lnet/lnet/api-ni.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  */
  31
  32 #define DEBUG_SUBSYSTEM S_LNET
  33
  34 #include <linux/ctype.h>
  35 #include <linux/log2.h>
  36 #include <linux/ktime.h>
  37 #include <linux/moduleparam.h>
  38 #include <linux/uaccess.h>
  39 #ifdef HAVE_SCHED_HEADERS
  40 #include <linux/sched/signal.h>
  41 #endif
  42
  43 #include <lnet/udsp.h>
  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;
  75 module_param(use_tcp_bonding, int, 0444);
  76 MODULE_PARM_DESC(use_tcp_bonding,
  77                  "use_tcp_bonding parameter has been removed");
  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                 "DEPRECATED - Interval to recover unhealthy interfaces in seconds");
 126
 127 unsigned int lnet_recovery_limit;
 128 module_param(lnet_recovery_limit, uint, 0644);
 129 MODULE_PARM_DESC(lnet_recovery_limit,
 130                  "How long to attempt recovery of unhealthy peer interfaces in seconds. Set to 0 to allow indefinite recovery");
 131
 132 static int lnet_interfaces_max = LNET_INTERFACES_MAX_DEFAULT;
 133 static int intf_max_set(const char *val, cfs_kernel_param_arg_t *kp);
 134
 135 static struct kernel_param_ops param_ops_interfaces_max = {
 136         .set = intf_max_set,
 137         .get = param_get_int,
 138 };
 139
 140 #define param_check_interfaces_max(name, p) \
 141                 __param_check(name, p, int)
 142
 143 #ifdef HAVE_KERNEL_PARAM_OPS
 144 module_param(lnet_interfaces_max, interfaces_max, 0644);
 145 #else
 146 module_param_call(lnet_interfaces_max, intf_max_set, param_get_int,
 147                   &param_ops_interfaces_max, 0644);
 148 #endif
 149 MODULE_PARM_DESC(lnet_interfaces_max,
 150                 "Maximum number of interfaces in a node.");
 151
 152 unsigned lnet_peer_discovery_disabled = 0;
 153 static int discovery_set(const char *val, cfs_kernel_param_arg_t *kp);
 154
 155 static struct kernel_param_ops param_ops_discovery_disabled = {
 156         .set = discovery_set,
 157         .get = param_get_int,
 158 };
 159
 160 #define param_check_discovery_disabled(name, p) \
 161                 __param_check(name, p, int)
 162 #ifdef HAVE_KERNEL_PARAM_OPS
 163 module_param(lnet_peer_discovery_disabled, discovery_disabled, 0644);
 164 #else
 165 module_param_call(lnet_peer_discovery_disabled, discovery_set, param_get_int,
 166                   &param_ops_discovery_disabled, 0644);
 167 #endif
 168 MODULE_PARM_DESC(lnet_peer_discovery_disabled,
 169                 "Set to 1 to disable peer discovery on this node.");
 170
 171 unsigned int lnet_drop_asym_route;
 172 static int drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp);
 173
 174 static struct kernel_param_ops param_ops_drop_asym_route = {
 175         .set = drop_asym_route_set,
 176         .get = param_get_int,
 177 };
 178
 179 #define param_check_drop_asym_route(name, p)    \
 180         __param_check(name, p, int)
 181 #ifdef HAVE_KERNEL_PARAM_OPS
 182 module_param(lnet_drop_asym_route, drop_asym_route, 0644);
 183 #else
 184 module_param_call(lnet_drop_asym_route, drop_asym_route_set, param_get_int,
 185                   &param_ops_drop_asym_route, 0644);
 186 #endif
 187 MODULE_PARM_DESC(lnet_drop_asym_route,
 188                  "Set to 1 to drop asymmetrical route messages.");
 189
 190 #define LNET_TRANSACTION_TIMEOUT_DEFAULT 50
 191 unsigned int lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_DEFAULT;
 192 static int transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp);
 193 #ifdef HAVE_KERNEL_PARAM_OPS
 194 static struct kernel_param_ops param_ops_transaction_timeout = {
 195         .set = transaction_to_set,
 196         .get = param_get_int,
 197 };
 198
 199 #define param_check_transaction_timeout(name, p) \
 200                 __param_check(name, p, int)
 201 module_param(lnet_transaction_timeout, transaction_timeout, S_IRUGO|S_IWUSR);
 202 #else
 203 module_param_call(lnet_transaction_timeout, transaction_to_set, param_get_int,
 204                   &lnet_transaction_timeout, S_IRUGO|S_IWUSR);
 205 #endif
 206 MODULE_PARM_DESC(lnet_transaction_timeout,
 207                 "Maximum number of seconds to wait for a peer response.");
 208
 209 #define LNET_RETRY_COUNT_DEFAULT 2
 210 unsigned int lnet_retry_count = LNET_RETRY_COUNT_DEFAULT;
 211 static int retry_count_set(const char *val, cfs_kernel_param_arg_t *kp);
 212 #ifdef HAVE_KERNEL_PARAM_OPS
 213 static struct kernel_param_ops param_ops_retry_count = {
 214         .set = retry_count_set,
 215         .get = param_get_int,
 216 };
 217
 218 #define param_check_retry_count(name, p) \
 219                 __param_check(name, p, int)
 220 module_param(lnet_retry_count, retry_count, S_IRUGO|S_IWUSR);
 221 #else
 222 module_param_call(lnet_retry_count, retry_count_set, param_get_int,
 223                   &lnet_retry_count, S_IRUGO|S_IWUSR);
 224 #endif
 225 MODULE_PARM_DESC(lnet_retry_count,
 226                  "Maximum number of times to retry transmitting a message");
 227
 228 unsigned int lnet_response_tracking = 3;
 229 static int response_tracking_set(const char *val, cfs_kernel_param_arg_t *kp);
 230
 231 #ifdef HAVE_KERNEL_PARAM_OPS
 232 static struct kernel_param_ops param_ops_response_tracking = {
 233         .set = response_tracking_set,
 234         .get = param_get_int,
 235 };
 236
 237 #define param_check_response_tracking(name, p)  \
 238         __param_check(name, p, int)
 239 module_param(lnet_response_tracking, response_tracking, 0644);
 240 #else
 241 module_param_call(lnet_response_tracking, response_tracking_set, param_get_int,
 242                   &lnet_response_tracking, 0644);
 243 #endif
 244 MODULE_PARM_DESC(lnet_response_tracking,
 245                  "(0|1|2|3) LNet Internal Only|GET Reply only|PUT ACK only|Full Tracking (default)");
 246
 247 #define LNET_LND_TIMEOUT_DEFAULT ((LNET_TRANSACTION_TIMEOUT_DEFAULT - 1) / \
 248                                   (LNET_RETRY_COUNT_DEFAULT + 1))
 249 unsigned int lnet_lnd_timeout = LNET_LND_TIMEOUT_DEFAULT;
 250 static void lnet_set_lnd_timeout(void)
 251 {
 252         lnet_lnd_timeout = (lnet_transaction_timeout - 1) /
 253                            (lnet_retry_count + 1);
 254 }
 255
 256 /*
 257  * This sequence number keeps track of how many times DLC was used to
 258  * update the local NIs. It is incremented when a NI is added or
 259  * removed and checked when sending a message to determine if there is
 260  * a need to re-run the selection algorithm. See lnet_select_pathway()
 261  * for more details on its usage.
 262  */
 263 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
 264
 265 static int lnet_ping(struct lnet_process_id id, signed long timeout,
 266                      struct lnet_process_id __user *ids, int n_ids);
 267
 268 static int lnet_discover(struct lnet_process_id id, __u32 force,
 269                          struct lnet_process_id __user *ids, int n_ids);
 270
 271 static int
 272 sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
 273 {
 274         int rc;
 275         unsigned *sensitivity = (unsigned *)kp->arg;
 276         unsigned long value;
 277
 278         rc = kstrtoul(val, 0, &value);
 279         if (rc) {
 280                 CERROR("Invalid module parameter value for 'lnet_health_sensitivity'\n");
 281                 return rc;
 282         }
 283
 284         /*
 285          * The purpose of locking the api_mutex here is to ensure that
 286          * the correct value ends up stored properly.
 287          */
 288         mutex_lock(&the_lnet.ln_api_mutex);
 289
 290         if (value > LNET_MAX_HEALTH_VALUE) {
 291                 mutex_unlock(&the_lnet.ln_api_mutex);
 292                 CERROR("Invalid health value. Maximum: %d value = %lu\n",
 293                        LNET_MAX_HEALTH_VALUE, value);
 294                 return -EINVAL;
 295         }
 296
 297         if (*sensitivity != 0 && value == 0 && lnet_retry_count != 0) {
 298                 lnet_retry_count = 0;
 299                 lnet_set_lnd_timeout();
 300         }
 301
 302         *sensitivity = value;
 303
 304         mutex_unlock(&the_lnet.ln_api_mutex);
 305
 306         return 0;
 307 }
 308
 309 static int
 310 recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp)
 311 {
 312         CWARN("'lnet_recovery_interval' has been deprecated\n");
 313
 314         return 0;
 315 }
 316
 317 static int
 318 discovery_set(const char *val, cfs_kernel_param_arg_t *kp)
 319 {
 320         int rc;
 321         unsigned *discovery_off = (unsigned *)kp->arg;
 322         unsigned long value;
 323         struct lnet_ping_buffer *pbuf;
 324
 325         rc = kstrtoul(val, 0, &value);
 326         if (rc) {
 327                 CERROR("Invalid module parameter value for 'lnet_peer_discovery_disabled'\n");
 328                 return rc;
 329         }
 330
 331         value = (value) ? 1 : 0;
 332
 333         /*
 334          * The purpose of locking the api_mutex here is to ensure that
 335          * the correct value ends up stored properly.
 336          */
 337         mutex_lock(&the_lnet.ln_api_mutex);
 338
 339         if (value == *discovery_off) {
 340                 mutex_unlock(&the_lnet.ln_api_mutex);
 341                 return 0;
 342         }
 343
 344         /*
 345          * We still want to set the discovery value even when LNet is not
 346          * running. This is the case when LNet is being loaded and we want
 347          * the module parameters to take effect. Otherwise if we're
 348          * changing the value dynamically, we want to set it after
 349          * updating the peers
 350          */
 351         if (the_lnet.ln_state != LNET_STATE_RUNNING) {
 352                 *discovery_off = value;
 353                 mutex_unlock(&the_lnet.ln_api_mutex);
 354                 return 0;
 355         }
 356
 357         /* tell peers that discovery setting has changed */
 358         lnet_net_lock(LNET_LOCK_EX);
 359         pbuf = the_lnet.ln_ping_target;
 360         if (value)
 361                 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
 362         else
 363                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
 364         lnet_net_unlock(LNET_LOCK_EX);
 365
 366         /* only send a push when we're turning off discovery */
 367         if (*discovery_off <= 0 && value > 0)
 368                 lnet_push_update_to_peers(1);
 369         *discovery_off = value;
 370
 371         mutex_unlock(&the_lnet.ln_api_mutex);
 372
 373         return 0;
 374 }
 375
 376 static int
 377 drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp)
 378 {
 379         int rc;
 380         unsigned int *drop_asym_route = (unsigned int *)kp->arg;
 381         unsigned long value;
 382
 383         rc = kstrtoul(val, 0, &value);
 384         if (rc) {
 385                 CERROR("Invalid module parameter value for "
 386                        "'lnet_drop_asym_route'\n");
 387                 return rc;
 388         }
 389
 390         /*
 391          * The purpose of locking the api_mutex here is to ensure that
 392          * the correct value ends up stored properly.
 393          */
 394         mutex_lock(&the_lnet.ln_api_mutex);
 395
 396         if (value == *drop_asym_route) {
 397                 mutex_unlock(&the_lnet.ln_api_mutex);
 398                 return 0;
 399         }
 400
 401         *drop_asym_route = value;
 402
 403         mutex_unlock(&the_lnet.ln_api_mutex);
 404
 405         return 0;
 406 }
 407
 408 static int
 409 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
 410 {
 411         int rc;
 412         unsigned *transaction_to = (unsigned *)kp->arg;
 413         unsigned long value;
 414
 415         rc = kstrtoul(val, 0, &value);
 416         if (rc) {
 417                 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
 418                 return rc;
 419         }
 420
 421         /*
 422          * The purpose of locking the api_mutex here is to ensure that
 423          * the correct value ends up stored properly.
 424          */
 425         mutex_lock(&the_lnet.ln_api_mutex);
 426
 427         if (value <= lnet_retry_count || value == 0) {
 428                 mutex_unlock(&the_lnet.ln_api_mutex);
 429                 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
 430                        "Has to be greater than lnet_retry_count (%u)\n",
 431                        value, lnet_retry_count);
 432                 return -EINVAL;
 433         }
 434
 435         if (value == *transaction_to) {
 436                 mutex_unlock(&the_lnet.ln_api_mutex);
 437                 return 0;
 438         }
 439
 440         *transaction_to = value;
 441         /* Update the lnet_lnd_timeout now that we've modified the
 442          * transaction timeout
 443          */
 444         lnet_set_lnd_timeout();
 445
 446         mutex_unlock(&the_lnet.ln_api_mutex);
 447
 448         return 0;
 449 }
 450
 451 static int
 452 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
 453 {
 454         int rc;
 455         unsigned *retry_count = (unsigned *)kp->arg;
 456         unsigned long value;
 457
 458         rc = kstrtoul(val, 0, &value);
 459         if (rc) {
 460                 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
 461                 return rc;
 462         }
 463
 464         /*
 465          * The purpose of locking the api_mutex here is to ensure that
 466          * the correct value ends up stored properly.
 467          */
 468         mutex_lock(&the_lnet.ln_api_mutex);
 469
 470         if (lnet_health_sensitivity == 0 && value > 0) {
 471                 mutex_unlock(&the_lnet.ln_api_mutex);
 472                 CERROR("Can not set lnet_retry_count when health feature is turned off\n");
 473                 return -EINVAL;
 474         }
 475
 476         if (value > lnet_transaction_timeout) {
 477                 mutex_unlock(&the_lnet.ln_api_mutex);
 478                 CERROR("Invalid value for lnet_retry_count (%lu). "
 479                        "Has to be smaller than lnet_transaction_timeout (%u)\n",
 480                        value, lnet_transaction_timeout);
 481                 return -EINVAL;
 482         }
 483
 484         *retry_count = value;
 485
 486         /* Update the lnet_lnd_timeout now that we've modified the
 487          * retry count
 488          */
 489         lnet_set_lnd_timeout();
 490
 491         mutex_unlock(&the_lnet.ln_api_mutex);
 492
 493         return 0;
 494 }
 495
 496 static int
 497 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
 498 {
 499         int value, rc;
 500
 501         rc = kstrtoint(val, 0, &value);
 502         if (rc) {
 503                 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
 504                 return rc;
 505         }
 506
 507         if (value < LNET_INTERFACES_MIN) {
 508                 CWARN("max interfaces provided are too small, setting to %d\n",
 509                       LNET_INTERFACES_MAX_DEFAULT);
 510                 value = LNET_INTERFACES_MAX_DEFAULT;
 511         }
 512
 513         *(int *)kp->arg = value;
 514
 515         return 0;
 516 }
 517
 518 static int
 519 response_tracking_set(const char *val, cfs_kernel_param_arg_t *kp)
 520 {
 521         int rc;
 522         unsigned long new_value;
 523
 524         rc = kstrtoul(val, 0, &new_value);
 525         if (rc) {
 526                 CERROR("Invalid value for 'lnet_response_tracking'\n");
 527                 return -EINVAL;
 528         }
 529
 530         if (new_value < 0 || new_value > 3) {
 531                 CWARN("Invalid value (%lu) for 'lnet_response_tracking'\n",
 532                       new_value);
 533                 return -EINVAL;
 534         }
 535
 536         lnet_response_tracking = new_value;
 537
 538         return 0;
 539 }
 540
 541 static const char *
 542 lnet_get_routes(void)
 543 {
 544         return routes;
 545 }
 546
 547 static const char *
 548 lnet_get_networks(void)
 549 {
 550         const char *nets;
 551         int rc;
 552
 553         if (*networks != 0 && *ip2nets != 0) {
 554                 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
 555                                    "'ip2nets' but not both at once\n");
 556                 return NULL;
 557         }
 558
 559         if (*ip2nets != 0) {
 560                 rc = lnet_parse_ip2nets(&nets, ip2nets);
 561                 return (rc == 0) ? nets : NULL;
 562         }
 563
 564         if (*networks != 0)
 565                 return networks;
 566
 567         return "tcp";
 568 }
 569
 570 static void
 571 lnet_init_locks(void)
 572 {
 573         spin_lock_init(&the_lnet.ln_eq_wait_lock);
 574         spin_lock_init(&the_lnet.ln_msg_resend_lock);
 575         init_completion(&the_lnet.ln_mt_wait_complete);
 576         mutex_init(&the_lnet.ln_lnd_mutex);
 577 }
 578
 579 struct kmem_cache *lnet_mes_cachep;        /* MEs kmem_cache */
 580 struct kmem_cache *lnet_small_mds_cachep;  /* <= LNET_SMALL_MD_SIZE bytes
 581                                             *  MDs kmem_cache */
 582 struct kmem_cache *lnet_udsp_cachep;       /* udsp cache */
 583 struct kmem_cache *lnet_rspt_cachep;       /* response tracker cache */
 584 struct kmem_cache *lnet_msg_cachep;
 585
 586 static int
 587 lnet_slab_setup(void)
 588 {
 589         /* create specific kmem_cache for MEs and small MDs (i.e., originally
 590          * allocated in <size-xxx> kmem_cache).
