key = nid;
1336         unsigned int    val;
1337
1338         LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1339
1340         if (number == 1)
1341                 return 0;
1342
1343         val = hash_long(key, LNET_CPT_BITS);
1344         /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1345         if (val < number)
1346                 return val;
1347
1348         return (unsigned int)(key + val + (val >> 1)) % number;
1349 }
1350
1351 int
1352 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1353 {
1354         struct lnet_net *net;
1355
1356         /* must called with hold of lnet_net_lock */
1357         if (LNET_CPT_NUMBER == 1)
1358                 return 0; /* the only one */
1359
1360         /*
1361          * If NI is provided then use the CPT identified in the NI cpt
1362          * list if one exists. If one doesn't exist, then that NI is
1363          * associated with all CPTs and it follows that the net it belongs
1364          * to is implicitly associated with all CPTs, so just hash the nid
1365          * and return that.
1366          */
1367         if (ni != NULL) {
1368                 if (ni->ni_cpts != NULL)
1369                         return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1370                                                              ni->ni_ncpts)];
1371                 else
1372                         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1373         }
1374
1375         /* no NI provided so look at the net */
1376         net = lnet_get_net_locked(LNET_NIDNET(nid));
1377
1378         if (net != NULL && net->net_cpts != NULL) {
1379                 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1380         }
1381
1382         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1383 }
1384
1385 int
1386 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1387 {
1388         int     cpt;
1389         int     cpt2;
1390
1391         if (LNET_CPT_NUMBER == 1)
1392                 return 0; /* the only one */
1393
1394         cpt = lnet_net_lock_current();
1395
1396         cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1397
1398         lnet_net_unlock(cpt);
1399
1400         return cpt2;
1401 }
1402 EXPORT_SYMBOL(lnet_cpt_of_nid);
1403
1404 int
1405 lnet_islocalnet_locked(__u32 net_id)
1406 {
1407         struct lnet_net *net;
1408         bool local;
1409
1410         net = lnet_get_net_locked(net_id);
1411
1412         local = net != NULL;
1413
1414         return local;
1415 }
1416
1417 int
1418 lnet_islocalnet(__u32 net_id)
1419 {
1420         int cpt;
1421         bool local;
1422
1423         cpt = lnet_net_lock_current();
1424
1425         local = lnet_islocalnet_locked(net_id);
1426
1427         lnet_net_unlock(cpt);
1428
1429         return local;
1430 }
1431
1432 struct lnet_ni  *
1433 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
1434 {
1435         struct lnet_net  *net;
1436         struct lnet_ni   *ni;
1437
1438         LASSERT(cpt != LNET_LOCK_EX);
1439
1440         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1441                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1442                         if (ni->ni_nid == nid)
1443                                 return ni;
1444                 }
1445         }
1446
1447         return NULL;
1448 }
1449
1450 struct lnet_ni *
1451 lnet_nid2ni_addref(lnet_nid_t nid)
1452 {
1453         struct lnet_ni *ni;
1454
1455         lnet_net_lock(0);
1456         ni = lnet_nid2ni_locked(nid, 0);
1457         if (ni)
1458                 lnet_ni_addref_locked(ni, 0);
1459         lnet_net_unlock(0);
1460
1461         return ni;
1462 }
1463 EXPORT_SYMBOL(lnet_nid2ni_addref);
1464
1465 int
1466 lnet_islocalnid(lnet_nid_t nid)
1467 {
1468         struct lnet_ni  *ni;
1469         int             cpt;
1470
1471         cpt = lnet_net_lock_current();
1472         ni = lnet_nid2ni_locked(nid, cpt);
1473         lnet_net_unlock(cpt);
1474
1475         return ni != NULL;
1476 }
1477
1478 int
1479 lnet_count_acceptor_nets(void)
1480 {
1481         /* Return the # of NIs that need the acceptor. */
1482         int              count = 0;
1483         struct lnet_net  *net;
1484         int              cpt;
1485
1486         cpt = lnet_net_lock_current();
1487         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1488                 /* all socklnd type networks should have the acceptor
1489                  * thread started */
1490                 if (net->net_lnd->lnd_accept != NULL)
1491                         count++;
1492         }
1493
1494         lnet_net_unlock(cpt);
1495
1496         return count;
1497 }
1498
1499 struct lnet_ping_buffer *
1500 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1501 {
1502         struct lnet_ping_buffer *pbuf;
1503
1504         LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1505         if (pbuf) {
1506                 pbuf->pb_nnis = nnis;
1507                 atomic_set(&pbuf->pb_refcnt, 1);
1508         }
1509
1510         return pbuf;
1511 }
1512
1513 void
1514 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1515 {
1516         LASSERT(atomic_read(&pbuf->pb_refcnt) == 0);
1517         LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1518 }
1519
1520 static struct lnet_ping_buffer *
1521 lnet_ping_target_create(int nnis)
1522 {
1523         struct lnet_ping_buffer *pbuf;
1524
1525         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1526         if (pbuf == NULL) {
1527                 CERROR("Can't allocate ping source [%d]\n", nnis);
1528                 return NULL;
1529         }
1530
1531         pbuf->pb_info.pi_nnis = nnis;
1532         pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1533         pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1534         pbuf->pb_info.pi_features =
1535                 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1536
1537         return pbuf;
1538 }
1539
1540 static inline int
1541 lnet_get_net_ni_count_locked(struct lnet_net *net)
1542 {
1543         struct lnet_ni  *ni;
1544         int             count = 0;
1545
1546         list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1547                 count++;
1548
1549         return count;
1550 }
1551
1552 static inline int
1553 lnet_get_net_ni_count_pre(struct lnet_net *net)
1554 {
1555         struct lnet_ni  *ni;
1556         int             count = 0;
1557
1558         list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1559                 count++;
1560
1561         return count;
1562 }
1563
1564 static inline int
1565 lnet_get_ni_count(void)
1566 {
1567         struct lnet_ni  *ni;
1568         struct lnet_net *net;
1569         int             count = 0;
1570
1571         lnet_net_lock(0);
1572
1573         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1574                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1575                         count++;
1576         }
1577
1578         lnet_net_unlock(0);
1579
1580         return count;
1581 }
1582
1583 int
1584 lnet_get_net_count(void)
1585 {
1586         struct lnet_net *net;
1587         int count = 0;
1588
1589         lnet_net_lock(0);
1590
1591         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1592                 count++;
1593         }
1594
1595         lnet_net_unlock(0);
1596
1597         return count;
1598 }
1599
1600 void
1601 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1602 {
1603         struct lnet_ni_status *stat;
1604         int nnis;
1605         int i;
1606
1607         __swab32s(&pbuf->pb_info.pi_magic);
1608         __swab32s(&pbuf->pb_info.pi_features);
1609         __swab32s(&pbuf->pb_info.pi_pid);
1610         __swab32s(&pbuf->pb_info.pi_nnis);
1611         nnis = pbuf->pb_info.pi_nnis;
1612         if (nnis > pbuf->pb_nnis)
1613                 nnis = pbuf->pb_nnis;
1614         for (i = 0; i < nnis; i++) {
1615                 stat = &pbuf->pb_info.pi_ni[i];
1616                 __swab64s(&stat->ns_nid);
1617                 __swab32s(&stat->ns_status);
1618         }
1619 }
1620
1621 int
1622 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1623 {
1624         if (!pinfo)
1625                 return -EINVAL;
1626         if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1627                 return -EPROTO;
1628         if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1629                 return -EPROTO;
1630         /* Loopback is guaranteed to be present */
1631         if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1632                 return -ERANGE;
1633         if (LNET_NETTYP(LNET_NIDNET(LNET_PING_INFO_LONI(pinfo))) != LOLND)
1634                 return -EPROTO;
1635         return 0;
1636 }
1637
1638 static void
1639 lnet_ping_target_destroy(void)
1640 {
1641         struct lnet_net *net;
1642         struct lnet_ni  *ni;
1643
1644         lnet_net_lock(LNET_LOCK_EX);
1645
1646         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1647                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1648                         lnet_ni_lock(ni);
1649                         ni->ni_status = NULL;
1650                         lnet_ni_unlock(ni);
1651                 }
1652         }
1653
1654         lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1655         the_lnet.ln_ping_target = NULL;
1656
1657         lnet_net_unlock(LNET_LOCK_EX);
1658 }
1659
1660 static void
1661 lnet_ping_target_event_handler(struct lnet_event *event)
1662 {
1663         struct lnet_ping_buffer *pbuf = event->md.user_ptr;
1664
1665         if (event->unlinked)
1666                 lnet_ping_buffer_decref(pbuf);
1667 }
1668
1669 static int
1670 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1671                        struct lnet_handle_md *ping_mdh,
1672                        int ni_count, bool set_eq)
1673 {
1674         struct lnet_process_id id = {
1675                 .nid = LNET_NID_ANY,
1676                 .pid = LNET_PID_ANY
1677         };
1678         struct lnet_me *me;
1679         struct lnet_md md = { NULL };
1680         int rc, rc2;
1681
1682         if (set_eq) {
1683                 the_lnet.ln_ping_target_eq =
1684                         LNetEQAlloc(0, lnet_ping_target_event_handler);
1685                 if (IS_ERR(the_lnet.ln_ping_target_eq)) {
1686                         rc = PTR_ERR(the_lnet.ln_ping_target_eq);
1687                         CERROR("Can't allocate ping buffer EQ: %d\n", rc);
1688                         return rc;
1689                 }
1690         }
1691
1692         *ppbuf = lnet_ping_target_create(ni_count);
1693         if (*ppbuf == NULL) {
1694                 rc = -ENOMEM;
1695                 goto fail_free_eq;
1696         }
1697
1698         /* Ping target ME/MD */
1699         me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1700                           LNET_PROTO_PING_MATCHBITS, 0,
1701                           LNET_UNLINK, LNET_INS_AFTER);
1702         if (IS_ERR(me)) {
1703                 rc = PTR_ERR(me);
1704                 CERROR("Can't create ping target ME: %d\n", rc);
1705                 goto fail_decref_ping_buffer;
1706         }
1707
1708         /* initialize md content */
1709         md.start     = &(*ppbuf)->pb_info;
1710         md.length    = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1711         md.threshold = LNET_MD_THRESH_INF;
1712         md.max_size  = 0;
1713         md.options   = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1714                        LNET_MD_MANAGE_REMOTE;
1715         md.eq_handle = the_lnet.ln_ping_target_eq;
1716         md.user_ptr  = *ppbuf;
1717
1718         rc = LNetMDAttach(me, md, LNET_RETAIN, ping_mdh);
1719         if (rc != 0) {
1720                 CERROR("Can't attach ping target MD: %d\n", rc);
1721                 goto fail_unlink_ping_me;
1722         }
1723         lnet_ping_buffer_addref(*ppbuf);
1724
1725         return 0;
1726
1727 fail_unlink_ping_me:
1728         LNetMEUnlink(me);
1729 fail_decref_ping_buffer:
1730         LASSERT(atomic_read(&(*ppbuf)->pb_refcnt) == 1);
1731         lnet_ping_buffer_decref(*ppbuf);
1732         *ppbuf = NULL;
1733 fail_free_eq:
1734         if (set_eq) {
1735                 rc2 = LNetEQFree(the_lnet.ln_ping_target_eq);
1736                 LASSERT(rc2 == 0);
1737         }
1738         return rc;
1739 }
1740
1741 static void
1742 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1743                     struct lnet_handle_md *ping_mdh)
1744 {
1745         LNetMDUnlink(*ping_mdh);
1746         LNetInvalidateMDHandle(ping_mdh);
1747
1748         /* NB the MD could be busy; this just starts the unlink */
1749         while (atomic_read(&pbuf->pb_refcnt) > 1) {
1750                 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1751                 schedule_timeout_uninterruptible(cfs_time_seconds(1));
1752         }
1753 }
1754
1755 static void
1756 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1757 {
1758         struct lnet_ni          *ni;
1759         struct lnet_net         *net;
1760         struct lnet_ni_status *ns;
1761         int                     i;
1762         int                     rc;
1763
1764         i = 0;
1765         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1766                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1767                         LASSERT(i < pbuf->pb_nnis);
1768
1769                         ns = &pbuf->pb_info.pi_ni[i];
1770
1771                         ns->ns_nid = ni->ni_nid;
1772
1773                         lnet_ni_lock(ni);
1774                         ns->ns_status = (ni->ni_status != NULL) ?
