lnet/lnet/api-ni.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  */
  31
  32 #define DEBUG_SUBSYSTEM S_LNET
  33
  34 #include <linux/ctype.h>
  35 #include <linux/log2.h>
  36 #include <linux/ktime.h>
  37 #include <linux/moduleparam.h>
  38 #include <linux/uaccess.h>
  39 #ifdef HAVE_SCHED_HEADERS
  40 #include <linux/sched/signal.h>
  41 #endif
  42 #include <lnet/udsp.h>
  43 #include <lnet/lib-lnet.h>
  44
  45 #define D_LNI D_CONSOLE
  46
  47 /*
  48  * initialize ln_api_mutex statically, since it needs to be used in
  49  * discovery_set callback. That module parameter callback can be called
  50  * before module init completes. The mutex needs to be ready for use then.
  51  */
  52 struct lnet the_lnet = {
  53         .ln_api_mutex = __MUTEX_INITIALIZER(the_lnet.ln_api_mutex),
  54 };              /* THE state of the network */
  55 EXPORT_SYMBOL(the_lnet);
  56
  57 static char *ip2nets = "";
  58 module_param(ip2nets, charp, 0444);
  59 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
  60
  61 static char *networks = "";
  62 module_param(networks, charp, 0444);
  63 MODULE_PARM_DESC(networks, "local networks");
  64
  65 static char *routes = "";
  66 module_param(routes, charp, 0444);
  67 MODULE_PARM_DESC(routes, "routes to non-local networks");
  68
  69 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
  70 module_param(rnet_htable_size, int, 0444);
  71 MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
  72
  73 static int use_tcp_bonding;
  74 module_param(use_tcp_bonding, int, 0444);
  75 MODULE_PARM_DESC(use_tcp_bonding,
  76                  "use_tcp_bonding parameter has been removed");
  77
  78 unsigned int lnet_numa_range = 0;
  79 module_param(lnet_numa_range, uint, 0444);
  80 MODULE_PARM_DESC(lnet_numa_range,
  81                 "NUMA range to consider during Multi-Rail selection");
  82
  83 /*
  84  * lnet_health_sensitivity determines by how much we decrement the health
  85  * value on sending error. The value defaults to 100, which means health
  86  * interface health is decremented by 100 points every failure.
  87  */
  88 unsigned int lnet_health_sensitivity = 100;
  89 static int sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp);
  90 #ifdef HAVE_KERNEL_PARAM_OPS
  91 static struct kernel_param_ops param_ops_health_sensitivity = {
  92         .set = sensitivity_set,
  93         .get = param_get_int,
  94 };
  95 #define param_check_health_sensitivity(name, p) \
  96                 __param_check(name, p, int)
  97 module_param(lnet_health_sensitivity, health_sensitivity, S_IRUGO|S_IWUSR);
  98 #else
  99 module_param_call(lnet_health_sensitivity, sensitivity_set, param_get_int,
 100                   &lnet_health_sensitivity, S_IRUGO|S_IWUSR);
 101 #endif
 102 MODULE_PARM_DESC(lnet_health_sensitivity,
 103                 "Value to decrement the health value by on error");
 104
 105 /*
 106  * lnet_recovery_interval determines how often we should perform recovery
 107  * on unhealthy interfaces.
 108  */
 109 unsigned int lnet_recovery_interval = 1;
 110 static int recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp);
 111 #ifdef HAVE_KERNEL_PARAM_OPS
 112 static struct kernel_param_ops param_ops_recovery_interval = {
 113         .set = recovery_interval_set,
 114         .get = param_get_int,
 115 };
 116 #define param_check_recovery_interval(name, p) \
 117                 __param_check(name, p, int)
 118 module_param(lnet_recovery_interval, recovery_interval, S_IRUGO|S_IWUSR);
 119 #else
 120 module_param_call(lnet_recovery_interval, recovery_interval_set, param_get_int,
 121                   &lnet_recovery_interval, S_IRUGO|S_IWUSR);
 122 #endif
 123 MODULE_PARM_DESC(lnet_recovery_interval,
 124                 "DEPRECATED - Interval to recover unhealthy interfaces in seconds");
 125
 126 unsigned int lnet_recovery_limit;
 127 module_param(lnet_recovery_limit, uint, 0644);
 128 MODULE_PARM_DESC(lnet_recovery_limit,
 129                  "How long to attempt recovery of unhealthy peer interfaces in seconds. Set to 0 to allow indefinite recovery");
 130
 131 static int lnet_interfaces_max = LNET_INTERFACES_MAX_DEFAULT;
 132 static int intf_max_set(const char *val, cfs_kernel_param_arg_t *kp);
 133
 134 static struct kernel_param_ops param_ops_interfaces_max = {
 135         .set = intf_max_set,
 136         .get = param_get_int,
 137 };
 138
 139 #define param_check_interfaces_max(name, p) \
 140                 __param_check(name, p, int)
 141
 142 #ifdef HAVE_KERNEL_PARAM_OPS
 143 module_param(lnet_interfaces_max, interfaces_max, 0644);
 144 #else
 145 module_param_call(lnet_interfaces_max, intf_max_set, param_get_int,
 146                   &param_ops_interfaces_max, 0644);
 147 #endif
 148 MODULE_PARM_DESC(lnet_interfaces_max,
 149                 "Maximum number of interfaces in a node.");
 150
 151 unsigned lnet_peer_discovery_disabled = 0;
 152 static int discovery_set(const char *val, cfs_kernel_param_arg_t *kp);
 153
 154 static struct kernel_param_ops param_ops_discovery_disabled = {
 155         .set = discovery_set,
 156         .get = param_get_int,
 157 };
 158
 159 #define param_check_discovery_disabled(name, p) \
 160                 __param_check(name, p, int)
 161 #ifdef HAVE_KERNEL_PARAM_OPS
 162 module_param(lnet_peer_discovery_disabled, discovery_disabled, 0644);
 163 #else
 164 module_param_call(lnet_peer_discovery_disabled, discovery_set, param_get_int,
 165                   &param_ops_discovery_disabled, 0644);
 166 #endif
 167 MODULE_PARM_DESC(lnet_peer_discovery_disabled,
 168                 "Set to 1 to disable peer discovery on this node.");
 169
 170 unsigned int lnet_drop_asym_route;
 171 static int drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp);
 172
 173 static struct kernel_param_ops param_ops_drop_asym_route = {
 174         .set = drop_asym_route_set,
 175         .get = param_get_int,
 176 };
 177
 178 #define param_check_drop_asym_route(name, p)    \
 179         __param_check(name, p, int)
 180 #ifdef HAVE_KERNEL_PARAM_OPS
 181 module_param(lnet_drop_asym_route, drop_asym_route, 0644);
 182 #else
 183 module_param_call(lnet_drop_asym_route, drop_asym_route_set, param_get_int,
 184                   &param_ops_drop_asym_route, 0644);
 185 #endif
 186 MODULE_PARM_DESC(lnet_drop_asym_route,
 187                  "Set to 1 to drop asymmetrical route messages.");
 188
 189 #define LNET_TRANSACTION_TIMEOUT_DEFAULT 50
 190 unsigned int lnet_transaction_timeout = LNET_TRANSACTION_TIMEOUT_DEFAULT;
 191 static int transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp);
 192 #ifdef HAVE_KERNEL_PARAM_OPS
 193 static struct kernel_param_ops param_ops_transaction_timeout = {
 194         .set = transaction_to_set,
 195         .get = param_get_int,
 196 };
 197
 198 #define param_check_transaction_timeout(name, p) \
 199                 __param_check(name, p, int)
 200 module_param(lnet_transaction_timeout, transaction_timeout, S_IRUGO|S_IWUSR);
 201 #else
 202 module_param_call(lnet_transaction_timeout, transaction_to_set, param_get_int,
 203                   &lnet_transaction_timeout, S_IRUGO|S_IWUSR);
 204 #endif
 205 MODULE_PARM_DESC(lnet_transaction_timeout,
 206                 "Maximum number of seconds to wait for a peer response.");
 207
 208 #define LNET_RETRY_COUNT_DEFAULT 2
 209 unsigned int lnet_retry_count = LNET_RETRY_COUNT_DEFAULT;
 210 static int retry_count_set(const char *val, cfs_kernel_param_arg_t *kp);
 211 #ifdef HAVE_KERNEL_PARAM_OPS
 212 static struct kernel_param_ops param_ops_retry_count = {
 213         .set = retry_count_set,
 214         .get = param_get_int,
 215 };
 216
 217 #define param_check_retry_count(name, p) \
 218                 __param_check(name, p, int)
 219 module_param(lnet_retry_count, retry_count, S_IRUGO|S_IWUSR);
 220 #else
 221 module_param_call(lnet_retry_count, retry_count_set, param_get_int,
 222                   &lnet_retry_count, S_IRUGO|S_IWUSR);
 223 #endif
 224 MODULE_PARM_DESC(lnet_retry_count,
 225                  "Maximum number of times to retry transmitting a message");
 226
 227 unsigned int lnet_response_tracking = 3;
 228 static int response_tracking_set(const char *val, cfs_kernel_param_arg_t *kp);
 229
 230 #ifdef HAVE_KERNEL_PARAM_OPS
 231 static struct kernel_param_ops param_ops_response_tracking = {
 232         .set = response_tracking_set,
 233         .get = param_get_int,
 234 };
 235
 236 #define param_check_response_tracking(name, p)  \
 237         __param_check(name, p, int)
 238 module_param(lnet_response_tracking, response_tracking, 0644);
 239 #else
 240 module_param_call(lnet_response_tracking, response_tracking_set, param_get_int,
 241                   &lnet_response_tracking, 0644);
 242 #endif
 243 MODULE_PARM_DESC(lnet_response_tracking,
 244                  "(0|1|2|3) LNet Internal Only|GET Reply only|PUT ACK only|Full Tracking (default)");
 245
 246 #define LNET_LND_TIMEOUT_DEFAULT ((LNET_TRANSACTION_TIMEOUT_DEFAULT - 1) / \
 247                                   (LNET_RETRY_COUNT_DEFAULT + 1))
 248 unsigned int lnet_lnd_timeout = LNET_LND_TIMEOUT_DEFAULT;
 249 static void lnet_set_lnd_timeout(void)
 250 {
 251         lnet_lnd_timeout = (lnet_transaction_timeout - 1) /
 252                            (lnet_retry_count + 1);
 253 }
 254
 255 /*
 256  * This sequence number keeps track of how many times DLC was used to
 257  * update the local NIs. It is incremented when a NI is added or
 258  * removed and checked when sending a message to determine if there is
 259  * a need to re-run the selection algorithm. See lnet_select_pathway()
 260  * for more details on its usage.
 261  */
 262 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
 263
 264 static int lnet_ping(struct lnet_process_id id, signed long timeout,
 265                      struct lnet_process_id __user *ids, int n_ids);
 266
 267 static int lnet_discover(struct lnet_process_id id, __u32 force,
 268                          struct lnet_process_id __user *ids, int n_ids);
 269
 270 static int
 271 sensitivity_set(const char *val, cfs_kernel_param_arg_t *kp)
 272 {
 273         int rc;
 274         unsigned *sensitivity = (unsigned *)kp->arg;
 275         unsigned long value;
 276
 277         rc = kstrtoul(val, 0, &value);
 278         if (rc) {
 279                 CERROR("Invalid module parameter value for 'lnet_health_sensitivity'\n");
 280                 return rc;
 281         }
 282
 283         /*
 284          * The purpose of locking the api_mutex here is to ensure that
 285          * the correct value ends up stored properly.