 591          */
 592         lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
 593                                             0, 0, NULL);
 594         if (!lnet_mes_cachep)
 595                 return -ENOMEM;
 596
 597         lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
 598                                                   LNET_SMALL_MD_SIZE, 0, 0,
 599                                                   NULL);
 600         if (!lnet_small_mds_cachep)
 601                 return -ENOMEM;
 602
 603         lnet_udsp_cachep = kmem_cache_create("lnet_udsp",
 604                                              sizeof(struct lnet_udsp),
 605                                              0, 0, NULL);
 606         if (!lnet_udsp_cachep)
 607                 return -ENOMEM;
 608
 609         lnet_rspt_cachep = kmem_cache_create("lnet_rspt", sizeof(struct lnet_rsp_tracker),
 610                                             0, 0, NULL);
 611         if (!lnet_rspt_cachep)
 612                 return -ENOMEM;
 613
 614         lnet_msg_cachep = kmem_cache_create("lnet_msg", sizeof(struct lnet_msg),
 615                                             0, 0, NULL);
 616         if (!lnet_msg_cachep)
 617                 return -ENOMEM;
 618
 619         return 0;
 620 }
 621
 622 static void
 623 lnet_slab_cleanup(void)
 624 {
 625         if (lnet_msg_cachep) {
 626                 kmem_cache_destroy(lnet_msg_cachep);
 627                 lnet_msg_cachep = NULL;
 628         }
 629
 630         if (lnet_rspt_cachep) {
 631                 kmem_cache_destroy(lnet_rspt_cachep);
 632                 lnet_rspt_cachep = NULL;
 633         }
 634
 635         if (lnet_udsp_cachep) {
 636                 kmem_cache_destroy(lnet_udsp_cachep);
 637                 lnet_udsp_cachep = NULL;
 638         }
 639
 640         if (lnet_small_mds_cachep) {
 641                 kmem_cache_destroy(lnet_small_mds_cachep);
 642                 lnet_small_mds_cachep = NULL;
 643         }
 644
 645         if (lnet_mes_cachep) {
 646                 kmem_cache_destroy(lnet_mes_cachep);
 647                 lnet_mes_cachep = NULL;
 648         }
 649 }
 650
 651 static int
 652 lnet_create_remote_nets_table(void)
 653 {
 654         int               i;
 655         struct list_head *hash;
 656
 657         LASSERT(the_lnet.ln_remote_nets_hash == NULL);
 658         LASSERT(the_lnet.ln_remote_nets_hbits > 0);
 659         CFS_ALLOC_PTR_ARRAY(hash, LNET_REMOTE_NETS_HASH_SIZE);
 660         if (hash == NULL) {
 661                 CERROR("Failed to create remote nets hash table\n");
 662                 return -ENOMEM;
 663         }
 664
 665         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 666                 INIT_LIST_HEAD(&hash[i]);
 667         the_lnet.ln_remote_nets_hash = hash;
 668         return 0;
 669 }
 670
 671 static void
 672 lnet_destroy_remote_nets_table(void)
 673 {
 674         int i;
 675
 676         if (the_lnet.ln_remote_nets_hash == NULL)
 677                 return;
 678
 679         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 680                 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
 681
 682         CFS_FREE_PTR_ARRAY(the_lnet.ln_remote_nets_hash,
 683                            LNET_REMOTE_NETS_HASH_SIZE);
 684         the_lnet.ln_remote_nets_hash = NULL;
 685 }
 686
 687 static void
 688 lnet_destroy_locks(void)
 689 {
 690         if (the_lnet.ln_res_lock != NULL) {
 691                 cfs_percpt_lock_free(the_lnet.ln_res_lock);
 692                 the_lnet.ln_res_lock = NULL;
 693         }
 694
 695         if (the_lnet.ln_net_lock != NULL) {
 696                 cfs_percpt_lock_free(the_lnet.ln_net_lock);
 697                 the_lnet.ln_net_lock = NULL;
 698         }
 699 }
 700
 701 static int
 702 lnet_create_locks(void)
 703 {
 704         lnet_init_locks();
 705
 706         the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 707         if (the_lnet.ln_res_lock == NULL)
 708                 goto failed;
 709
 710         the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 711         if (the_lnet.ln_net_lock == NULL)
 712                 goto failed;
 713
 714         return 0;
 715
 716  failed:
 717         lnet_destroy_locks();
 718         return -ENOMEM;
 719 }
 720
 721 static void lnet_assert_wire_constants(void)
 722 {
 723         /* Wire protocol assertions generated by 'wirecheck'
 724          * running on Linux robert.bartonsoftware.com 2.6.8-1.521
 725          * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
 726          * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7)
 727          */
 728
 729         /* Constants... */
 730         BUILD_BUG_ON(LNET_PROTO_TCP_MAGIC != 0xeebc0ded);
 731         BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MAJOR != 1);
 732         BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MINOR != 0);
 733         BUILD_BUG_ON(LNET_MSG_ACK != 0);
 734         BUILD_BUG_ON(LNET_MSG_PUT != 1);
 735         BUILD_BUG_ON(LNET_MSG_GET != 2);
 736         BUILD_BUG_ON(LNET_MSG_REPLY != 3);
 737         BUILD_BUG_ON(LNET_MSG_HELLO != 4);
 738
 739         BUILD_BUG_ON((int)sizeof(lnet_nid_t) != 8);
 740         BUILD_BUG_ON((int)sizeof(lnet_pid_t) != 4);
 741
 742         /* Checks for struct lnet_process_id_packed */
 743         BUILD_BUG_ON((int)sizeof(struct lnet_process_id_packed) != 12);
 744         BUILD_BUG_ON((int)offsetof(struct lnet_process_id_packed, nid) != 0);
 745         BUILD_BUG_ON((int)sizeof(((struct lnet_process_id_packed *)0)->nid) != 8);
 746         BUILD_BUG_ON((int)offsetof(struct lnet_process_id_packed, pid) != 8);
 747         BUILD_BUG_ON((int)sizeof(((struct lnet_process_id_packed *)0)->pid) != 4);
 748
 749         /* Checks for struct lnet_handle_wire */
 750         BUILD_BUG_ON((int)sizeof(struct lnet_handle_wire) != 16);
 751         BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
 752                                    wh_interface_cookie) != 0);
 753         BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) != 8);
 754         BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
 755                                    wh_object_cookie) != 8);
 756         BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) != 8);
 757
 758         /* Checks for struct struct lnet_magicversion */
 759         BUILD_BUG_ON((int)sizeof(struct lnet_magicversion) != 8);
 760         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, magic) != 0);
 761         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->magic) != 4);
 762         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, version_major) != 4);
 763         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_major) != 2);
 764         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion,
 765                                    version_minor) != 6);
 766         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_minor) != 2);
 767
 768         /* Checks for struct struct lnet_hdr */
 769         BUILD_BUG_ON((int)sizeof(struct lnet_hdr) != 72);
 770         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_nid) != 0);
 771         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_nid) != 8);
 772         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_nid) != 8);
 773         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_nid) != 8);
 774         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_pid) != 16);
 775         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_pid) != 4);
 776         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_pid) != 20);
 777         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_pid) != 4);
 778         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, type) != 24);
 779         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->type) != 4);
 780         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, payload_length) != 28);
 781         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->payload_length) != 4);
 782         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg) != 32);
 783         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg) != 40);
 784
 785         /* Ack */
 786         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) != 32);
 787         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) != 16);
 788         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.match_bits) != 48);
 789         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) != 8);
 790         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.mlength) != 56);
 791         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) != 4);
 792
 793         /* Put */
 794         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) != 32);
 795         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) != 16);
 796         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.match_bits) != 48);
 797         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) != 8);
 798         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.hdr_data) != 56);
 799         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) != 8);
 800         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ptl_index) != 64);
 801         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) != 4);
 802         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.offset) != 68);
 803         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) != 4);
 804
 805         /* Get */
 806         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.return_wmd) != 32);
 807         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) != 16);
 808         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.match_bits) != 48);
 809         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) != 8);
 810         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.ptl_index) != 56);
 811         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) != 4);
 812         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.src_offset) != 60);
 813         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) != 4);
 814         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.sink_length) != 64);
 815         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) != 4);
 816
 817         /* Reply */
 818         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) != 32);
 819         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) != 16);
 820
 821         /* Hello */
 822         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.incarnation) != 32);
 823         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) != 8);
 824         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.type) != 40);
 825         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) != 4);
 826
 827         /* Checks for struct lnet_ni_status and related constants */
 828         BUILD_BUG_ON(LNET_NI_STATUS_INVALID != 0x00000000);
 829         BUILD_BUG_ON(LNET_NI_STATUS_UP != 0x15aac0de);
 830         BUILD_BUG_ON(LNET_NI_STATUS_DOWN != 0xdeadface);
 831
 832         /* Checks for struct lnet_ni_status */
 833         BUILD_BUG_ON((int)sizeof(struct lnet_ni_status) != 16);
 834         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_nid) != 0);
 835         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) != 8);
 836         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_status) != 8);
 837         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_status) != 4);
 838         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_unused) != 12);
 839         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) != 4);
 840
 841         /* Checks for struct lnet_ping_info and related constants */
 842         BUILD_BUG_ON(LNET_PROTO_PING_MAGIC != 0x70696E67);
 843         BUILD_BUG_ON(LNET_PING_FEAT_INVAL != 0);
 844         BUILD_BUG_ON(LNET_PING_FEAT_BASE != 1);
 845         BUILD_BUG_ON(LNET_PING_FEAT_NI_STATUS != 2);
 846         BUILD_BUG_ON(LNET_PING_FEAT_RTE_DISABLED != 4);
 847         BUILD_BUG_ON(LNET_PING_FEAT_MULTI_RAIL != 8);
 848         BUILD_BUG_ON(LNET_PING_FEAT_DISCOVERY != 16);
 849         BUILD_BUG_ON(LNET_PING_FEAT_BITS != 31);
 850
 851         /* Checks for struct lnet_ping_info */
 852         BUILD_BUG_ON((int)sizeof(struct lnet_ping_info) != 16);
 853         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_magic) != 0);
 854         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) != 4);
 855         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_features) != 4);
 856         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_features) != 4);
 857         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_pid) != 8);
 858         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) != 4);
 859         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_nnis) != 12);
 860         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) != 4);
 861         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_ni) != 16);
 862         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) != 0);
 863
 864         /* Acceptor connection request */
 865         BUILD_BUG_ON(LNET_PROTO_ACCEPTOR_VERSION != 1);
 866
 867         /* Checks for struct lnet_acceptor_connreq */
 868         BUILD_BUG_ON((int)sizeof(struct lnet_acceptor_connreq) != 16);
 869         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_magic) != 0);
 870         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_magic) != 4);
 871         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_version) != 4);
 872         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_version) != 4);
 873         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_nid) != 8);
 874         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_nid) != 8);
 875
 876         /* Checks for struct lnet_counters_common */
 877         BUILD_BUG_ON((int)sizeof(struct lnet_counters_common) != 60);
 878         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_msgs_alloc) != 0);
 879         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_msgs_alloc) != 4);
 880         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_msgs_max) != 4);
 881         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_msgs_max) != 4);
 882         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_errors) != 8);
 883         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_errors) != 4);
 884         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_send_count) != 12);
 885         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_send_count) != 4);
 886         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_recv_count) != 16);
 887         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_recv_count) != 4);
 888         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_route_count) != 20);
 889         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_route_count) != 4);
 890         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_drop_count) != 24);
 891         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_drop_count) != 4);
 892         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_send_length) != 28);
 893         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_send_length) != 8);
 894         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_recv_length) != 36);
 895         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_recv_length) != 8);
 896         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_route_length) != 44);
 897         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_route_length) != 8);
 898         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_drop_length) != 52);
 899         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_drop_length) != 8);
 900 }
 901
 902 static const struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
 903 {
 904         const struct lnet_lnd *lnd;
 905
 906         /* holding lnd mutex */
 907         if (type >= NUM_LNDS)
 908                 return NULL;
 909         lnd = the_lnet.ln_lnds[type];
 910         LASSERT(!lnd || lnd->lnd_type == type);
 911
 912         return lnd;
 913 }
 914
 915 unsigned int
 916 lnet_get_lnd_timeout(void)
 917 {
 918         return lnet_lnd_timeout;
 919 }
 920 EXPORT_SYMBOL(lnet_get_lnd_timeout);
 921
 922 void
 923 lnet_register_lnd(const struct lnet_lnd *lnd)
 924 {
 925         mutex_lock(&the_lnet.ln_lnd_mutex);
 926
 927         LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
 928         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
 929
 930         the_lnet.ln_lnds[lnd->lnd_type] = lnd;
 931
 932         CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
 933
 934         mutex_unlock(&the_lnet.ln_lnd_mutex);
 935 }
 936 EXPORT_SYMBOL(lnet_register_lnd);
 937
 938 void
 939 lnet_unregister_lnd(const struct lnet_lnd *lnd)
 940 {
 941         mutex_lock(&the_lnet.ln_lnd_mutex);
 942
 943         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
 944
 945         the_lnet.ln_lnds[lnd->lnd_type] = NULL;
 946         CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
 947
 948         mutex_unlock(&the_lnet.ln_lnd_mutex);
 949 }
 950 EXPORT_SYMBOL(lnet_unregister_lnd);
 951
 952 static void
 953 lnet_counters_get_common_locked(struct lnet_counters_common *common)
 954 {
 955         struct lnet_counters *ctr;
 956         int i;
 957
 958         /* FIXME !!! Their is no assert_lnet_net_locked() to ensure this
 959          * actually called under the protection of the lnet_net_lock.
 960          */
 961         memset(common, 0, sizeof(*common));
 962
 963         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
 964                 common->lcc_msgs_max     += ctr->lct_common.lcc_msgs_max;
 965                 common->lcc_msgs_alloc   += ctr->lct_common.lcc_msgs_alloc;
 966                 common->lcc_errors       += ctr->lct_common.lcc_errors;
 967                 common->lcc_send_count   += ctr->lct_common.lcc_send_count;
 968                 common->lcc_recv_count   += ctr->lct_common.lcc_recv_count;
 969                 common->lcc_route_count  += ctr->lct_common.lcc_route_count;
 970                 common->lcc_drop_count   += ctr->lct_common.lcc_drop_count;
 971                 common->lcc_send_length  += ctr->lct_common.lcc_send_length;
 972                 common->lcc_recv_length  += ctr->lct_common.lcc_recv_length;
 973                 common->lcc_route_length += ctr->lct_common.lcc_route_length;
 974                 common->lcc_drop_length  += ctr->lct_common.lcc_drop_length;
 975         }
 976 }
 977
 978 void
 979 lnet_counters_get_common(struct lnet_counters_common *common)
 980 {
 981         lnet_net_lock(LNET_LOCK_EX);
 982         lnet_counters_get_common_locked(common);
 983         lnet_net_unlock(LNET_LOCK_EX);
 984 }
 985 EXPORT_SYMBOL(lnet_counters_get_common);
 986
 987 int
 988 lnet_counters_get(struct lnet_counters *counters)
 989 {
 990         struct lnet_counters *ctr;
 991         struct lnet_counters_health *health = &counters->lct_health;
 992         int i, rc = 0;
 993
 994         memset(counters, 0, sizeof(*counters));
 995
 996         lnet_net_lock(LNET_LOCK_EX);
 997
 998         if (the_lnet.ln_state != LNET_STATE_RUNNING)
 999                 GOTO(out_unlock, rc = -ENODEV);
1000
1001         lnet_counters_get_common_locked(&counters->lct_common);
1002
1003         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
1004                 health->lch_rst_alloc    += ctr->lct_health.lch_rst_alloc;
1005                 health->lch_resend_count += ctr->lct_health.lch_resend_count;
1006                 health->lch_response_timeout_count +=
1007                                 ctr->lct_health.lch_response_timeout_count;
1008                 health->lch_local_interrupt_count +=
1009                                 ctr->lct_health.lch_local_interrupt_count;
1010                 health->lch_local_dropped_count +=
1011                                 ctr->lct_health.lch_local_dropped_count;
1012                 health->lch_local_aborted_count +=
1013                                 ctr->lct_health.lch_local_aborted_count;
1014                 health->lch_local_no_route_count +=
1015                                 ctr->lct_health.lch_local_no_route_count;
1016                 health->lch_local_timeout_count +=
1017                                 ctr->lct_health.lch_local_timeout_count;
1018                 health->lch_local_error_count +=
1019                                 ctr->lct_health.lch_local_error_count;
1020                 health->lch_remote_dropped_count +=
1021                                 ctr->lct_health.lch_remote_dropped_count;
1022                 health->lch_remote_error_count +=
1023                                 ctr->lct_health.lch_remote_error_count;
1024                 health->lch_remote_timeout_count +=
1025                                 ctr->lct_health.lch_remote_timeout_count;
1026                 health->lch_network_timeout_count +=
1027                                 ctr->lct_health.lch_network_timeout_count;
1028         }
1029 out_unlock:
1030         lnet_net_unlock(LNET_LOCK_EX);
1031         return rc;
1032 }
1033 EXPORT_SYMBOL(lnet_counters_get);
1034
1035 void
1036 lnet_counters_reset(void)
1037 {
1038         struct lnet_counters *counters;
1039         int             i;
1040
1041         lnet_net_lock(LNET_LOCK_EX);
1042
1043         if (the_lnet.ln_state != LNET_STATE_RUNNING)
1044                 goto avoid_reset;
1045
1046         cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
1047                 memset(counters, 0, sizeof(struct lnet_counters));
1048 avoid_reset:
1049         lnet_net_unlock(LNET_LOCK_EX);
1050 }
1051
1052 static char *
1053 lnet_res_type2str(int type)
1054 {
1055         switch (type) {
1056         default:
1057                 LBUG();
1058         case LNET_COOKIE_TYPE_MD:
1059                 return "MD";
1060         case LNET_COOKIE_TYPE_ME:
1061                 return "ME";
1062         case LNET_COOKIE_TYPE_EQ:
1063                 return "EQ";
1064         }
1065 }
1066
1067 static void
1068 lnet_res_container_cleanup(struct lnet_res_container *rec)
1069 {
1070         int     count = 0;
1071
1072         if (rec->rec_type == 0) /* not set yet, it's uninitialized */
1073                 return;
1074
1075         while (!list_empty(&rec->rec_active)) {
1076                 struct list_head *e = rec->rec_active.next;
1077
1078                 list_del_init(e);
1079                 if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
1080                         lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
1081
1082                 } else { /* NB: Active MEs should be attached on portals */
1083                         LBUG();
1084                 }
1085                 count++;
1086         }
1087
1088         if (count > 0) {
1089                 /* Found alive MD/ME/EQ, user really should unlink/free
1090                  * all of them before finalize LNet, but if someone didn't,
1091                  * we have to recycle garbage for him */
1092                 CERROR("%d active elements on exit of %s container\n",
1093                        count, lnet_res_type2str(rec->rec_type));
1094         }
1095
1096         if (rec->rec_lh_hash != NULL) {
1097                 CFS_FREE_PTR_ARRAY(rec->rec_lh_hash, LNET_LH_HASH_SIZE);
1098                 rec->rec_lh_hash = NULL;
1099         }
1100
1101         rec->rec_type = 0; /* mark it as finalized */
1102 }
1103
1104 static int
1105 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
1106 {
1107         int     rc = 0;
1108         int     i;
1109
1110         LASSERT(rec->rec_type == 0);
1111
1112         rec->rec_type = type;
1113         INIT_LIST_HEAD(&rec->rec_active);
1114
1115         rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1116
1117         /* Arbitrary choice of hash table size */
1118         LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1119                          LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1120         if (rec->rec_lh_hash == NULL) {
1121                 rc = -ENOMEM;
1122                 goto out;
1123         }
1124
1125         for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1126                 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1127
1128         return 0;
1129
1130 out:
1131         CERROR("Failed to setup %s resource container\n",
1132                lnet_res_type2str(type));
1133         lnet_res_container_cleanup(rec);
1134         return rc;
1135 }
1136
1137 static void
1138 lnet_res_containers_destroy(struct lnet_res_container **recs)
1139 {
1140         struct lnet_res_container       *rec;
1141         int                             i;
1142
1143         cfs_percpt_for_each(rec, i, recs)
1144                 lnet_res_container_cleanup(rec);
1145
1146         cfs_percpt_free(recs);
1147 }
1148
1149 static struct lnet_res_container **
1150 lnet_res_containers_create(int type)
1151 {
1152         struct lnet_res_container       **recs;
1153         struct lnet_res_container       *rec;
1154         int                             rc;
1155         int                             i;
1156
1157         recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1158         if (recs == NULL) {
1159                 CERROR("Failed to allocate %s resource containers\n",
1160                        lnet_res_type2str(type));
1161                 return NULL;
1162         }
1163
1164         cfs_percpt_for_each(rec, i, recs) {
1165                 rc = lnet_res_container_setup(rec, i, type);
1166                 if (rc != 0) {
1167                         lnet_res_containers_destroy(recs);
1168                         return NULL;
1169                 }
1170         }
1171
1172         return recs;
1173 }
1174
1175 struct lnet_libhandle *
1176 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1177 {
1178         /* ALWAYS called with lnet_res_lock held */
1179         struct list_head        *head;
1180         struct lnet_libhandle   *lh;
1181         unsigned int            hash;
1182
1183         if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1184                 return NULL;
1185
1186         hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1187         head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1188
1189         list_for_each_entry(lh, head, lh_hash_chain) {
1190                 if (lh->lh_cookie == cookie)
1191                         return lh;
1192         }
1193
1194         return NULL;
1195 }
1196
1197 void
1198 lnet_res_lh_initialize(struct lnet_res_container *rec,
1199                        struct lnet_libhandle *lh)
1200 {
1201         /* ALWAYS called with lnet_res_lock held */
1202         unsigned int    ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1203         unsigned int    hash;
1204
1205         lh->lh_cookie = rec->rec_lh_cookie;
1206         rec->rec_lh_cookie += 1 << ibits;
1207
1208         hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1209
1210         list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1211 }
1212
1213 struct list_head **
1214 lnet_create_array_of_queues(void)
1215 {
1216         struct list_head **qs;
1217         struct list_head *q;
1218         int i;
1219
1220         qs = cfs_percpt_alloc(lnet_cpt_table(),
1221                               sizeof(struct list_head));
1222         if (!qs) {
1223                 CERROR("Failed to allocate queues\n");
1224                 return NULL;
1225         }
1226
1227         cfs_percpt_for_each(q, i, qs)
1228                 INIT_LIST_HEAD(q);
1229
1230         return qs;
1231 }
1232
1233 static int lnet_unprepare(void);
1234
1235 static int
1236 lnet_prepare(lnet_pid_t requested_pid)
1237 {
1238         /* Prepare to bring up the network */
1239         struct lnet_res_container **recs;
1240         int                       rc = 0;
1241
1242         if (requested_pid == LNET_PID_ANY) {
1243                 /* Don't instantiate LNET just for me */
1244                 return -ENETDOWN;
1245         }
1246
1247         LASSERT(the_lnet.ln_refcount == 0);
1248
1249         the_lnet.ln_routing = 0;
1250
1251         LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1252         the_lnet.ln_pid = requested_pid;
1253
1254         INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1255         INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1256         INIT_LIST_HEAD(&the_lnet.ln_nets);
1257         INIT_LIST_HEAD(&the_lnet.ln_routers);
1258         INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1259         INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1260         INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1261         INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1262         INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1263         INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1264         INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1265         INIT_LIST_HEAD(&the_lnet.ln_udsp_list);
1266         init_waitqueue_head(&the_lnet.ln_dc_waitq);
1267         the_lnet.ln_mt_handler = NULL;
1268         init_completion(&the_lnet.ln_started);
1269
1270         rc = lnet_slab_setup();
1271         if (rc != 0)
1272                 goto failed;
1273
1274         rc = lnet_create_remote_nets_table();
1275         if (rc != 0)
1276                 goto failed;
1277
1278         /*
1279          * NB the interface cookie in wire handles guards against delayed
1280          * replies and ACKs appearing valid after reboot.