1775                                          ni->ni_status->ns_status :
1776                                                 LNET_NI_STATUS_UP;
1777                         ni->ni_status = ns;
1778                         lnet_ni_unlock(ni);
1779
1780                         i++;
1781                 }
1782         }
1783         /*
1784          * We (ab)use the ns_status of the loopback interface to
1785          * transmit the sequence number. The first interface listed
1786          * must be the loopback interface.
1787          */
1788         rc = lnet_ping_info_validate(&pbuf->pb_info);
1789         if (rc) {
1790                 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1791                 LBUG();
1792         }
1793         LNET_PING_BUFFER_SEQNO(pbuf) =
1794                 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1795 }
1796
1797 static void
1798 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1799                         struct lnet_handle_md ping_mdh)
1800 {
1801         struct lnet_ping_buffer *old_pbuf = NULL;
1802         struct lnet_handle_md old_ping_md;
1803
1804         /* switch the NIs to point to the new ping info created */
1805         lnet_net_lock(LNET_LOCK_EX);
1806
1807         if (!the_lnet.ln_routing)
1808                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1809         if (!lnet_peer_discovery_disabled)
1810                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1811
1812         /* Ensure only known feature bits have been set. */
1813         LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1814         LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1815
1816         lnet_ping_target_install_locked(pbuf);
1817
1818         if (the_lnet.ln_ping_target) {
1819                 old_pbuf = the_lnet.ln_ping_target;
1820                 old_ping_md = the_lnet.ln_ping_target_md;
1821         }
1822         the_lnet.ln_ping_target_md = ping_mdh;
1823         the_lnet.ln_ping_target = pbuf;
1824
1825         lnet_net_unlock(LNET_LOCK_EX);
1826
1827         if (old_pbuf) {
1828                 /* unlink and free the old ping info */
1829                 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1830                 lnet_ping_buffer_decref(old_pbuf);
1831         }
1832
1833         lnet_push_update_to_peers(0);
1834 }
1835
1836 static void
1837 lnet_ping_target_fini(void)
1838 {
1839         int             rc;
1840
1841         lnet_ping_md_unlink(the_lnet.ln_ping_target,
1842                             &the_lnet.ln_ping_target_md);
1843
1844         rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1845         LASSERT(rc == 0);
1846
1847         lnet_ping_target_destroy();
1848 }
1849
1850 /* Resize the push target. */
1851 int lnet_push_target_resize(void)
1852 {
1853         struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
1854         struct lnet_md md = { NULL };
1855         struct lnet_me *me;
1856         struct lnet_handle_md mdh;
1857         struct lnet_handle_md old_mdh;
1858         struct lnet_ping_buffer *pbuf;
1859         struct lnet_ping_buffer *old_pbuf;
1860         int nnis = the_lnet.ln_push_target_nnis;
1861         int rc;
1862
1863         if (nnis <= 0) {
1864                 rc = -EINVAL;
1865                 goto fail_return;
1866         }
1867 again:
1868         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1869         if (!pbuf) {
1870                 rc = -ENOMEM;
1871                 goto fail_return;
1872         }
1873
1874         me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1875                           LNET_PROTO_PING_MATCHBITS, 0,
1876                           LNET_UNLINK, LNET_INS_AFTER);
1877
1878         if (IS_ERR(me)) {
1879                 rc = PTR_ERR(me);
1880                 CERROR("Can't create push target ME: %d\n", rc);
1881                 goto fail_decref_pbuf;
1882         }
1883
1884         /* initialize md content */
1885         md.start     = &pbuf->pb_info;
1886         md.length    = LNET_PING_INFO_SIZE(nnis);
1887         md.threshold = LNET_MD_THRESH_INF;
1888         md.max_size  = 0;
1889         md.options   = LNET_MD_OP_PUT | LNET_MD_TRUNCATE |
1890                        LNET_MD_MANAGE_REMOTE;
1891         md.user_ptr  = pbuf;
1892         md.eq_handle = the_lnet.ln_push_target_eq;
1893
1894         rc = LNetMDAttach(me, md, LNET_RETAIN, &mdh);
1895         if (rc) {
1896                 CERROR("Can't attach push MD: %d\n", rc);
1897                 goto fail_unlink_me;
1898         }
1899         lnet_ping_buffer_addref(pbuf);
1900
1901         lnet_net_lock(LNET_LOCK_EX);
1902         old_pbuf = the_lnet.ln_push_target;
1903         old_mdh = the_lnet.ln_push_target_md;
1904         the_lnet.ln_push_target = pbuf;
1905         the_lnet.ln_push_target_md = mdh;
1906         lnet_net_unlock(LNET_LOCK_EX);
1907
1908         if (old_pbuf) {
1909                 LNetMDUnlink(old_mdh);
1910                 lnet_ping_buffer_decref(old_pbuf);
1911         }
1912
1913         if (nnis < the_lnet.ln_push_target_nnis)
1914                 goto again;
1915
1916         CDEBUG(D_NET, "nnis %d success\n", nnis);
1917
1918         return 0;
1919
1920 fail_unlink_me:
1921         LNetMEUnlink(me);
1922 fail_decref_pbuf:
1923         lnet_ping_buffer_decref(pbuf);
1924 fail_return:
1925         CDEBUG(D_NET, "nnis %d error %d\n", nnis, rc);
1926         return rc;
1927 }
1928
1929 static void lnet_push_target_event_handler(struct lnet_event *ev)
1930 {
1931         struct lnet_ping_buffer *pbuf = ev->md.user_ptr;
1932
1933         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
1934                 lnet_swap_pinginfo(pbuf);
1935
1936         lnet_peer_push_event(ev);
1937         if (ev->unlinked)
1938                 lnet_ping_buffer_decref(pbuf);
1939 }
1940
1941 /* Initialize the push target. */
1942 static int lnet_push_target_init(void)
1943 {
1944         int rc;
1945
1946         if (the_lnet.ln_push_target)
1947                 return -EALREADY;
1948
1949         the_lnet.ln_push_target_eq =
1950                 LNetEQAlloc(0, lnet_push_target_event_handler);
1951         if (IS_ERR(the_lnet.ln_push_target_eq)) {
1952                 rc = PTR_ERR(the_lnet.ln_push_target_eq);
1953                 CERROR("Can't allocated push target EQ: %d\n", rc);
1954                 return rc;
1955         }
1956
1957         /* Start at the required minimum, we'll enlarge if required. */
1958         the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
1959
1960         rc = lnet_push_target_resize();
1961
1962         if (rc) {
1963                 LNetEQFree(the_lnet.ln_push_target_eq);
1964                 the_lnet.ln_push_target_eq = NULL;
1965         }
1966
1967         return rc;
1968 }
1969
1970 /* Clean up the push target. */
1971 static void lnet_push_target_fini(void)
1972 {
1973         if (!the_lnet.ln_push_target)
1974                 return;
1975
1976         /* Unlink and invalidate to prevent new references. */
1977         LNetMDUnlink(the_lnet.ln_push_target_md);
1978         LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
1979
1980         /* Wait for the unlink to complete. */
1981         while (atomic_read(&the_lnet.ln_push_target->pb_refcnt) > 1) {
1982                 CDEBUG(D_NET, "Still waiting for ping data MD to unlink\n");
1983                 schedule_timeout_uninterruptible(cfs_time_seconds(1));
1984         }
1985
1986         lnet_ping_buffer_decref(the_lnet.ln_push_target);
1987         the_lnet.ln_push_target = NULL;
1988         the_lnet.ln_push_target_nnis = 0;
1989
1990         LNetEQFree(the_lnet.ln_push_target_eq);
1991         the_lnet.ln_push_target_eq = NULL;
1992 }
1993
1994 static int
1995 lnet_ni_tq_credits(struct lnet_ni *ni)
1996 {
1997         int     credits;
1998
1999         LASSERT(ni->ni_ncpts >= 1);
2000
2001         if (ni->ni_ncpts == 1)
2002                 return ni->ni_net->net_tunables.lct_max_tx_credits;
2003
2004         credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2005         credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2006         credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2007
2008         return credits;
2009 }
2010
2011 static void
2012 lnet_ni_unlink_locked(struct lnet_ni *ni)
2013 {
2014         /* move it to zombie list and nobody can find it anymore */
2015         LASSERT(!list_empty(&ni->ni_netlist));
2016         list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2017         lnet_ni_decref_locked(ni, 0);
2018 }
2019
2020 static void
2021 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2022 {
2023         int             i;
2024         int             islo;
2025         struct lnet_ni  *ni;
2026         struct list_head *zombie_list = &net->net_ni_zombie;
2027
2028         /*
2029          * Now wait for the NIs I just nuked to show up on the zombie
2030          * list and shut them down in guaranteed thread context
2031          */
2032         i = 2;
2033         while (!list_empty(zombie_list)) {
2034                 int     *ref;
2035                 int     j;
2036
2037                 ni = list_entry(zombie_list->next,
2038                                 struct lnet_ni, ni_netlist);
2039                 list_del_init(&ni->ni_netlist);
2040                 /* the ni should be in deleting state. If it's not it's
2041                  * a bug */
2042                 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2043                 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2044                         if (*ref == 0)
2045                                 continue;
2046                         /* still busy, add it back to zombie list */
2047                         list_add(&ni->ni_netlist, zombie_list);
2048                         break;
2049                 }
2050
2051                 if (!list_empty(&ni->ni_netlist)) {
2052                         lnet_net_unlock(LNET_LOCK_EX);
2053                         ++i;
2054                         if ((i & (-i)) == i) {
2055                                 CDEBUG(D_WARNING,
2056                                        "Waiting for zombie LNI %s\n",
2057                                        libcfs_nid2str(ni->ni_nid));
2058                         }
2059                         schedule_timeout_uninterruptible(cfs_time_seconds(1));
2060                         lnet_net_lock(LNET_LOCK_EX);
2061                         continue;
2062                 }
2063
2064                 lnet_net_unlock(LNET_LOCK_EX);
2065
2066                 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2067
2068                 LASSERT(!in_interrupt());
2069                 /* Holding the mutex makes it safe for lnd_shutdown
2070                  * to call module_put(). Module unload cannot finish
2071                  * until lnet_unregister_lnd() completes, and that
2072                  * requires the mutex.