 286          */
 287         mutex_lock(&the_lnet.ln_api_mutex);
 288
 289         if (value > LNET_MAX_HEALTH_VALUE) {
 290                 mutex_unlock(&the_lnet.ln_api_mutex);
 291                 CERROR("Invalid health value. Maximum: %d value = %lu\n",
 292                        LNET_MAX_HEALTH_VALUE, value);
 293                 return -EINVAL;
 294         }
 295
 296         if (*sensitivity != 0 && value == 0 && lnet_retry_count != 0) {
 297                 lnet_retry_count = 0;
 298                 lnet_set_lnd_timeout();
 299         }
 300
 301         *sensitivity = value;
 302
 303         mutex_unlock(&the_lnet.ln_api_mutex);
 304
 305         return 0;
 306 }
 307
 308 static int
 309 recovery_interval_set(const char *val, cfs_kernel_param_arg_t *kp)
 310 {
 311         CWARN("'lnet_recovery_interval' has been deprecated\n");
 312
 313         return 0;
 314 }
 315
 316 static int
 317 discovery_set(const char *val, cfs_kernel_param_arg_t *kp)
 318 {
 319         int rc;
 320         unsigned *discovery_off = (unsigned *)kp->arg;
 321         unsigned long value;
 322         struct lnet_ping_buffer *pbuf;
 323
 324         rc = kstrtoul(val, 0, &value);
 325         if (rc) {
 326                 CERROR("Invalid module parameter value for 'lnet_peer_discovery_disabled'\n");
 327                 return rc;
 328         }
 329
 330         value = (value) ? 1 : 0;
 331
 332         /*
 333          * The purpose of locking the api_mutex here is to ensure that
 334          * the correct value ends up stored properly.
 335          */
 336         mutex_lock(&the_lnet.ln_api_mutex);
 337
 338         if (value == *discovery_off) {
 339                 mutex_unlock(&the_lnet.ln_api_mutex);
 340                 return 0;
 341         }
 342
 343         /*
 344          * We still want to set the discovery value even when LNet is not
 345          * running. This is the case when LNet is being loaded and we want
 346          * the module parameters to take effect. Otherwise if we're
 347          * changing the value dynamically, we want to set it after
 348          * updating the peers
 349          */
 350         if (the_lnet.ln_state != LNET_STATE_RUNNING) {
 351                 *discovery_off = value;
 352                 mutex_unlock(&the_lnet.ln_api_mutex);
 353                 return 0;
 354         }
 355
 356         /* tell peers that discovery setting has changed */
 357         lnet_net_lock(LNET_LOCK_EX);
 358         pbuf = the_lnet.ln_ping_target;
 359         if (value)
 360                 pbuf->pb_info.pi_features &= ~LNET_PING_FEAT_DISCOVERY;
 361         else
 362                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
 363         lnet_net_unlock(LNET_LOCK_EX);
 364
 365         /* only send a push when we're turning off discovery */
 366         if (*discovery_off <= 0 && value > 0)
 367                 lnet_push_update_to_peers(1);
 368         *discovery_off = value;
 369
 370         mutex_unlock(&the_lnet.ln_api_mutex);
 371
 372         return 0;
 373 }
 374
 375 static int
 376 drop_asym_route_set(const char *val, cfs_kernel_param_arg_t *kp)
 377 {
 378         int rc;
 379         unsigned int *drop_asym_route = (unsigned int *)kp->arg;
 380         unsigned long value;
 381
 382         rc = kstrtoul(val, 0, &value);
 383         if (rc) {
 384                 CERROR("Invalid module parameter value for "
 385                        "'lnet_drop_asym_route'\n");
 386                 return rc;
 387         }
 388
 389         /*
 390          * The purpose of locking the api_mutex here is to ensure that
 391          * the correct value ends up stored properly.
 392          */
 393         mutex_lock(&the_lnet.ln_api_mutex);
 394
 395         if (value == *drop_asym_route) {
 396                 mutex_unlock(&the_lnet.ln_api_mutex);
 397                 return 0;
 398         }
 399
 400         *drop_asym_route = value;
 401
 402         mutex_unlock(&the_lnet.ln_api_mutex);
 403
 404         return 0;
 405 }
 406
 407 static int
 408 transaction_to_set(const char *val, cfs_kernel_param_arg_t *kp)
 409 {
 410         int rc;
 411         unsigned *transaction_to = (unsigned *)kp->arg;
 412         unsigned long value;
 413
 414         rc = kstrtoul(val, 0, &value);
 415         if (rc) {
 416                 CERROR("Invalid module parameter value for 'lnet_transaction_timeout'\n");
 417                 return rc;
 418         }
 419
 420         /*
 421          * The purpose of locking the api_mutex here is to ensure that
 422          * the correct value ends up stored properly.
 423          */
 424         mutex_lock(&the_lnet.ln_api_mutex);
 425
 426         if (value <= lnet_retry_count || value == 0) {
 427                 mutex_unlock(&the_lnet.ln_api_mutex);
 428                 CERROR("Invalid value for lnet_transaction_timeout (%lu). "
 429                        "Has to be greater than lnet_retry_count (%u)\n",
 430                        value, lnet_retry_count);
 431                 return -EINVAL;
 432         }
 433
 434         if (value == *transaction_to) {
 435                 mutex_unlock(&the_lnet.ln_api_mutex);
 436                 return 0;
 437         }
 438
 439         *transaction_to = value;
 440         /* Update the lnet_lnd_timeout now that we've modified the
 441          * transaction timeout
 442          */
 443         lnet_set_lnd_timeout();
 444
 445         mutex_unlock(&the_lnet.ln_api_mutex);
 446
 447         return 0;
 448 }
 449
 450 static int
 451 retry_count_set(const char *val, cfs_kernel_param_arg_t *kp)
 452 {
 453         int rc;
 454         unsigned *retry_count = (unsigned *)kp->arg;
 455         unsigned long value;
 456
 457         rc = kstrtoul(val, 0, &value);
 458         if (rc) {
 459                 CERROR("Invalid module parameter value for 'lnet_retry_count'\n");
 460                 return rc;
 461         }
 462
 463         /*
 464          * The purpose of locking the api_mutex here is to ensure that
 465          * the correct value ends up stored properly.
 466          */
 467         mutex_lock(&the_lnet.ln_api_mutex);
 468
 469         if (lnet_health_sensitivity == 0 && value > 0) {
 470                 mutex_unlock(&the_lnet.ln_api_mutex);
 471                 CERROR("Can not set lnet_retry_count when health feature is turned off\n");
 472                 return -EINVAL;
 473         }
 474
 475         if (value > lnet_transaction_timeout) {
 476                 mutex_unlock(&the_lnet.ln_api_mutex);
 477                 CERROR("Invalid value for lnet_retry_count (%lu). "
 478                        "Has to be smaller than lnet_transaction_timeout (%u)\n",
 479                        value, lnet_transaction_timeout);
 480                 return -EINVAL;
 481         }
 482
 483         *retry_count = value;
 484
 485         /* Update the lnet_lnd_timeout now that we've modified the
 486          * retry count
 487          */
 488         lnet_set_lnd_timeout();
 489
 490         mutex_unlock(&the_lnet.ln_api_mutex);
 491
 492         return 0;
 493 }
 494
 495 static int
 496 intf_max_set(const char *val, cfs_kernel_param_arg_t *kp)
 497 {
 498         int value, rc;
 499
 500         rc = kstrtoint(val, 0, &value);
 501         if (rc) {
 502                 CERROR("Invalid module parameter value for 'lnet_interfaces_max'\n");
 503                 return rc;
 504         }
 505
 506         if (value < LNET_INTERFACES_MIN) {
 507                 CWARN("max interfaces provided are too small, setting to %d\n",
 508                       LNET_INTERFACES_MAX_DEFAULT);
 509                 value = LNET_INTERFACES_MAX_DEFAULT;
 510         }
 511
 512         *(int *)kp->arg = value;
 513
 514         return 0;
 515 }
 516
 517 static int
 518 response_tracking_set(const char *val, cfs_kernel_param_arg_t *kp)
 519 {
 520         int rc;
 521         unsigned long new_value;
 522
 523         rc = kstrtoul(val, 0, &new_value);
 524         if (rc) {
 525                 CERROR("Invalid value for 'lnet_response_tracking'\n");
 526                 return -EINVAL;
 527         }
 528
 529         if (new_value < 0 || new_value > 3) {
 530                 CWARN("Invalid value (%lu) for 'lnet_response_tracking'\n",
 531                       new_value);
 532                 return -EINVAL;
 533         }
 534
 535         lnet_response_tracking = new_value;
 536
 537         return 0;
 538 }
 539
 540 static const char *
 541 lnet_get_routes(void)
 542 {
 543         return routes;
 544 }
 545
 546 static const char *
 547 lnet_get_networks(void)
 548 {
 549         const char *nets;
 550         int rc;
 551
 552         if (*networks != 0 && *ip2nets != 0) {
 553                 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
 554                                    "'ip2nets' but not both at once\n");
 555                 return NULL;
 556         }
 557
 558         if (*ip2nets != 0) {
 559                 rc = lnet_parse_ip2nets(&nets, ip2nets);
 560                 return (rc == 0) ? nets : NULL;
 561         }
 562
 563         if (*networks != 0)
 564                 return networks;
 565
 566         return "tcp";
 567 }
 568
 569 static void
 570 lnet_init_locks(void)
 571 {
 572         spin_lock_init(&the_lnet.ln_eq_wait_lock);
 573         spin_lock_init(&the_lnet.ln_msg_resend_lock);
 574         init_completion(&the_lnet.ln_mt_wait_complete);
 575         mutex_init(&the_lnet.ln_lnd_mutex);
 576 }
 577
 578 struct kmem_cache *lnet_mes_cachep;        /* MEs kmem_cache */
 579 struct kmem_cache *lnet_small_mds_cachep;  /* <= LNET_SMALL_MD_SIZE bytes
 580                                             *  MDs kmem_cache */
 581 struct kmem_cache *lnet_udsp_cachep;       /* udsp cache */
 582 struct kmem_cache *lnet_rspt_cachep;       /* response tracker cache */
 583 struct kmem_cache *lnet_msg_cachep;
 584
 585 static int
 586 lnet_slab_setup(void)
 587 {
 588         /* create specific kmem_cache for MEs and small MDs (i.e., originally
 589          * allocated in <size-xxx> kmem_cache).