1281          */
1282         the_lnet.ln_interface_cookie = ktime_get_real_ns();
1283
1284         the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1285                                                 sizeof(struct lnet_counters));
1286         if (the_lnet.ln_counters == NULL) {
1287                 CERROR("Failed to allocate counters for LNet\n");
1288                 rc = -ENOMEM;
1289                 goto failed;
1290         }
1291
1292         rc = lnet_peer_tables_create();
1293         if (rc != 0)
1294                 goto failed;
1295
1296         rc = lnet_msg_containers_create();
1297         if (rc != 0)
1298                 goto failed;
1299
1300         rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1301                                       LNET_COOKIE_TYPE_EQ);
1302         if (rc != 0)
1303                 goto failed;
1304
1305         recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1306         if (recs == NULL) {
1307                 rc = -ENOMEM;
1308                 goto failed;
1309         }
1310
1311         the_lnet.ln_md_containers = recs;
1312
1313         rc = lnet_portals_create();
1314         if (rc != 0) {
1315                 CERROR("Failed to create portals for LNet: %d\n", rc);
1316                 goto failed;
1317         }
1318
1319         the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1320         if (!the_lnet.ln_mt_zombie_rstqs) {
1321                 rc = -ENOMEM;
1322                 goto failed;
1323         }
1324
1325         return 0;
1326
1327  failed:
1328         lnet_unprepare();
1329         return rc;
1330 }
1331
1332 static int
1333 lnet_unprepare (void)
1334 {
1335         /* NB no LNET_LOCK since this is the last reference.  All LND instances
1336          * have shut down already, so it is safe to unlink and free all
1337          * descriptors, even those that appear committed to a network op (eg MD
1338          * with non-zero pending count) */
1339
1340         lnet_fail_nid(LNET_NID_ANY, 0);
1341
1342         LASSERT(the_lnet.ln_refcount == 0);
1343         LASSERT(list_empty(&the_lnet.ln_test_peers));
1344         LASSERT(list_empty(&the_lnet.ln_nets));
1345
1346         if (the_lnet.ln_mt_zombie_rstqs) {
1347                 lnet_clean_zombie_rstqs();
1348                 the_lnet.ln_mt_zombie_rstqs = NULL;
1349         }
1350
1351         lnet_assert_handler_unused(the_lnet.ln_mt_handler);
1352         the_lnet.ln_mt_handler = NULL;
1353
1354         lnet_portals_destroy();
1355
1356         if (the_lnet.ln_md_containers != NULL) {
1357                 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1358                 the_lnet.ln_md_containers = NULL;
1359         }
1360
1361         lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1362
1363         lnet_msg_containers_destroy();
1364         lnet_peer_uninit();
1365         lnet_rtrpools_free(0);
1366
1367         if (the_lnet.ln_counters != NULL) {
1368                 cfs_percpt_free(the_lnet.ln_counters);
1369                 the_lnet.ln_counters = NULL;
1370         }
1371         lnet_destroy_remote_nets_table();
1372         lnet_udsp_destroy(true);
1373         lnet_slab_cleanup();
1374
1375         return 0;
1376 }
1377
1378 struct lnet_ni  *
1379 lnet_net2ni_locked(__u32 net_id, int cpt)
1380 {
1381         struct lnet_ni   *ni;
1382         struct lnet_net  *net;
1383
1384         LASSERT(cpt != LNET_LOCK_EX);
1385
1386         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1387                 if (net->net_id == net_id) {
1388                         ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1389                                         ni_netlist);
1390                         return ni;
1391                 }
1392         }
1393
1394         return NULL;
1395 }
1396
1397 struct lnet_ni *
1398 lnet_net2ni_addref(__u32 net)
1399 {
1400         struct lnet_ni *ni;
1401
1402         lnet_net_lock(0);
1403         ni = lnet_net2ni_locked(net, 0);
1404         if (ni)
1405                 lnet_ni_addref_locked(ni, 0);
1406         lnet_net_unlock(0);
1407
1408         return ni;
1409 }
1410 EXPORT_SYMBOL(lnet_net2ni_addref);
1411
1412 struct lnet_net *
1413 lnet_get_net_locked(__u32 net_id)
1414 {
1415         struct lnet_net  *net;
1416
1417         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1418                 if (net->net_id == net_id)
1419                         return net;
1420         }
1421
1422         return NULL;
1423 }
1424
1425 void
1426 lnet_net_clr_pref_rtrs(struct lnet_net *net)
1427 {
1428         struct list_head zombies;
1429         struct lnet_nid_list *ne;
1430         struct lnet_nid_list *tmp;
1431
1432         INIT_LIST_HEAD(&zombies);
1433
1434         lnet_net_lock(LNET_LOCK_EX);
1435         list_splice_init(&net->net_rtr_pref_nids, &zombies);
1436         lnet_net_unlock(LNET_LOCK_EX);
1437
1438         list_for_each_entry_safe(ne, tmp, &zombies, nl_list) {
1439                 list_del_init(&ne->nl_list);
1440                 LIBCFS_FREE(ne, sizeof(*ne));
1441         }
1442 }
1443
1444 int
1445 lnet_net_add_pref_rtr(struct lnet_net *net,
1446                       lnet_nid_t gw_nid)
1447 __must_hold(&the_lnet.ln_api_mutex)
1448 {
1449         struct lnet_nid_list *ne;
1450
1451         /* This function is called with api_mutex held. When the api_mutex
1452          * is held the list can not be modified, as it is only modified as
1453          * a result of applying a UDSP and that happens under api_mutex
1454          * lock.
1455          */
1456         list_for_each_entry(ne, &net->net_rtr_pref_nids, nl_list) {
1457                 if (ne->nl_nid == gw_nid)
1458                         return -EEXIST;
1459         }
1460
1461         LIBCFS_ALLOC(ne, sizeof(*ne));
1462         if (!ne)
1463                 return -ENOMEM;
1464
1465         ne->nl_nid = gw_nid;
1466
1467         /* Lock the cpt to protect against addition and checks in the
1468          * selection algorithm
1469          */
1470         lnet_net_lock(LNET_LOCK_EX);
1471         list_add(&ne->nl_list, &net->net_rtr_pref_nids);
1472         lnet_net_unlock(LNET_LOCK_EX);
1473
1474         return 0;
1475 }
1476
1477 bool
1478 lnet_net_is_pref_rtr_locked(struct lnet_net *net, lnet_nid_t rtr_nid)
1479 {
1480         struct lnet_nid_list *ne;
1481
1482         CDEBUG(D_NET, "%s: rtr pref emtpy: %d\n",
1483                libcfs_net2str(net->net_id),
1484                list_empty(&net->net_rtr_pref_nids));
1485
1486         if (list_empty(&net->net_rtr_pref_nids))
1487                 return false;
1488
1489         list_for_each_entry(ne, &net->net_rtr_pref_nids, nl_list) {
1490                 CDEBUG(D_NET, "Comparing pref %s with gw %s\n",
1491                        libcfs_nid2str(ne->nl_nid),
1492                        libcfs_nid2str(rtr_nid));
1493                 if (rtr_nid == ne->nl_nid)
1494                         return true;
1495         }
1496
1497         return false;
1498 }
1499
1500 unsigned int
1501 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1502 {
1503         __u64           key = nid;
1504         unsigned int    val;
1505
1506         LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1507
1508         if (number == 1)
1509                 return 0;
1510
1511         val = hash_long(key, LNET_CPT_BITS);
1512         /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1513         if (val < number)
1514                 return val;
1515
1516         return (unsigned int)(key + val + (val >> 1)) % number;
1517 }
1518
1519 int
1520 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1521 {
1522         struct lnet_net *net;
1523
1524         /* must called with hold of lnet_net_lock */
1525         if (LNET_CPT_NUMBER == 1)
1526                 return 0; /* the only one */
1527
1528         /*
1529          * If NI is provided then use the CPT identified in the NI cpt
1530          * list if one exists. If one doesn't exist, then that NI is
1531          * associated with all CPTs and it follows that the net it belongs
1532          * to is implicitly associated with all CPTs, so just hash the nid
1533          * and return that.
1534          */
1535         if (ni != NULL) {
1536                 if (ni->ni_cpts != NULL)
1537                         return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1538                                                              ni->ni_ncpts)];
1539                 else
1540                         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1541         }
1542
1543         /* no NI provided so look at the net */
1544         net = lnet_get_net_locked(LNET_NIDNET(nid));
1545
1546         if (net != NULL && net->net_cpts != NULL) {
1547                 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1548         }
1549
1550         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1551 }
1552
1553 int
1554 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1555 {
1556         int     cpt;
1557         int     cpt2;
1558
1559         if (LNET_CPT_NUMBER == 1)
1560                 return 0; /* the only one */
1561
1562         cpt = lnet_net_lock_current();
1563
1564         cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1565
1566         lnet_net_unlock(cpt);
1567
1568         return cpt2;
1569 }
1570 EXPORT_SYMBOL(lnet_cpt_of_nid);
1571
1572 int
1573 lnet_islocalnet_locked(__u32 net_id)
1574 {
1575         struct lnet_net *net;
1576         bool local;
1577
1578         net = lnet_get_net_locked(net_id);
1579
1580         local = net != NULL;
1581
1582         return local;
1583 }
1584
1585 int
1586 lnet_islocalnet(__u32 net_id)
1587 {
1588         int cpt;
1589         bool local;
1590
1591         cpt = lnet_net_lock_current();
1592
1593         local = lnet_islocalnet_locked(net_id);
1594
1595         lnet_net_unlock(cpt);
1596
1597         return local;
1598 }
1599
1600 struct lnet_ni  *
1601 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
1602 {
1603         struct lnet_net  *net;
1604         struct lnet_ni   *ni;
1605
1606         LASSERT(cpt != LNET_LOCK_EX);
1607
1608         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1609                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1610                         if (ni->ni_nid == nid)
1611                                 return ni;
1612                 }
1613         }
1614
1615         return NULL;
1616 }
1617
1618 struct lnet_ni *
1619 lnet_nid2ni_addref(lnet_nid_t nid)
1620 {
1621         struct lnet_ni *ni;
1622
1623         lnet_net_lock(0);
1624         ni = lnet_nid2ni_locked(nid, 0);
1625         if (ni)
1626                 lnet_ni_addref_locked(ni, 0);
1627         lnet_net_unlock(0);
1628
1629         return ni;
1630 }
1631 EXPORT_SYMBOL(lnet_nid2ni_addref);
1632
1633 int
1634 lnet_islocalnid(lnet_nid_t nid)
1635 {
1636         struct lnet_ni  *ni;
1637         int             cpt;
1638
1639         cpt = lnet_net_lock_current();
1640         ni = lnet_nid2ni_locked(nid, cpt);
1641         lnet_net_unlock(cpt);
1642
1643         return ni != NULL;
1644 }
1645
1646 int
1647 lnet_count_acceptor_nets(void)
1648 {
1649         /* Return the # of NIs that need the acceptor. */
1650         int              count = 0;
1651         struct lnet_net  *net;
1652         int              cpt;
1653
1654         cpt = lnet_net_lock_current();
1655         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1656                 /* all socklnd type networks should have the acceptor
1657                  * thread started */
1658                 if (net->net_lnd->lnd_accept != NULL)
1659                         count++;
1660         }
1661
1662         lnet_net_unlock(cpt);
1663
1664         return count;
1665 }
1666
1667 struct lnet_ping_buffer *
1668 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1669 {
1670         struct lnet_ping_buffer *pbuf;
1671
1672         LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1673         if (pbuf) {
1674                 pbuf->pb_nnis = nnis;
1675                 pbuf->pb_needs_post = false;
1676                 atomic_set(&pbuf->pb_refcnt, 1);
1677         }
1678
1679         return pbuf;
1680 }
1681
1682 void
1683 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1684 {
1685         LASSERT(atomic_read(&pbuf->pb_refcnt) == 0);
1686         LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1687 }
1688
1689 static struct lnet_ping_buffer *
1690 lnet_ping_target_create(int nnis)
1691 {
1692         struct lnet_ping_buffer *pbuf;
1693
1694         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1695         if (pbuf == NULL) {
1696                 CERROR("Can't allocate ping source [%d]\n", nnis);
1697                 return NULL;
1698         }
1699
1700         pbuf->pb_info.pi_nnis = nnis;
1701         pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1702         pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1703         pbuf->pb_info.pi_features =
1704                 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1705
1706         return pbuf;
1707 }
1708
1709 static inline int
1710 lnet_get_net_ni_count_locked(struct lnet_net *net)
1711 {
1712         struct lnet_ni  *ni;
1713         int             count = 0;
1714
1715         list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1716                 count++;
1717
1718         return count;
1719 }
1720
1721 static inline int
1722 lnet_get_net_ni_count_pre(struct lnet_net *net)
1723 {
1724         struct lnet_ni  *ni;
1725         int             count = 0;
1726
1727         list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1728                 count++;
1729
1730         return count;
1731 }
1732
1733 static inline int
1734 lnet_get_ni_count(void)
1735 {
1736         struct lnet_ni  *ni;
1737         struct lnet_net *net;
1738         int             count = 0;
1739
1740         lnet_net_lock(0);
1741
1742         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1743                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1744                         count++;
1745         }
1746
1747         lnet_net_unlock(0);
1748
1749         return count;
1750 }
1751
1752 void
1753 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1754 {
1755         struct lnet_ni_status *stat;
1756         int nnis;
1757         int i;
1758
1759         __swab32s(&pbuf->pb_info.pi_magic);
1760         __swab32s(&pbuf->pb_info.pi_features);
1761         __swab32s(&pbuf->pb_info.pi_pid);
1762         __swab32s(&pbuf->pb_info.pi_nnis);
1763         nnis = pbuf->pb_info.pi_nnis;
1764         if (nnis > pbuf->pb_nnis)
1765                 nnis = pbuf->pb_nnis;
1766         for (i = 0; i < nnis; i++) {
1767                 stat = &pbuf->pb_info.pi_ni[i];
1768                 __swab64s(&stat->ns_nid);
1769                 __swab32s(&stat->ns_status);
1770         }
1771 }
1772
1773 int
1774 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1775 {
1776         if (!pinfo)
1777                 return -EINVAL;
1778         if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1779                 return -EPROTO;
1780         if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1781                 return -EPROTO;
1782         /* Loopback is guaranteed to be present */
1783         if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1784                 return -ERANGE;
1785         if (LNET_PING_INFO_LONI(pinfo) != LNET_NID_LO_0)
1786                 return -EPROTO;
1787         return 0;
1788 }
1789
1790 static void
1791 lnet_ping_target_destroy(void)
1792 {
1793         struct lnet_net *net;
1794         struct lnet_ni  *ni;
1795
1796         lnet_net_lock(LNET_LOCK_EX);
1797
1798         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1799                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1800                         lnet_ni_lock(ni);
1801                         ni->ni_status = NULL;
1802                         lnet_ni_unlock(ni);
1803                 }
1804         }
1805
1806         lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1807         the_lnet.ln_ping_target = NULL;
1808
1809         lnet_net_unlock(LNET_LOCK_EX);
1810 }
1811
1812 static void
1813 lnet_ping_target_event_handler(struct lnet_event *event)
1814 {
1815         struct lnet_ping_buffer *pbuf = event->md_user_ptr;
1816
1817         if (event->unlinked)
1818                 lnet_ping_buffer_decref(pbuf);
1819 }
1820
1821 static int
1822 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1823                        struct lnet_handle_md *ping_mdh,
1824                        int ni_count, bool set_eq)
1825 {
1826         struct lnet_process_id id = {
1827                 .nid = LNET_NID_ANY,
1828                 .pid = LNET_PID_ANY
1829         };
1830         struct lnet_me *me;
1831         struct lnet_md md = { NULL };
1832         int rc;
1833
1834         if (set_eq)
1835                 the_lnet.ln_ping_target_handler =
1836                         lnet_ping_target_event_handler;
1837
1838         *ppbuf = lnet_ping_target_create(ni_count);
1839         if (*ppbuf == NULL) {
1840                 rc = -ENOMEM;
1841                 goto fail_free_eq;
1842         }
1843
1844         /* Ping target ME/MD */
1845         me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1846                           LNET_PROTO_PING_MATCHBITS, 0,
1847                           LNET_UNLINK, LNET_INS_AFTER);
1848         if (IS_ERR(me)) {
1849                 rc = PTR_ERR(me);
1850                 CERROR("Can't create ping target ME: %d\n", rc);
1851                 goto fail_decref_ping_buffer;
1852         }
1853
1854         /* initialize md content */
1855         md.start     = &(*ppbuf)->pb_info;
1856         md.length    = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1857         md.threshold = LNET_MD_THRESH_INF;
1858         md.max_size  = 0;
1859         md.options   = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1860                        LNET_MD_MANAGE_REMOTE;
1861         md.handler   = the_lnet.ln_ping_target_handler;
1862         md.user_ptr  = *ppbuf;
1863
1864         rc = LNetMDAttach(me, &md, LNET_RETAIN, ping_mdh);
1865         if (rc != 0) {
1866                 CERROR("Can't attach ping target MD: %d\n", rc);
1867                 goto fail_decref_ping_buffer;
1868         }
1869         lnet_ping_buffer_addref(*ppbuf);
1870
1871         return 0;
1872
1873 fail_decref_ping_buffer:
1874         LASSERT(atomic_read(&(*ppbuf)->pb_refcnt) == 1);
1875         lnet_ping_buffer_decref(*ppbuf);
1876         *ppbuf = NULL;
1877 fail_free_eq:
1878         return rc;
1879 }
1880
1881 static void
1882 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1883                     struct lnet_handle_md *ping_mdh)
1884 {
1885         LNetMDUnlink(*ping_mdh);
1886         LNetInvalidateMDHandle(ping_mdh);
1887
1888         /* NB the MD could be busy; this just starts the unlink */
1889         wait_var_event_warning(&pbuf->pb_refcnt,
1890                                atomic_read(&pbuf->pb_refcnt) <= 1,
1891                                "Still waiting for ping data MD to unlink\n");
1892 }
1893
1894 static void
1895 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1896 {
1897         struct lnet_ni          *ni;
1898         struct lnet_net         *net;
1899         struct lnet_ni_status *ns;
1900         int                     i;
1901         int                     rc;
1902
1903         i = 0;
1904         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1905                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1906                         LASSERT(i < pbuf->pb_nnis);
1907
1908                         ns = &pbuf->pb_info.pi_ni[i];
1909
1910                         ns->ns_nid = ni->ni_nid;
1911
1912                         lnet_ni_lock(ni);
1913                         ns->ns_status = (ni->ni_status != NULL) ?