2073                  */
2074                 mutex_lock(&the_lnet.ln_lnd_mutex);
2075                 (net->net_lnd->lnd_shutdown)(ni);
2076                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2077
2078                 if (!islo)
2079                         CDEBUG(D_LNI, "Removed LNI %s\n",
2080                               libcfs_nid2str(ni->ni_nid));
2081
2082                 lnet_ni_free(ni);
2083                 i = 2;
2084                 lnet_net_lock(LNET_LOCK_EX);
2085         }
2086 }
2087
2088 /* shutdown down the NI and release refcount */
2089 static void
2090 lnet_shutdown_lndni(struct lnet_ni *ni)
2091 {
2092         int i;
2093         struct lnet_net *net = ni->ni_net;
2094
2095         lnet_net_lock(LNET_LOCK_EX);
2096         lnet_ni_lock(ni);
2097         ni->ni_state = LNET_NI_STATE_DELETING;
2098         lnet_ni_unlock(ni);
2099         lnet_ni_unlink_locked(ni);
2100         lnet_incr_dlc_seq();
2101         lnet_net_unlock(LNET_LOCK_EX);
2102
2103         /* clear messages for this NI on the lazy portal */
2104         for (i = 0; i < the_lnet.ln_nportals; i++)
2105                 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2106
2107         lnet_net_lock(LNET_LOCK_EX);
2108         lnet_clear_zombies_nis_locked(net);
2109         lnet_net_unlock(LNET_LOCK_EX);
2110 }
2111
2112 static void
2113 lnet_shutdown_lndnet(struct lnet_net *net)
2114 {
2115         struct lnet_ni *ni;
2116
2117         lnet_net_lock(LNET_LOCK_EX);
2118
2119         list_del_init(&net->net_list);
2120
2121         while (!list_empty(&net->net_ni_list)) {
2122                 ni = list_entry(net->net_ni_list.next,
2123                                 struct lnet_ni, ni_netlist);
2124                 lnet_net_unlock(LNET_LOCK_EX);
2125                 lnet_shutdown_lndni(ni);
2126                 lnet_net_lock(LNET_LOCK_EX);
2127         }
2128
2129         lnet_net_unlock(LNET_LOCK_EX);
2130
2131         /* Do peer table cleanup for this net */
2132         lnet_peer_tables_cleanup(net);
2133
2134         lnet_net_free(net);
2135 }
2136
2137 static void
2138 lnet_shutdown_lndnets(void)
2139 {
2140         struct lnet_net *net;
2141         LIST_HEAD(resend);
2142         struct lnet_msg *msg, *tmp;
2143
2144         /* NB called holding the global mutex */
2145
2146         /* All quiet on the API front */
2147         LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2148         LASSERT(the_lnet.ln_refcount == 0);
2149
2150         lnet_net_lock(LNET_LOCK_EX);
2151         the_lnet.ln_state = LNET_STATE_STOPPING;
2152
2153         /*
2154          * move the nets to the zombie list to avoid them being
2155          * picked up for new work. LONET is also included in the
2156          * Nets that will be moved to the zombie list
2157          */
2158         list_splice_init(&the_lnet.ln_nets, &the_lnet.ln_net_zombie);
2159
2160         /* Drop the cached loopback Net. */
2161         if (the_lnet.ln_loni != NULL) {
2162                 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2163                 the_lnet.ln_loni = NULL;
2164         }
2165         lnet_net_unlock(LNET_LOCK_EX);
2166
2167         /* iterate through the net zombie list and delete each net */
2168         while (!list_empty(&the_lnet.ln_net_zombie)) {
2169                 net = list_entry(the_lnet.ln_net_zombie.next,
2170                                  struct lnet_net, net_list);
2171                 lnet_shutdown_lndnet(net);
2172         }
2173
2174         spin_lock(&the_lnet.ln_msg_resend_lock);
2175         list_splice(&the_lnet.ln_msg_resend, &resend);
2176         spin_unlock(&the_lnet.ln_msg_resend_lock);
2177
2178         list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2179                 list_del_init(&msg->msg_list);
2180                 msg->msg_no_resend = true;
2181                 lnet_finalize(msg, -ECANCELED);
2182         }
2183
2184         lnet_net_lock(LNET_LOCK_EX);
2185         the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2186         lnet_net_unlock(LNET_LOCK_EX);
2187 }
2188
2189 static int
2190 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2191 {
2192         int                     rc = -EINVAL;
2193         struct lnet_tx_queue    *tq;
2194         int                     i;
2195         struct lnet_net         *net = ni->ni_net;
2196
2197         mutex_lock(&the_lnet.ln_lnd_mutex);
2198
2199         if (tun) {
2200                 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2201                 ni->ni_lnd_tunables_set = true;
2202         }
2203
2204         rc = (net->net_lnd->lnd_startup)(ni);
2205
2206         mutex_unlock(&the_lnet.ln_lnd_mutex);
2207
2208         if (rc != 0) {
2209                 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2210                                    rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2211                 goto failed0;
2212         }
2213
2214         lnet_ni_lock(ni);
2215         ni->ni_state = LNET_NI_STATE_ACTIVE;
2216         lnet_ni_unlock(ni);
2217
2218         /* We keep a reference on the loopback net through the loopback NI */
2219         if (net->net_lnd->lnd_type == LOLND) {
2220                 lnet_ni_addref(ni);
2221                 LASSERT(the_lnet.ln_loni == NULL);
2222                 the_lnet.ln_loni = ni;
2223                 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2224                 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2225                 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2226                 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2227                 return 0;
2228         }
2229
2230         if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2231             ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2232                 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2233                                    libcfs_lnd2str(net->net_lnd->lnd_type),
2234                                    ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2235                                         "" : "per-peer ");
2236                 /* shutdown the NI since if we get here then it must've already
2237                  * been started
2238                  */
2239                 lnet_shutdown_lndni(ni);
2240                 return -EINVAL;
2241         }
2242
2243         cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2244                 tq->tq_credits_min =
2245                 tq->tq_credits_max =
2246                 tq->tq_credits = lnet_ni_tq_credits(ni);
2247         }
2248
2249         atomic_set(&ni->ni_tx_credits,
2250                    lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2251         atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2252
2253         CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2254                 libcfs_nid2str(ni->ni_nid),
2255                 ni->ni_net->net_tunables.lct_peer_tx_credits,
2256                 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2257                 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2258                 ni->ni_net->net_tunables.lct_peer_timeout);
2259
2260         return 0;
2261 failed0:
2262         lnet_ni_free(ni);
2263         return rc;
2264 }
2265
2266 static int
2267 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2268 {
2269         struct lnet_ni *ni;
2270         struct lnet_net *net_l = NULL;
2271         LIST_HEAD(local_ni_list);
2272         int rc;
2273         int ni_count = 0;
2274         __u32 lnd_type;
2275         const struct lnet_lnd  *lnd;
2276         int peer_timeout =
2277                 net->net_tunables.lct_peer_timeout;
2278         int maxtxcredits =
2279                 net->net_tunables.lct_max_tx_credits;
2280         int peerrtrcredits =
2281                 net->net_tunables.lct_peer_rtr_credits;
2282
2283         /*
2284          * make sure that this net is unique. If it isn't then
2285          * we are adding interfaces to an already existing network, and
2286          * 'net' is just a convenient way to pass in the list.
2287          * if it is unique we need to find the LND and load it if
2288          * necessary.
2289          */
2290         if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2291                 lnd_type = LNET_NETTYP(net->net_id);
2292
2293                 mutex_lock(&the_lnet.ln_lnd_mutex);
2294                 lnd = lnet_find_lnd_by_type(lnd_type);
2295
2296                 if (lnd == NULL) {
2297                         mutex_unlock(&the_lnet.ln_lnd_mutex);
2298                         rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2299                         mutex_lock(&the_lnet.ln_lnd_mutex);
2300
2301                         lnd = lnet_find_lnd_by_type(lnd_type);
2302                         if (lnd == NULL) {
2303                                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2304                                 CERROR("Can't load LND %s, module %s, rc=%d\n",
2305                                 libcfs_lnd2str(lnd_type),
2306                                 libcfs_lnd2modname(lnd_type), rc);
2307 #ifndef HAVE_MODULE_LOADING_SUPPORT
2308                                 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2309                                                 "compiled with kernel module "
2310                                                 "loading support.");
2311 #endif
2312                                 rc = -EINVAL;
2313                                 goto failed0;
2314                         }
2315                 }
2316
2317                 net->net_lnd = lnd;
2318
2319                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2320
2321                 net_l = net;
2322         }
2323
2324         /*
2325          * net_l: if the network being added is unique then net_l
2326          *        will point to that network
2327          *        if the network being added is not unique then
2328          *        net_l points to the existing network.
2329          *
2330          * When we enter the loop below, we'll pick NIs off he
2331          * network beign added and start them up, then add them to
2332          * a local ni list. Once we've successfully started all
2333          * the NIs then we join the local NI list (of started up
2334          * networks) with the net_l->net_ni_list, which should
2335          * point to the correct network to add the new ni list to
2336          *
2337          * If any of the new NIs fail to start up, then we want to
2338          * iterate through the local ni list, which should include
2339          * any NIs which were successfully started up, and shut
2340          * them down.
2341          *
2342          * After than we want to delete the network being added,
2343          * to avoid a memory leak.
2344          */
2345
2346         /*
2347          * When a network uses TCP bonding then all its interfaces
2348          * must be specified when the network is first defined: the
2349          * TCP bonding code doesn't allow for interfaces to be added
2350          * or removed.
2351          */
2352         if (net_l != net && net_l != NULL && use_tcp_bonding &&
2353             LNET_NETTYP(net_l->net_id) == SOCKLND) {
2354                 rc = -EINVAL;
2355                 goto failed0;
2356         }
2357
2358         while (!list_empty(&net->net_ni_added)) {
2359                 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2360                                 ni_netlist);
2361                 list_del_init(&ni->ni_netlist);
2362
2363                 /* make sure that the the NI we're about to start
2364                  * up is actually unique. if it's not fail. */
2365                 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2366                                         ni->ni_interfaces[0])) {
2367                         rc = -EEXIST;
2368                         goto failed1;
2369                 }
2370
2371                 /* adjust the pointer the parent network, just in case it
2372                  * the net is a duplicate */
2373                 ni->ni_net = net_l;
2374
2375                 rc = lnet_startup_lndni(ni, tun);
2376
2377                 if (rc < 0)
2378                         goto failed1;
2379
2380                 lnet_ni_addref(ni);
2381                 list_add_tail(&ni->ni_netlist, &local_ni_list);
2382
2383                 ni_count++;
2384         }
2385
2386         lnet_net_lock(LNET_LOCK_EX);
2387         list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2388         lnet_incr_dlc_seq();
2389         lnet_net_unlock(LNET_LOCK_EX);
2390
2391         /* if the network is not unique then we don't want to keep
2392          * it around after we're done. Free it. Otherwise add that
2393          * net to the global the_lnet.ln_nets */
2394         if (net_l != net && net_l != NULL) {
2395                 /*
2396                  * TODO - note. currently the tunables can not be updated
2397                  * once added
2398                  */
2399                 lnet_net_free(net);
2400         } else {
2401                 /*
2402                  * restore tunables after it has been overwitten by the
2403                  * lnd
2404                  */
2405                 if (peer_timeout != -1)
2406                         net->net_tunables.lct_peer_timeout = peer_timeout;
2407                 if (maxtxcredits != -1)
2408                         net->net_tunables.lct_max_tx_credits = maxtxcredits;
2409                 if (peerrtrcredits != -1)
2410                         net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2411
2412                 lnet_net_lock(LNET_LOCK_EX);
2413                 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2414                 lnet_net_unlock(LNET_LOCK_EX);
2415         }
2416
2417         /* update net count */
2418         lnet_current_net_count = lnet_get_net_count();
2419
2420         return ni_count;
2421
2422 failed1:
2423         /*
2424          * shutdown the new NIs that are being started up
2425          * free the NET being started
2426          */
2427         while (!list_empty(&local_ni_list)) {
2428                 ni = list_entry(local_ni_list.next, struct lnet_ni,
2429                                 ni_netlist);
2430
2431                 lnet_shutdown_lndni(ni);
2432         }
2433
2434 failed0:
2435         lnet_net_free(net);
2436
2437         return rc;
2438 }
2439
2440 static int
2441 lnet_startup_lndnets(struct list_head *netlist)
2442 {
2443         struct lnet_net         *net;
2444         int                     rc;
2445         int                     ni_count = 0;
2446
2447         /*
2448          * Change to running state before bringing up the LNDs. This
2449          * allows lnet_shutdown_lndnets() to assert that we've passed
2450          * through here.
2451          */
2452         lnet_net_lock(LNET_LOCK_EX);
2453         the_lnet.ln_state = LNET_STATE_RUNNING;
2454         lnet_net_unlock(LNET_LOCK_EX);
2455
2456         while (!list_empty(netlist)) {
2457                 net = list_entry(netlist->next, struct lnet_net, net_list);
2458                 list_del_init(&net->net_list);
2459
2460                 rc = lnet_startup_lndnet(net, NULL);
2461
2462                 if (rc < 0)
2463                         goto failed;
2464
2465                 ni_count += rc;
2466         }
2467
2468         return ni_count;
2469 failed:
2470         lnet_shutdown_lndnets();
2471
2472         return rc;
2473 }
2474
2475 /**
2476  * Initialize LNet library.
2477  *
2478  * Automatically called at module loading time. Caller has to call
2479  * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2480  * latter returned 0. It must be called exactly once.
2481  *
2482  * \retval 0 on success
2483  * \retval -ve on failures.