 590          */
 591         lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(struct lnet_me),
 592                                             0, 0, NULL);
 593         if (!lnet_mes_cachep)
 594                 return -ENOMEM;
 595
 596         lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
 597                                                   LNET_SMALL_MD_SIZE, 0, 0,
 598                                                   NULL);
 599         if (!lnet_small_mds_cachep)
 600                 return -ENOMEM;
 601
 602         lnet_udsp_cachep = kmem_cache_create("lnet_udsp",
 603                                              sizeof(struct lnet_udsp),
 604                                              0, 0, NULL);
 605         if (!lnet_udsp_cachep)
 606                 return -ENOMEM;
 607
 608         lnet_rspt_cachep = kmem_cache_create("lnet_rspt", sizeof(struct lnet_rsp_tracker),
 609                                             0, 0, NULL);
 610         if (!lnet_rspt_cachep)
 611                 return -ENOMEM;
 612
 613         lnet_msg_cachep = kmem_cache_create("lnet_msg", sizeof(struct lnet_msg),
 614                                             0, 0, NULL);
 615         if (!lnet_msg_cachep)
 616                 return -ENOMEM;
 617
 618         return 0;
 619 }
 620
 621 static void
 622 lnet_slab_cleanup(void)
 623 {
 624         if (lnet_msg_cachep) {
 625                 kmem_cache_destroy(lnet_msg_cachep);
 626                 lnet_msg_cachep = NULL;
 627         }
 628
 629         if (lnet_rspt_cachep) {
 630                 kmem_cache_destroy(lnet_rspt_cachep);
 631                 lnet_rspt_cachep = NULL;
 632         }
 633
 634         if (lnet_udsp_cachep) {
 635                 kmem_cache_destroy(lnet_udsp_cachep);
 636                 lnet_udsp_cachep = NULL;
 637         }
 638
 639         if (lnet_small_mds_cachep) {
 640                 kmem_cache_destroy(lnet_small_mds_cachep);
 641                 lnet_small_mds_cachep = NULL;
 642         }
 643
 644         if (lnet_mes_cachep) {
 645                 kmem_cache_destroy(lnet_mes_cachep);
 646                 lnet_mes_cachep = NULL;
 647         }
 648 }
 649
 650 static int
 651 lnet_create_remote_nets_table(void)
 652 {
 653         int               i;
 654         struct list_head *hash;
 655
 656         LASSERT(the_lnet.ln_remote_nets_hash == NULL);
 657         LASSERT(the_lnet.ln_remote_nets_hbits > 0);
 658         CFS_ALLOC_PTR_ARRAY(hash, LNET_REMOTE_NETS_HASH_SIZE);
 659         if (hash == NULL) {
 660                 CERROR("Failed to create remote nets hash table\n");
 661                 return -ENOMEM;
 662         }
 663
 664         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 665                 INIT_LIST_HEAD(&hash[i]);
 666         the_lnet.ln_remote_nets_hash = hash;
 667         return 0;
 668 }
 669
 670 static void
 671 lnet_destroy_remote_nets_table(void)
 672 {
 673         int i;
 674
 675         if (the_lnet.ln_remote_nets_hash == NULL)
 676                 return;
 677
 678         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 679                 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
 680
 681         CFS_FREE_PTR_ARRAY(the_lnet.ln_remote_nets_hash,
 682                            LNET_REMOTE_NETS_HASH_SIZE);
 683         the_lnet.ln_remote_nets_hash = NULL;
 684 }
 685
 686 static void
 687 lnet_destroy_locks(void)
 688 {
 689         if (the_lnet.ln_res_lock != NULL) {
 690                 cfs_percpt_lock_free(the_lnet.ln_res_lock);
 691                 the_lnet.ln_res_lock = NULL;
 692         }
 693
 694         if (the_lnet.ln_net_lock != NULL) {
 695                 cfs_percpt_lock_free(the_lnet.ln_net_lock);
 696                 the_lnet.ln_net_lock = NULL;
 697         }
 698 }
 699
 700 static int
 701 lnet_create_locks(void)
 702 {
 703         lnet_init_locks();
 704
 705         the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 706         if (the_lnet.ln_res_lock == NULL)
 707                 goto failed;
 708
 709         the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 710         if (the_lnet.ln_net_lock == NULL)
 711                 goto failed;
 712
 713         return 0;
 714
 715  failed:
 716         lnet_destroy_locks();
 717         return -ENOMEM;
 718 }
 719
 720 static void lnet_assert_wire_constants(void)
 721 {
 722         /* Wire protocol assertions generated by 'wirecheck'
 723          * running on Linux robert.bartonsoftware.com 2.6.8-1.521
 724          * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
 725          * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7)
 726          */
 727
 728         /* Constants... */
 729         BUILD_BUG_ON(LNET_PROTO_TCP_MAGIC != 0xeebc0ded);
 730         BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MAJOR != 1);
 731         BUILD_BUG_ON(LNET_PROTO_TCP_VERSION_MINOR != 0);
 732         BUILD_BUG_ON(LNET_MSG_ACK != 0);
 733         BUILD_BUG_ON(LNET_MSG_PUT != 1);
 734         BUILD_BUG_ON(LNET_MSG_GET != 2);
 735         BUILD_BUG_ON(LNET_MSG_REPLY != 3);
 736         BUILD_BUG_ON(LNET_MSG_HELLO != 4);
 737
 738         BUILD_BUG_ON((int)sizeof(lnet_nid_t) != 8);
 739         BUILD_BUG_ON((int)sizeof(lnet_pid_t) != 4);
 740
 741         /* Checks for struct lnet_nid */
 742         BUILD_BUG_ON((int)sizeof(struct lnet_nid) != 20);
 743         BUILD_BUG_ON((int)offsetof(struct lnet_nid, nid_size) != 0);
 744         BUILD_BUG_ON((int)sizeof(((struct lnet_nid *)0)->nid_size) != 1);
 745         BUILD_BUG_ON((int)offsetof(struct lnet_nid, nid_type) != 1);
 746         BUILD_BUG_ON((int)sizeof(((struct lnet_nid *)0)->nid_type) != 1);
 747         BUILD_BUG_ON((int)offsetof(struct lnet_nid, nid_num) != 2);
 748         BUILD_BUG_ON((int)sizeof(((struct lnet_nid *)0)->nid_num) != 2);
 749         BUILD_BUG_ON((int)offsetof(struct lnet_nid, nid_addr) != 4);
 750         BUILD_BUG_ON((int)sizeof(((struct lnet_nid *)0)->nid_addr) != 16);
 751
 752         /* Checks for struct lnet_process_id_packed */
 753         BUILD_BUG_ON((int)sizeof(struct lnet_process_id_packed) != 12);
 754         BUILD_BUG_ON((int)offsetof(struct lnet_process_id_packed, nid) != 0);
 755         BUILD_BUG_ON((int)sizeof(((struct lnet_process_id_packed *)0)->nid) != 8);
 756         BUILD_BUG_ON((int)offsetof(struct lnet_process_id_packed, pid) != 8);
 757         BUILD_BUG_ON((int)sizeof(((struct lnet_process_id_packed *)0)->pid) != 4);
 758
 759         /* Checks for struct lnet_handle_wire */
 760         BUILD_BUG_ON((int)sizeof(struct lnet_handle_wire) != 16);
 761         BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
 762                                    wh_interface_cookie) != 0);
 763         BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) != 8);
 764         BUILD_BUG_ON((int)offsetof(struct lnet_handle_wire,
 765                                    wh_object_cookie) != 8);
 766         BUILD_BUG_ON((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) != 8);
 767
 768         /* Checks for struct struct lnet_magicversion */
 769         BUILD_BUG_ON((int)sizeof(struct lnet_magicversion) != 8);
 770         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, magic) != 0);
 771         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->magic) != 4);
 772         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion, version_major) != 4);
 773         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_major) != 2);
 774         BUILD_BUG_ON((int)offsetof(struct lnet_magicversion,
 775                                    version_minor) != 6);
 776         BUILD_BUG_ON((int)sizeof(((struct lnet_magicversion *)0)->version_minor) != 2);
 777
 778         /* Checks for struct struct lnet_hdr */
 779         BUILD_BUG_ON((int)sizeof(struct lnet_hdr) != 72);
 780         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_nid) != 0);
 781         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_nid) != 8);
 782         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_nid) != 8);
 783         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_nid) != 8);
 784         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, dest_pid) != 16);
 785         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->dest_pid) != 4);
 786         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, src_pid) != 20);
 787         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->src_pid) != 4);
 788         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, type) != 24);
 789         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->type) != 4);
 790         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, payload_length) != 28);
 791         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->payload_length) != 4);
 792         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg) != 32);
 793         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg) != 40);
 794
 795         /* Ack */
 796         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.dst_wmd) != 32);
 797         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.dst_wmd) != 16);
 798         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.match_bits) != 48);
 799         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.match_bits) != 8);
 800         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.ack.mlength) != 56);
 801         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.ack.mlength) != 4);
 802
 803         /* Put */
 804         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ack_wmd) != 32);
 805         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ack_wmd) != 16);
 806         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.match_bits) != 48);
 807         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.match_bits) != 8);
 808         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.hdr_data) != 56);
 809         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.hdr_data) != 8);
 810         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.ptl_index) != 64);
 811         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.ptl_index) != 4);
 812         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.put.offset) != 68);
 813         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.put.offset) != 4);
 814
 815         /* Get */
 816         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.return_wmd) != 32);
 817         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.return_wmd) != 16);
 818         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.match_bits) != 48);
 819         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.match_bits) != 8);
 820         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.ptl_index) != 56);
 821         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.ptl_index) != 4);
 822         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.src_offset) != 60);
 823         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.src_offset) != 4);
 824         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.get.sink_length) != 64);
 825         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.get.sink_length) != 4);
 826
 827         /* Reply */
 828         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.reply.dst_wmd) != 32);
 829         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.reply.dst_wmd) != 16);
 830
 831         /* Hello */
 832         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.incarnation) != 32);
 833         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.incarnation) != 8);
 834         BUILD_BUG_ON((int)offsetof(struct lnet_hdr, msg.hello.type) != 40);
 835         BUILD_BUG_ON((int)sizeof(((struct lnet_hdr *)0)->msg.hello.type) != 4);
 836
 837         /* Checks for struct lnet_ni_status and related constants */
 838         BUILD_BUG_ON(LNET_NI_STATUS_INVALID != 0x00000000);
 839         BUILD_BUG_ON(LNET_NI_STATUS_UP != 0x15aac0de);
 840         BUILD_BUG_ON(LNET_NI_STATUS_DOWN != 0xdeadface);
 841
 842         /* Checks for struct lnet_ni_status */
 843         BUILD_BUG_ON((int)sizeof(struct lnet_ni_status) != 16);
 844         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_nid) != 0);
 845         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_nid) != 8);
 846         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_status) != 8);
 847         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_status) != 4);
 848         BUILD_BUG_ON((int)offsetof(struct lnet_ni_status, ns_unused) != 12);
 849         BUILD_BUG_ON((int)sizeof(((struct lnet_ni_status *)0)->ns_unused) != 4);
 850
 851         /* Checks for struct lnet_ping_info and related constants */
 852         BUILD_BUG_ON(LNET_PROTO_PING_MAGIC != 0x70696E67);
 853         BUILD_BUG_ON(LNET_PING_FEAT_INVAL != 0);
 854         BUILD_BUG_ON(LNET_PING_FEAT_BASE != 1);
 855         BUILD_BUG_ON(LNET_PING_FEAT_NI_STATUS != 2);
 856         BUILD_BUG_ON(LNET_PING_FEAT_RTE_DISABLED != 4);
 857         BUILD_BUG_ON(LNET_PING_FEAT_MULTI_RAIL != 8);
 858         BUILD_BUG_ON(LNET_PING_FEAT_DISCOVERY != 16);
 859         BUILD_BUG_ON(LNET_PING_FEAT_BITS != 31);
 860
 861         /* Checks for struct lnet_ping_info */
 862         BUILD_BUG_ON((int)sizeof(struct lnet_ping_info) != 16);
 863         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_magic) != 0);
 864         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_magic) != 4);
 865         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_features) != 4);
 866         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_features) != 4);
 867         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_pid) != 8);
 868         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_pid) != 4);
 869         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_nnis) != 12);
 870         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_nnis) != 4);
 871         BUILD_BUG_ON((int)offsetof(struct lnet_ping_info, pi_ni) != 16);
 872         BUILD_BUG_ON((int)sizeof(((struct lnet_ping_info *)0)->pi_ni) != 0);
 873
 874         /* Acceptor connection request */
 875         BUILD_BUG_ON(LNET_PROTO_ACCEPTOR_VERSION != 1);
 876
 877         /* Checks for struct lnet_acceptor_connreq */
 878         BUILD_BUG_ON((int)sizeof(struct lnet_acceptor_connreq) != 16);
 879         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_magic) != 0);
 880         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_magic) != 4);
 881         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_version) != 4);
 882         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_version) != 4);
 883         BUILD_BUG_ON((int)offsetof(struct lnet_acceptor_connreq, acr_nid) != 8);
 884         BUILD_BUG_ON((int)sizeof(((struct lnet_acceptor_connreq *)0)->acr_nid) != 8);
 885
 886         /* Checks for struct lnet_counters_common */
 887         BUILD_BUG_ON((int)sizeof(struct lnet_counters_common) != 60);
 888         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_msgs_alloc) != 0);
 889         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_msgs_alloc) != 4);
 890         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_msgs_max) != 4);
 891         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_msgs_max) != 4);
 892         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_errors) != 8);
 893         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_errors) != 4);
 894         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_send_count) != 12);
 895         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_send_count) != 4);
 896         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_recv_count) != 16);
 897         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_recv_count) != 4);
 898         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_route_count) != 20);
 899         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_route_count) != 4);
 900         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_drop_count) != 24);
 901         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_drop_count) != 4);
 902         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_send_length) != 28);
 903         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_send_length) != 8);
 904         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_recv_length) != 36);
 905         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_recv_length) != 8);
 906         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_route_length) != 44);
 907         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_route_length) != 8);
 908         BUILD_BUG_ON((int)offsetof(struct lnet_counters_common, lcc_drop_length) != 52);
 909         BUILD_BUG_ON((int)sizeof(((struct lnet_counters_common *)0)->lcc_drop_length) != 8);
 910 }
 911
 912 static const struct lnet_lnd *lnet_find_lnd_by_type(__u32 type)
 913 {
 914         const struct lnet_lnd *lnd;
 915
 916         /* holding lnd mutex */
 917         if (type >= NUM_LNDS)
 918                 return NULL;
 919         lnd = the_lnet.ln_lnds[type];
 920         LASSERT(!lnd || lnd->lnd_type == type);
 921
 922         return lnd;
 923 }
 924
 925 unsigned int
 926 lnet_get_lnd_timeout(void)
 927 {
 928         return lnet_lnd_timeout;
 929 }
 930 EXPORT_SYMBOL(lnet_get_lnd_timeout);
 931
 932 void
 933 lnet_register_lnd(const struct lnet_lnd *lnd)
 934 {
 935         mutex_lock(&the_lnet.ln_lnd_mutex);
 936
 937         LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
 938         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
 939
 940         the_lnet.ln_lnds[lnd->lnd_type] = lnd;
 941
 942         CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
 943
 944         mutex_unlock(&the_lnet.ln_lnd_mutex);
 945 }
 946 EXPORT_SYMBOL(lnet_register_lnd);
 947
 948 void
 949 lnet_unregister_lnd(const struct lnet_lnd *lnd)
 950 {
 951         mutex_lock(&the_lnet.ln_lnd_mutex);
 952
 953         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
 954
 955         the_lnet.ln_lnds[lnd->lnd_type] = NULL;
 956         CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
 957
 958         mutex_unlock(&the_lnet.ln_lnd_mutex);
 959 }
 960 EXPORT_SYMBOL(lnet_unregister_lnd);
 961
 962 static void
 963 lnet_counters_get_common_locked(struct lnet_counters_common *common)
 964 {
 965         struct lnet_counters *ctr;
 966         int i;
 967
 968         /* FIXME !!! Their is no assert_lnet_net_locked() to ensure this
 969          * actually called under the protection of the lnet_net_lock.