1914                                          ni->ni_status->ns_status :
1915                                                 LNET_NI_STATUS_UP;
1916                         ni->ni_status = ns;
1917                         lnet_ni_unlock(ni);
1918
1919                         i++;
1920                 }
1921         }
1922         /*
1923          * We (ab)use the ns_status of the loopback interface to
1924          * transmit the sequence number. The first interface listed
1925          * must be the loopback interface.
1926          */
1927         rc = lnet_ping_info_validate(&pbuf->pb_info);
1928         if (rc) {
1929                 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1930                 LBUG();
1931         }
1932         LNET_PING_BUFFER_SEQNO(pbuf) =
1933                 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1934 }
1935
1936 static void
1937 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1938                         struct lnet_handle_md ping_mdh)
1939 {
1940         struct lnet_ping_buffer *old_pbuf = NULL;
1941         struct lnet_handle_md old_ping_md;
1942
1943         /* switch the NIs to point to the new ping info created */
1944         lnet_net_lock(LNET_LOCK_EX);
1945
1946         if (!the_lnet.ln_routing)
1947                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1948         if (!lnet_peer_discovery_disabled)
1949                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1950
1951         /* Ensure only known feature bits have been set. */
1952         LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1953         LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1954
1955         lnet_ping_target_install_locked(pbuf);
1956
1957         if (the_lnet.ln_ping_target) {
1958                 old_pbuf = the_lnet.ln_ping_target;
1959                 old_ping_md = the_lnet.ln_ping_target_md;
1960         }
1961         the_lnet.ln_ping_target_md = ping_mdh;
1962         the_lnet.ln_ping_target = pbuf;
1963
1964         lnet_net_unlock(LNET_LOCK_EX);
1965
1966         if (old_pbuf) {
1967                 /* unlink and free the old ping info */
1968                 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1969                 lnet_ping_buffer_decref(old_pbuf);
1970         }
1971
1972         lnet_push_update_to_peers(0);
1973 }
1974
1975 static void
1976 lnet_ping_target_fini(void)
1977 {
1978         lnet_ping_md_unlink(the_lnet.ln_ping_target,
1979                             &the_lnet.ln_ping_target_md);
1980
1981         lnet_assert_handler_unused(the_lnet.ln_ping_target_handler);
1982         lnet_ping_target_destroy();
1983 }
1984
1985 /* Resize the push target. */
1986 int lnet_push_target_resize(void)
1987 {
1988         struct lnet_handle_md mdh;
1989         struct lnet_handle_md old_mdh;
1990         struct lnet_ping_buffer *pbuf;
1991         struct lnet_ping_buffer *old_pbuf;
1992         int nnis;
1993         int rc;
1994
1995 again:
1996         nnis = the_lnet.ln_push_target_nnis;
1997         if (nnis <= 0) {
1998                 CDEBUG(D_NET, "Invalid nnis %d\n", nnis);
1999                 return -EINVAL;
2000         }
2001
2002         /* NB: lnet_ping_buffer_alloc() sets pbuf refcount to 1. That ref is
2003          * dropped when we need to resize again (see "old_pbuf" below) or when
2004          * LNet is shutdown (see lnet_push_target_fini())
2005          */
2006         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
2007         if (!pbuf) {
2008                 CDEBUG(D_NET, "Can't allocate pbuf for nnis %d\n", nnis);
2009                 return -ENOMEM;
2010         }
2011
2012         rc = lnet_push_target_post(pbuf, &mdh);
2013         if (rc) {
2014                 CDEBUG(D_NET, "Failed to post push target: %d\n", rc);
2015                 lnet_ping_buffer_decref(pbuf);
2016                 return rc;
2017         }
2018
2019         lnet_net_lock(LNET_LOCK_EX);
2020         old_pbuf = the_lnet.ln_push_target;
2021         old_mdh = the_lnet.ln_push_target_md;
2022         the_lnet.ln_push_target = pbuf;
2023         the_lnet.ln_push_target_md = mdh;
2024         lnet_net_unlock(LNET_LOCK_EX);
2025
2026         if (old_pbuf) {
2027                 LNetMDUnlink(old_mdh);
2028                 /* Drop ref set by lnet_ping_buffer_alloc() */
2029                 lnet_ping_buffer_decref(old_pbuf);
2030         }
2031
2032         /* Received another push or reply that requires a larger buffer */
2033         if (nnis < the_lnet.ln_push_target_nnis)
2034                 goto again;
2035
2036         CDEBUG(D_NET, "nnis %d success\n", nnis);
2037         return 0;
2038 }
2039
2040 int lnet_push_target_post(struct lnet_ping_buffer *pbuf,
2041                           struct lnet_handle_md *mdhp)
2042 {
2043         struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
2044         struct lnet_md md = { NULL };
2045         struct lnet_me *me;
2046         int rc;
2047
2048         me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
2049                           LNET_PROTO_PING_MATCHBITS, 0,
2050                           LNET_UNLINK, LNET_INS_AFTER);
2051         if (IS_ERR(me)) {
2052                 rc = PTR_ERR(me);
2053                 CERROR("Can't create push target ME: %d\n", rc);
2054                 return rc;
2055         }
2056
2057         pbuf->pb_needs_post = false;
2058
2059         /* This reference is dropped by lnet_push_target_event_handler() */
2060         lnet_ping_buffer_addref(pbuf);
2061
2062         /* initialize md content */
2063         md.start     = &pbuf->pb_info;
2064         md.length    = LNET_PING_INFO_SIZE(pbuf->pb_nnis);
2065         md.threshold = 1;
2066         md.max_size  = 0;
2067         md.options   = LNET_MD_OP_PUT | LNET_MD_TRUNCATE;
2068         md.user_ptr  = pbuf;
2069         md.handler   = the_lnet.ln_push_target_handler;
2070
2071         rc = LNetMDAttach(me, &md, LNET_UNLINK, mdhp);
2072         if (rc) {
2073                 CERROR("Can't attach push MD: %d\n", rc);
2074                 lnet_ping_buffer_decref(pbuf);
2075                 pbuf->pb_needs_post = true;
2076                 return rc;
2077         }
2078
2079         CDEBUG(D_NET, "posted push target %p\n", pbuf);
2080
2081         return 0;
2082 }
2083
2084 static void lnet_push_target_event_handler(struct lnet_event *ev)
2085 {
2086         struct lnet_ping_buffer *pbuf = ev->md_user_ptr;
2087
2088         CDEBUG(D_NET, "type %d status %d unlinked %d\n", ev->type, ev->status,
2089                ev->unlinked);
2090
2091         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
2092                 lnet_swap_pinginfo(pbuf);
2093
2094         if (ev->type == LNET_EVENT_UNLINK) {
2095                 /* Drop ref added by lnet_push_target_post() */
2096                 lnet_ping_buffer_decref(pbuf);
2097                 return;
2098         }
2099
2100         lnet_peer_push_event(ev);
2101         if (ev->unlinked)
2102                 /* Drop ref added by lnet_push_target_post */
2103                 lnet_ping_buffer_decref(pbuf);
2104 }
2105
2106 /* Initialize the push target. */
2107 static int lnet_push_target_init(void)
2108 {
2109         int rc;
2110
2111         if (the_lnet.ln_push_target)
2112                 return -EALREADY;
2113
2114         the_lnet.ln_push_target_handler =
2115                 lnet_push_target_event_handler;
2116
2117         rc = LNetSetLazyPortal(LNET_RESERVED_PORTAL);
2118         LASSERT(rc == 0);
2119
2120         /* Start at the required minimum, we'll enlarge if required. */
2121         the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
2122
2123         rc = lnet_push_target_resize();
2124
2125         if (rc) {
2126                 LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2127                 the_lnet.ln_push_target_handler = NULL;
2128         }
2129
2130         return rc;
2131 }
2132
2133 /* Clean up the push target. */
2134 static void lnet_push_target_fini(void)
2135 {
2136         if (!the_lnet.ln_push_target)
2137                 return;
2138
2139         /* Unlink and invalidate to prevent new references. */
2140         LNetMDUnlink(the_lnet.ln_push_target_md);
2141         LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
2142
2143         /* Wait for the unlink to complete. */
2144         wait_var_event_warning(&the_lnet.ln_push_target->pb_refcnt,
2145                                atomic_read(&the_lnet.ln_push_target->pb_refcnt) <= 1,
2146                                "Still waiting for ping data MD to unlink\n");
2147
2148         /* Drop ref set by lnet_ping_buffer_alloc() */
2149         lnet_ping_buffer_decref(the_lnet.ln_push_target);
2150         the_lnet.ln_push_target = NULL;
2151         the_lnet.ln_push_target_nnis = 0;
2152
2153         LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2154         lnet_assert_handler_unused(the_lnet.ln_push_target_handler);
2155         the_lnet.ln_push_target_handler = NULL;
2156 }
2157
2158 static int
2159 lnet_ni_tq_credits(struct lnet_ni *ni)
2160 {
2161         int     credits;
2162
2163         LASSERT(ni->ni_ncpts >= 1);
2164
2165         if (ni->ni_ncpts == 1)
2166                 return ni->ni_net->net_tunables.lct_max_tx_credits;
2167
2168         credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2169         credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2170         credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2171
2172         return credits;
2173 }
2174
2175 static void
2176 lnet_ni_unlink_locked(struct lnet_ni *ni)
2177 {
2178         /* move it to zombie list and nobody can find it anymore */
2179         LASSERT(!list_empty(&ni->ni_netlist));
2180         list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2181         lnet_ni_decref_locked(ni, 0);
2182 }
2183
2184 static void
2185 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2186 {
2187         int             i;
2188         int             islo;
2189         struct lnet_ni  *ni;
2190         struct list_head *zombie_list = &net->net_ni_zombie;
2191
2192         /*
2193          * Now wait for the NIs I just nuked to show up on the zombie
2194          * list and shut them down in guaranteed thread context
2195          */
2196         i = 2;
2197         while (!list_empty(zombie_list)) {
2198                 int     *ref;
2199                 int     j;
2200
2201                 ni = list_entry(zombie_list->next,
2202                                 struct lnet_ni, ni_netlist);
2203                 list_del_init(&ni->ni_netlist);
2204                 /* the ni should be in deleting state. If it's not it's
2205                  * a bug */
2206                 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2207                 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2208                         if (*ref == 0)
2209                                 continue;
2210                         /* still busy, add it back to zombie list */
2211                         list_add(&ni->ni_netlist, zombie_list);
2212                         break;
2213                 }
2214
2215                 if (!list_empty(&ni->ni_netlist)) {
2216                         /* Unlock mutex while waiting to allow other
2217                          * threads to read the LNet state and fall through
2218                          * to avoid deadlock
2219                          */
2220                         lnet_net_unlock(LNET_LOCK_EX);
2221                         mutex_unlock(&the_lnet.ln_api_mutex);
2222
2223                         ++i;
2224                         if ((i & (-i)) == i) {
2225                                 CDEBUG(D_WARNING,
2226                                        "Waiting for zombie LNI %s\n",
2227                                        libcfs_nid2str(ni->ni_nid));
2228                         }
2229                         schedule_timeout_uninterruptible(cfs_time_seconds(1));
2230
2231                         mutex_lock(&the_lnet.ln_api_mutex);
2232                         lnet_net_lock(LNET_LOCK_EX);
2233                         continue;
2234                 }
2235
2236                 lnet_net_unlock(LNET_LOCK_EX);
2237
2238                 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2239
2240                 LASSERT(!in_interrupt());
2241                 /* Holding the mutex makes it safe for lnd_shutdown
2242                  * to call module_put(). Module unload cannot finish
2243                  * until lnet_unregister_lnd() completes, and that
2244                  * requires the mutex.
2245                  */
2246                 mutex_lock(&the_lnet.ln_lnd_mutex);
2247                 (net->net_lnd->lnd_shutdown)(ni);
2248                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2249
2250                 if (!islo)
2251                         CDEBUG(D_LNI, "Removed LNI %s\n",
2252                               libcfs_nid2str(ni->ni_nid));
2253
2254                 lnet_ni_free(ni);
2255                 i = 2;
2256                 lnet_net_lock(LNET_LOCK_EX);
2257         }
2258 }
2259
2260 /* shutdown down the NI and release refcount */
2261 static void
2262 lnet_shutdown_lndni(struct lnet_ni *ni)
2263 {
2264         int i;
2265         struct lnet_net *net = ni->ni_net;
2266
2267         lnet_net_lock(LNET_LOCK_EX);
2268         lnet_ni_lock(ni);
2269         ni->ni_state = LNET_NI_STATE_DELETING;
2270         lnet_ni_unlock(ni);
2271         lnet_ni_unlink_locked(ni);
2272         lnet_incr_dlc_seq();
2273         lnet_net_unlock(LNET_LOCK_EX);
2274
2275         /* clear messages for this NI on the lazy portal */
2276         for (i = 0; i < the_lnet.ln_nportals; i++)
2277                 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2278
2279         lnet_net_lock(LNET_LOCK_EX);
2280         lnet_clear_zombies_nis_locked(net);
2281         lnet_net_unlock(LNET_LOCK_EX);
2282 }
2283
2284 static void
2285 lnet_shutdown_lndnet(struct lnet_net *net)
2286 {
2287         struct lnet_ni *ni;
2288
2289         lnet_net_lock(LNET_LOCK_EX);
2290
2291         list_del_init(&net->net_list);
2292
2293         while (!list_empty(&net->net_ni_list)) {
2294                 ni = list_entry(net->net_ni_list.next,
2295                                 struct lnet_ni, ni_netlist);
2296                 lnet_net_unlock(LNET_LOCK_EX);
2297                 lnet_shutdown_lndni(ni);
2298                 lnet_net_lock(LNET_LOCK_EX);
2299         }
2300
2301         lnet_net_unlock(LNET_LOCK_EX);
2302
2303         /* Do peer table cleanup for this net */
2304         lnet_peer_tables_cleanup(net);
2305
2306         lnet_net_free(net);
2307 }
2308
2309 static void
2310 lnet_shutdown_lndnets(void)
2311 {
2312         struct lnet_net *net;
2313         LIST_HEAD(resend);
2314         struct lnet_msg *msg, *tmp;
2315
2316         /* NB called holding the global mutex */
2317
2318         /* All quiet on the API front */
2319         LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2320         LASSERT(the_lnet.ln_refcount == 0);
2321
2322         lnet_net_lock(LNET_LOCK_EX);
2323         the_lnet.ln_state = LNET_STATE_STOPPING;
2324
2325         /*
2326          * move the nets to the zombie list to avoid them being
2327          * picked up for new work. LONET is also included in the
2328          * Nets that will be moved to the zombie list
2329          */
2330         list_splice_init(&the_lnet.ln_nets, &the_lnet.ln_net_zombie);
2331
2332         /* Drop the cached loopback Net. */
2333         if (the_lnet.ln_loni != NULL) {
2334                 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2335                 the_lnet.ln_loni = NULL;
2336         }
2337         lnet_net_unlock(LNET_LOCK_EX);
2338
2339         /* iterate through the net zombie list and delete each net */
2340         while (!list_empty(&the_lnet.ln_net_zombie)) {
2341                 net = list_entry(the_lnet.ln_net_zombie.next,
2342                                  struct lnet_net, net_list);
2343                 lnet_shutdown_lndnet(net);
2344         }
2345
2346         spin_lock(&the_lnet.ln_msg_resend_lock);
2347         list_splice(&the_lnet.ln_msg_resend, &resend);
2348         spin_unlock(&the_lnet.ln_msg_resend_lock);
2349
2350         list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2351                 list_del_init(&msg->msg_list);
2352                 msg->msg_no_resend = true;
2353                 lnet_finalize(msg, -ECANCELED);
2354         }
2355
2356         lnet_net_lock(LNET_LOCK_EX);
2357         the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2358         lnet_net_unlock(LNET_LOCK_EX);
2359 }
2360
2361 static int
2362 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2363 {
2364         int                     rc = -EINVAL;
2365         struct lnet_tx_queue    *tq;
2366         int                     i;
2367         struct lnet_net         *net = ni->ni_net;
2368
2369         mutex_lock(&the_lnet.ln_lnd_mutex);
2370
2371         if (tun) {
2372                 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2373                 ni->ni_lnd_tunables_set = true;
2374         }
2375
2376         rc = (net->net_lnd->lnd_startup)(ni);
2377
2378         mutex_unlock(&the_lnet.ln_lnd_mutex);
2379
2380         if (rc != 0) {
2381                 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2382                                    rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2383                 goto failed0;
2384         }
2385
2386         lnet_ni_lock(ni);
2387         ni->ni_state = LNET_NI_STATE_ACTIVE;
2388         lnet_ni_unlock(ni);
2389
2390         /* We keep a reference on the loopback net through the loopback NI */
2391         if (net->net_lnd->lnd_type == LOLND) {
2392                 lnet_ni_addref(ni);
2393                 LASSERT(the_lnet.ln_loni == NULL);
2394                 the_lnet.ln_loni = ni;
2395                 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2396                 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2397                 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2398                 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2399                 return 0;
2400         }
2401
2402         if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2403             ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2404                 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2405                                    libcfs_lnd2str(net->net_lnd->lnd_type),
2406                                    ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2407                                         "" : "per-peer ");
2408                 /* shutdown the NI since if we get here then it must've already
2409                  * been started
2410                  */
2411                 lnet_shutdown_lndni(ni);
2412                 return -EINVAL;
2413         }
2414
2415         cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2416                 tq->tq_credits_min =
2417                 tq->tq_credits_max =
2418                 tq->tq_credits = lnet_ni_tq_credits(ni);
2419         }
2420
2421         atomic_set(&ni->ni_tx_credits,
2422                    lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2423         atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2424
2425         CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2426                 libcfs_nid2str(ni->ni_nid),
2427                 ni->ni_net->net_tunables.lct_peer_tx_credits,
2428                 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2429                 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2430                 ni->ni_net->net_tunables.lct_peer_timeout);
2431
2432         return 0;
2433 failed0:
2434         lnet_ni_free(ni);
2435         return rc;
2436 }
2437
2438 static int
2439 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2440 {
2441         struct lnet_ni *ni;
2442         struct lnet_net *net_l = NULL;
2443         LIST_HEAD(local_ni_list);
2444         int rc;
2445         int ni_count = 0;
2446         __u32 lnd_type;
2447         const struct lnet_lnd  *lnd;
2448         int peer_timeout =
2449                 net->net_tunables.lct_peer_timeout;
2450         int maxtxcredits =
2451                 net->net_tunables.lct_max_tx_credits;
2452         int peerrtrcredits =
2453                 net->net_tunables.lct_peer_rtr_credits;
2454
2455         /*
2456          * make sure that this net is unique. If it isn't then
2457          * we are adding interfaces to an already existing network, and
2458          * 'net' is just a convenient way to pass in the list.