2484  */
2485 int lnet_lib_init(void)
2486 {
2487         int rc;
2488
2489         lnet_assert_wire_constants();
2490
2491         /* refer to global cfs_cpt_table for now */
2492         the_lnet.ln_cpt_table = cfs_cpt_tab;
2493         the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_tab);
2494
2495         LASSERT(the_lnet.ln_cpt_number > 0);
2496         if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2497                 /* we are under risk of consuming all lh_cookie */
2498                 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2499                        "please change setting of CPT-table and retry\n",
2500                        the_lnet.ln_cpt_number, LNET_CPT_MAX);
2501                 return -E2BIG;
2502         }
2503
2504         while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2505                 the_lnet.ln_cpt_bits++;
2506
2507         rc = lnet_create_locks();
2508         if (rc != 0) {
2509                 CERROR("Can't create LNet global locks: %d\n", rc);
2510                 return rc;
2511         }
2512
2513         the_lnet.ln_refcount = 0;
2514         INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2515         INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2516
2517         /* The hash table size is the number of bits it takes to express the set
2518          * ln_num_routes, minus 1 (better to under estimate than over so we
2519          * don't waste memory). */
2520         if (rnet_htable_size <= 0)
2521                 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2522         else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2523                 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2524         the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2525                                            order_base_2(rnet_htable_size) - 1);
2526
2527         /* All LNDs apart from the LOLND are in separate modules.  They
2528          * register themselves when their module loads, and unregister
2529          * themselves when their module is unloaded. */
2530         lnet_register_lnd(&the_lolnd);
2531         return 0;
2532 }
2533
2534 /**
2535  * Finalize LNet library.
2536  *
2537  * \pre lnet_lib_init() called with success.
2538  * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2539  *
2540  * As this happens at module-unload, all lnds must already be unloaded,
2541  * so they must already be unregistered.
2542  */
2543 void lnet_lib_exit(void)
2544 {
2545         int i;
2546
2547         LASSERT(the_lnet.ln_refcount == 0);
2548         lnet_unregister_lnd(&the_lolnd);
2549         for (i = 0; i < NUM_LNDS; i++)
2550                 LASSERT(!the_lnet.ln_lnds[i]);
2551         lnet_destroy_locks();
2552 }
2553
2554 /**
2555  * Set LNet PID and start LNet interfaces, routing, and forwarding.
2556  *
2557  * Users must call this function at least once before any other functions.
2558  * For each successful call there must be a corresponding call to
2559  * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2560  * ignored.
2561  *
2562  * The PID used by LNet may be different from the one requested.
2563  * See LNetGetId().
2564  *
2565  * \param requested_pid PID requested by the caller.
2566  *
2567  * \return >= 0 on success, and < 0 error code on failures.
2568  */
2569 int
2570 LNetNIInit(lnet_pid_t requested_pid)
2571 {
2572         int                     im_a_router = 0;
2573         int                     rc;
2574         int                     ni_count;
2575         struct lnet_ping_buffer *pbuf;
2576         struct lnet_handle_md   ping_mdh;
2577         LIST_HEAD(net_head);
2578         struct lnet_net         *net;
2579
2580         mutex_lock(&the_lnet.ln_api_mutex);
2581
2582         CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2583
2584         if (the_lnet.ln_refcount > 0) {
2585                 rc = the_lnet.ln_refcount++;
2586                 mutex_unlock(&the_lnet.ln_api_mutex);
2587                 return rc;
2588         }
2589
2590         rc = lnet_prepare(requested_pid);
2591         if (rc != 0) {
2592                 mutex_unlock(&the_lnet.ln_api_mutex);
2593                 return rc;
2594         }
2595
2596         /* create a network for Loopback network */
2597         net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2598         if (net == NULL) {
2599                 rc = -ENOMEM;
2600                 goto err_empty_list;
2601         }
2602
2603         /* Add in the loopback NI */
2604         if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2605                 rc = -ENOMEM;
2606                 goto err_empty_list;
2607         }
2608
2609         /* If LNet is being initialized via DLC it is possible
2610          * that the user requests not to load module parameters (ones which
2611          * are supported by DLC) on initialization.  Therefore, make sure not
2612          * to load networks, routes and forwarding from module parameters
2613          * in this case.  On cleanup in case of failure only clean up
2614          * routes if it has been loaded */
2615         if (!the_lnet.ln_nis_from_mod_params) {
2616                 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
2617                                          use_tcp_bonding);
2618                 if (rc < 0)
2619                         goto err_empty_list;
2620         }
2621
2622         ni_count = lnet_startup_lndnets(&net_head);
2623         if (ni_count < 0) {
2624                 rc = ni_count;
2625                 goto err_empty_list;
2626         }
2627
2628         if (!the_lnet.ln_nis_from_mod_params) {
2629                 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2630                 if (rc != 0)
2631                         goto err_shutdown_lndnis;
2632
2633                 rc = lnet_rtrpools_alloc(im_a_router);
2634                 if (rc != 0)
2635                         goto err_destroy_routes;
2636         }
2637
2638         rc = lnet_acceptor_start();
2639         if (rc != 0)
2640                 goto err_destroy_routes;
2641
2642         the_lnet.ln_refcount = 1;
2643         /* Now I may use my own API functions... */
2644
2645         rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2646         if (rc != 0)
2647                 goto err_acceptor_stop;
2648
2649         lnet_ping_target_update(pbuf, ping_mdh);
2650
2651         the_lnet.ln_mt_eq = LNetEQAlloc(0, lnet_mt_event_handler);
2652         if (IS_ERR(the_lnet.ln_mt_eq)) {
2653                 rc = PTR_ERR(the_lnet.ln_mt_eq);
2654                 CERROR("Can't allocate monitor thread EQ: %d\n", rc);
2655                 goto err_stop_ping;
2656         }
2657
2658         rc = lnet_push_target_init();
2659         if (rc != 0)
2660                 goto err_stop_ping;
2661
2662         rc = lnet_peer_discovery_start();
2663         if (rc != 0)
2664                 goto err_destroy_push_target;
2665
2666         rc = lnet_monitor_thr_start();
2667         if (rc != 0)
2668                 goto err_stop_discovery_thr;
2669
2670         lnet_fault_init();
2671         lnet_router_debugfs_init();
2672
2673         mutex_unlock(&the_lnet.ln_api_mutex);
2674
2675         complete_all(&the_lnet.ln_started);
2676
2677         /* wait for all routers to start */
2678         lnet_wait_router_start();
2679
2680         return 0;
2681
2682 err_stop_discovery_thr:
2683         lnet_peer_discovery_stop();
2684 err_destroy_push_target:
2685         lnet_push_target_fini();
2686 err_stop_ping:
2687         lnet_ping_target_fini();
2688 err_acceptor_stop:
2689         the_lnet.ln_refcount = 0;
2690         lnet_acceptor_stop();
2691 err_destroy_routes:
2692         if (!the_lnet.ln_nis_from_mod_params)
2693                 lnet_destroy_routes();
2694 err_shutdown_lndnis:
2695         lnet_shutdown_lndnets();
2696 err_empty_list:
2697         lnet_unprepare();
2698         LASSERT(rc < 0);
2699         mutex_unlock(&the_lnet.ln_api_mutex);
2700         while (!list_empty(&net_head)) {
2701                 struct lnet_net *net;
2702
2703                 net = list_entry(net_head.next, struct lnet_net, net_list);
2704                 list_del_init(&net->net_list);
2705                 lnet_net_free(net);
2706         }
2707         return rc;
2708 }
2709 EXPORT_SYMBOL(LNetNIInit);
2710
2711 /**
2712  * Stop LNet interfaces, routing, and forwarding.
2713  *
2714  * Users must call this function once for each successful call to LNetNIInit().
2715  * Once the LNetNIFini() operation has been started, the results of pending
2716  * API operations are undefined.
2717  *
2718  * \return always 0 for current implementation.
2719  */
2720 int
2721 LNetNIFini(void)
2722 {
2723         mutex_lock(&the_lnet.ln_api_mutex);
2724
2725         LASSERT(the_lnet.ln_refcount > 0);
2726
2727         if (the_lnet.ln_refcount != 1) {
2728                 the_lnet.ln_refcount--;
2729         } else {
2730                 LASSERT(!the_lnet.ln_niinit_self);
2731
2732                 lnet_fault_fini();
2733
2734                 lnet_router_debugfs_fini();
2735                 lnet_monitor_thr_stop();
2736                 lnet_peer_discovery_stop();
2737                 lnet_push_target_fini();
2738                 lnet_ping_target_fini();
2739
2740                 /* Teardown fns that use my own API functions BEFORE here */
2741                 the_lnet.ln_refcount = 0;
2742
2743                 lnet_acceptor_stop();
2744                 lnet_destroy_routes();
2745                 lnet_shutdown_lndnets();
2746                 lnet_unprepare();
2747         }
2748
2749         mutex_unlock(&the_lnet.ln_api_mutex);
2750         return 0;
2751 }
2752 EXPORT_SYMBOL(LNetNIFini);
2753
2754 /**
2755  * Grabs the ni data from the ni structure and fills the out
2756  * parameters
2757  *
2758  * \param[in] ni network        interface structure
2759  * \param[out] cfg_ni           NI config information
2760  * \param[out] tun              network and LND tunables
2761  */
2762 static void
2763 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2764                    struct lnet_ioctl_config_lnd_tunables *tun,
2765                    struct lnet_ioctl_element_stats *stats,
2766                    __u32 tun_size)
2767 {
2768         size_t min_size = 0;
2769         int i;
2770
2771         if (!ni || !cfg_ni || !tun)
2772                 return;
2773
2774         if (ni->ni_interfaces[0] != NULL) {
2775                 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2776                         if (ni->ni_interfaces[i] != NULL) {
2777                                 strncpy(cfg_ni->lic_ni_intf[i],
2778                                         ni->ni_interfaces[i],
2779                                         sizeof(cfg_ni->lic_ni_intf[i]));
2780                         }
2781                 }
2782         }
2783
2784         cfg_ni->lic_nid = ni->ni_nid;
2785         if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2786                 cfg_ni->lic_status = LNET_NI_STATUS_UP;
2787         else
2788                 cfg_ni->lic_status = ni->ni_status->ns_status;
2789         cfg_ni->lic_tcp_bonding = use_tcp_bonding;
2790         cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
2791
2792         memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
2793
2794         if (stats) {
2795                 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
2796                                                        LNET_STATS_TYPE_SEND);
2797                 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
2798                                                        LNET_STATS_TYPE_RECV);
2799                 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
2800                                                        LNET_STATS_TYPE_DROP);
2801         }
2802
2803         /*
2804          * tun->lt_tun will always be present, but in order to be
2805          * backwards compatible, we need to deal with the cases when
2806          * tun->lt_tun is smaller than what the kernel has, because it
2807          * comes from an older version of a userspace program, then we'll
2808          * need to copy as much information as we have available space.
2809          */
2810         min_size = tun_size - sizeof(tun->lt_cmn);
2811         memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
2812
2813         /* copy over the cpts */
2814         if (ni->ni_ncpts == LNET_CPT_NUMBER &&
2815             ni->ni_cpts == NULL)  {
2816                 for (i = 0; i < ni->ni_ncpts; i++)
2817                         cfg_ni->lic_cpts[i] = i;
2818         } else {
2819                 for (i = 0;
2820                      ni->ni_cpts != NULL && i < ni->ni_ncpts &&
2821                      i < LNET_MAX_SHOW_NUM_CPT;
2822                      i++)
2823                         cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2824         }
2825         cfg_ni->lic_ncpts = ni->ni_ncpts;
2826 }
2827
2828 /**
2829  * NOTE: This is a legacy function left in the code to be backwards
2830  * compatible with older userspace programs. It should eventually be
2831  * removed.