 970          */
 971         memset(common, 0, sizeof(*common));
 972
 973         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
 974                 common->lcc_msgs_max     += ctr->lct_common.lcc_msgs_max;
 975                 common->lcc_msgs_alloc   += ctr->lct_common.lcc_msgs_alloc;
 976                 common->lcc_errors       += ctr->lct_common.lcc_errors;
 977                 common->lcc_send_count   += ctr->lct_common.lcc_send_count;
 978                 common->lcc_recv_count   += ctr->lct_common.lcc_recv_count;
 979                 common->lcc_route_count  += ctr->lct_common.lcc_route_count;
 980                 common->lcc_drop_count   += ctr->lct_common.lcc_drop_count;
 981                 common->lcc_send_length  += ctr->lct_common.lcc_send_length;
 982                 common->lcc_recv_length  += ctr->lct_common.lcc_recv_length;
 983                 common->lcc_route_length += ctr->lct_common.lcc_route_length;
 984                 common->lcc_drop_length  += ctr->lct_common.lcc_drop_length;
 985         }
 986 }
 987
 988 void
 989 lnet_counters_get_common(struct lnet_counters_common *common)
 990 {
 991         lnet_net_lock(LNET_LOCK_EX);
 992         lnet_counters_get_common_locked(common);
 993         lnet_net_unlock(LNET_LOCK_EX);
 994 }
 995 EXPORT_SYMBOL(lnet_counters_get_common);
 996
 997 int
 998 lnet_counters_get(struct lnet_counters *counters)
 999 {
1000         struct lnet_counters *ctr;
1001         struct lnet_counters_health *health = &counters->lct_health;
1002         int i, rc = 0;
1003
1004         memset(counters, 0, sizeof(*counters));
1005
1006         lnet_net_lock(LNET_LOCK_EX);
1007
1008         if (the_lnet.ln_state != LNET_STATE_RUNNING)
1009                 GOTO(out_unlock, rc = -ENODEV);
1010
1011         lnet_counters_get_common_locked(&counters->lct_common);
1012
1013         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
1014                 health->lch_rst_alloc    += ctr->lct_health.lch_rst_alloc;
1015                 health->lch_resend_count += ctr->lct_health.lch_resend_count;
1016                 health->lch_response_timeout_count +=
1017                                 ctr->lct_health.lch_response_timeout_count;
1018                 health->lch_local_interrupt_count +=
1019                                 ctr->lct_health.lch_local_interrupt_count;
1020                 health->lch_local_dropped_count +=
1021                                 ctr->lct_health.lch_local_dropped_count;
1022                 health->lch_local_aborted_count +=
1023                                 ctr->lct_health.lch_local_aborted_count;
1024                 health->lch_local_no_route_count +=
1025                                 ctr->lct_health.lch_local_no_route_count;
1026                 health->lch_local_timeout_count +=
1027                                 ctr->lct_health.lch_local_timeout_count;
1028                 health->lch_local_error_count +=
1029                                 ctr->lct_health.lch_local_error_count;
1030                 health->lch_remote_dropped_count +=
1031                                 ctr->lct_health.lch_remote_dropped_count;
1032                 health->lch_remote_error_count +=
1033                                 ctr->lct_health.lch_remote_error_count;
1034                 health->lch_remote_timeout_count +=
1035                                 ctr->lct_health.lch_remote_timeout_count;
1036                 health->lch_network_timeout_count +=
1037                                 ctr->lct_health.lch_network_timeout_count;
1038         }
1039 out_unlock:
1040         lnet_net_unlock(LNET_LOCK_EX);
1041         return rc;
1042 }
1043 EXPORT_SYMBOL(lnet_counters_get);
1044
1045 void
1046 lnet_counters_reset(void)
1047 {
1048         struct lnet_counters *counters;
1049         int             i;
1050
1051         lnet_net_lock(LNET_LOCK_EX);
1052
1053         if (the_lnet.ln_state != LNET_STATE_RUNNING)
1054                 goto avoid_reset;
1055
1056         cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
1057                 memset(counters, 0, sizeof(struct lnet_counters));
1058 avoid_reset:
1059         lnet_net_unlock(LNET_LOCK_EX);
1060 }
1061
1062 static char *
1063 lnet_res_type2str(int type)
1064 {
1065         switch (type) {
1066         default:
1067                 LBUG();
1068         case LNET_COOKIE_TYPE_MD:
1069                 return "MD";
1070         case LNET_COOKIE_TYPE_ME:
1071                 return "ME";
1072         case LNET_COOKIE_TYPE_EQ:
1073                 return "EQ";
1074         }
1075 }
1076
1077 static void
1078 lnet_res_container_cleanup(struct lnet_res_container *rec)
1079 {
1080         int     count = 0;
1081
1082         if (rec->rec_type == 0) /* not set yet, it's uninitialized */
1083                 return;
1084
1085         while (!list_empty(&rec->rec_active)) {
1086                 struct list_head *e = rec->rec_active.next;
1087
1088                 list_del_init(e);
1089                 if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
1090                         lnet_md_free(list_entry(e, struct lnet_libmd, md_list));
1091
1092                 } else { /* NB: Active MEs should be attached on portals */
1093                         LBUG();
1094                 }
1095                 count++;
1096         }
1097
1098         if (count > 0) {
1099                 /* Found alive MD/ME/EQ, user really should unlink/free
1100                  * all of them before finalize LNet, but if someone didn't,
1101                  * we have to recycle garbage for him */
1102                 CERROR("%d active elements on exit of %s container\n",
1103                        count, lnet_res_type2str(rec->rec_type));
1104         }
1105
1106         if (rec->rec_lh_hash != NULL) {
1107                 CFS_FREE_PTR_ARRAY(rec->rec_lh_hash, LNET_LH_HASH_SIZE);
1108                 rec->rec_lh_hash = NULL;
1109         }
1110
1111         rec->rec_type = 0; /* mark it as finalized */
1112 }
1113
1114 static int
1115 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
1116 {
1117         int     rc = 0;
1118         int     i;
1119
1120         LASSERT(rec->rec_type == 0);
1121
1122         rec->rec_type = type;
1123         INIT_LIST_HEAD(&rec->rec_active);
1124
1125         rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
1126
1127         /* Arbitrary choice of hash table size */
1128         LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
1129                          LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
1130         if (rec->rec_lh_hash == NULL) {
1131                 rc = -ENOMEM;
1132                 goto out;
1133         }
1134
1135         for (i = 0; i < LNET_LH_HASH_SIZE; i++)
1136                 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
1137
1138         return 0;
1139
1140 out:
1141         CERROR("Failed to setup %s resource container\n",
1142                lnet_res_type2str(type));
1143         lnet_res_container_cleanup(rec);
1144         return rc;
1145 }
1146
1147 static void
1148 lnet_res_containers_destroy(struct lnet_res_container **recs)
1149 {
1150         struct lnet_res_container       *rec;
1151         int                             i;
1152
1153         cfs_percpt_for_each(rec, i, recs)
1154                 lnet_res_container_cleanup(rec);
1155
1156         cfs_percpt_free(recs);
1157 }
1158
1159 static struct lnet_res_container **
1160 lnet_res_containers_create(int type)
1161 {
1162         struct lnet_res_container       **recs;
1163         struct lnet_res_container       *rec;
1164         int                             rc;
1165         int                             i;
1166
1167         recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
1168         if (recs == NULL) {
1169                 CERROR("Failed to allocate %s resource containers\n",
1170                        lnet_res_type2str(type));
1171                 return NULL;
1172         }
1173
1174         cfs_percpt_for_each(rec, i, recs) {
1175                 rc = lnet_res_container_setup(rec, i, type);
1176                 if (rc != 0) {
1177                         lnet_res_containers_destroy(recs);
1178                         return NULL;
1179                 }
1180         }
1181
1182         return recs;
1183 }
1184
1185 struct lnet_libhandle *
1186 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
1187 {
1188         /* ALWAYS called with lnet_res_lock held */
1189         struct list_head        *head;
1190         struct lnet_libhandle   *lh;
1191         unsigned int            hash;
1192
1193         if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
1194                 return NULL;
1195
1196         hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
1197         head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
1198
1199         list_for_each_entry(lh, head, lh_hash_chain) {
1200                 if (lh->lh_cookie == cookie)
1201                         return lh;
1202         }
1203
1204         return NULL;
1205 }
1206
1207 void
1208 lnet_res_lh_initialize(struct lnet_res_container *rec,
1209                        struct lnet_libhandle *lh)
1210 {
1211         /* ALWAYS called with lnet_res_lock held */
1212         unsigned int    ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
1213         unsigned int    hash;
1214
1215         lh->lh_cookie = rec->rec_lh_cookie;
1216         rec->rec_lh_cookie += 1 << ibits;
1217
1218         hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
1219
1220         list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
1221 }
1222
1223 struct list_head **
1224 lnet_create_array_of_queues(void)
1225 {
1226         struct list_head **qs;
1227         struct list_head *q;
1228         int i;
1229
1230         qs = cfs_percpt_alloc(lnet_cpt_table(),
1231                               sizeof(struct list_head));
1232         if (!qs) {
1233                 CERROR("Failed to allocate queues\n");
1234                 return NULL;
1235         }
1236
1237         cfs_percpt_for_each(q, i, qs)
1238                 INIT_LIST_HEAD(q);
1239
1240         return qs;
1241 }
1242
1243 static int lnet_unprepare(void);
1244
1245 static int
1246 lnet_prepare(lnet_pid_t requested_pid)
1247 {
1248         /* Prepare to bring up the network */
1249         struct lnet_res_container **recs;
1250         int                       rc = 0;
1251
1252         if (requested_pid == LNET_PID_ANY) {
1253                 /* Don't instantiate LNET just for me */
1254                 return -ENETDOWN;
1255         }
1256
1257         LASSERT(the_lnet.ln_refcount == 0);
1258
1259         the_lnet.ln_routing = 0;
1260
1261         LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
1262         the_lnet.ln_pid = requested_pid;
1263
1264         INIT_LIST_HEAD(&the_lnet.ln_test_peers);
1265         INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
1266         INIT_LIST_HEAD(&the_lnet.ln_nets);
1267         INIT_LIST_HEAD(&the_lnet.ln_routers);
1268         INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
1269         INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
1270         INIT_LIST_HEAD(&the_lnet.ln_dc_request);
1271         INIT_LIST_HEAD(&the_lnet.ln_dc_working);
1272         INIT_LIST_HEAD(&the_lnet.ln_dc_expired);
1273         INIT_LIST_HEAD(&the_lnet.ln_mt_localNIRecovq);
1274         INIT_LIST_HEAD(&the_lnet.ln_mt_peerNIRecovq);
1275         INIT_LIST_HEAD(&the_lnet.ln_udsp_list);
1276         init_waitqueue_head(&the_lnet.ln_dc_waitq);
1277         the_lnet.ln_mt_handler = NULL;
1278         init_completion(&the_lnet.ln_started);
1279
1280         rc = lnet_slab_setup();
1281         if (rc != 0)
1282                 goto failed;
1283
1284         rc = lnet_create_remote_nets_table();
1285         if (rc != 0)
1286                 goto failed;
1287
1288         /*
1289          * NB the interface cookie in wire handles guards against delayed
1290          * replies and ACKs appearing valid after reboot.