2459          * if it is unique we need to find the LND and load it if
2460          * necessary.
2461          */
2462         if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2463                 lnd_type = LNET_NETTYP(net->net_id);
2464
2465                 mutex_lock(&the_lnet.ln_lnd_mutex);
2466                 lnd = lnet_find_lnd_by_type(lnd_type);
2467
2468                 if (lnd == NULL) {
2469                         mutex_unlock(&the_lnet.ln_lnd_mutex);
2470                         rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2471                         mutex_lock(&the_lnet.ln_lnd_mutex);
2472
2473                         lnd = lnet_find_lnd_by_type(lnd_type);
2474                         if (lnd == NULL) {
2475                                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2476                                 CERROR("Can't load LND %s, module %s, rc=%d\n",
2477                                 libcfs_lnd2str(lnd_type),
2478                                 libcfs_lnd2modname(lnd_type), rc);
2479 #ifndef HAVE_MODULE_LOADING_SUPPORT
2480                                 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2481                                                 "compiled with kernel module "
2482                                                 "loading support.");
2483 #endif
2484                                 rc = -EINVAL;
2485                                 goto failed0;
2486                         }
2487                 }
2488
2489                 net->net_lnd = lnd;
2490
2491                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2492
2493                 net_l = net;
2494         }
2495
2496         /*
2497          * net_l: if the network being added is unique then net_l
2498          *        will point to that network
2499          *        if the network being added is not unique then
2500          *        net_l points to the existing network.
2501          *
2502          * When we enter the loop below, we'll pick NIs off he
2503          * network beign added and start them up, then add them to
2504          * a local ni list. Once we've successfully started all
2505          * the NIs then we join the local NI list (of started up
2506          * networks) with the net_l->net_ni_list, which should
2507          * point to the correct network to add the new ni list to
2508          *
2509          * If any of the new NIs fail to start up, then we want to
2510          * iterate through the local ni list, which should include
2511          * any NIs which were successfully started up, and shut
2512          * them down.
2513          *
2514          * After than we want to delete the network being added,
2515          * to avoid a memory leak.
2516          */
2517         while (!list_empty(&net->net_ni_added)) {
2518                 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2519                                 ni_netlist);
2520                 list_del_init(&ni->ni_netlist);
2521
2522                 /* make sure that the the NI we're about to start
2523                  * up is actually unique. if it's not fail. */
2524                 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2525                                         ni->ni_interface)) {
2526                         rc = -EEXIST;
2527                         goto failed1;
2528                 }
2529
2530                 /* adjust the pointer the parent network, just in case it
2531                  * the net is a duplicate */
2532                 ni->ni_net = net_l;
2533
2534                 rc = lnet_startup_lndni(ni, tun);
2535
2536                 if (rc < 0)
2537                         goto failed1;
2538
2539                 lnet_ni_addref(ni);
2540                 list_add_tail(&ni->ni_netlist, &local_ni_list);
2541
2542                 ni_count++;
2543         }
2544
2545         lnet_net_lock(LNET_LOCK_EX);
2546         list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2547         lnet_incr_dlc_seq();
2548         lnet_net_unlock(LNET_LOCK_EX);
2549
2550         /* if the network is not unique then we don't want to keep
2551          * it around after we're done. Free it. Otherwise add that
2552          * net to the global the_lnet.ln_nets */
2553         if (net_l != net && net_l != NULL) {
2554                 /*
2555                  * TODO - note. currently the tunables can not be updated
2556                  * once added
2557                  */
2558                 lnet_net_free(net);
2559         } else {
2560                 /*
2561                  * restore tunables after it has been overwitten by the
2562                  * lnd
2563                  */
2564                 if (peer_timeout != -1)
2565                         net->net_tunables.lct_peer_timeout = peer_timeout;
2566                 if (maxtxcredits != -1)
2567                         net->net_tunables.lct_max_tx_credits = maxtxcredits;
2568                 if (peerrtrcredits != -1)
2569                         net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2570
2571                 lnet_net_lock(LNET_LOCK_EX);
2572                 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2573                 lnet_net_unlock(LNET_LOCK_EX);
2574         }
2575
2576         return ni_count;
2577
2578 failed1:
2579         /*
2580          * shutdown the new NIs that are being started up
2581          * free the NET being started
2582          */
2583         while (!list_empty(&local_ni_list)) {
2584                 ni = list_entry(local_ni_list.next, struct lnet_ni,
2585                                 ni_netlist);
2586
2587                 lnet_shutdown_lndni(ni);
2588         }
2589
2590 failed0:
2591         lnet_net_free(net);
2592
2593         return rc;
2594 }
2595
2596 static int
2597 lnet_startup_lndnets(struct list_head *netlist)
2598 {
2599         struct lnet_net         *net;
2600         int                     rc;
2601         int                     ni_count = 0;
2602
2603         /*
2604          * Change to running state before bringing up the LNDs. This
2605          * allows lnet_shutdown_lndnets() to assert that we've passed
2606          * through here.
2607          */
2608         lnet_net_lock(LNET_LOCK_EX);
2609         the_lnet.ln_state = LNET_STATE_RUNNING;
2610         lnet_net_unlock(LNET_LOCK_EX);
2611
2612         while (!list_empty(netlist)) {
2613                 net = list_entry(netlist->next, struct lnet_net, net_list);
2614                 list_del_init(&net->net_list);
2615
2616                 rc = lnet_startup_lndnet(net, NULL);
2617
2618                 if (rc < 0)
2619                         goto failed;
2620
2621                 ni_count += rc;
2622         }
2623
2624         return ni_count;
2625 failed:
2626         lnet_shutdown_lndnets();
2627
2628         return rc;
2629 }
2630
2631 static int lnet_genl_parse_list(struct sk_buff *msg,
2632                                 const struct ln_key_list *data[], u16 idx)
2633 {
2634         const struct ln_key_list *list = data[idx];
2635         const struct ln_key_props *props;
2636         struct nlattr *node;
2637         u16 count;
2638
2639         if (!list)
2640                 return 0;
2641
2642         if (!list->lkl_maxattr)
2643                 return -ERANGE;
2644
2645         props = list->lkl_list;
2646         if (!props)
2647                 return -EINVAL;
2648
2649         node = nla_nest_start(msg, LN_SCALAR_ATTR_LIST);
2650         if (!node)
2651                 return -ENOBUFS;
2652
2653         for (count = 1; count <= list->lkl_maxattr; count++) {
2654                 struct nlattr *key = nla_nest_start(msg, count);
2655
2656                 if (count == 1)
2657                         nla_put_u16(msg, LN_SCALAR_ATTR_LIST_SIZE,
2658                                     list->lkl_maxattr);
2659
2660                 nla_put_u16(msg, LN_SCALAR_ATTR_INDEX, count);
2661                 if (props[count].lkp_values)
2662                         nla_put_string(msg, LN_SCALAR_ATTR_VALUE,
2663                                        props[count].lkp_values);
2664                 if (props[count].lkp_key_format)
2665                         nla_put_u16(msg, LN_SCALAR_ATTR_KEY_FORMAT,
2666                                     props[count].lkp_key_format);
2667                 nla_put_u16(msg, LN_SCALAR_ATTR_NLA_TYPE,
2668                             props[count].lkp_data_type);
2669                 if (props[count].lkp_data_type == NLA_NESTED) {
2670                         int rc;
2671
2672                         rc = lnet_genl_parse_list(msg, data, ++idx);
2673                         if (rc < 0)
2674                                 return rc;
2675                 }
2676
2677                 nla_nest_end(msg, key);
2678         }
2679
2680         nla_nest_end(msg, node);
2681         return 0;
2682 }
2683
2684 int lnet_genl_send_scalar_list(struct sk_buff *msg, u32 portid, u32 seq,
2685                                const struct genl_family *family, int flags,
2686                                u8 cmd, const struct ln_key_list *data[])
2687 {
2688         int rc = 0;
2689         void *hdr;
2690
2691         if (!data[0])
2692                 return -EINVAL;
2693
2694         hdr = genlmsg_put(msg, portid, seq, family, flags, cmd);
2695         if (!hdr)
2696                 GOTO(canceled, rc = -EMSGSIZE);
2697
2698         rc = lnet_genl_parse_list(msg, data, 0);
2699         if (rc < 0)
2700                 GOTO(canceled, rc);
2701
2702         genlmsg_end(msg, hdr);
2703 canceled:
2704         if (rc < 0)
2705                 genlmsg_cancel(msg, hdr);
2706         return rc;
2707 }
2708 EXPORT_SYMBOL(lnet_genl_send_scalar_list);
2709
2710 /**
2711  * Initialize LNet library.
2712  *
2713  * Automatically called at module loading time. Caller has to call
2714  * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2715  * latter returned 0. It must be called exactly once.
2716  *
2717  * \retval 0 on success
2718  * \retval -ve on failures.
2719  */
2720 int lnet_lib_init(void)
2721 {
2722         int rc;
2723
2724         lnet_assert_wire_constants();
2725
2726         /* refer to global cfs_cpt_table for now */
2727         the_lnet.ln_cpt_table = cfs_cpt_tab;
2728         the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_tab);
2729
2730         LASSERT(the_lnet.ln_cpt_number > 0);
2731         if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2732                 /* we are under risk of consuming all lh_cookie */
2733                 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2734                        "please change setting of CPT-table and retry\n",
2735                        the_lnet.ln_cpt_number, LNET_CPT_MAX);
2736                 return -E2BIG;
2737         }
2738
2739         while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2740                 the_lnet.ln_cpt_bits++;
2741
2742         rc = lnet_create_locks();
2743         if (rc != 0) {
2744                 CERROR("Can't create LNet global locks: %d\n", rc);
2745                 return rc;
2746         }
2747
2748         the_lnet.ln_refcount = 0;
2749         INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2750         INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2751
2752         /* The hash table size is the number of bits it takes to express the set
2753          * ln_num_routes, minus 1 (better to under estimate than over so we
2754          * don't waste memory). */
2755         if (rnet_htable_size <= 0)
2756                 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2757         else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2758                 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2759         the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2760                                            order_base_2(rnet_htable_size) - 1);
2761
2762         /* All LNDs apart from the LOLND are in separate modules.  They
2763          * register themselves when their module loads, and unregister
2764          * themselves when their module is unloaded. */
2765         lnet_register_lnd(&the_lolnd);
2766         return 0;
2767 }
2768
2769 /**
2770  * Finalize LNet library.
2771  *
2772  * \pre lnet_lib_init() called with success.
2773  * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2774  *
2775  * As this happens at module-unload, all lnds must already be unloaded,
2776  * so they must already be unregistered.
2777  */
2778 void lnet_lib_exit(void)
2779 {
2780         int i;
2781
2782         LASSERT(the_lnet.ln_refcount == 0);
2783         lnet_unregister_lnd(&the_lolnd);
2784         for (i = 0; i < NUM_LNDS; i++)
2785                 LASSERT(!the_lnet.ln_lnds[i]);
2786         lnet_destroy_locks();
2787 }
2788
2789 /**
2790  * Set LNet PID and start LNet interfaces, routing, and forwarding.
2791  *
2792  * Users must call this function at least once before any other functions.
2793  * For each successful call there must be a corresponding call to
2794  * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2795  * ignored.
2796  *
2797  * The PID used by LNet may be different from the one requested.
2798  * See LNetGetId().
2799  *
2800  * \param requested_pid PID requested by the caller.
2801  *
2802  * \return >= 0 on success, and < 0 error code on failures.
2803  */
2804 int
2805 LNetNIInit(lnet_pid_t requested_pid)
2806 {
2807         int                     im_a_router = 0;
2808         int                     rc;
2809         int                     ni_count;
2810         struct lnet_ping_buffer *pbuf;
2811         struct lnet_handle_md   ping_mdh;
2812         LIST_HEAD(net_head);
2813         struct lnet_net         *net;
2814
2815         mutex_lock(&the_lnet.ln_api_mutex);
2816
2817         CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2818
2819         if (the_lnet.ln_refcount > 0) {
2820                 rc = the_lnet.ln_refcount++;
2821                 mutex_unlock(&the_lnet.ln_api_mutex);
2822                 return rc;
2823         }
2824
2825         rc = lnet_prepare(requested_pid);
2826         if (rc != 0) {
2827                 mutex_unlock(&the_lnet.ln_api_mutex);
2828                 return rc;
2829         }
2830
2831         /* create a network for Loopback network */
2832         net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2833         if (net == NULL) {
2834                 rc = -ENOMEM;
2835                 goto err_empty_list;
2836         }
2837
2838         /* Add in the loopback NI */
2839         if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2840                 rc = -ENOMEM;
2841                 goto err_empty_list;
2842         }
2843
2844         if (use_tcp_bonding)
2845                 CWARN("use_tcp_bonding has been removed. Use Multi-Rail and Dynamic Discovery instead, see LU-13641\n");
2846
2847         /* If LNet is being initialized via DLC it is possible
2848          * that the user requests not to load module parameters (ones which
2849          * are supported by DLC) on initialization.  Therefore, make sure not
2850          * to load networks, routes and forwarding from module parameters
2851          * in this case.  On cleanup in case of failure only clean up
2852          * routes if it has been loaded */
2853         if (!the_lnet.ln_nis_from_mod_params) {
2854                 rc = lnet_parse_networks(&net_head, lnet_get_networks());
2855                 if (rc < 0)
2856                         goto err_empty_list;
2857         }
2858
2859         ni_count = lnet_startup_lndnets(&net_head);
2860         if (ni_count < 0) {
2861                 rc = ni_count;
2862                 goto err_empty_list;
2863         }
2864
2865         if (!the_lnet.ln_nis_from_mod_params) {
2866                 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2867                 if (rc != 0)
2868                         goto err_shutdown_lndnis;
2869
2870                 rc = lnet_rtrpools_alloc(im_a_router);
2871                 if (rc != 0)
2872                         goto err_destroy_routes;
2873         }
2874
2875         rc = lnet_acceptor_start();
2876         if (rc != 0)
2877                 goto err_destroy_routes;
2878
2879         the_lnet.ln_refcount = 1;
2880         /* Now I may use my own API functions... */
2881
2882         rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2883         if (rc != 0)
2884                 goto err_acceptor_stop;
2885
2886         lnet_ping_target_update(pbuf, ping_mdh);
2887
2888         the_lnet.ln_mt_handler = lnet_mt_event_handler;
2889
2890         rc = lnet_push_target_init();
2891         if (rc != 0)
2892                 goto err_stop_ping;
2893
2894         rc = lnet_peer_discovery_start();
2895         if (rc != 0)
2896                 goto err_destroy_push_target;
2897
2898         rc = lnet_monitor_thr_start();
2899         if (rc != 0)
2900                 goto err_stop_discovery_thr;
2901
2902         lnet_fault_init();
2903         lnet_router_debugfs_init();
2904
2905         mutex_unlock(&the_lnet.ln_api_mutex);
2906
2907         complete_all(&the_lnet.ln_started);
2908
2909         /* wait for all routers to start */
2910         lnet_wait_router_start();
2911
2912         return 0;
2913
2914 err_stop_discovery_thr:
2915         lnet_peer_discovery_stop();
2916 err_destroy_push_target:
2917         lnet_push_target_fini();
2918 err_stop_ping:
2919         lnet_ping_target_fini();
2920 err_acceptor_stop:
2921         the_lnet.ln_refcount = 0;
2922         lnet_acceptor_stop();
2923 err_destroy_routes:
2924         if (!the_lnet.ln_nis_from_mod_params)
2925                 lnet_destroy_routes();
2926 err_shutdown_lndnis:
2927         lnet_shutdown_lndnets();
2928 err_empty_list:
2929         lnet_unprepare();
2930         LASSERT(rc < 0);
2931         mutex_unlock(&the_lnet.ln_api_mutex);
2932         while (!list_empty(&net_head)) {
2933                 struct lnet_net *net;
2934
2935                 net = list_entry(net_head.next, struct lnet_net, net_list);
2936                 list_del_init(&net->net_list);
2937                 lnet_net_free(net);
2938         }
2939         return rc;
2940 }
2941 EXPORT_SYMBOL(LNetNIInit);
2942
2943 /**
2944  * Stop LNet interfaces, routing, and forwarding.
2945  *
2946  * Users must call this function once for each successful call to LNetNIInit().
2947  * Once the LNetNIFini() operation has been started, the results of pending
2948  * API operations are undefined.
2949  *
2950  * \return always 0 for current implementation.