2832  *
2833  * Grabs the ni data from the ni structure and fills the out
2834  * parameters
2835  *
2836  * \param[in] ni network        interface structure
2837  * \param[out] config           config information
2838  */
2839 static void
2840 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
2841                          struct lnet_ioctl_config_data *config)
2842 {
2843         struct lnet_ioctl_net_config *net_config;
2844         struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
2845         size_t min_size, tunable_size = 0;
2846         int i;
2847
2848         if (!ni || !config)
2849                 return;
2850
2851         net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
2852         if (!net_config)
2853                 return;
2854
2855         BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
2856                      ARRAY_SIZE(net_config->ni_interfaces));
2857
2858         for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2859                 if (!ni->ni_interfaces[i])
2860                         break;
2861
2862                 strncpy(net_config->ni_interfaces[i],
2863                         ni->ni_interfaces[i],
2864                         sizeof(net_config->ni_interfaces[i]));
2865         }
2866
2867         config->cfg_nid = ni->ni_nid;
2868         config->cfg_config_u.cfg_net.net_peer_timeout =
2869                 ni->ni_net->net_tunables.lct_peer_timeout;
2870         config->cfg_config_u.cfg_net.net_max_tx_credits =
2871                 ni->ni_net->net_tunables.lct_max_tx_credits;
2872         config->cfg_config_u.cfg_net.net_peer_tx_credits =
2873                 ni->ni_net->net_tunables.lct_peer_tx_credits;
2874         config->cfg_config_u.cfg_net.net_peer_rtr_credits =
2875                 ni->ni_net->net_tunables.lct_peer_rtr_credits;
2876
2877         if (LNET_NETTYP(LNET_NIDNET(ni->ni_nid)) == LOLND)
2878                 net_config->ni_status = LNET_NI_STATUS_UP;
2879         else
2880                 net_config->ni_status = ni->ni_status->ns_status;
2881
2882         if (ni->ni_cpts) {
2883                 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
2884
2885                 for (i = 0; i < num_cpts; i++)
2886                         net_config->ni_cpts[i] = ni->ni_cpts[i];
2887
2888                 config->cfg_ncpts = num_cpts;
2889         }
2890
2891         /*
2892          * See if user land tools sent in a newer and larger version
2893          * of struct lnet_tunables than what the kernel uses.
2894          */
2895         min_size = sizeof(*config) + sizeof(*net_config);
2896
2897         if (config->cfg_hdr.ioc_len > min_size)
2898                 tunable_size = config->cfg_hdr.ioc_len - min_size;
2899
2900         /* Don't copy too much data to user space */
2901         min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
2902         lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
2903
2904         if (lnd_cfg && min_size) {
2905                 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
2906                 config->cfg_config_u.cfg_net.net_interface_count = 1;
2907
2908                 /* Tell user land that kernel side has less data */
2909                 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
2910                         min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
2911                         config->cfg_hdr.ioc_len -= min_size;
2912                 }
2913         }
2914 }
2915
2916 struct lnet_ni *
2917 lnet_get_ni_idx_locked(int idx)
2918 {
2919         struct lnet_ni          *ni;
2920         struct lnet_net         *net;
2921
2922         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2923                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2924                         if (idx-- == 0)
2925                                 return ni;
2926                 }
2927         }
2928
2929         return NULL;
2930 }
2931
2932 struct lnet_ni *
2933 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2934 {
2935         struct lnet_ni          *ni;
2936         struct lnet_net         *net = mynet;
2937
2938         /*
2939          * It is possible that the net has been cleaned out while there is
2940          * a message being sent. This function accessed the net without
2941          * checking if the list is empty
2942          */
2943         if (prev == NULL) {
2944                 if (net == NULL)
2945                         net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2946                                         net_list);
2947                 if (list_empty(&net->net_ni_list))
2948                         return NULL;
2949                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2950                                 ni_netlist);
2951
2952                 return ni;
2953         }
2954
2955         if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
2956                 /* if you reached the end of the ni list and the net is
2957                  * specified, then there are no more nis in that net */
2958                 if (net != NULL)
2959                         return NULL;
2960
2961                 /* we reached the end of this net ni list. move to the
2962                  * next net */
2963                 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
2964                         /* no more nets and no more NIs. */
2965                         return NULL;
2966
2967                 /* get the next net */
2968                 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
2969                                  net_list);
2970                 if (list_empty(&net->net_ni_list))
2971                         return NULL;
2972                 /* get the ni on it */
2973                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2974                                 ni_netlist);
2975
2976                 return ni;
2977         }
2978
2979         if (list_empty(&prev->ni_netlist))
2980                 return NULL;
2981
2982         /* there are more nis left */
2983         ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
2984
2985         return ni;
2986 }
2987
2988 int
2989 lnet_get_net_config(struct lnet_ioctl_config_data *config)
2990 {
2991         struct lnet_ni *ni;
2992         int cpt;
2993         int rc = -ENOENT;
2994         int idx = config->cfg_count;
2995
2996         cpt = lnet_net_lock_current();
2997
2998         ni = lnet_get_ni_idx_locked(idx);
2999
3000         if (ni != NULL) {
3001                 rc = 0;
3002                 lnet_ni_lock(ni);
3003                 lnet_fill_ni_info_legacy(ni, config);
3004                 lnet_ni_unlock(ni);
3005         }
3006
3007         lnet_net_unlock(cpt);
3008         return rc;
3009 }
3010
3011 int
3012 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3013                    struct lnet_ioctl_config_lnd_tunables *tun,
3014                    struct lnet_ioctl_element_stats *stats,
3015                    __u32 tun_size)
3016 {
3017         struct lnet_ni          *ni;
3018         int                     cpt;
3019         int                     rc = -ENOENT;
3020
3021         if (!cfg_ni || !tun || !stats)
3022                 return -EINVAL;
3023
3024         cpt = lnet_net_lock_current();
3025
3026         ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3027
3028         if (ni) {
3029                 rc = 0;
3030                 lnet_ni_lock(ni);
3031                 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3032                 lnet_ni_unlock(ni);
3033         }
3034
3035         lnet_net_unlock(cpt);
3036         return rc;
3037 }
3038
3039 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3040 {
3041         struct lnet_ni *ni;
3042         int cpt;
3043         int rc = -ENOENT;
3044
3045         if (!msg_stats)
3046                 return -EINVAL;
3047
3048         cpt = lnet_net_lock_current();
3049
3050         ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3051
3052         if (ni) {
3053                 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3054                 rc = 0;
3055         }
3056
3057         lnet_net_unlock(cpt);
3058
3059         return rc;
3060 }
3061
3062 static int lnet_add_net_common(struct lnet_net *net,
3063                                struct lnet_ioctl_config_lnd_tunables *tun)
3064 {
3065         __u32                   net_id;
3066         struct lnet_ping_buffer *pbuf;
3067         struct lnet_handle_md   ping_mdh;
3068         int                     rc;
3069         struct lnet_remotenet *rnet;
3070         int                     net_ni_count;
3071         int                     num_acceptor_nets;
3072
3073         lnet_net_lock(LNET_LOCK_EX);
3074         rnet = lnet_find_rnet_locked(net->net_id);
3075         lnet_net_unlock(LNET_LOCK_EX);
3076         /*
3077          * make sure that the net added doesn't invalidate the current
3078          * configuration LNet is keeping
3079          */
3080         if (rnet) {
3081                 CERROR("Adding net %s will invalidate routing configuration\n",
3082                        libcfs_net2str(net->net_id));
3083                 lnet_net_free(net);
3084                 return -EUSERS;
3085         }
3086
3087         /*
3088          * make sure you calculate the correct number of slots in the ping
3089          * buffer. Since the ping info is a flattened list of all the NIs,
3090          * we should allocate enough slots to accomodate the number of NIs
3091          * which will be added.
3092          *
3093          * since ni hasn't been configured yet, use
3094          * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3095          */
3096         net_ni_count = lnet_get_net_ni_count_pre(net);
3097
3098         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3099                                     net_ni_count + lnet_get_ni_count(),
3100                                     false);
3101         if (rc < 0) {
3102                 lnet_net_free(net);
3103                 return rc;
3104         }
3105
3106         if (tun)
3107                 memcpy(&net->net_tunables,
3108                        &tun->lt_cmn, sizeof(net->net_tunables));
3109         else
3110                 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3111
3112         /*
3113          * before starting this network get a count of the current TCP
3114          * networks which require the acceptor thread running. If that
3115          * count is == 0 before we start up this network, then we'd want to
3116          * start up the acceptor thread after starting up this network
3117          */
3118         num_acceptor_nets = lnet_count_acceptor_nets();
3119
3120         net_id = net->net_id;
3121
3122         rc = lnet_startup_lndnet(net,
3123                                  (tun) ? &tun->lt_tun : NULL);
3124         if (rc < 0)
3125                 goto failed;
3126
3127         lnet_net_lock(LNET_LOCK_EX);
3128         net = lnet_get_net_locked(net_id);
3129         lnet_net_unlock(LNET_LOCK_EX);
3130
3131         LASSERT(net);
3132
3133         /*
3134          * Start the acceptor thread if this is the first network
3135          * being added that requires the thread.
3136          */
3137         if (net->net_lnd->lnd_accept && num_acceptor_nets == 0) {
3138                 rc = lnet_acceptor_start();
3139                 if (rc < 0) {
3140                         /* shutdown the net that we just started */
3141                         CERROR("Failed to start up acceptor thread\n");
3142                         lnet_shutdown_lndnet(net);
3143                         goto failed;
3144                 }
3145         }
3146
3147         lnet_net_lock(LNET_LOCK_EX);
3148         lnet_peer_net_added(net);
3149         lnet_net_unlock(LNET_LOCK_EX);
3150
3151         lnet_ping_target_update(pbuf, ping_mdh);
3152
3153         return 0;
3154
3155 failed:
3156         lnet_ping_md_unlink(pbuf, &ping_mdh);
3157         lnet_ping_buffer_decref(pbuf);
3158         return rc;
3159 }
3160
3161 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3162                                       struct lnet_ioctl_config_lnd_tunables *tun)
3163 {
3164         struct lnet_net *net;
3165         char *nets;
3166         int rc;
3167         LIST_HEAD(net_head);
3168
3169         rc = lnet_parse_ip2nets(&nets, ip2nets);
3170         if (rc < 0)
3171                 return rc;
3172
3173         rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3174         if (rc < 0)
3175                 return rc;
3176
3177         mutex_lock(&the_lnet.ln_api_mutex);
3178         while (!list_empty(&net_head)) {
3179                 net = list_entry(net_head.next, struct lnet_net, net_list);
3180                 list_del_init(&net->net_list);
3181                 rc = lnet_add_net_common(net, tun);
3182                 if (rc < 0)
3183                         goto out;
3184         }
3185
3186 out:
3187         mutex_unlock(&the_lnet.ln_api_mutex);
3188
3189         while (!list_empty(&net_head)) {
3190                 net = list_entry(net_head.next, struct lnet_net, net_list);
3191                 list_del_init(&net->net_list);
3192                 lnet_net_free(net);
3193         }
3194         return rc;
3195 }
3196
3197 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3198 {
3199         struct lnet_net *net;
3200         struct lnet_ni *ni;
3201         struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3202         int rc, i;
3203         __u32 net_id, lnd_type;
3204
3205         /* get the tunables if they are available */
3206         if (conf->lic_cfg_hdr.ioc_len >=
3207             sizeof(*conf) + sizeof(*tun))
3208                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3209                         conf->lic_bulk;
3210
3211         /* handle legacy ip2nets from DLC */
3212         if (conf->lic_legacy_ip2nets[0] != '\0')
3213                 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3214                                                   tun);
3215
3216         net_id = LNET_NIDNET(conf->lic_nid);
3217         lnd_type = LNET_NETTYP(net_id);
3218
3219         if (!libcfs_isknown_lnd(lnd_type)) {
3220                 CERROR("No valid net and lnd information provided\n");
3221                 return -EINVAL;
3222         }
3223
3224         net = lnet_net_alloc(net_id, NULL);
3225         if (!net)
3226                 return -ENOMEM;
3227
3228         for (i = 0; i < conf->lic_ncpts; i++) {
3229                 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3230                         return -EINVAL;
3231         }
3232
3233         ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3234                                        conf->lic_ni_intf[0]);
3235         if (!ni)
3236                 return -ENOMEM;
3237
3238         mutex_lock(&the_lnet.ln_api_mutex);
3239
3240         rc = lnet_add_net_common(net, tun);
3241
3242         mutex_unlock(&the_lnet.ln_api_mutex);
3243
3244         return rc;
3245 }
3246
3247 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3248 {
3249         struct lnet_net  *net;
3250         struct lnet_ni *ni;
3251         __u32 net_id = LNET_NIDNET(conf->lic_nid);
3252         struct lnet_ping_buffer *pbuf;
3253         struct lnet_handle_md  ping_mdh;
3254         int               rc;
3255         int               net_count;
3256         __u32             addr;
3257
3258         /* don't allow userspace to shutdown the LOLND */
3259         if (LNET_NETTYP(net_id) == LOLND)
3260                 return -EINVAL;
3261
3262         mutex_lock(&the_lnet.ln_api_mutex);
3263
3264         lnet_net_lock(0);
3265
3266         net = lnet_get_net_locked(net_id);
3267         if (!net) {
3268                 CERROR("net %s not found\n",
3269                        libcfs_net2str(net_id));
3270                 rc = -ENOENT;
3271                 goto unlock_net;
3272         }
3273
3274         addr = LNET_NIDADDR(conf->lic_nid);
3275         if (addr == 0) {
3276                 /* remove the entire net */
3277                 net_count = lnet_get_net_ni_count_locked(net);
3278
3279                 lnet_net_unlock(0);
3280
3281                 /* create and link a new ping info, before removing the old one */
3282                 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3283                                         lnet_get_ni_count() - net_count,
3284                                         false);
3285                 if (rc != 0)
3286                         goto unlock_api_mutex;
3287
3288                 lnet_shutdown_lndnet(net);
3289
3290                 if (lnet_count_acceptor_nets() == 0)
3291                         lnet_acceptor_stop();
3292
3293                 lnet_ping_target_update(pbuf, ping_mdh);
3294
3295                 goto unlock_api_mutex;
3296         }
3297
3298         ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3299         if (!ni) {
3300                 CERROR("nid %s not found\n",
3301                        libcfs_nid2str(conf->lic_nid));
3302                 rc = -ENOENT;
3303                 goto unlock_net;
3304         }
3305
3306         net_count = lnet_get_net_ni_count_locked(net);
3307
3308         lnet_net_unlock(0);
3309
3310         /* create and link a new ping info, before removing the old one */
3311         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3312                                   lnet_get_ni_count() - 1, false);
3313         if (rc != 0)
3314                 goto unlock_api_mutex;
3315
3316         lnet_shutdown_lndni(ni);
3317
3318         if (lnet_count_acceptor_nets() == 0)
3319                 lnet_acceptor_stop();
3320
3321         lnet_ping_target_update(pbuf, ping_mdh);
3322
3323         /* check if the net is empty and remove it if it is */
3324         if (net_count == 1)
3325                 lnet_shutdown_lndnet(net);
3326
3327         goto unlock_api_mutex;
3328
3329 unlock_net:
3330         lnet_net_unlock(0);
3331 unlock_api_mutex:
3332         mutex_unlock(&the_lnet.ln_api_mutex);
3333
3334         return rc;
3335 }
3336
3337 /*
3338  * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3339  * They are only expected to be called for unique networks.