1291          */
1292         the_lnet.ln_interface_cookie = ktime_get_real_ns();
1293
1294         the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
1295                                                 sizeof(struct lnet_counters));
1296         if (the_lnet.ln_counters == NULL) {
1297                 CERROR("Failed to allocate counters for LNet\n");
1298                 rc = -ENOMEM;
1299                 goto failed;
1300         }
1301
1302         rc = lnet_peer_tables_create();
1303         if (rc != 0)
1304                 goto failed;
1305
1306         rc = lnet_msg_containers_create();
1307         if (rc != 0)
1308                 goto failed;
1309
1310         rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
1311                                       LNET_COOKIE_TYPE_EQ);
1312         if (rc != 0)
1313                 goto failed;
1314
1315         recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
1316         if (recs == NULL) {
1317                 rc = -ENOMEM;
1318                 goto failed;
1319         }
1320
1321         the_lnet.ln_md_containers = recs;
1322
1323         rc = lnet_portals_create();
1324         if (rc != 0) {
1325                 CERROR("Failed to create portals for LNet: %d\n", rc);
1326                 goto failed;
1327         }
1328
1329         the_lnet.ln_mt_zombie_rstqs = lnet_create_array_of_queues();
1330         if (!the_lnet.ln_mt_zombie_rstqs) {
1331                 rc = -ENOMEM;
1332                 goto failed;
1333         }
1334
1335         return 0;
1336
1337  failed:
1338         lnet_unprepare();
1339         return rc;
1340 }
1341
1342 static int
1343 lnet_unprepare (void)
1344 {
1345         /* NB no LNET_LOCK since this is the last reference.  All LND instances
1346          * have shut down already, so it is safe to unlink and free all
1347          * descriptors, even those that appear committed to a network op (eg MD
1348          * with non-zero pending count) */
1349
1350         lnet_fail_nid(LNET_NID_ANY, 0);
1351
1352         LASSERT(the_lnet.ln_refcount == 0);
1353         LASSERT(list_empty(&the_lnet.ln_test_peers));
1354         LASSERT(list_empty(&the_lnet.ln_nets));
1355
1356         if (the_lnet.ln_mt_zombie_rstqs) {
1357                 lnet_clean_zombie_rstqs();
1358                 the_lnet.ln_mt_zombie_rstqs = NULL;
1359         }
1360
1361         lnet_assert_handler_unused(the_lnet.ln_mt_handler);
1362         the_lnet.ln_mt_handler = NULL;
1363
1364         lnet_portals_destroy();
1365
1366         if (the_lnet.ln_md_containers != NULL) {
1367                 lnet_res_containers_destroy(the_lnet.ln_md_containers);
1368                 the_lnet.ln_md_containers = NULL;
1369         }
1370
1371         lnet_res_container_cleanup(&the_lnet.ln_eq_container);
1372
1373         lnet_msg_containers_destroy();
1374         lnet_peer_uninit();
1375         lnet_rtrpools_free(0);
1376
1377         if (the_lnet.ln_counters != NULL) {
1378                 cfs_percpt_free(the_lnet.ln_counters);
1379                 the_lnet.ln_counters = NULL;
1380         }
1381         lnet_destroy_remote_nets_table();
1382         lnet_udsp_destroy(true);
1383         lnet_slab_cleanup();
1384
1385         return 0;
1386 }
1387
1388 struct lnet_ni  *
1389 lnet_net2ni_locked(__u32 net_id, int cpt)
1390 {
1391         struct lnet_ni   *ni;
1392         struct lnet_net  *net;
1393
1394         LASSERT(cpt != LNET_LOCK_EX);
1395
1396         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1397                 if (net->net_id == net_id) {
1398                         ni = list_entry(net->net_ni_list.next, struct lnet_ni,
1399                                         ni_netlist);
1400                         return ni;
1401                 }
1402         }
1403
1404         return NULL;
1405 }
1406
1407 struct lnet_ni *
1408 lnet_net2ni_addref(__u32 net)
1409 {
1410         struct lnet_ni *ni;
1411
1412         lnet_net_lock(0);
1413         ni = lnet_net2ni_locked(net, 0);
1414         if (ni)
1415                 lnet_ni_addref_locked(ni, 0);
1416         lnet_net_unlock(0);
1417
1418         return ni;
1419 }
1420 EXPORT_SYMBOL(lnet_net2ni_addref);
1421
1422 struct lnet_net *
1423 lnet_get_net_locked(__u32 net_id)
1424 {
1425         struct lnet_net  *net;
1426
1427         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1428                 if (net->net_id == net_id)
1429                         return net;
1430         }
1431
1432         return NULL;
1433 }
1434
1435 void
1436 lnet_net_clr_pref_rtrs(struct lnet_net *net)
1437 {
1438         struct list_head zombies;
1439         struct lnet_nid_list *ne;
1440         struct lnet_nid_list *tmp;
1441
1442         INIT_LIST_HEAD(&zombies);
1443
1444         lnet_net_lock(LNET_LOCK_EX);
1445         list_splice_init(&net->net_rtr_pref_nids, &zombies);
1446         lnet_net_unlock(LNET_LOCK_EX);
1447
1448         list_for_each_entry_safe(ne, tmp, &zombies, nl_list) {
1449                 list_del_init(&ne->nl_list);
1450                 LIBCFS_FREE(ne, sizeof(*ne));
1451         }
1452 }
1453
1454 int
1455 lnet_net_add_pref_rtr(struct lnet_net *net,
1456                       lnet_nid_t gw_nid)
1457 __must_hold(&the_lnet.ln_api_mutex)
1458 {
1459         struct lnet_nid_list *ne;
1460
1461         /* This function is called with api_mutex held. When the api_mutex
1462          * is held the list can not be modified, as it is only modified as
1463          * a result of applying a UDSP and that happens under api_mutex
1464          * lock.
1465          */
1466         list_for_each_entry(ne, &net->net_rtr_pref_nids, nl_list) {
1467                 if (ne->nl_nid == gw_nid)
1468                         return -EEXIST;
1469         }
1470
1471         LIBCFS_ALLOC(ne, sizeof(*ne));
1472         if (!ne)
1473                 return -ENOMEM;
1474
1475         ne->nl_nid = gw_nid;
1476
1477         /* Lock the cpt to protect against addition and checks in the
1478          * selection algorithm
1479          */
1480         lnet_net_lock(LNET_LOCK_EX);
1481         list_add(&ne->nl_list, &net->net_rtr_pref_nids);
1482         lnet_net_unlock(LNET_LOCK_EX);
1483
1484         return 0;
1485 }
1486
1487 bool
1488 lnet_net_is_pref_rtr_locked(struct lnet_net *net, lnet_nid_t rtr_nid)
1489 {
1490         struct lnet_nid_list *ne;
1491
1492         CDEBUG(D_NET, "%s: rtr pref emtpy: %d\n",
1493                libcfs_net2str(net->net_id),
1494                list_empty(&net->net_rtr_pref_nids));
1495
1496         if (list_empty(&net->net_rtr_pref_nids))
1497                 return false;
1498
1499         list_for_each_entry(ne, &net->net_rtr_pref_nids, nl_list) {
1500                 CDEBUG(D_NET, "Comparing pref %s with gw %s\n",
1501                        libcfs_nid2str(ne->nl_nid),
1502                        libcfs_nid2str(rtr_nid));
1503                 if (rtr_nid == ne->nl_nid)
1504                         return true;
1505         }
1506
1507         return false;
1508 }
1509
1510 unsigned int
1511 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
1512 {
1513         __u64           key = nid;
1514         unsigned int    val;
1515
1516         LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
1517
1518         if (number == 1)
1519                 return 0;
1520
1521         val = hash_long(key, LNET_CPT_BITS);
1522         /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
1523         if (val < number)
1524                 return val;
1525
1526         return (unsigned int)(key + val + (val >> 1)) % number;
1527 }
1528
1529 int
1530 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
1531 {
1532         struct lnet_net *net;
1533
1534         /* must called with hold of lnet_net_lock */
1535         if (LNET_CPT_NUMBER == 1)
1536                 return 0; /* the only one */
1537
1538         /*
1539          * If NI is provided then use the CPT identified in the NI cpt
1540          * list if one exists. If one doesn't exist, then that NI is
1541          * associated with all CPTs and it follows that the net it belongs
1542          * to is implicitly associated with all CPTs, so just hash the nid
1543          * and return that.