2951  */
2952 int
2953 LNetNIFini(void)
2954 {
2955         mutex_lock(&the_lnet.ln_api_mutex);
2956
2957         LASSERT(the_lnet.ln_refcount > 0);
2958
2959         if (the_lnet.ln_refcount != 1) {
2960                 the_lnet.ln_refcount--;
2961         } else {
2962                 LASSERT(!the_lnet.ln_niinit_self);
2963
2964                 lnet_fault_fini();
2965
2966                 lnet_router_debugfs_fini();
2967                 lnet_monitor_thr_stop();
2968                 lnet_peer_discovery_stop();
2969                 lnet_push_target_fini();
2970                 lnet_ping_target_fini();
2971
2972                 /* Teardown fns that use my own API functions BEFORE here */
2973                 the_lnet.ln_refcount = 0;
2974
2975                 lnet_acceptor_stop();
2976                 lnet_destroy_routes();
2977                 lnet_shutdown_lndnets();
2978                 lnet_unprepare();
2979         }
2980
2981         mutex_unlock(&the_lnet.ln_api_mutex);
2982         return 0;
2983 }
2984 EXPORT_SYMBOL(LNetNIFini);
2985
2986 /**
2987  * Grabs the ni data from the ni structure and fills the out
2988  * parameters
2989  *
2990  * \param[in] ni network        interface structure
2991  * \param[out] cfg_ni           NI config information
2992  * \param[out] tun              network and LND tunables
2993  */
2994 static void
2995 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
2996                    struct lnet_ioctl_config_lnd_tunables *tun,
2997                    struct lnet_ioctl_element_stats *stats,
2998                    __u32 tun_size)
2999 {
3000         size_t min_size = 0;
3001         int i;
3002
3003         if (!ni || !cfg_ni || !tun)
3004                 return;
3005
3006         if (ni->ni_interface != NULL) {
3007                 strncpy(cfg_ni->lic_ni_intf,
3008                         ni->ni_interface,
3009                         sizeof(cfg_ni->lic_ni_intf));
3010         }
3011
3012         cfg_ni->lic_nid = ni->ni_nid;
3013         if (ni->ni_nid == LNET_NID_LO_0)
3014                 cfg_ni->lic_status = LNET_NI_STATUS_UP;
3015         else
3016                 cfg_ni->lic_status = ni->ni_status->ns_status;
3017         cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
3018
3019         memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
3020
3021         if (stats) {
3022                 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
3023                                                        LNET_STATS_TYPE_SEND);
3024                 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
3025                                                        LNET_STATS_TYPE_RECV);
3026                 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
3027                                                        LNET_STATS_TYPE_DROP);
3028         }
3029
3030         /*
3031          * tun->lt_tun will always be present, but in order to be
3032          * backwards compatible, we need to deal with the cases when
3033          * tun->lt_tun is smaller than what the kernel has, because it
3034          * comes from an older version of a userspace program, then we'll
3035          * need to copy as much information as we have available space.
3036          */
3037         min_size = tun_size - sizeof(tun->lt_cmn);
3038         memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
3039
3040         /* copy over the cpts */
3041         if (ni->ni_ncpts == LNET_CPT_NUMBER &&
3042             ni->ni_cpts == NULL)  {
3043                 for (i = 0; i < ni->ni_ncpts; i++)
3044                         cfg_ni->lic_cpts[i] = i;
3045         } else {
3046                 for (i = 0;
3047                      ni->ni_cpts != NULL && i < ni->ni_ncpts &&
3048                      i < LNET_MAX_SHOW_NUM_CPT;
3049                      i++)
3050                         cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
3051         }
3052         cfg_ni->lic_ncpts = ni->ni_ncpts;
3053 }
3054
3055 /**
3056  * NOTE: This is a legacy function left in the code to be backwards
3057  * compatible with older userspace programs. It should eventually be
3058  * removed.
3059  *
3060  * Grabs the ni data from the ni structure and fills the out
3061  * parameters
3062  *
3063  * \param[in] ni network        interface structure
3064  * \param[out] config           config information
3065  */
3066 static void
3067 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
3068                          struct lnet_ioctl_config_data *config)
3069 {
3070         struct lnet_ioctl_net_config *net_config;
3071         struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
3072         size_t min_size, tunable_size = 0;
3073         int i;
3074
3075         if (!ni || !config)
3076                 return;
3077
3078         net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
3079         if (!net_config)
3080                 return;
3081
3082         if (!ni->ni_interface)
3083                 return;
3084
3085         strncpy(net_config->ni_interface,
3086                 ni->ni_interface,
3087                 sizeof(net_config->ni_interface));
3088
3089         config->cfg_nid = ni->ni_nid;
3090         config->cfg_config_u.cfg_net.net_peer_timeout =
3091                 ni->ni_net->net_tunables.lct_peer_timeout;
3092         config->cfg_config_u.cfg_net.net_max_tx_credits =
3093                 ni->ni_net->net_tunables.lct_max_tx_credits;
3094         config->cfg_config_u.cfg_net.net_peer_tx_credits =
3095                 ni->ni_net->net_tunables.lct_peer_tx_credits;
3096         config->cfg_config_u.cfg_net.net_peer_rtr_credits =
3097                 ni->ni_net->net_tunables.lct_peer_rtr_credits;
3098
3099         if (ni->ni_nid == LNET_NID_LO_0)
3100                 net_config->ni_status = LNET_NI_STATUS_UP;
3101         else
3102                 net_config->ni_status = ni->ni_status->ns_status;
3103
3104         if (ni->ni_cpts) {
3105                 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
3106
3107                 for (i = 0; i < num_cpts; i++)
3108                         net_config->ni_cpts[i] = ni->ni_cpts[i];
3109
3110                 config->cfg_ncpts = num_cpts;
3111         }
3112
3113         /*
3114          * See if user land tools sent in a newer and larger version
3115          * of struct lnet_tunables than what the kernel uses.
3116          */
3117         min_size = sizeof(*config) + sizeof(*net_config);
3118
3119         if (config->cfg_hdr.ioc_len > min_size)
3120                 tunable_size = config->cfg_hdr.ioc_len - min_size;
3121
3122         /* Don't copy too much data to user space */
3123         min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
3124         lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
3125
3126         if (lnd_cfg && min_size) {
3127                 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
3128                 config->cfg_config_u.cfg_net.net_interface_count = 1;
3129
3130                 /* Tell user land that kernel side has less data */
3131                 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
3132                         min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
3133                         config->cfg_hdr.ioc_len -= min_size;
3134                 }
3135         }
3136 }
3137
3138 struct lnet_ni *
3139 lnet_get_ni_idx_locked(int idx)
3140 {
3141         struct lnet_ni          *ni;
3142         struct lnet_net         *net;
3143
3144         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3145                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3146                         if (idx-- == 0)
3147                                 return ni;
3148                 }
3149         }
3150
3151         return NULL;
3152 }
3153
3154 int lnet_get_net_healthv_locked(struct lnet_net *net)
3155 {
3156         struct lnet_ni *ni;
3157         int best_healthv = 0;
3158         int healthv, ni_fatal;
3159
3160         list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3161                 healthv = atomic_read(&ni->ni_healthv);
3162                 ni_fatal = atomic_read(&ni->ni_fatal_error_on);
3163                 if (!ni_fatal && healthv > best_healthv)
3164                         best_healthv = healthv;
3165         }
3166
3167         return best_healthv;
3168 }
3169
3170 struct lnet_ni *
3171 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
3172 {
3173         struct lnet_ni          *ni;
3174         struct lnet_net         *net = mynet;
3175
3176         /*
3177          * It is possible that the net has been cleaned out while there is
3178          * a message being sent. This function accessed the net without
3179          * checking if the list is empty
3180          */
3181         if (prev == NULL) {
3182                 if (net == NULL)
3183                         net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
3184                                         net_list);
3185                 if (list_empty(&net->net_ni_list))
3186                         return NULL;
3187                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3188                                 ni_netlist);
3189
3190                 return ni;
3191         }
3192
3193         if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
3194                 /* if you reached the end of the ni list and the net is
3195                  * specified, then there are no more nis in that net */
3196                 if (net != NULL)
3197                         return NULL;
3198
3199                 /* we reached the end of this net ni list. move to the
3200                  * next net */
3201                 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
3202                         /* no more nets and no more NIs. */
3203                         return NULL;
3204
3205                 /* get the next net */
3206                 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
3207                                  net_list);
3208                 if (list_empty(&net->net_ni_list))
3209                         return NULL;
3210                 /* get the ni on it */
3211                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3212                                 ni_netlist);
3213
3214                 return ni;
3215         }
3216
3217         if (list_empty(&prev->ni_netlist))
3218                 return NULL;
3219
3220         /* there are more nis left */
3221         ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
3222
3223         return ni;
3224 }
3225
3226 int
3227 lnet_get_net_config(struct lnet_ioctl_config_data *config)
3228 {
3229         struct lnet_ni *ni;
3230         int cpt;
3231         int rc = -ENOENT;
3232         int idx = config->cfg_count;
3233
3234         cpt = lnet_net_lock_current();
3235
3236         ni = lnet_get_ni_idx_locked(idx);
3237
3238         if (ni != NULL) {
3239                 rc = 0;
3240                 lnet_ni_lock(ni);
3241                 lnet_fill_ni_info_legacy(ni, config);
3242                 lnet_ni_unlock(ni);
3243         }
3244
3245         lnet_net_unlock(cpt);
3246         return rc;
3247 }
3248
3249 int
3250 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3251                    struct lnet_ioctl_config_lnd_tunables *tun,
3252                    struct lnet_ioctl_element_stats *stats,
3253                    __u32 tun_size)
3254 {
3255         struct lnet_ni          *ni;
3256         int                     cpt;
3257         int                     rc = -ENOENT;
3258
3259         if (!cfg_ni || !tun || !stats)
3260                 return -EINVAL;
3261
3262         cpt = lnet_net_lock_current();
3263
3264         ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3265
3266         if (ni) {
3267                 rc = 0;
3268                 lnet_ni_lock(ni);
3269                 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3270                 lnet_ni_unlock(ni);
3271         }
3272
3273         lnet_net_unlock(cpt);
3274         return rc;
3275 }
3276
3277 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3278 {
3279         struct lnet_ni *ni;
3280         int cpt;
3281         int rc = -ENOENT;
3282
3283         if (!msg_stats)
3284                 return -EINVAL;
3285
3286         cpt = lnet_net_lock_current();
3287
3288         ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3289
3290         if (ni) {
3291                 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3292                 rc = 0;
3293         }
3294
3295         lnet_net_unlock(cpt);
3296
3297         return rc;
3298 }
3299
3300 static int lnet_add_net_common(struct lnet_net *net,
3301                                struct lnet_ioctl_config_lnd_tunables *tun)
3302 {
3303         struct lnet_handle_md ping_mdh;
3304         struct lnet_ping_buffer *pbuf;
3305         struct lnet_remotenet *rnet;
3306         struct lnet_ni *ni;
3307         int net_ni_count;
3308         __u32 net_id;
3309         int rc;
3310
3311         lnet_net_lock(LNET_LOCK_EX);
3312         rnet = lnet_find_rnet_locked(net->net_id);
3313         lnet_net_unlock(LNET_LOCK_EX);
3314         /*
3315          * make sure that the net added doesn't invalidate the current
3316          * configuration LNet is keeping
3317          */
3318         if (rnet) {
3319                 CERROR("Adding net %s will invalidate routing configuration\n",
3320                        libcfs_net2str(net->net_id));
3321                 lnet_net_free(net);
3322                 return -EUSERS;
3323         }
3324
3325         /*
3326          * make sure you calculate the correct number of slots in the ping
3327          * buffer. Since the ping info is a flattened list of all the NIs,
3328          * we should allocate enough slots to accomodate the number of NIs
3329          * which will be added.
3330          *
3331          * since ni hasn't been configured yet, use
3332          * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3333          */
3334         net_ni_count = lnet_get_net_ni_count_pre(net);
3335
3336         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3337                                     net_ni_count + lnet_get_ni_count(),
3338                                     false);
3339         if (rc < 0) {
3340                 lnet_net_free(net);
3341                 return rc;
3342         }
3343
3344         if (tun)
3345                 memcpy(&net->net_tunables,
3346                        &tun->lt_cmn, sizeof(net->net_tunables));
3347         else
3348                 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3349
3350         net_id = net->net_id;
3351
3352         rc = lnet_startup_lndnet(net,
3353                                  (tun) ? &tun->lt_tun : NULL);
3354         if (rc < 0)
3355                 goto failed;
3356
3357         lnet_net_lock(LNET_LOCK_EX);
3358         net = lnet_get_net_locked(net_id);
3359         LASSERT(net);
3360
3361         /* apply the UDSPs */
3362         rc = lnet_udsp_apply_policies_on_net(net);
3363         if (rc)
3364                 CERROR("Failed to apply UDSPs on local net %s\n",
3365                        libcfs_net2str(net->net_id));
3366
3367         /* At this point we lost track of which NI was just added, so we
3368          * just re-apply the policies on all of the NIs on this net
3369          */
3370         list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3371                 rc = lnet_udsp_apply_policies_on_ni(ni);
3372                 if (rc)
3373                         CERROR("Failed to apply UDSPs on ni %s\n",
3374                                libcfs_nid2str(ni->ni_nid));
3375         }
3376         lnet_net_unlock(LNET_LOCK_EX);
3377
3378         /*
3379          * Start the acceptor thread if this is the first network
3380          * being added that requires the thread.
3381          */
3382         if (net->net_lnd->lnd_accept) {
3383                 rc = lnet_acceptor_start();
3384                 if (rc < 0) {
3385                         /* shutdown the net that we just started */
3386                         CERROR("Failed to start up acceptor thread\n");
3387                         lnet_shutdown_lndnet(net);
3388                         goto failed;
3389                 }
3390         }
3391
3392         lnet_net_lock(LNET_LOCK_EX);
3393         lnet_peer_net_added(net);
3394         lnet_net_unlock(LNET_LOCK_EX);
3395
3396         lnet_ping_target_update(pbuf, ping_mdh);
3397
3398         return 0;
3399
3400 failed:
3401         lnet_ping_md_unlink(pbuf, &ping_mdh);
3402         lnet_ping_buffer_decref(pbuf);
3403         return rc;
3404 }
3405
3406 static void
3407 lnet_set_tune_defaults(struct lnet_ioctl_config_lnd_tunables *tun)
3408 {
3409         if (tun) {
3410                 if (!tun->lt_cmn.lct_peer_timeout)
3411                         tun->lt_cmn.lct_peer_timeout = DEFAULT_PEER_TIMEOUT;
3412                 if (!tun->lt_cmn.lct_peer_tx_credits)
3413                         tun->lt_cmn.lct_peer_tx_credits = DEFAULT_PEER_CREDITS;
3414                 if (!tun->lt_cmn.lct_max_tx_credits)
3415                         tun->lt_cmn.lct_max_tx_credits = DEFAULT_CREDITS;
3416         }
3417 }
3418
3419 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3420                                       struct lnet_ioctl_config_lnd_tunables *tun)
3421 {
3422         struct lnet_net *net;
3423         const char *nets;
3424         int rc;
3425         LIST_HEAD(net_head);
3426
3427         rc = lnet_parse_ip2nets(&nets, ip2nets);
3428         if (rc < 0)
3429                 return rc;
3430
3431         rc = lnet_parse_networks(&net_head, nets);
3432         if (rc < 0)
3433                 return rc;
3434
3435         lnet_set_tune_defaults(tun);
3436
3437         mutex_lock(&the_lnet.ln_api_mutex);
3438         while (!list_empty(&net_head)) {
3439                 net = list_entry(net_head.next, struct lnet_net, net_list);
3440                 list_del_init(&net->net_list);
3441                 rc = lnet_add_net_common(net, tun);
3442                 if (rc < 0)
3443                         goto out;
3444         }
3445
3446 out:
3447         mutex_unlock(&the_lnet.ln_api_mutex);
3448
3449         while (!list_empty(&net_head)) {
3450                 net = list_entry(net_head.next, struct lnet_net, net_list);
3451                 list_del_init(&net->net_list);
3452                 lnet_net_free(net);
3453         }
3454         return rc;
3455 }
3456
3457 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3458 {
3459         struct lnet_net *net;
3460         struct lnet_ni *ni;
3461         struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3462         int rc, i;
3463         __u32 net_id, lnd_type;
3464
3465         /* get the tunables if they are available */
3466         if (conf->lic_cfg_hdr.ioc_len >=
3467             sizeof(*conf) + sizeof(*tun))
3468                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3469                         conf->lic_bulk;
3470
3471         /* handle legacy ip2nets from DLC */
3472         if (conf->lic_legacy_ip2nets[0] != '\0')
3473                 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3474                                                   tun);
3475
3476         net_id = LNET_NIDNET(conf->lic_nid);
3477         lnd_type = LNET_NETTYP(net_id);
3478
3479         if (!libcfs_isknown_lnd(lnd_type)) {
3480                 CERROR("No valid net and lnd information provided\n");
3481                 return -EINVAL;
3482         }
3483
3484         net = lnet_net_alloc(net_id, NULL);
3485         if (!net)
3486                 return -ENOMEM;
3487
3488         for (i = 0; i < conf->lic_ncpts; i++) {
3489                 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3490                         return -EINVAL;
3491         }
3492
3493         ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3494                                        conf->lic_ni_intf);
3495         if (!ni)
3496                 return -ENOMEM;
3497
3498         lnet_set_tune_defaults(tun);
3499
3500         mutex_lock(&the_lnet.ln_api_mutex);
3501
3502         rc = lnet_add_net_common(net, tun);
3503
3504         mutex_unlock(&the_lnet.ln_api_mutex);
3505
3506         return rc;
3507 }
3508
3509 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3510 {
3511         struct lnet_net  *net;
3512         struct lnet_ni *ni;
3513         __u32 net_id = LNET_NIDNET(conf->lic_nid);
3514         struct lnet_ping_buffer *pbuf;
3515         struct lnet_handle_md  ping_mdh;
3516         int               rc;
3517         int               net_count;
3518         __u32             addr;
3519
3520         /* don't allow userspace to shutdown the LOLND */
3521         if (LNET_NETTYP(net_id) == LOLND)
3522                 return -EINVAL;
3523
3524         mutex_lock(&the_lnet.ln_api_mutex);
3525
3526         lnet_net_lock(0);
3527
3528         net = lnet_get_net_locked(net_id);
3529         if (!net) {
3530                 CERROR("net %s not found\n",
3531                        libcfs_net2str(net_id));
3532                 rc = -ENOENT;
3533                 goto unlock_net;
3534         }
3535
3536         addr = LNET_NIDADDR(conf->lic_nid);
3537         if (addr == 0) {
3538                 /* remove the entire net */
3539                 net_count = lnet_get_net_ni_count_locked(net);
3540
3541                 lnet_net_unlock(0);
3542
3543                 /* create and link a new ping info, before removing the old one */
3544                 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3545                                         lnet_get_ni_count() - net_count,
3546                                         false);
3547                 if (rc != 0)
3548                         goto unlock_api_mutex;
3549
3550                 lnet_shutdown_lndnet(net);
3551
3552                 lnet_acceptor_stop();
3553
3554                 lnet_ping_target_update(pbuf, ping_mdh);
3555
3556                 goto unlock_api_mutex;
3557         }
3558
3559         ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3560         if (!ni) {
3561                 CERROR("nid %s not found\n",
3562                        libcfs_nid2str(conf->lic_nid));
3563                 rc = -ENOENT;
3564                 goto unlock_net;
3565         }
3566
3567         net_count = lnet_get_net_ni_count_locked(net);
3568
3569         lnet_net_unlock(0);
3570
3571         /* create and link a new ping info, before removing the old one */
3572         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3573                                   lnet_get_ni_count() - 1, false);
3574         if (rc != 0)
3575                 goto unlock_api_mutex;
3576
3577         lnet_shutdown_lndni(ni);
3578
3579         lnet_acceptor_stop();
3580
3581         lnet_ping_target_update(pbuf, ping_mdh);
3582
3583         /* check if the net is empty and remove it if it is */
3584         if (net_count == 1)
3585                 lnet_shutdown_lndnet(net);
3586
3587         goto unlock_api_mutex;
3588
3589 unlock_net:
3590         lnet_net_unlock(0);
3591 unlock_api_mutex:
3592         mutex_unlock(&the_lnet.ln_api_mutex);
3593
3594         return rc;
3595 }
3596
3597 /*
3598  * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3599  * They are only expected to be called for unique networks.