3340  * That can be as a result of older DLC library
3341  * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3342  */
3343 int
3344 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3345 {
3346         struct lnet_net *net;
3347         LIST_HEAD(net_head);
3348         int rc;
3349         struct lnet_ioctl_config_lnd_tunables tun;
3350         char *nets = conf->cfg_config_u.cfg_net.net_intf;
3351
3352         /* Create a net/ni structures for the network string */
3353         rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
3354         if (rc <= 0)
3355                 return rc == 0 ? -EINVAL : rc;
3356
3357         mutex_lock(&the_lnet.ln_api_mutex);
3358
3359         if (rc > 1) {
3360                 rc = -EINVAL; /* only add one network per call */
3361                 goto out_unlock_clean;
3362         }
3363
3364         net = list_entry(net_head.next, struct lnet_net, net_list);
3365         list_del_init(&net->net_list);
3366
3367         LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3368
3369         memset(&tun, 0, sizeof(tun));
3370
3371         tun.lt_cmn.lct_peer_timeout =
3372           conf->cfg_config_u.cfg_net.net_peer_timeout;
3373         tun.lt_cmn.lct_peer_tx_credits =
3374           conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3375         tun.lt_cmn.lct_peer_rtr_credits =
3376           conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3377         tun.lt_cmn.lct_max_tx_credits =
3378           conf->cfg_config_u.cfg_net.net_max_tx_credits;
3379
3380         rc = lnet_add_net_common(net, &tun);
3381
3382 out_unlock_clean:
3383         mutex_unlock(&the_lnet.ln_api_mutex);
3384         while (!list_empty(&net_head)) {
3385                 /* net_head list is empty in success case */
3386                 net = list_entry(net_head.next, struct lnet_net, net_list);
3387                 list_del_init(&net->net_list);
3388                 lnet_net_free(net);
3389         }
3390         return rc;
3391 }
3392
3393 int
3394 lnet_dyn_del_net(__u32 net_id)
3395 {
3396         struct lnet_net  *net;
3397         struct lnet_ping_buffer *pbuf;
3398         struct lnet_handle_md ping_mdh;
3399         int               rc;
3400         int               net_ni_count;
3401
3402         /* don't allow userspace to shutdown the LOLND */
3403         if (LNET_NETTYP(net_id) == LOLND)
3404                 return -EINVAL;
3405
3406         mutex_lock(&the_lnet.ln_api_mutex);
3407
3408         lnet_net_lock(0);
3409
3410         net = lnet_get_net_locked(net_id);
3411         if (net == NULL) {
3412                 lnet_net_unlock(0);
3413                 rc = -EINVAL;
3414                 goto out;
3415         }
3416
3417         net_ni_count = lnet_get_net_ni_count_locked(net);
3418
3419         lnet_net_unlock(0);
3420
3421         /* create and link a new ping info, before removing the old one */
3422         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3423                                     lnet_get_ni_count() - net_ni_count, false);
3424         if (rc != 0)
3425                 goto out;
3426
3427         lnet_shutdown_lndnet(net);
3428
3429         if (lnet_count_acceptor_nets() == 0)
3430                 lnet_acceptor_stop();
3431
3432         lnet_ping_target_update(pbuf, ping_mdh);
3433
3434 out:
3435         mutex_unlock(&the_lnet.ln_api_mutex);
3436
3437         return rc;
3438 }
3439
3440 void lnet_incr_dlc_seq(void)
3441 {
3442         atomic_inc(&lnet_dlc_seq_no);
3443 }
3444
3445 __u32 lnet_get_dlc_seq_locked(void)
3446 {
3447         return atomic_read(&lnet_dlc_seq_no);
3448 }
3449
3450 static void
3451 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3452 {
3453         struct lnet_net *net;
3454         struct lnet_ni *ni;
3455
3456         lnet_net_lock(LNET_LOCK_EX);
3457         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3458                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3459                         if (ni->ni_nid == nid || all) {
3460                                 atomic_set(&ni->ni_healthv, value);
3461                                 if (list_empty(&ni->ni_recovery) &&
3462                                     value < LNET_MAX_HEALTH_VALUE) {
3463                                         CERROR("manually adding local NI %s to recovery\n",
3464                                                libcfs_nid2str(ni->ni_nid));
3465                                         list_add_tail(&ni->ni_recovery,
3466                                                       &the_lnet.ln_mt_localNIRecovq);
3467                                         lnet_ni_addref_locked(ni, 0);
3468                                 }
3469                                 if (!all) {
3470                                         lnet_net_unlock(LNET_LOCK_EX);
3471                                         return;
3472                                 }
3473                         }
3474                 }
3475         }
3476         lnet_net_unlock(LNET_LOCK_EX);
3477 }
3478
3479 static int
3480 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3481 {
3482         int cpt, rc = 0;
3483         struct lnet_ni *ni;
3484         lnet_nid_t nid = stats->hlni_nid;
3485
3486         cpt = lnet_net_lock_current();
3487         ni = lnet_nid2ni_locked(nid, cpt);
3488
3489         if (!ni) {
3490                 rc = -ENOENT;
3491                 goto unlock;
3492         }
3493
3494         stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3495         stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3496         stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3497         stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3498         stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3499         stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3500         stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3501
3502 unlock:
3503         lnet_net_unlock(cpt);
3504
3505         return rc;
3506 }
3507
3508 static int
3509 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3510 {
3511         struct lnet_ni *ni;
3512         int i = 0;
3513
3514         lnet_net_lock(LNET_LOCK_EX);
3515         list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3516                 list->rlst_nid_array[i] = ni->ni_nid;
3517                 i++;
3518                 if (i >= LNET_MAX_SHOW_NUM_NID)
3519                         break;
3520         }
3521         lnet_net_unlock(LNET_LOCK_EX);
3522         list->rlst_num_nids = i;
3523
3524         return 0;
3525 }
3526
3527 static int
3528 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3529 {
3530         struct lnet_peer_ni *lpni;
3531         int i = 0;
3532
3533         lnet_net_lock(LNET_LOCK_EX);
3534         list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3535                 list->rlst_nid_array[i] = lpni->lpni_nid;
3536                 i++;
3537                 if (i >= LNET_MAX_SHOW_NUM_NID)
3538                         break;
3539         }
3540         lnet_net_unlock(LNET_LOCK_EX);
3541         list->rlst_num_nids = i;
3542
3543         return 0;
3544 }
3545
3546 /**
3547  * LNet ioctl handler.
3548  *
3549  */
3550 int
3551 LNetCtl(unsigned int cmd, void *arg)
3552 {
3553         struct libcfs_ioctl_data *data = arg;
3554         struct lnet_ioctl_config_data *config;
3555         struct lnet_process_id    id = {0};
3556         struct lnet_ni           *ni;
3557         int                       rc;
3558
3559         BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3560                      sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3561
3562         switch (cmd) {
3563         case IOC_LIBCFS_GET_NI:
3564                 rc = LNetGetId(data->ioc_count, &id);
3565                 data->ioc_nid = id.nid;
3566                 return rc;
3567
3568         case IOC_LIBCFS_FAIL_NID:
3569                 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3570
3571         case IOC_LIBCFS_ADD_ROUTE: {
3572                 /* default router sensitivity to 1 */
3573                 unsigned int sensitivity = 1;
3574                 config = arg;
3575
3576                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3577                         return -EINVAL;
3578
3579                 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3580                         sensitivity =
3581                           config->cfg_config_u.cfg_route.rtr_sensitivity;
3582                 }
3583
3584                 mutex_lock(&the_lnet.ln_api_mutex);
3585                 rc = lnet_add_route(config->cfg_net,
3586                                     config->cfg_config_u.cfg_route.rtr_hop,
3587                                     config->cfg_nid,
3588                                     config->cfg_config_u.cfg_route.
3589                                         rtr_priority, sensitivity);
3590                 mutex_unlock(&the_lnet.ln_api_mutex);
3591                 return rc;
3592         }
3593
3594         case IOC_LIBCFS_DEL_ROUTE:
3595                 config = arg;
3596
3597                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3598                         return -EINVAL;
3599
3600                 mutex_lock(&the_lnet.ln_api_mutex);
3601                 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3602                 mutex_unlock(&the_lnet.ln_api_mutex);
3603                 return rc;
3604
3605         case IOC_LIBCFS_GET_ROUTE:
3606                 config = arg;
3607
3608                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3609                         return -EINVAL;
3610
3611                 mutex_lock(&the_lnet.ln_api_mutex);
3612                 rc = lnet_get_route(config->cfg_count,
3613                                     &config->cfg_net,
3614                                     &config->cfg_config_u.cfg_route.rtr_hop,
3615                                     &config->cfg_nid,
3616                                     &config->cfg_config_u.cfg_route.rtr_flags,
3617                                     &config->cfg_config_u.cfg_route.
3618                                         rtr_priority,
3619                                     &config->cfg_config_u.cfg_route.