1544          */
1545         if (ni != NULL) {
1546                 if (ni->ni_cpts != NULL)
1547                         return ni->ni_cpts[lnet_nid_cpt_hash(nid,
1548                                                              ni->ni_ncpts)];
1549                 else
1550                         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1551         }
1552
1553         /* no NI provided so look at the net */
1554         net = lnet_get_net_locked(LNET_NIDNET(nid));
1555
1556         if (net != NULL && net->net_cpts != NULL) {
1557                 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
1558         }
1559
1560         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
1561 }
1562
1563 int
1564 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
1565 {
1566         int     cpt;
1567         int     cpt2;
1568
1569         if (LNET_CPT_NUMBER == 1)
1570                 return 0; /* the only one */
1571
1572         cpt = lnet_net_lock_current();
1573
1574         cpt2 = lnet_cpt_of_nid_locked(nid, ni);
1575
1576         lnet_net_unlock(cpt);
1577
1578         return cpt2;
1579 }
1580 EXPORT_SYMBOL(lnet_cpt_of_nid);
1581
1582 int
1583 lnet_islocalnet_locked(__u32 net_id)
1584 {
1585         struct lnet_net *net;
1586         bool local;
1587
1588         net = lnet_get_net_locked(net_id);
1589
1590         local = net != NULL;
1591
1592         return local;
1593 }
1594
1595 int
1596 lnet_islocalnet(__u32 net_id)
1597 {
1598         int cpt;
1599         bool local;
1600
1601         cpt = lnet_net_lock_current();
1602
1603         local = lnet_islocalnet_locked(net_id);
1604
1605         lnet_net_unlock(cpt);
1606
1607         return local;
1608 }
1609
1610 struct lnet_ni  *
1611 lnet_nid2ni_locked(lnet_nid_t nid4, int cpt)
1612 {
1613         struct lnet_net *net;
1614         struct lnet_ni *ni;
1615         struct lnet_nid nid;
1616
1617         LASSERT(cpt != LNET_LOCK_EX);
1618         lnet_nid4_to_nid(nid4, &nid);
1619
1620         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1621                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1622                         if (nid_same(&ni->ni_nid, &nid))
1623                                 return ni;
1624                 }
1625         }
1626
1627         return NULL;
1628 }
1629
1630 struct lnet_ni *
1631 lnet_nid2ni_addref(lnet_nid_t nid)
1632 {
1633         struct lnet_ni *ni;
1634
1635         lnet_net_lock(0);
1636         ni = lnet_nid2ni_locked(nid, 0);
1637         if (ni)
1638                 lnet_ni_addref_locked(ni, 0);
1639         lnet_net_unlock(0);
1640
1641         return ni;
1642 }
1643 EXPORT_SYMBOL(lnet_nid2ni_addref);
1644
1645 int
1646 lnet_islocalnid(lnet_nid_t nid)
1647 {
1648         struct lnet_ni  *ni;
1649         int             cpt;
1650
1651         cpt = lnet_net_lock_current();
1652         ni = lnet_nid2ni_locked(nid, cpt);
1653         lnet_net_unlock(cpt);
1654
1655         return ni != NULL;
1656 }
1657
1658 int
1659 lnet_count_acceptor_nets(void)
1660 {
1661         /* Return the # of NIs that need the acceptor. */
1662         int              count = 0;
1663         struct lnet_net  *net;
1664         int              cpt;
1665
1666         cpt = lnet_net_lock_current();
1667         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1668                 /* all socklnd type networks should have the acceptor
1669                  * thread started */
1670                 if (net->net_lnd->lnd_accept != NULL)
1671                         count++;
1672         }
1673
1674         lnet_net_unlock(cpt);
1675
1676         return count;
1677 }
1678
1679 struct lnet_ping_buffer *
1680 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1681 {
1682         struct lnet_ping_buffer *pbuf;
1683
1684         LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1685         if (pbuf) {
1686                 pbuf->pb_nnis = nnis;
1687                 pbuf->pb_needs_post = false;
1688                 atomic_set(&pbuf->pb_refcnt, 1);
1689         }
1690
1691         return pbuf;
1692 }
1693
1694 void
1695 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1696 {
1697         LASSERT(atomic_read(&pbuf->pb_refcnt) == 0);
1698         LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1699 }
1700
1701 static struct lnet_ping_buffer *
1702 lnet_ping_target_create(int nnis)
1703 {
1704         struct lnet_ping_buffer *pbuf;
1705
1706         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1707         if (pbuf == NULL) {
1708                 CERROR("Can't allocate ping source [%d]\n", nnis);
1709                 return NULL;
1710         }
1711
1712         pbuf->pb_info.pi_nnis = nnis;
1713         pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1714         pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1715         pbuf->pb_info.pi_features =
1716                 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1717
1718         return pbuf;
1719 }
1720
1721 static inline int
1722 lnet_get_net_ni_count_locked(struct lnet_net *net)
1723 {
1724         struct lnet_ni  *ni;
1725         int             count = 0;
1726
1727         list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1728                 count++;
1729
1730         return count;
1731 }
1732
1733 static inline int
1734 lnet_get_net_ni_count_pre(struct lnet_net *net)
1735 {
1736         struct lnet_ni  *ni;
1737         int             count = 0;
1738
1739         list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1740                 count++;
1741
1742         return count;
1743 }
1744
1745 static inline int
1746 lnet_get_ni_count(void)
1747 {
1748         struct lnet_ni  *ni;
1749         struct lnet_net *net;
1750         int             count = 0;
1751
1752         lnet_net_lock(0);
1753
1754         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1755                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1756                         count++;
1757         }
1758
1759         lnet_net_unlock(0);
1760
1761         return count;
1762 }
1763
1764 void
1765 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1766 {
1767         struct lnet_ni_status *stat;
1768         int nnis;
1769         int i;
1770
1771         __swab32s(&pbuf->pb_info.pi_magic);
1772         __swab32s(&pbuf->pb_info.pi_features);
1773         __swab32s(&pbuf->pb_info.pi_pid);
1774         __swab32s(&pbuf->pb_info.pi_nnis);
1775         nnis = pbuf->pb_info.pi_nnis;
1776         if (nnis > pbuf->pb_nnis)
1777                 nnis = pbuf->pb_nnis;
1778         for (i = 0; i < nnis; i++) {
1779                 stat = &pbuf->pb_info.pi_ni[i];
1780                 __swab64s(&stat->ns_nid);
1781                 __swab32s(&stat->ns_status);
1782         }
1783 }
1784
1785 int
1786 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1787 {
1788         if (!pinfo)
1789                 return -EINVAL;
1790         if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1791                 return -EPROTO;
1792         if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1793                 return -EPROTO;
1794         /* Loopback is guaranteed to be present */
1795         if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1796                 return -ERANGE;
1797         if (LNET_PING_INFO_LONI(pinfo) != LNET_NID_LO_0)
1798                 return -EPROTO;
1799         return 0;
1800 }
1801
1802 static void
1803 lnet_ping_target_destroy(void)
1804 {
1805         struct lnet_net *net;
1806         struct lnet_ni  *ni;
1807
1808         lnet_net_lock(LNET_LOCK_EX);
1809
1810         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1811                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1812                         lnet_ni_lock(ni);
1813                         ni->ni_status = NULL;
1814                         lnet_ni_unlock(ni);
1815                 }
1816         }
1817
1818         lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1819         the_lnet.ln_ping_target = NULL;
1820
1821         lnet_net_unlock(LNET_LOCK_EX);
1822 }
1823
1824 static void
1825 lnet_ping_target_event_handler(struct lnet_event *event)
1826 {
1827         struct lnet_ping_buffer *pbuf = event->md_user_ptr;
1828
1829         if (event->unlinked)
1830                 lnet_ping_buffer_decref(pbuf);
1831 }
1832
1833 static int
1834 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1835                        struct lnet_handle_md *ping_mdh,
1836                        int ni_count, bool set_eq)
1837 {
1838         struct lnet_process_id id = {
1839                 .nid = LNET_NID_ANY,
1840                 .pid = LNET_PID_ANY
1841         };
1842         struct lnet_me *me;
1843         struct lnet_md md = { NULL };
1844         int rc;
1845
1846         if (set_eq)
1847                 the_lnet.ln_ping_target_handler =
1848                         lnet_ping_target_event_handler;
1849
1850         *ppbuf = lnet_ping_target_create(ni_count);
1851         if (*ppbuf == NULL) {
1852                 rc = -ENOMEM;
1853                 goto fail_free_eq;
1854         }
1855
1856         /* Ping target ME/MD */
1857         me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1858                           LNET_PROTO_PING_MATCHBITS, 0,
1859                           LNET_UNLINK, LNET_INS_AFTER);
1860         if (IS_ERR(me)) {
1861                 rc = PTR_ERR(me);
1862                 CERROR("Can't create ping target ME: %d\n", rc);
1863                 goto fail_decref_ping_buffer;
1864         }
1865
1866         /* initialize md content */
1867         md.start     = &(*ppbuf)->pb_info;
1868         md.length    = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1869         md.threshold = LNET_MD_THRESH_INF;
1870         md.max_size  = 0;
1871         md.options   = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1872                        LNET_MD_MANAGE_REMOTE;
1873         md.handler   = the_lnet.ln_ping_target_handler;
1874         md.user_ptr  = *ppbuf;
1875
1876         rc = LNetMDAttach(me, &md, LNET_RETAIN, ping_mdh);
1877         if (rc != 0) {
1878                 CERROR("Can't attach ping target MD: %d\n", rc);
1879                 goto fail_decref_ping_buffer;
1880         }
1881         lnet_ping_buffer_addref(*ppbuf);
1882
1883         return 0;
1884
1885 fail_decref_ping_buffer:
1886         LASSERT(atomic_read(&(*ppbuf)->pb_refcnt) == 1);
1887         lnet_ping_buffer_decref(*ppbuf);
1888         *ppbuf = NULL;
1889 fail_free_eq:
1890         return rc;
1891 }
1892
1893 static void
1894 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1895                     struct lnet_handle_md *ping_mdh)
1896 {
1897         LNetMDUnlink(*ping_mdh);
1898         LNetInvalidateMDHandle(ping_mdh);
1899
1900         /* NB the MD could be busy; this just starts the unlink */
1901         wait_var_event_warning(&pbuf->pb_refcnt,
1902                                atomic_read(&pbuf->pb_refcnt) <= 1,
1903                                "Still waiting for ping data MD to unlink\n");
1904 }
1905
1906 static void
1907 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1908 {
1909         struct lnet_ni          *ni;
1910         struct lnet_net         *net;
1911         struct lnet_ni_status *ns;
1912         int                     i;
1913         int                     rc;
1914
1915         i = 0;
1916         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1917                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1918                         LASSERT(i < pbuf->pb_nnis);
1919
1920                         ns = &pbuf->pb_info.pi_ni[i];
1921
1922                         if (!nid_is_nid4(&ni->ni_nid))
1923                                 continue;
1924                         ns->ns_nid = lnet_nid_to_nid4(&ni->ni_nid);
1925
1926                         lnet_ni_lock(ni);
1927                         ns->ns_status = lnet_ni_get_status_locked(ni);
1928                         ni->ni_status = ns;
1929                         lnet_ni_unlock(ni);
1930
1931                         i++;
1932                 }
1933         }
1934         /*
1935          * We (ab)use the ns_status of the loopback interface to
1936          * transmit the sequence number. The first interface listed
1937          * must be the loopback interface.