3600  * That can be as a result of older DLC library
3601  * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3602  */
3603 int
3604 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3605 {
3606         struct lnet_net *net;
3607         LIST_HEAD(net_head);
3608         int rc;
3609         struct lnet_ioctl_config_lnd_tunables tun;
3610         const char *nets = conf->cfg_config_u.cfg_net.net_intf;
3611
3612         /* Create a net/ni structures for the network string */
3613         rc = lnet_parse_networks(&net_head, nets);
3614         if (rc <= 0)
3615                 return rc == 0 ? -EINVAL : rc;
3616
3617         mutex_lock(&the_lnet.ln_api_mutex);
3618
3619         if (rc > 1) {
3620                 rc = -EINVAL; /* only add one network per call */
3621                 goto out_unlock_clean;
3622         }
3623
3624         net = list_entry(net_head.next, struct lnet_net, net_list);
3625         list_del_init(&net->net_list);
3626
3627         LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3628
3629         memset(&tun, 0, sizeof(tun));
3630
3631         tun.lt_cmn.lct_peer_timeout =
3632           (!conf->cfg_config_u.cfg_net.net_peer_timeout) ? DEFAULT_PEER_TIMEOUT :
3633                 conf->cfg_config_u.cfg_net.net_peer_timeout;
3634         tun.lt_cmn.lct_peer_tx_credits =
3635           (!conf->cfg_config_u.cfg_net.net_peer_tx_credits) ? DEFAULT_PEER_CREDITS :
3636                 conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3637         tun.lt_cmn.lct_peer_rtr_credits =
3638           conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3639         tun.lt_cmn.lct_max_tx_credits =
3640           (!conf->cfg_config_u.cfg_net.net_max_tx_credits) ? DEFAULT_CREDITS :
3641                 conf->cfg_config_u.cfg_net.net_max_tx_credits;
3642
3643         rc = lnet_add_net_common(net, &tun);
3644
3645 out_unlock_clean:
3646         mutex_unlock(&the_lnet.ln_api_mutex);
3647         while (!list_empty(&net_head)) {
3648                 /* net_head list is empty in success case */
3649                 net = list_entry(net_head.next, struct lnet_net, net_list);
3650                 list_del_init(&net->net_list);
3651                 lnet_net_free(net);
3652         }
3653         return rc;
3654 }
3655
3656 int
3657 lnet_dyn_del_net(__u32 net_id)
3658 {
3659         struct lnet_net  *net;
3660         struct lnet_ping_buffer *pbuf;
3661         struct lnet_handle_md ping_mdh;
3662         int               rc;
3663         int               net_ni_count;
3664
3665         /* don't allow userspace to shutdown the LOLND */
3666         if (LNET_NETTYP(net_id) == LOLND)
3667                 return -EINVAL;
3668
3669         mutex_lock(&the_lnet.ln_api_mutex);
3670
3671         lnet_net_lock(0);
3672
3673         net = lnet_get_net_locked(net_id);
3674         if (net == NULL) {
3675                 lnet_net_unlock(0);
3676                 rc = -EINVAL;
3677                 goto out;
3678         }
3679
3680         net_ni_count = lnet_get_net_ni_count_locked(net);
3681
3682         lnet_net_unlock(0);
3683
3684         /* create and link a new ping info, before removing the old one */
3685         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3686                                     lnet_get_ni_count() - net_ni_count, false);
3687         if (rc != 0)
3688                 goto out;
3689
3690         lnet_shutdown_lndnet(net);
3691
3692         lnet_acceptor_stop();
3693
3694         lnet_ping_target_update(pbuf, ping_mdh);
3695
3696 out:
3697         mutex_unlock(&the_lnet.ln_api_mutex);
3698
3699         return rc;
3700 }
3701
3702 void lnet_incr_dlc_seq(void)
3703 {
3704         atomic_inc(&lnet_dlc_seq_no);
3705 }
3706
3707 __u32 lnet_get_dlc_seq_locked(void)
3708 {
3709         return atomic_read(&lnet_dlc_seq_no);
3710 }
3711
3712 static void
3713 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3714 {
3715         struct lnet_net *net;
3716         struct lnet_ni *ni;
3717
3718         lnet_net_lock(LNET_LOCK_EX);
3719         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3720                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3721                         if (ni->ni_nid == nid || all) {
3722                                 atomic_set(&ni->ni_healthv, value);
3723                                 if (list_empty(&ni->ni_recovery) &&
3724                                     value < LNET_MAX_HEALTH_VALUE) {
3725                                         CERROR("manually adding local NI %s to recovery\n",
3726                                                libcfs_nid2str(ni->ni_nid));
3727                                         list_add_tail(&ni->ni_recovery,
3728                                                       &the_lnet.ln_mt_localNIRecovq);
3729                                         lnet_ni_addref_locked(ni, 0);
3730                                 }
3731                                 if (!all) {
3732                                         lnet_net_unlock(LNET_LOCK_EX);
3733                                         return;
3734                                 }
3735                         }
3736                 }
3737         }
3738         lnet_net_unlock(LNET_LOCK_EX);
3739 }
3740
3741 static int
3742 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3743 {
3744         int cpt, rc = 0;
3745         struct lnet_ni *ni;
3746         lnet_nid_t nid = stats->hlni_nid;
3747
3748         cpt = lnet_net_lock_current();
3749         ni = lnet_nid2ni_locked(nid, cpt);
3750
3751         if (!ni) {
3752                 rc = -ENOENT;
3753                 goto unlock;
3754         }
3755
3756         stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3757         stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3758         stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3759         stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3760         stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3761         stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3762         stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3763         stats->hlni_ping_count = ni->ni_ping_count;
3764         stats->hlni_next_ping = ni->ni_next_ping;
3765
3766 unlock:
3767         lnet_net_unlock(cpt);
3768
3769         return rc;
3770 }
3771
3772 static int
3773 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3774 {
3775         struct lnet_ni *ni;
3776         int i = 0;
3777
3778         lnet_net_lock(LNET_LOCK_EX);
3779         list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3780                 list->rlst_nid_array[i] = ni->ni_nid;
3781                 i++;
3782                 if (i >= LNET_MAX_SHOW_NUM_NID)
3783                         break;
3784         }
3785         lnet_net_unlock(LNET_LOCK_EX);
3786         list->rlst_num_nids = i;
3787
3788         return 0;
3789 }
3790
3791 static int
3792 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3793 {
3794         struct lnet_peer_ni *lpni;
3795         int i = 0;
3796
3797         lnet_net_lock(LNET_LOCK_EX);
3798         list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3799                 list->rlst_nid_array[i] = lpni->lpni_nid;
3800                 i++;
3801                 if (i >= LNET_MAX_SHOW_NUM_NID)
3802                         break;
3803         }
3804         lnet_net_unlock(LNET_LOCK_EX);
3805         list->rlst_num_nids = i;
3806
3807         return 0;
3808 }
3809
3810 /**
3811  * LNet ioctl handler.
3812  *
3813  */
3814 int
3815 LNetCtl(unsigned int cmd, void *arg)
3816 {
3817         struct libcfs_ioctl_data *data = arg;
3818         struct lnet_ioctl_config_data *config;
3819         struct lnet_process_id    id = {0};
3820         struct lnet_ni           *ni;
3821         int                       rc;
3822
3823         BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3824                      sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3825
3826         switch (cmd) {
3827         case IOC_LIBCFS_GET_NI:
3828                 rc = LNetGetId(data->ioc_count, &id);
3829                 data->ioc_nid = id.nid;
3830                 return rc;
3831
3832         case IOC_LIBCFS_FAIL_NID:
3833                 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3834
3835         case IOC_LIBCFS_ADD_ROUTE: {
3836                 /* default router sensitivity to 1 */
3837                 unsigned int sensitivity = 1;
3838                 config = arg;
3839
3840                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3841                         return -EINVAL;
3842
3843                 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3844                         sensitivity =
3845                           config->cfg_config_u.cfg_route.rtr_sensitivity;
3846                 }
3847
3848                 mutex_lock(&the_lnet.ln_api_mutex);
3849                 rc = lnet_add_route(config->cfg_net,
3850                                     config->cfg_config_u.cfg_route.rtr_hop,
3851                                     config->cfg_nid,
3852                                     config->cfg_config_u.cfg_route.
3853                                         rtr_priority, sensitivity);
3854                 mutex_unlock(&the_lnet.ln_api_mutex);
3855                 return rc;
3856         }
3857
3858         case IOC_LIBCFS_DEL_ROUTE:
3859                 config = arg;
3860
3861                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3862                         return -EINVAL;
3863
3864                 mutex_lock(&the_lnet.ln_api_mutex);
3865                 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3866                 mutex_unlock(&the_lnet.ln_api_mutex);
3867                 return rc;
3868
3869         case IOC_LIBCFS_GET_ROUTE:
3870                 config = arg;
3871
3872                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3873                         return -EINVAL;
3874
3875                 mutex_lock(&the_lnet.ln_api_mutex);
3876                 rc = lnet_get_route(config->cfg_count,
3877                                     &config->cfg_net,
3878                                     &config->cfg_config_u.cfg_route.rtr_hop,
3879                                     &config->cfg_nid,
3880                                     &config->cfg_config_u.cfg_route.rtr_flags,
3881                                     &config->cfg_config_u.cfg_route.
3882                                         rtr_priority,
3883                                     &config->cfg_config_u.cfg_route.
3884                                         rtr_sensitivity);
3885                 mutex_unlock(&the_lnet.ln_api_mutex);
3886                 return rc;
3887
3888         case IOC_LIBCFS_GET_LOCAL_NI: {
3889                 struct lnet_ioctl_config_ni *cfg_ni;
3890                 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3891                 struct lnet_ioctl_element_stats *stats;
3892                 __u32 tun_size;
3893
3894                 cfg_ni = arg;
3895
3896                 /* get the tunables if they are available */
3897                 if (cfg_ni->lic_cfg_hdr.ioc_len <
3898                     sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3899                         return -EINVAL;
3900
3901                 stats = (struct lnet_ioctl_element_stats *)
3902                         cfg_ni->lic_bulk;
3903                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3904                                 (cfg_ni->lic_bulk + sizeof(*stats));
3905
3906                 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3907                         sizeof(*stats);
3908
3909                 mutex_lock(&the_lnet.ln_api_mutex);
3910                 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3911                 mutex_unlock(&the_lnet.ln_api_mutex);
3912                 return rc;
3913         }
3914
3915         case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3916                 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3917
3918                 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3919                         return -EINVAL;
3920
3921                 mutex_lock(&the_lnet.ln_api_mutex);
3922                 rc = lnet_get_ni_stats(msg_stats);
3923                 mutex_unlock(&the_lnet.ln_api_mutex);
3924
3925                 return rc;
3926         }
3927
3928         case IOC_LIBCFS_GET_NET: {
3929                 size_t total = sizeof(*config) +
3930                                sizeof(struct lnet_ioctl_net_config);
3931                 config = arg;
3932
3933                 if (config->cfg_hdr.ioc_len < total)
3934                         return -EINVAL;
3935
3936                 mutex_lock(&the_lnet.ln_api_mutex);
3937                 rc = lnet_get_net_config(config);
3938                 mutex_unlock(&the_lnet.ln_api_mutex);
3939                 return rc;
3940         }
3941
3942         case IOC_LIBCFS_GET_LNET_STATS:
3943         {
3944                 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
3945
3946                 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
3947                         return -EINVAL;
3948
3949                 mutex_lock(&the_lnet.ln_api_mutex);
3950                 rc = lnet_counters_get(&lnet_stats->st_cntrs);
3951                 mutex_unlock(&the_lnet.ln_api_mutex);
3952                 return rc;
3953         }
3954
3955         case IOC_LIBCFS_CONFIG_RTR:
3956                 config = arg;
3957
3958                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3959                         return -EINVAL;
3960
3961                 mutex_lock(&the_lnet.ln_api_mutex);
3962                 if (config->cfg_config_u.cfg_buffers.buf_enable) {
3963                         rc = lnet_rtrpools_enable();
3964                         mutex_unlock(&the_lnet.ln_api_mutex);
3965                         return rc;
3966                 }
3967                 lnet_rtrpools_disable();
3968                 mutex_unlock(&the_lnet.ln_api_mutex);
3969                 return 0;
3970
3971         case IOC_LIBCFS_ADD_BUF:
3972                 config = arg;
3973
3974                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3975                         return -EINVAL;
3976
3977                 mutex_lock(&the_lnet.ln_api_mutex);
3978                 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
3979                                                 buf_tiny,
3980                                           config->cfg_config_u.cfg_buffers.
3981                                                 buf_small,
3982                                           config->cfg_config_u.cfg_buffers.
3983                                                 buf_large);
3984                 mutex_unlock(&the_lnet.ln_api_mutex);
3985                 return rc;
3986
3987         case IOC_LIBCFS_SET_NUMA_RANGE: {
3988                 struct lnet_ioctl_set_value *numa;
3989                 numa = arg;
3990                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
3991                         return -EINVAL;
3992                 lnet_net_lock(LNET_LOCK_EX);
3993                 lnet_numa_range = numa->sv_value;
3994                 lnet_net_unlock(LNET_LOCK_EX);
3995                 return 0;
3996         }
3997
3998         case IOC_LIBCFS_GET_NUMA_RANGE: {
3999                 struct lnet_ioctl_set_value *numa;
4000                 numa = arg;
4001                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
4002                         return -EINVAL;
4003                 numa->sv_value = lnet_numa_range;
4004                 return 0;
4005         }
4006
4007         case IOC_LIBCFS_GET_BUF: {
4008                 struct lnet_ioctl_pool_cfg *pool_cfg;
4009                 size_t total = sizeof(*config) + sizeof(*pool_cfg);
4010
4011                 config = arg;
4012
4013                 if (config->cfg_hdr.ioc_len < total)
4014                         return -EINVAL;
4015
4016                 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
4017
4018                 mutex_lock(&the_lnet.ln_api_mutex);
4019                 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
4020                 mutex_unlock(&the_lnet.ln_api_mutex);
4021                 return rc;
4022         }
4023
4024         case IOC_LIBCFS_GET_LOCAL_HSTATS: {
4025                 struct lnet_ioctl_local_ni_hstats *stats = arg;
4026
4027                 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
4028                         return -EINVAL;
4029
4030                 mutex_lock(&the_lnet.ln_api_mutex);
4031                 rc = lnet_get_local_ni_hstats(stats);
4032                 mutex_unlock(&the_lnet.ln_api_mutex);
4033
4034                 return rc;
4035         }
4036
4037         case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
4038                 struct lnet_ioctl_recovery_list *list = arg;
4039                 if (list->rlst_hdr.ioc_len < sizeof(*list))
4040                         return -EINVAL;
4041
4042                 mutex_lock(&the_lnet.ln_api_mutex);
4043                 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
4044                         rc = lnet_get_local_ni_recovery_list(list);
4045                 else
4046                         rc = lnet_get_peer_ni_recovery_list(list);
4047                 mutex_unlock(&the_lnet.ln_api_mutex);
4048                 return rc;
4049         }
4050
4051         case IOC_LIBCFS_ADD_PEER_NI: {
4052                 struct lnet_ioctl_peer_cfg *cfg = arg;
4053
4054                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4055                         return -EINVAL;
4056
4057                 mutex_lock(&the_lnet.ln_api_mutex);
4058                 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
4059                                       cfg->prcfg_cfg_nid,
4060                                       cfg->prcfg_mr);
4061                 mutex_unlock(&the_lnet.ln_api_mutex);
4062                 return rc;
4063         }
4064
4065         case IOC_LIBCFS_DEL_PEER_NI: {
4066                 struct lnet_ioctl_peer_cfg *cfg = arg;
4067
4068                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4069                         return -EINVAL;
4070
4071                 mutex_lock(&the_lnet.ln_api_mutex);
4072                 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
4073                                       cfg->prcfg_cfg_nid);
4074                 mutex_unlock(&the_lnet.ln_api_mutex);
4075                 return rc;
4076         }
4077
4078         case IOC_LIBCFS_GET_PEER_INFO: {
4079                 struct lnet_ioctl_peer *peer_info = arg;
4080
4081                 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
4082                         return -EINVAL;
4083
4084                 mutex_lock(&the_lnet.ln_api_mutex);
4085                 rc = lnet_get_peer_ni_info(
4086                    peer_info->pr_count,
4087                    &peer_info->pr_nid,
4088                    peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
4089                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
4090                    &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
4091                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
4092                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
4093                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
4094                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
4095                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
4096                 mutex_unlock(&the_lnet.ln_api_mutex);
4097                 return rc;
4098         }
4099
4100         case IOC_LIBCFS_GET_PEER_NI: {
4101                 struct lnet_ioctl_peer_cfg *cfg = arg;
4102
4103                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4104                         return -EINVAL;
4105
4106                 mutex_lock(&the_lnet.ln_api_mutex);
4107                 rc = lnet_get_peer_info(cfg,
4108                                         (void __user *)cfg->prcfg_bulk);
4109                 mutex_unlock(&the_lnet.ln_api_mutex);
4110                 return rc;
4111         }
4112
4113         case IOC_LIBCFS_GET_PEER_LIST: {
4114                 struct lnet_ioctl_peer_cfg *cfg = arg;
4115
4116                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4117                         return -EINVAL;
4118
4119                 mutex_lock(&the_lnet.ln_api_mutex);
4120                 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
4121                                 (struct lnet_process_id __user *)cfg->prcfg_bulk);
4122                 mutex_unlock(&the_lnet.ln_api_mutex);
4123                 return rc;
4124         }
4125
4126         case IOC_LIBCFS_SET_HEALHV: {
4127                 struct lnet_ioctl_reset_health_cfg *cfg = arg;
4128                 int value;
4129                 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
4130                         return -EINVAL;
4131                 if (cfg->rh_value < 0 ||
4132                     cfg->rh_value > LNET_MAX_HEALTH_VALUE)
4133                         value = LNET_MAX_HEALTH_VALUE;
4134                 else
4135                         value = cfg->rh_value;
4136                 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
4137                        value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
4138                        "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
4139                 mutex_lock(&the_lnet.ln_api_mutex);
4140                 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
4141                         lnet_ni_set_healthv(cfg->rh_nid, value,
4142                                              cfg->rh_all);
4143                 else
4144                         lnet_peer_ni_set_healthv(cfg->rh_nid, value,
4145                                                   cfg->rh_all);
4146                 mutex_unlock(&the_lnet.ln_api_mutex);
4147                 return 0;
4148         }
4149
4150         case IOC_LIBCFS_NOTIFY_ROUTER: {
4151                 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
4152
4153                 /* The deadline passed in by the user should be some time in
4154                  * seconds in the future since the UNIX epoch. We have to map
4155                  * that deadline to the wall clock.