3620                                         rtr_sensitivity);
3621                 mutex_unlock(&the_lnet.ln_api_mutex);
3622                 return rc;
3623
3624         case IOC_LIBCFS_GET_LOCAL_NI: {
3625                 struct lnet_ioctl_config_ni *cfg_ni;
3626                 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3627                 struct lnet_ioctl_element_stats *stats;
3628                 __u32 tun_size;
3629
3630                 cfg_ni = arg;
3631
3632                 /* get the tunables if they are available */
3633                 if (cfg_ni->lic_cfg_hdr.ioc_len <
3634                     sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3635                         return -EINVAL;
3636
3637                 stats = (struct lnet_ioctl_element_stats *)
3638                         cfg_ni->lic_bulk;
3639                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3640                                 (cfg_ni->lic_bulk + sizeof(*stats));
3641
3642                 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3643                         sizeof(*stats);
3644
3645                 mutex_lock(&the_lnet.ln_api_mutex);
3646                 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3647                 mutex_unlock(&the_lnet.ln_api_mutex);
3648                 return rc;
3649         }
3650
3651         case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3652                 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3653
3654                 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3655                         return -EINVAL;
3656
3657                 mutex_lock(&the_lnet.ln_api_mutex);
3658                 rc = lnet_get_ni_stats(msg_stats);
3659                 mutex_unlock(&the_lnet.ln_api_mutex);
3660
3661                 return rc;
3662         }
3663
3664         case IOC_LIBCFS_GET_NET: {
3665                 size_t total = sizeof(*config) +
3666                                sizeof(struct lnet_ioctl_net_config);
3667                 config = arg;
3668
3669                 if (config->cfg_hdr.ioc_len < total)
3670                         return -EINVAL;
3671
3672                 mutex_lock(&the_lnet.ln_api_mutex);
3673                 rc = lnet_get_net_config(config);
3674                 mutex_unlock(&the_lnet.ln_api_mutex);
3675                 return rc;
3676         }
3677
3678         case IOC_LIBCFS_GET_LNET_STATS:
3679         {
3680                 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
3681
3682                 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
3683                         return -EINVAL;
3684
3685                 mutex_lock(&the_lnet.ln_api_mutex);
3686                 lnet_counters_get(&lnet_stats->st_cntrs);
3687                 mutex_unlock(&the_lnet.ln_api_mutex);
3688                 return 0;
3689         }
3690
3691         case IOC_LIBCFS_CONFIG_RTR:
3692                 config = arg;
3693
3694                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3695                         return -EINVAL;
3696
3697                 mutex_lock(&the_lnet.ln_api_mutex);
3698                 if (config->cfg_config_u.cfg_buffers.buf_enable) {
3699                         rc = lnet_rtrpools_enable();
3700                         mutex_unlock(&the_lnet.ln_api_mutex);
3701                         return rc;
3702                 }
3703                 lnet_rtrpools_disable();
3704                 mutex_unlock(&the_lnet.ln_api_mutex);
3705                 return 0;
3706
3707         case IOC_LIBCFS_ADD_BUF:
3708                 config = arg;
3709
3710                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3711                         return -EINVAL;
3712
3713                 mutex_lock(&the_lnet.ln_api_mutex);
3714                 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
3715                                                 buf_tiny,
3716                                           config->cfg_config_u.cfg_buffers.
3717                                                 buf_small,
3718                                           config->cfg_config_u.cfg_buffers.
3719                                                 buf_large);
3720                 mutex_unlock(&the_lnet.ln_api_mutex);
3721                 return rc;
3722
3723         case IOC_LIBCFS_SET_NUMA_RANGE: {
3724                 struct lnet_ioctl_set_value *numa;
3725                 numa = arg;
3726                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3727                         return -EINVAL;
3728                 lnet_net_lock(LNET_LOCK_EX);
3729                 lnet_numa_range = numa->sv_value;
3730                 lnet_net_unlock(LNET_LOCK_EX);
3731                 return 0;
3732         }
3733
3734         case IOC_LIBCFS_GET_NUMA_RANGE: {
3735                 struct lnet_ioctl_set_value *numa;
3736                 numa = arg;
3737                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3738                         return -EINVAL;
3739                 numa->sv_value = lnet_numa_range;
3740                 return 0;
3741         }
3742
3743         case IOC_LIBCFS_GET_BUF: {
3744                 struct lnet_ioctl_pool_cfg *pool_cfg;
3745                 size_t total = sizeof(*config) + sizeof(*pool_cfg);
3746
3747                 config = arg;
3748
3749                 if (config->cfg_hdr.ioc_len < total)
3750                         return -EINVAL;
3751
3752                 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
3753
3754                 mutex_lock(&the_lnet.ln_api_mutex);
3755                 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
3756                 mutex_unlock(&the_lnet.ln_api_mutex);
3757                 return rc;
3758         }
3759
3760         case IOC_LIBCFS_GET_LOCAL_HSTATS: {
3761                 struct lnet_ioctl_local_ni_hstats *stats = arg;
3762
3763                 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
3764                         return -EINVAL;
3765
3766                 mutex_lock(&the_lnet.ln_api_mutex);
3767                 rc = lnet_get_local_ni_hstats(stats);
3768                 mutex_unlock(&the_lnet.ln_api_mutex);
3769
3770                 return rc;
3771         }
3772
3773         case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
3774                 struct lnet_ioctl_recovery_list *list = arg;
3775                 if (list->rlst_hdr.ioc_len < sizeof(*list))
3776                         return -EINVAL;
3777
3778                 mutex_lock(&the_lnet.ln_api_mutex);
3779                 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
3780                         rc = lnet_get_local_ni_recovery_list(list);
3781                 else
3782                         rc = lnet_get_peer_ni_recovery_list(list);
3783                 mutex_unlock(&the_lnet.ln_api_mutex);
3784                 return rc;
3785         }
3786
3787         case IOC_LIBCFS_ADD_PEER_NI: {
3788                 struct lnet_ioctl_peer_cfg *cfg = arg;
3789
3790                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3791                         return -EINVAL;
3792
3793                 mutex_lock(&the_lnet.ln_api_mutex);
3794                 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
3795                                       cfg->prcfg_cfg_nid,
3796                                       cfg->prcfg_mr);
3797                 mutex_unlock(&the_lnet.ln_api_mutex);
3798                 return rc;
3799         }
3800
3801         case IOC_LIBCFS_DEL_PEER_NI: {
3802                 struct lnet_ioctl_peer_cfg *cfg = arg;
3803
3804                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3805                         return -EINVAL;
3806
3807                 mutex_lock(&the_lnet.ln_api_mutex);
3808                 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
3809                                       cfg->prcfg_cfg_nid);
3810                 mutex_unlock(&the_lnet.ln_api_mutex);
3811                 return rc;
3812         }
3813
3814         case IOC_LIBCFS_GET_PEER_INFO: {
3815                 struct lnet_ioctl_peer *peer_info = arg;
3816
3817                 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
3818                         return -EINVAL;
3819
3820                 mutex_lock(&the_lnet.ln_api_mutex);
3821                 rc = lnet_get_peer_ni_info(
3822                    peer_info->pr_count,
3823                    &peer_info->pr_nid,
3824                    peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
3825                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
3826                    &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
3827                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
3828                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
3829                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
3830                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
3831                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
3832                 mutex_unlock(&the_lnet.ln_api_mutex);
3833                 return rc;
3834         }
3835
3836         case IOC_LIBCFS_GET_PEER_NI: {
3837                 struct lnet_ioctl_peer_cfg *cfg = arg;
3838
3839                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3840                         return -EINVAL;
3841
3842                 mutex_lock(&the_lnet.ln_api_mutex);
3843                 rc = lnet_get_peer_info(cfg,
3844                                         (void __user *)cfg->prcfg_bulk);
3845                 mutex_unlock(&the_lnet.ln_api_mutex);
3846                 return rc;
3847         }
3848
3849         case IOC_LIBCFS_GET_PEER_LIST: {
3850                 struct lnet_ioctl_peer_cfg *cfg = arg;
3851
3852                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
3853                         return -EINVAL;
3854
3855                 mutex_lock(&the_lnet.ln_api_mutex);
3856                 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
3857                                 (struct lnet_process_id __user *)cfg->prcfg_bulk);
3858                 mutex_unlock(&the_lnet.ln_api_mutex);
3859                 return rc;
3860         }
3861
3862         case IOC_LIBCFS_SET_HEALHV: {
3863                 struct lnet_ioctl_reset_health_cfg *cfg = arg;
3864                 int value;
3865                 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
3866                         return -EINVAL;
3867                 if (cfg->rh_value < 0 ||
3868                     cfg->rh_value > LNET_MAX_HEALTH_VALUE)
3869                         value = LNET_MAX_HEALTH_VALUE;
3870                 else
3871                         value = cfg->rh_value;
3872                 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
3873                        value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
3874                        "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
3875                 mutex_lock(&the_lnet.ln_api_mutex);
3876                 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
3877                         lnet_ni_set_healthv(cfg->rh_nid, value,
3878                                              cfg->rh_all);
3879                 else
3880                         lnet_peer_ni_set_healthv(cfg->rh_nid, value,
3881                                                   cfg->rh_all);
3882                 mutex_unlock(&the_lnet.ln_api_mutex);
3883                 return 0;
3884         }
3885
3886         case IOC_LIBCFS_NOTIFY_ROUTER: {
3887                 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
3888
3889                 /* The deadline passed in by the user should be some time in
3890                  * seconds in the future since the UNIX epoch. We have to map
3891                  * that deadline to the wall clock.
3892                  */
3893                 deadline += ktime_get_seconds();
3894                 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
3895                                    deadline);
3896         }
3897
3898         case IOC_LIBCFS_LNET_DIST:
3899                 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
3900                 if (rc < 0 && rc != -EHOSTUNREACH)
3901                         return rc;
3902
3903                 data->ioc_u32[0] = rc;
3904                 return 0;
3905
3906         case IOC_LIBCFS_TESTPROTOCOMPAT:
3907                 the_lnet.ln_testprotocompat = data->ioc_flags;
3908                 return 0;
3909
3910         case IOC_LIBCFS_LNET_FAULT:
3911                 return lnet_fault_ctl(data->ioc_flags, data);
3912
3913         case IOC_LIBCFS_PING: {
3914                 signed long timeout;
3915
3916                 id.nid = data->ioc_nid;
3917                 id.pid = data->ioc_u32[0];
3918
3919                 /* If timeout is negative then set default of 3 minutes */
3920                 if (((s32)data->ioc_u32[1] <= 0) ||
3921                     data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3922                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3923                 else
3924                         timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
3925
3926                 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
3927                                data->ioc_plen1 / sizeof(struct lnet_process_id));
3928
3929                 if (rc < 0)
3930                         return rc;
3931
3932                 data->ioc_count = rc;
3933                 return 0;
3934         }
3935
3936         case IOC_LIBCFS_PING_PEER: {
3937                 struct lnet_ioctl_ping_data *ping = arg;
3938                 struct lnet_peer *lp;
3939                 signed long timeout;
3940
3941                 /* If timeout is negative then set default of 3 minutes */
3942                 if (((s32)ping->op_param) <= 0 ||
3943                     ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
3944                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
3945                 else
3946                         timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
3947
3948                 rc = lnet_ping(ping->ping_id, timeout,
3949                                ping->ping_buf,
3950                                ping->ping_count);
3951                 if (rc < 0)
3952                         return rc;
3953
3954                 mutex_lock(&the_lnet.ln_api_mutex);
3955                 lp = lnet_find_peer(ping->ping_id.nid);
3956                 if (lp) {
3957                         ping->ping_id.nid = lp->lp_primary_nid;
3958                         ping->mr_info = lnet_peer_is_multi_rail(lp);
3959                         lnet_peer_decref_locked(lp);
3960                 }
3961                 mutex_unlock(&the_lnet.ln_api_mutex);
3962
3963                 ping->ping_count = rc;
3964                 return 0;
3965         }
3966
3967         case IOC_LIBCFS_DISCOVER: {
3968                 struct lnet_ioctl_ping_data *discover = arg;
3969                 struct lnet_peer *lp;
3970
3971                 rc = lnet_discover(discover->ping_id, discover->op_param,
3972                                    discover->ping_buf,
3973                                    discover->ping_count);
3974                 if (rc < 0)
3975                         return rc;
3976
3977                 mutex_lock(&the_lnet.ln_api_mutex);
3978                 lp = lnet_find_peer(discover->ping_id.nid);
3979                 if (lp) {
3980                         discover->ping_id.nid = lp->lp_primary_nid;
3981                         discover->mr_info = lnet_peer_is_multi_rail(lp);
3982                         lnet_peer_decref_locked(lp);
3983                 }
3984                 mutex_unlock(&the_lnet.ln_api_mutex);
3985
3986                 discover->ping_count = rc;
3987                 return 0;
3988         }
3989
3990         default:
3991                 ni = lnet_net2ni_addref(data->ioc_net);
3992                 if (ni == NULL)
3993                         return -EINVAL;
3994
3995                 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
3996                         rc = -EINVAL;
3997                 else
3998                         rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
3999
4000                 lnet_ni_decref(ni);
4001                 return rc;
4002         }
4003         /* not reached */
4004 }
4005 EXPORT_SYMBOL(LNetCtl);
4006
4007 void LNetDebugPeer(struct lnet_process_id id)
4008 {
4009         lnet_debug_peer(id.nid);
4010 }
4011 EXPORT_SYMBOL(LNetDebugPeer);
4012
4013 /**
4014  * Determine if the specified peer \a nid is on the local node.