1938          */
1939         rc = lnet_ping_info_validate(&pbuf->pb_info);
1940         if (rc) {
1941                 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1942                 LBUG();
1943         }
1944         LNET_PING_BUFFER_SEQNO(pbuf) =
1945                 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1946 }
1947
1948 static void
1949 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1950                         struct lnet_handle_md ping_mdh)
1951 {
1952         struct lnet_ping_buffer *old_pbuf = NULL;
1953         struct lnet_handle_md old_ping_md;
1954
1955         /* switch the NIs to point to the new ping info created */
1956         lnet_net_lock(LNET_LOCK_EX);
1957
1958         if (!the_lnet.ln_routing)
1959                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1960         if (!lnet_peer_discovery_disabled)
1961                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1962
1963         /* Ensure only known feature bits have been set. */
1964         LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1965         LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1966
1967         lnet_ping_target_install_locked(pbuf);
1968
1969         if (the_lnet.ln_ping_target) {
1970                 old_pbuf = the_lnet.ln_ping_target;
1971                 old_ping_md = the_lnet.ln_ping_target_md;
1972         }
1973         the_lnet.ln_ping_target_md = ping_mdh;
1974         the_lnet.ln_ping_target = pbuf;
1975
1976         lnet_net_unlock(LNET_LOCK_EX);
1977
1978         if (old_pbuf) {
1979                 /* unlink and free the old ping info */
1980                 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
1981                 lnet_ping_buffer_decref(old_pbuf);
1982         }
1983
1984         lnet_push_update_to_peers(0);
1985 }
1986
1987 static void
1988 lnet_ping_target_fini(void)
1989 {
1990         lnet_ping_md_unlink(the_lnet.ln_ping_target,
1991                             &the_lnet.ln_ping_target_md);
1992
1993         lnet_assert_handler_unused(the_lnet.ln_ping_target_handler);
1994         lnet_ping_target_destroy();
1995 }
1996
1997 /* Resize the push target. */
1998 int lnet_push_target_resize(void)
1999 {
2000         struct lnet_handle_md mdh;
2001         struct lnet_handle_md old_mdh;
2002         struct lnet_ping_buffer *pbuf;
2003         struct lnet_ping_buffer *old_pbuf;
2004         int nnis;
2005         int rc;
2006
2007 again:
2008         nnis = the_lnet.ln_push_target_nnis;
2009         if (nnis <= 0) {
2010                 CDEBUG(D_NET, "Invalid nnis %d\n", nnis);
2011                 return -EINVAL;
2012         }
2013
2014         /* NB: lnet_ping_buffer_alloc() sets pbuf refcount to 1. That ref is
2015          * dropped when we need to resize again (see "old_pbuf" below) or when
2016          * LNet is shutdown (see lnet_push_target_fini())
2017          */
2018         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
2019         if (!pbuf) {
2020                 CDEBUG(D_NET, "Can't allocate pbuf for nnis %d\n", nnis);
2021                 return -ENOMEM;
2022         }
2023
2024         rc = lnet_push_target_post(pbuf, &mdh);
2025         if (rc) {
2026                 CDEBUG(D_NET, "Failed to post push target: %d\n", rc);
2027                 lnet_ping_buffer_decref(pbuf);
2028                 return rc;
2029         }
2030
2031         lnet_net_lock(LNET_LOCK_EX);
2032         old_pbuf = the_lnet.ln_push_target;
2033         old_mdh = the_lnet.ln_push_target_md;
2034         the_lnet.ln_push_target = pbuf;
2035         the_lnet.ln_push_target_md = mdh;
2036         lnet_net_unlock(LNET_LOCK_EX);
2037
2038         if (old_pbuf) {
2039                 LNetMDUnlink(old_mdh);
2040                 /* Drop ref set by lnet_ping_buffer_alloc() */
2041                 lnet_ping_buffer_decref(old_pbuf);
2042         }
2043
2044         /* Received another push or reply that requires a larger buffer */
2045         if (nnis < the_lnet.ln_push_target_nnis)
2046                 goto again;
2047
2048         CDEBUG(D_NET, "nnis %d success\n", nnis);
2049         return 0;
2050 }
2051
2052 int lnet_push_target_post(struct lnet_ping_buffer *pbuf,
2053                           struct lnet_handle_md *mdhp)
2054 {
2055         struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
2056         struct lnet_md md = { NULL };
2057         struct lnet_me *me;
2058         int rc;
2059
2060         me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
2061                           LNET_PROTO_PING_MATCHBITS, 0,
2062                           LNET_UNLINK, LNET_INS_AFTER);
2063         if (IS_ERR(me)) {
2064                 rc = PTR_ERR(me);
2065                 CERROR("Can't create push target ME: %d\n", rc);
2066                 return rc;
2067         }
2068
2069         pbuf->pb_needs_post = false;
2070
2071         /* This reference is dropped by lnet_push_target_event_handler() */
2072         lnet_ping_buffer_addref(pbuf);
2073
2074         /* initialize md content */
2075         md.start     = &pbuf->pb_info;
2076         md.length    = LNET_PING_INFO_SIZE(pbuf->pb_nnis);
2077         md.threshold = 1;
2078         md.max_size  = 0;
2079         md.options   = LNET_MD_OP_PUT | LNET_MD_TRUNCATE;
2080         md.user_ptr  = pbuf;
2081         md.handler   = the_lnet.ln_push_target_handler;
2082
2083         rc = LNetMDAttach(me, &md, LNET_UNLINK, mdhp);
2084         if (rc) {
2085                 CERROR("Can't attach push MD: %d\n", rc);
2086                 lnet_ping_buffer_decref(pbuf);
2087                 pbuf->pb_needs_post = true;
2088                 return rc;
2089         }
2090
2091         CDEBUG(D_NET, "posted push target %p\n", pbuf);
2092
2093         return 0;
2094 }
2095
2096 static void lnet_push_target_event_handler(struct lnet_event *ev)
2097 {
2098         struct lnet_ping_buffer *pbuf = ev->md_user_ptr;
2099
2100         CDEBUG(D_NET, "type %d status %d unlinked %d\n", ev->type, ev->status,
2101                ev->unlinked);
2102
2103         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
2104                 lnet_swap_pinginfo(pbuf);
2105
2106         if (ev->type == LNET_EVENT_UNLINK) {
2107                 /* Drop ref added by lnet_push_target_post() */
2108                 lnet_ping_buffer_decref(pbuf);
2109                 return;
2110         }
2111
2112         lnet_peer_push_event(ev);
2113         if (ev->unlinked)
2114                 /* Drop ref added by lnet_push_target_post */
2115                 lnet_ping_buffer_decref(pbuf);
2116 }
2117
2118 /* Initialize the push target. */
2119 static int lnet_push_target_init(void)
2120 {
2121         int rc;
2122
2123         if (the_lnet.ln_push_target)
2124                 return -EALREADY;
2125
2126         the_lnet.ln_push_target_handler =
2127                 lnet_push_target_event_handler;
2128
2129         rc = LNetSetLazyPortal(LNET_RESERVED_PORTAL);
2130         LASSERT(rc == 0);
2131
2132         /* Start at the required minimum, we'll enlarge if required. */
2133         the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
2134
2135         rc = lnet_push_target_resize();
2136
2137         if (rc) {
2138                 LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2139                 the_lnet.ln_push_target_handler = NULL;
2140         }
2141
2142         return rc;
2143 }
2144
2145 /* Clean up the push target. */
2146 static void lnet_push_target_fini(void)
2147 {
2148         if (!the_lnet.ln_push_target)
2149                 return;
2150
2151         /* Unlink and invalidate to prevent new references. */
2152         LNetMDUnlink(the_lnet.ln_push_target_md);
2153         LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
2154
2155         /* Wait for the unlink to complete. */
2156         wait_var_event_warning(&the_lnet.ln_push_target->pb_refcnt,
2157                                atomic_read(&the_lnet.ln_push_target->pb_refcnt) <= 1,
2158                                "Still waiting for ping data MD to unlink\n");
2159
2160         /* Drop ref set by lnet_ping_buffer_alloc() */
2161         lnet_ping_buffer_decref(the_lnet.ln_push_target);
2162         the_lnet.ln_push_target = NULL;
2163         the_lnet.ln_push_target_nnis = 0;
2164
2165         LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2166         lnet_assert_handler_unused(the_lnet.ln_push_target_handler);
2167         the_lnet.ln_push_target_handler = NULL;
2168 }
2169
2170 static int
2171 lnet_ni_tq_credits(struct lnet_ni *ni)
2172 {
2173         int     credits;
2174
2175         LASSERT(ni->ni_ncpts >= 1);
2176
2177         if (ni->ni_ncpts == 1)
2178                 return ni->ni_net->net_tunables.lct_max_tx_credits;
2179
2180         credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2181         credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2182         credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2183
2184         return credits;
2185 }
2186
2187 static void
2188 lnet_ni_unlink_locked(struct lnet_ni *ni)
2189 {
2190         /* move it to zombie list and nobody can find it anymore */
2191         LASSERT(!list_empty(&ni->ni_netlist));
2192         list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2193         lnet_ni_decref_locked(ni, 0);
2194 }
2195
2196 static void
2197 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2198 {
2199         int             i;
2200         int             islo;
2201         struct lnet_ni  *ni;
2202         struct list_head *zombie_list = &net->net_ni_zombie;
2203
2204         /*
2205          * Now wait for the NIs I just nuked to show up on the zombie
2206          * list and shut them down in guaranteed thread context
2207          */
2208         i = 2;
2209         while (!list_empty(zombie_list)) {
2210                 int     *ref;
2211                 int     j;
2212
2213                 ni = list_entry(zombie_list->next,
2214                                 struct lnet_ni, ni_netlist);
2215                 list_del_init(&ni->ni_netlist);
2216                 /* the ni should be in deleting state. If it's not it's
2217                  * a bug */
2218                 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2219                 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2220                         if (*ref == 0)
2221                                 continue;
2222                         /* still busy, add it back to zombie list */
2223                         list_add(&ni->ni_netlist, zombie_list);
2224                         break;
2225                 }
2226
2227                 if (!list_empty(&ni->ni_netlist)) {
2228                         /* Unlock mutex while waiting to allow other
2229                          * threads to read the LNet state and fall through
2230                          * to avoid deadlock
2231                          */
2232                         lnet_net_unlock(LNET_LOCK_EX);
2233                         mutex_unlock(&the_lnet.ln_api_mutex);
2234
2235                         ++i;
2236                         if ((i & (-i)) == i) {
2237                                 CDEBUG(D_WARNING,
2238                                        "Waiting for zombie LNI %s\n",
2239                                        libcfs_nidstr(&ni->ni_nid));
2240                         }
2241                         schedule_timeout_uninterruptible(cfs_time_seconds(1));
2242
2243                         mutex_lock(&the_lnet.ln_api_mutex);
2244                         lnet_net_lock(LNET_LOCK_EX);
2245                         continue;
2246                 }
2247
2248                 lnet_net_unlock(LNET_LOCK_EX);
2249
2250                 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2251
2252                 LASSERT(!in_interrupt());
2253                 /* Holding the mutex makes it safe for lnd_shutdown
2254                  * to call module_put(). Module unload cannot finish
2255                  * until lnet_unregister_lnd() completes, and that
2256                  * requires the mutex.