4156                  */
4157                 deadline += ktime_get_seconds();
4158                 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
4159                                    deadline);
4160         }
4161
4162         case IOC_LIBCFS_LNET_DIST:
4163                 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
4164                 if (rc < 0 && rc != -EHOSTUNREACH)
4165                         return rc;
4166
4167                 data->ioc_u32[0] = rc;
4168                 return 0;
4169
4170         case IOC_LIBCFS_TESTPROTOCOMPAT:
4171                 the_lnet.ln_testprotocompat = data->ioc_flags;
4172                 return 0;
4173
4174         case IOC_LIBCFS_LNET_FAULT:
4175                 return lnet_fault_ctl(data->ioc_flags, data);
4176
4177         case IOC_LIBCFS_PING: {
4178                 signed long timeout;
4179
4180                 id.nid = data->ioc_nid;
4181                 id.pid = data->ioc_u32[0];
4182
4183                 /* If timeout is negative then set default of 3 minutes */
4184                 if (((s32)data->ioc_u32[1] <= 0) ||
4185                     data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
4186                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
4187                 else
4188                         timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
4189
4190                 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
4191                                data->ioc_plen1 / sizeof(struct lnet_process_id));
4192
4193                 if (rc < 0)
4194                         return rc;
4195
4196                 data->ioc_count = rc;
4197                 return 0;
4198         }
4199
4200         case IOC_LIBCFS_PING_PEER: {
4201                 struct lnet_ioctl_ping_data *ping = arg;
4202                 struct lnet_peer *lp;
4203                 signed long timeout;
4204
4205                 /* If timeout is negative then set default of 3 minutes */
4206                 if (((s32)ping->op_param) <= 0 ||
4207                     ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
4208                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
4209                 else
4210                         timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
4211
4212                 rc = lnet_ping(ping->ping_id, timeout,
4213                                ping->ping_buf,
4214                                ping->ping_count);
4215                 if (rc < 0)
4216                         return rc;
4217
4218                 mutex_lock(&the_lnet.ln_api_mutex);
4219                 lp = lnet_find_peer(ping->ping_id.nid);
4220                 if (lp) {
4221                         ping->ping_id.nid = lp->lp_primary_nid;
4222                         ping->mr_info = lnet_peer_is_multi_rail(lp);
4223                         lnet_peer_decref_locked(lp);
4224                 }
4225                 mutex_unlock(&the_lnet.ln_api_mutex);
4226
4227                 ping->ping_count = rc;
4228                 return 0;
4229         }
4230
4231         case IOC_LIBCFS_DISCOVER: {
4232                 struct lnet_ioctl_ping_data *discover = arg;
4233                 struct lnet_peer *lp;
4234
4235                 rc = lnet_discover(discover->ping_id, discover->op_param,
4236                                    discover->ping_buf,
4237                                    discover->ping_count);
4238                 if (rc < 0)
4239                         return rc;
4240
4241                 mutex_lock(&the_lnet.ln_api_mutex);
4242                 lp = lnet_find_peer(discover->ping_id.nid);
4243                 if (lp) {
4244                         discover->ping_id.nid = lp->lp_primary_nid;
4245                         discover->mr_info = lnet_peer_is_multi_rail(lp);
4246                         lnet_peer_decref_locked(lp);
4247                 }
4248                 mutex_unlock(&the_lnet.ln_api_mutex);
4249
4250                 discover->ping_count = rc;
4251                 return 0;
4252         }
4253
4254         case IOC_LIBCFS_ADD_UDSP: {
4255                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4256                 __u32 bulk_size = ioc_udsp->iou_hdr.ioc_len;
4257
4258                 mutex_lock(&the_lnet.ln_api_mutex);
4259                 rc = lnet_udsp_demarshal_add(arg, bulk_size);
4260                 if (!rc) {
4261                         rc = lnet_udsp_apply_policies(NULL, false);
4262                         CDEBUG(D_NET, "policy application returned %d\n", rc);
4263                         rc = 0;
4264                 }
4265                 mutex_unlock(&the_lnet.ln_api_mutex);
4266
4267                 return rc;
4268         }
4269
4270         case IOC_LIBCFS_DEL_UDSP: {
4271                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4272                 int idx = ioc_udsp->iou_idx;
4273
4274                 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4275                         return -EINVAL;
4276
4277                 mutex_lock(&the_lnet.ln_api_mutex);
4278                 rc = lnet_udsp_del_policy(idx);
4279                 if (!rc) {
4280                         rc = lnet_udsp_apply_policies(NULL, false);
4281                         CDEBUG(D_NET, "policy re-application returned %d\n",
4282                                rc);
4283                         rc = 0;
4284                 }
4285                 mutex_unlock(&the_lnet.ln_api_mutex);
4286
4287                 return rc;
4288         }
4289
4290         case IOC_LIBCFS_GET_UDSP_SIZE: {
4291                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4292                 struct lnet_udsp *udsp;
4293
4294                 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4295                         return -EINVAL;
4296
4297                 rc = 0;
4298
4299                 mutex_lock(&the_lnet.ln_api_mutex);
4300                 udsp = lnet_udsp_get_policy(ioc_udsp->iou_idx);
4301                 if (!udsp) {
4302                         rc = -ENOENT;
4303                 } else {
4304                         /* coming in iou_idx will hold the idx of the udsp
4305                          * to get the size of. going out the iou_idx will
4306                          * hold the size of the UDSP found at the passed
4307                          * in index.
4308                          */
4309                         ioc_udsp->iou_idx = lnet_get_udsp_size(udsp);
4310                         if (ioc_udsp->iou_idx < 0)
4311                                 rc = -EINVAL;
4312                 }
4313                 mutex_unlock(&the_lnet.ln_api_mutex);
4314
4315                 return rc;
4316         }
4317
4318         case IOC_LIBCFS_GET_UDSP: {
4319                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4320                 struct lnet_udsp *udsp;
4321
4322                 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4323                         return -EINVAL;
4324
4325                 rc = 0;
4326
4327                 mutex_lock(&the_lnet.ln_api_mutex);
4328                 udsp = lnet_udsp_get_policy(ioc_udsp->iou_idx);
4329                 if (!udsp)
4330                         rc = -ENOENT;
4331                 else
4332                         rc = lnet_udsp_marshal(udsp, ioc_udsp);
4333                 mutex_unlock(&the_lnet.ln_api_mutex);
4334
4335                 return rc;
4336         }
4337
4338         case IOC_LIBCFS_GET_CONST_UDSP_INFO: {
4339                 struct lnet_ioctl_construct_udsp_info *info = arg;
4340
4341                 if (info->cud_hdr.ioc_len < sizeof(*info))
4342                         return -EINVAL;
4343
4344                 CDEBUG(D_NET, "GET_UDSP_INFO for %s\n",
4345                        libcfs_nid2str(info->cud_nid));
4346
4347                 mutex_lock(&the_lnet.ln_api_mutex);
4348                 lnet_udsp_get_construct_info(info);
4349                 mutex_unlock(&the_lnet.ln_api_mutex);
4350
4351                 return 0;
4352         }
4353
4354         default:
4355                 ni = lnet_net2ni_addref(data->ioc_net);
4356                 if (ni == NULL)
4357                         return -EINVAL;
4358
4359                 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
4360                         rc = -EINVAL;
4361                 else
4362                         rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
4363
4364                 lnet_ni_decref(ni);
4365                 return rc;
4366         }
4367         /* not reached */
4368 }
4369 EXPORT_SYMBOL(LNetCtl);
4370
4371 void LNetDebugPeer(struct lnet_process_id id)
4372 {
4373         lnet_debug_peer(id.nid);
4374 }
4375 EXPORT_SYMBOL(LNetDebugPeer);
4376
4377 /**
4378  * Determine if the specified peer \a nid is on the local node.
4379  *
4380  * \param nid   peer nid to check
4381  *
4382  * \retval true         If peer NID is on the local node.
4383  * \retval false        If peer NID is not on the local node.
4384  */
4385 bool LNetIsPeerLocal(lnet_nid_t nid)
4386 {
4387         struct lnet_net *net;
4388         struct lnet_ni *ni;
4389         int cpt;
4390
4391         cpt = lnet_net_lock_current();
4392         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4393                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4394                         if (ni->ni_nid == nid) {
4395                                 lnet_net_unlock(cpt);
4396                                 return true;
4397                         }
4398                 }
4399         }
4400         lnet_net_unlock(cpt);
4401
4402         return false;
4403 }
4404 EXPORT_SYMBOL(LNetIsPeerLocal);
4405
4406 /**
4407  * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4408  * Note that all interfaces share a same PID, as requested by LNetNIInit().
4409  *
4410  * \param index Index of the interface to look up.
4411  * \param id On successful return, this location will hold the
4412  * struct lnet_process_id ID of the interface.
4413  *
4414  * \retval 0 If an interface exists at \a index.
4415  * \retval -ENOENT If no interface has been found.
4416  */
4417 int
4418 LNetGetId(unsigned int index, struct lnet_process_id *id)
4419 {
4420         struct lnet_ni   *ni;
4421         struct lnet_net  *net;
4422         int               cpt;
4423         int               rc = -ENOENT;
4424
4425         LASSERT(the_lnet.ln_refcount > 0);
4426
4427         cpt = lnet_net_lock_current();
4428
4429         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4430                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4431                         if (index-- != 0)
4432                                 continue;
4433
4434                         id->nid = ni->ni_nid;
4435                         id->pid = the_lnet.ln_pid;
4436                         rc = 0;
4437                         break;
4438                 }
4439         }
4440
4441         lnet_net_unlock(cpt);
4442         return rc;
4443 }
4444 EXPORT_SYMBOL(LNetGetId);
4445
4446 struct ping_data {
4447         int rc;
4448         int replied;
4449         struct lnet_handle_md mdh;
4450         struct completion completion;
4451 };
4452
4453 static void
4454 lnet_ping_event_handler(struct lnet_event *event)
4455 {
4456         struct ping_data *pd = event->md_user_ptr;
4457
4458         CDEBUG(D_NET, "ping event (%d %d)%s\n",
4459                event->type, event->status,
4460                event->unlinked ? " unlinked" : "");
4461
4462         if (event->status) {
4463                 if (!pd->rc)
4464                         pd->rc = event->status;
4465         } else if (event->type == LNET_EVENT_REPLY) {
4466                 pd->replied = 1;
4467                 pd->rc = event->mlength;
4468         }
4469         if (event->unlinked)
4470                 complete(&pd->completion);
4471 }
4472
4473 static int lnet_ping(struct lnet_process_id id, signed long timeout,
4474                      struct lnet_process_id __user *ids, int n_ids)
4475 {
4476         struct lnet_md md = { NULL };
4477         struct ping_data pd = { 0 };
4478         struct lnet_ping_buffer *pbuf;
4479         struct lnet_process_id tmpid;
4480         int i;
4481         int nob;
4482         int rc;
4483         int rc2;
4484
4485         /* n_ids limit is arbitrary */
4486         if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4487                 return -EINVAL;
4488
4489         /*
4490          * if the user buffer has more space than the lnet_interfaces_max
4491          * then only fill it up to lnet_interfaces_max
4492          */
4493         if (n_ids > lnet_interfaces_max)
4494                 n_ids = lnet_interfaces_max;
4495
4496         if (id.pid == LNET_PID_ANY)
4497                 id.pid = LNET_PID_LUSTRE;
4498
4499         pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4500         if (!pbuf)
4501                 return -ENOMEM;
4502
4503         /* initialize md content */
4504         md.start     = &pbuf->pb_info;
4505         md.length    = LNET_PING_INFO_SIZE(n_ids);
4506         md.threshold = 2; /* GET/REPLY */
4507         md.max_size  = 0;
4508         md.options   = LNET_MD_TRUNCATE;
4509         md.user_ptr  = &pd;
4510         md.handler   = lnet_ping_event_handler;
4511
4512         init_completion(&pd.completion);
4513
4514         rc = LNetMDBind(&md, LNET_UNLINK, &pd.mdh);
4515         if (rc != 0) {
4516                 CERROR("Can't bind MD: %d\n", rc);
4517                 goto fail_ping_buffer_decref;
4518         }
4519
4520         rc = LNetGet(LNET_NID_ANY, pd.mdh, id,
4521                      LNET_RESERVED_PORTAL,
4522                      LNET_PROTO_PING_MATCHBITS, 0, false);
4523
4524         if (rc != 0) {
4525                 /* Don't CERROR; this could be deliberate! */
4526                 rc2 = LNetMDUnlink(pd.mdh);
4527                 LASSERT(rc2 == 0);
4528
4529                 /* NB must wait for the UNLINK event below... */
4530         }
4531
4532         if (wait_for_completion_timeout(&pd.completion, timeout) == 0) {
4533                 /* Ensure completion in finite time... */
4534                 LNetMDUnlink(pd.mdh);
4535                 wait_for_completion(&pd.completion);
4536         }
4537         if (!pd.replied) {
4538                 rc = -EIO;
4539                 goto fail_ping_buffer_decref;
4540         }
4541
4542         nob = pd.rc;
4543         LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4544
4545         rc = -EPROTO;           /* if I can't parse... */
4546
4547         if (nob < 8) {
4548                 CERROR("%s: ping info too short %d\n",
4549                        libcfs_id2str(id), nob);
4550                 goto fail_ping_buffer_decref;
4551         }
4552
4553         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4554                 lnet_swap_pinginfo(pbuf);
4555         } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4556                 CERROR("%s: Unexpected magic %08x\n",
4557                        libcfs_id2str(id), pbuf->pb_info.pi_magic);
4558                 goto fail_ping_buffer_decref;
4559         }
4560
4561         if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4562                 CERROR("%s: ping w/o NI status: 0x%x\n",
4563                        libcfs_id2str(id), pbuf->pb_info.pi_features);
4564                 goto fail_ping_buffer_decref;
4565         }
4566
4567         if (nob < LNET_PING_INFO_SIZE(0)) {
4568                 CERROR("%s: Short reply %d(%d min)\n",
4569                        libcfs_id2str(id),
4570                        nob, (int)LNET_PING_INFO_SIZE(0));
4571                 goto fail_ping_buffer_decref;
4572         }
4573
4574         if (pbuf->pb_info.pi_nnis < n_ids)
4575                 n_ids = pbuf->pb_info.pi_nnis;
4576
4577         if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4578                 CERROR("%s: Short reply %d(%d expected)\n",
4579                        libcfs_id2str(id),
4580                        nob, (int)LNET_PING_INFO_SIZE(n_ids));
4581                 goto fail_ping_buffer_decref;
4582         }
4583
4584         rc = -EFAULT;           /* if I segv in copy_to_user()... */
4585
4586         memset(&tmpid, 0, sizeof(tmpid));
4587         for (i = 0; i < n_ids; i++) {
4588                 tmpid.pid = pbuf->pb_info.pi_pid;
4589                 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4590                 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4591                         goto fail_ping_buffer_decref;
4592         }
4593         rc = pbuf->pb_info.pi_nnis;
4594
4595  fail_ping_buffer_decref:
4596         lnet_ping_buffer_decref(pbuf);
4597         return rc;
4598 }
4599
4600 static int
4601 lnet_discover(struct lnet_process_id id, __u32 force,
4602               struct lnet_process_id __user *ids, int n_ids)
4603 {
4604         struct lnet_peer_ni *lpni;
4605         struct lnet_peer_ni *p;
4606         struct lnet_peer *lp;
4607         struct lnet_process_id *buf;
4608         int cpt;
4609         int i;
4610         int rc;
4611
4612         if (n_ids <= 0 ||
4613             id.nid == LNET_NID_ANY)
4614                 return -EINVAL;
4615
4616         if (id.pid == LNET_PID_ANY)
4617                 id.pid = LNET_PID_LUSTRE;
4618
4619         /*
4620          * If the user buffer has more space than the lnet_interfaces_max,
4621          * then only fill it up to lnet_interfaces_max.
4622          */
4623         if (n_ids > lnet_interfaces_max)
4624                 n_ids = lnet_interfaces_max;
4625
4626         CFS_ALLOC_PTR_ARRAY(buf, n_ids);
4627         if (!buf)
4628                 return -ENOMEM;
4629
4630         cpt = lnet_net_lock_current();
4631         lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4632         if (IS_ERR(lpni)) {
4633                 rc = PTR_ERR(lpni);
4634                 goto out;
4635         }
4636
4637         /*
4638          * Clearing the NIDS_UPTODATE flag ensures the peer will
4639          * be discovered, provided discovery has not been disabled.
4640          */
4641         lp = lpni->lpni_peer_net->lpn_peer;
4642         spin_lock(&lp->lp_lock);
4643         lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4644         /* If the force flag is set, force a PING and PUSH as well. */
4645         if (force)
4646                 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4647         spin_unlock(&lp->lp_lock);
4648         rc = lnet_discover_peer_locked(lpni, cpt, true);
4649         if (rc)
4650                 goto out_decref;
4651
4652         /* The lpni (or lp) for this NID may have changed and our ref is
4653          * the only thing keeping the old one around. Release the ref
4654          * and lookup the lpni again
4655          */
4656         lnet_peer_ni_decref_locked(lpni);
4657         lpni = lnet_find_peer_ni_locked(id.nid);
4658         if (!lpni) {
4659                 rc = -ENOENT;
4660                 goto out;
4661         }
4662         lp = lpni->lpni_peer_net->lpn_peer;
4663
4664         i = 0;
4665         p = NULL;
4666         while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4667                 buf[i].pid = id.pid;
4668                 buf[i].nid = p->lpni_nid;
4669                 if (++i >= n_ids)
4670                         break;
4671         }
4672         rc = i;
4673
4674 out_decref:
4675         lnet_peer_ni_decref_locked(lpni);
4676 out:
4677         lnet_net_unlock(cpt);
4678
4679         if (rc >= 0)
4680                 if (copy_to_user(ids, buf, rc * sizeof(*buf)))
4681                         rc = -EFAULT;
4682         CFS_FREE_PTR_ARRAY(buf, n_ids);
4683
4684         return rc;
4685 }
4686
4687 /**
4688  * Retrieve peer discovery status.
4689  *
4690  * \retval 1 if lnet_peer_discovery_disabled is 0
4691  * \retval 0 if lnet_peer_discovery_disabled is 1
4692  */
4693 int
4694 LNetGetPeerDiscoveryStatus(void)
4695 {
4696         return !lnet_peer_discovery_disabled;
4697 }
4698 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);