4015  *
4016  * \param nid   peer nid to check
4017  *
4018  * \retval true         If peer NID is on the local node.
4019  * \retval false        If peer NID is not on the local node.
4020  */
4021 bool LNetIsPeerLocal(lnet_nid_t nid)
4022 {
4023         struct lnet_net *net;
4024         struct lnet_ni *ni;
4025         int cpt;
4026
4027         cpt = lnet_net_lock_current();
4028         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4029                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4030                         if (ni->ni_nid == nid) {
4031                                 lnet_net_unlock(cpt);
4032                                 return true;
4033                         }
4034                 }
4035         }
4036         lnet_net_unlock(cpt);
4037
4038         return false;
4039 }
4040 EXPORT_SYMBOL(LNetIsPeerLocal);
4041
4042 /**
4043  * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4044  * Note that all interfaces share a same PID, as requested by LNetNIInit().
4045  *
4046  * \param index Index of the interface to look up.
4047  * \param id On successful return, this location will hold the
4048  * struct lnet_process_id ID of the interface.
4049  *
4050  * \retval 0 If an interface exists at \a index.
4051  * \retval -ENOENT If no interface has been found.
4052  */
4053 int
4054 LNetGetId(unsigned int index, struct lnet_process_id *id)
4055 {
4056         struct lnet_ni   *ni;
4057         struct lnet_net  *net;
4058         int               cpt;
4059         int               rc = -ENOENT;
4060
4061         LASSERT(the_lnet.ln_refcount > 0);
4062
4063         cpt = lnet_net_lock_current();
4064
4065         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4066                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4067                         if (index-- != 0)
4068                                 continue;
4069
4070                         id->nid = ni->ni_nid;
4071                         id->pid = the_lnet.ln_pid;
4072                         rc = 0;
4073                         break;
4074                 }
4075         }
4076
4077         lnet_net_unlock(cpt);
4078         return rc;
4079 }
4080 EXPORT_SYMBOL(LNetGetId);
4081
4082 static int lnet_ping(struct lnet_process_id id, signed long timeout,
4083                      struct lnet_process_id __user *ids, int n_ids)
4084 {
4085         struct lnet_eq *eq;
4086         struct lnet_handle_md mdh;
4087         struct lnet_event event;
4088         struct lnet_md md = { NULL };
4089         int which;
4090         int unlinked = 0;
4091         int replied = 0;
4092         const signed long a_long_time = cfs_time_seconds(60);
4093         struct lnet_ping_buffer *pbuf;
4094         struct lnet_process_id tmpid;
4095         int i;
4096         int nob;
4097         int rc;
4098         int rc2;
4099         sigset_t blocked;
4100
4101         /* n_ids limit is arbitrary */
4102         if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4103                 return -EINVAL;
4104
4105         /*
4106          * if the user buffer has more space than the lnet_interfaces_max
4107          * then only fill it up to lnet_interfaces_max
4108          */
4109         if (n_ids > lnet_interfaces_max)
4110                 n_ids = lnet_interfaces_max;
4111
4112         if (id.pid == LNET_PID_ANY)
4113                 id.pid = LNET_PID_LUSTRE;
4114
4115         pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4116         if (!pbuf)
4117                 return -ENOMEM;
4118
4119         /* NB 2 events max (including any unlink event) */
4120         eq = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE);
4121         if (IS_ERR(eq)) {
4122                 rc = PTR_ERR(eq);
4123                 CERROR("Can't allocate EQ: %d\n", rc);
4124                 goto fail_ping_buffer_decref;
4125         }
4126
4127         /* initialize md content */
4128         md.start     = &pbuf->pb_info;
4129         md.length    = LNET_PING_INFO_SIZE(n_ids);
4130         md.threshold = 2; /* GET/REPLY */
4131         md.max_size  = 0;
4132         md.options   = LNET_MD_TRUNCATE;
4133         md.user_ptr  = NULL;
4134         md.eq_handle = eq;
4135
4136         rc = LNetMDBind(md, LNET_UNLINK, &mdh);
4137         if (rc != 0) {
4138                 CERROR("Can't bind MD: %d\n", rc);
4139                 goto fail_free_eq;
4140         }
4141
4142         rc = LNetGet(LNET_NID_ANY, mdh, id,
4143                      LNET_RESERVED_PORTAL,
4144                      LNET_PROTO_PING_MATCHBITS, 0, false);
4145
4146         if (rc != 0) {
4147                 /* Don't CERROR; this could be deliberate! */
4148                 rc2 = LNetMDUnlink(mdh);
4149                 LASSERT(rc2 == 0);
4150
4151                 /* NB must wait for the UNLINK event below... */
4152                 unlinked = 1;
4153                 timeout = a_long_time;
4154         }
4155
4156         do {
4157                 /* MUST block for unlink to complete */
4158                 if (unlinked) {
4159                         sigset_t set;
4160
4161                         sigfillset(&set);
4162                         sigprocmask(SIG_SETMASK, &set, &blocked);
4163                 }
4164
4165                 rc2 = LNetEQPoll(&eq, 1, timeout, &event, &which);
4166
4167                 if (unlinked)
4168                         cfs_restore_sigs(blocked);
4169
4170                 CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
4171                        (rc2 <= 0) ? -1 : event.type,
4172                        (rc2 <= 0) ? -1 : event.status,
4173                        (rc2 > 0 && event.unlinked) ? " unlinked" : "");
4174
4175                 LASSERT(rc2 != -EOVERFLOW);     /* can't miss anything */
4176
4177                 if (rc2 <= 0 || event.status != 0) {
4178                         /* timeout or error */
4179                         if (!replied && rc == 0)
4180                                 rc = (rc2 < 0) ? rc2 :
4181                                      (rc2 == 0) ? -ETIMEDOUT :
4182                                      event.status;
4183
4184                         if (!unlinked) {
4185                                 /* Ensure completion in finite time... */
4186                                 LNetMDUnlink(mdh);
4187                                 /* No assertion (racing with network) */
4188                                 unlinked = 1;
4189                                 timeout = a_long_time;
4190                         } else if (rc2 == 0) {
4191                                 /* timed out waiting for unlink */
4192                                 CWARN("ping %s: late network completion\n",
4193                                       libcfs_id2str(id));
4194                         }
4195                 } else if (event.type == LNET_EVENT_REPLY) {
4196                         replied = 1;
4197                         rc = event.mlength;
4198                 }
4199         } while (rc2 <= 0 || !event.unlinked);
4200
4201         if (!replied) {
4202                 if (rc >= 0)
4203                         CWARN("%s: Unexpected rc >= 0 but no reply!\n",
4204                               libcfs_id2str(id));
4205                 rc = -EIO;
4206                 goto fail_free_eq;
4207         }
4208
4209         nob = rc;
4210         LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4211
4212         rc = -EPROTO;           /* if I can't parse... */
4213
4214         if (nob < 8) {
4215                 CERROR("%s: ping info too short %d\n",
4216                        libcfs_id2str(id), nob);
4217                 goto fail_free_eq;
4218         }
4219
4220         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4221                 lnet_swap_pinginfo(pbuf);
4222         } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4223                 CERROR("%s: Unexpected magic %08x\n",
4224                        libcfs_id2str(id), pbuf->pb_info.pi_magic);
4225                 goto fail_free_eq;
4226         }
4227
4228         if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4229                 CERROR("%s: ping w/o NI status: 0x%x\n",
4230                        libcfs_id2str(id), pbuf->pb_info.pi_features);
4231                 goto fail_free_eq;
4232         }
4233
4234         if (nob < LNET_PING_INFO_SIZE(0)) {
4235                 CERROR("%s: Short reply %d(%d min)\n",
4236                        libcfs_id2str(id),
4237                        nob, (int)LNET_PING_INFO_SIZE(0));
4238                 goto fail_free_eq;
4239         }
4240
4241         if (pbuf->pb_info.pi_nnis < n_ids)
4242                 n_ids = pbuf->pb_info.pi_nnis;
4243
4244         if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4245                 CERROR("%s: Short reply %d(%d expected)\n",
4246                        libcfs_id2str(id),
4247                        nob, (int)LNET_PING_INFO_SIZE(n_ids));
4248                 goto fail_free_eq;
4249         }
4250
4251         rc = -EFAULT;           /* if I segv in copy_to_user()... */
4252
4253         memset(&tmpid, 0, sizeof(tmpid));
4254         for (i = 0; i < n_ids; i++) {
4255                 tmpid.pid = pbuf->pb_info.pi_pid;
4256                 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4257                 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4258                         goto fail_free_eq;
4259         }
4260         rc = pbuf->pb_info.pi_nnis;
4261
4262  fail_free_eq:
4263         rc2 = LNetEQFree(eq);
4264         if (rc2 != 0)
4265                 CERROR("rc2 %d\n", rc2);
4266         LASSERT(rc2 == 0);
4267
4268  fail_ping_buffer_decref:
4269         lnet_ping_buffer_decref(pbuf);
4270         return rc;
4271 }
4272
4273 static int
4274 lnet_discover(struct lnet_process_id id, __u32 force,
4275               struct lnet_process_id __user *ids, int n_ids)
4276 {
4277         struct lnet_peer_ni *lpni;
4278         struct lnet_peer_ni *p;
4279         struct lnet_peer *lp;
4280         struct lnet_process_id *buf;
4281         int cpt;
4282         int i;
4283         int rc;
4284         int max_intf = lnet_interfaces_max;
4285         size_t buf_size;
4286
4287         if (n_ids <= 0 ||
4288             id.nid == LNET_NID_ANY)
4289                 return -EINVAL;
4290
4291         if (id.pid == LNET_PID_ANY)
4292                 id.pid = LNET_PID_LUSTRE;
4293
4294         /*
4295          * if the user buffer has more space than the max_intf
4296          * then only fill it up to max_intf
4297          */
4298         if (n_ids > max_intf)
4299                 n_ids = max_intf;
4300
4301         buf_size = n_ids * sizeof(*buf);
4302
4303         LIBCFS_ALLOC(buf, buf_size);
4304         if (!buf)
4305                 return -ENOMEM;
4306
4307         cpt = lnet_net_lock_current();
4308         lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4309         if (IS_ERR(lpni)) {
4310                 rc = PTR_ERR(lpni);
4311                 goto out;
4312         }
4313
4314         /*
4315          * Clearing the NIDS_UPTODATE flag ensures the peer will
4316          * be discovered, provided discovery has not been disabled.
4317          */
4318         lp = lpni->lpni_peer_net->lpn_peer;
4319         spin_lock(&lp->lp_lock);
4320         lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4321         /* If the force flag is set, force a PING and PUSH as well. */
4322         if (force)
4323                 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4324         spin_unlock(&lp->lp_lock);
4325         rc = lnet_discover_peer_locked(lpni, cpt, true);
4326         if (rc)
4327                 goto out_decref;
4328
4329         /* Peer may have changed. */
4330         lp = lpni->lpni_peer_net->lpn_peer;
4331         if (lp->lp_nnis < n_ids)
4332                 n_ids = lp->lp_nnis;
4333
4334         i = 0;
4335         p = NULL;
4336         while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4337                 buf[i].pid = id.pid;
4338                 buf[i].nid = p->lpni_nid;
4339                 if (++i >= n_ids)
4340                         break;
4341         }
4342
4343         lnet_net_unlock(cpt);
4344
4345         rc = -EFAULT;
4346         if (copy_to_user(ids, buf, n_ids * sizeof(*buf)))
4347                 goto out_relock;
4348         rc = n_ids;
4349 out_relock:
4350         lnet_net_lock(cpt);
4351 out_decref:
4352         lnet_peer_ni_decref_locked(lpni);
4353 out:
4354         lnet_net_unlock(cpt);
4355
4356         LIBCFS_FREE(buf, buf_size);
4357
4358         return rc;
4359 }
4360
4361 /**
4362  * Retrieve peer discovery status.
4363  *
4364  * \retval 1 if lnet_peer_discovery_disabled is 0
4365  * \retval 0 if lnet_peer_discovery_disabled is 1
4366  */
4367 int
4368 LNetGetPeerDiscoveryStatus(void)
4369 {
4370         return !lnet_peer_discovery_disabled;
4371 }
4372 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);