2257                  */
2258                 mutex_lock(&the_lnet.ln_lnd_mutex);
2259                 (net->net_lnd->lnd_shutdown)(ni);
2260                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2261
2262                 if (!islo)
2263                         CDEBUG(D_LNI, "Removed LNI %s\n",
2264                               libcfs_nidstr(&ni->ni_nid));
2265
2266                 lnet_ni_free(ni);
2267                 i = 2;
2268                 lnet_net_lock(LNET_LOCK_EX);
2269         }
2270 }
2271
2272 /* shutdown down the NI and release refcount */
2273 static void
2274 lnet_shutdown_lndni(struct lnet_ni *ni)
2275 {
2276         int i;
2277         struct lnet_net *net = ni->ni_net;
2278
2279         lnet_net_lock(LNET_LOCK_EX);
2280         lnet_ni_lock(ni);
2281         ni->ni_state = LNET_NI_STATE_DELETING;
2282         lnet_ni_unlock(ni);
2283         lnet_ni_unlink_locked(ni);
2284         lnet_incr_dlc_seq();
2285         lnet_net_unlock(LNET_LOCK_EX);
2286
2287         /* clear messages for this NI on the lazy portal */
2288         for (i = 0; i < the_lnet.ln_nportals; i++)
2289                 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2290
2291         lnet_net_lock(LNET_LOCK_EX);
2292         lnet_clear_zombies_nis_locked(net);
2293         lnet_net_unlock(LNET_LOCK_EX);
2294 }
2295
2296 static void
2297 lnet_shutdown_lndnet(struct lnet_net *net)
2298 {
2299         struct lnet_ni *ni;
2300
2301         lnet_net_lock(LNET_LOCK_EX);
2302
2303         list_del_init(&net->net_list);
2304
2305         while (!list_empty(&net->net_ni_list)) {
2306                 ni = list_entry(net->net_ni_list.next,
2307                                 struct lnet_ni, ni_netlist);
2308                 lnet_net_unlock(LNET_LOCK_EX);
2309                 lnet_shutdown_lndni(ni);
2310                 lnet_net_lock(LNET_LOCK_EX);
2311         }
2312
2313         lnet_net_unlock(LNET_LOCK_EX);
2314
2315         /* Do peer table cleanup for this net */
2316         lnet_peer_tables_cleanup(net);
2317
2318         lnet_net_free(net);
2319 }
2320
2321 static void
2322 lnet_shutdown_lndnets(void)
2323 {
2324         struct lnet_net *net;
2325         LIST_HEAD(resend);
2326         struct lnet_msg *msg, *tmp;
2327
2328         /* NB called holding the global mutex */
2329
2330         /* All quiet on the API front */
2331         LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2332         LASSERT(the_lnet.ln_refcount == 0);
2333
2334         lnet_net_lock(LNET_LOCK_EX);
2335         the_lnet.ln_state = LNET_STATE_STOPPING;
2336
2337         /*
2338          * move the nets to the zombie list to avoid them being
2339          * picked up for new work. LONET is also included in the
2340          * Nets that will be moved to the zombie list
2341          */
2342         list_splice_init(&the_lnet.ln_nets, &the_lnet.ln_net_zombie);
2343
2344         /* Drop the cached loopback Net. */
2345         if (the_lnet.ln_loni != NULL) {
2346                 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2347                 the_lnet.ln_loni = NULL;
2348         }
2349         lnet_net_unlock(LNET_LOCK_EX);
2350
2351         /* iterate through the net zombie list and delete each net */
2352         while (!list_empty(&the_lnet.ln_net_zombie)) {
2353                 net = list_entry(the_lnet.ln_net_zombie.next,
2354                                  struct lnet_net, net_list);
2355                 lnet_shutdown_lndnet(net);
2356         }
2357
2358         spin_lock(&the_lnet.ln_msg_resend_lock);
2359         list_splice(&the_lnet.ln_msg_resend, &resend);
2360         spin_unlock(&the_lnet.ln_msg_resend_lock);
2361
2362         list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2363                 list_del_init(&msg->msg_list);
2364                 msg->msg_no_resend = true;
2365                 lnet_finalize(msg, -ECANCELED);
2366         }
2367
2368         lnet_net_lock(LNET_LOCK_EX);
2369         the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2370         lnet_net_unlock(LNET_LOCK_EX);
2371 }
2372
2373 static int
2374 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2375 {
2376         int                     rc = -EINVAL;
2377         struct lnet_tx_queue    *tq;
2378         int                     i;
2379         struct lnet_net         *net = ni->ni_net;
2380
2381         mutex_lock(&the_lnet.ln_lnd_mutex);
2382
2383         if (tun) {
2384                 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2385                 ni->ni_lnd_tunables_set = true;
2386         }
2387
2388         rc = (net->net_lnd->lnd_startup)(ni);
2389
2390         mutex_unlock(&the_lnet.ln_lnd_mutex);
2391
2392         if (rc != 0) {
2393                 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2394                                    rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2395                 goto failed0;
2396         }
2397
2398         lnet_ni_lock(ni);
2399         ni->ni_state = LNET_NI_STATE_ACTIVE;
2400         lnet_ni_unlock(ni);
2401
2402         /* We keep a reference on the loopback net through the loopback NI */
2403         if (net->net_lnd->lnd_type == LOLND) {
2404                 lnet_ni_addref(ni);
2405                 LASSERT(the_lnet.ln_loni == NULL);
2406                 the_lnet.ln_loni = ni;
2407                 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2408                 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2409                 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2410                 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2411                 return 0;
2412         }
2413
2414         if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2415             ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2416                 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2417                                    libcfs_lnd2str(net->net_lnd->lnd_type),
2418                                    ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2419                                         "" : "per-peer ");
2420                 /* shutdown the NI since if we get here then it must've already
2421                  * been started
2422                  */
2423                 lnet_shutdown_lndni(ni);
2424                 return -EINVAL;
2425         }
2426
2427         cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2428                 tq->tq_credits_min =
2429                 tq->tq_credits_max =
2430                 tq->tq_credits = lnet_ni_tq_credits(ni);
2431         }
2432
2433         atomic_set(&ni->ni_tx_credits,
2434                    lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2435         atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2436
2437         CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2438                 libcfs_nidstr(&ni->ni_nid),
2439                 ni->ni_net->net_tunables.lct_peer_tx_credits,
2440                 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2441                 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2442                 ni->ni_net->net_tunables.lct_peer_timeout);
2443
2444         return 0;
2445 failed0:
2446         lnet_ni_free(ni);
2447         return rc;
2448 }
2449
2450 static int
2451 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2452 {
2453         struct lnet_ni *ni;
2454         struct lnet_net *net_l = NULL;
2455         LIST_HEAD(local_ni_list);
2456         int rc;
2457         int ni_count = 0;
2458         __u32 lnd_type;
2459         const struct lnet_lnd  *lnd;
2460         int peer_timeout =
2461                 net->net_tunables.lct_peer_timeout;
2462         int maxtxcredits =
2463                 net->net_tunables.lct_max_tx_credits;
2464         int peerrtrcredits =
2465                 net->net_tunables.lct_peer_rtr_credits;
2466
2467         /*
2468          * make sure that this net is unique. If it isn't then
2469          * we are adding interfaces to an already existing network, and
2470          * 'net' is just a convenient way to pass in the list.
2471          * if it is unique we need to find the LND and load it if
2472          * necessary.
2473          */
2474         if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2475                 lnd_type = LNET_NETTYP(net->net_id);
2476
2477                 mutex_lock(&the_lnet.ln_lnd_mutex);
2478                 lnd = lnet_find_lnd_by_type(lnd_type);
2479
2480                 if (lnd == NULL) {
2481                         mutex_unlock(&the_lnet.ln_lnd_mutex);
2482                         rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2483                         mutex_lock(&the_lnet.ln_lnd_mutex);
2484
2485                         lnd = lnet_find_lnd_by_type(lnd_type);
2486                         if (lnd == NULL) {
2487                                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2488                                 CERROR("Can't load LND %s, module %s, rc=%d\n",
2489                                 libcfs_lnd2str(lnd_type),
2490                                 libcfs_lnd2modname(lnd_type), rc);
2491 #ifndef HAVE_MODULE_LOADING_SUPPORT
2492                                 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2493                                                 "compiled with kernel module "
2494                                                 "loading support.");
2495 #endif
2496                                 rc = -EINVAL;
2497                                 goto failed0;
2498                         }
2499                 }
2500
2501                 net->net_lnd = lnd;
2502
2503                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2504
2505                 net_l = net;
2506         }
2507
2508         /*
2509          * net_l: if the network being added is unique then net_l
2510          *        will point to that network
2511          *        if the network being added is not unique then
2512          *        net_l points to the existing network.
2513          *
2514          * When we enter the loop below, we'll pick NIs off he
2515          * network beign added and start them up, then add them to
2516          * a local ni list. Once we've successfully started all
2517          * the NIs then we join the local NI list (of started up
2518          * networks) with the net_l->net_ni_list, which should
2519          * point to the correct network to add the new ni list to
2520          *
2521          * If any of the new NIs fail to start up, then we want to
2522          * iterate through the local ni list, which should include
2523          * any NIs which were successfully started up, and shut
2524          * them down.
2525          *
2526          * After than we want to delete the network being added,
2527          * to avoid a memory leak.
2528          */
2529         while (!list_empty(&net->net_ni_added)) {
2530                 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2531                                 ni_netlist);
2532                 list_del_init(&ni->ni_netlist);
2533
2534                 /* make sure that the the NI we're about to start
2535                  * up is actually unique. if it's not fail. */
2536                 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2537                                         ni->ni_interface)) {
2538                         rc = -EEXIST;
2539                         goto failed1;
2540                 }
2541
2542                 /* adjust the pointer the parent network, just in case it
2543                  * the net is a duplicate */
2544                 ni->ni_net = net_l;
2545
2546                 rc = lnet_startup_lndni(ni, tun);
2547
2548                 if (rc < 0)
2549                         goto failed1;
2550
2551                 lnet_ni_addref(ni);
2552                 list_add_tail(&ni->ni_netlist, &local_ni_list);
2553
2554                 ni_count++;
2555         }
2556
2557         lnet_net_lock(LNET_LOCK_EX);
2558         list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2559         lnet_incr_dlc_seq();
2560         lnet_net_unlock(LNET_LOCK_EX);
2561
2562         /* if the network is not unique then we don't want to keep
2563          * it around after we're done. Free it. Otherwise add that
2564          * net to the global the_lnet.ln_nets */
2565         if (net_l != net && net_l != NULL) {
2566                 /*
2567                  * TODO - note. currently the tunables can not be updated
2568                  * once added
2569                  */
2570                 lnet_net_free(net);
2571         } else {
2572                 /*
2573                  * restore tunables after it has been overwitten by the
2574                  * lnd
2575                  */
2576                 if (peer_timeout != -1)
2577                         net->net_tunables.lct_peer_timeout = peer_timeout;
2578                 if (maxtxcredits != -1)
2579                         net->net_tunables.lct_max_tx_credits = maxtxcredits;
2580                 if (peerrtrcredits != -1)
2581                         net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2582
2583                 lnet_net_lock(LNET_LOCK_EX);
2584                 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2585                 lnet_net_unlock(LNET_LOCK_EX);
2586         }
2587
2588         return ni_count;
2589
2590 failed1:
2591         /*
2592          * shutdown the new NIs that are being started up
2593          * free the NET being started
2594          */
2595         while (!list_empty(&local_ni_list)) {
2596                 ni = list_entry(local_ni_list.next, struct lnet_ni,
2597                                 ni_netlist);
2598
2599                 lnet_shutdown_lndni(ni);
2600         }
2601
2602 failed0:
2603         lnet_net_free(net);
2604
2605         return rc;
2606 }
2607
2608 static int
2609 lnet_startup_lndnets(struct list_head *netlist)
2610 {
2611         struct lnet_net         *net;
2612         int                     rc;
2613         int                     ni_count = 0;
2614
2615         /*
2616          * Change to running state before bringing up the LNDs. This
2617          * allows lnet_shutdown_lndnets() to assert that we've passed
2618          * through here.
2619          */
2620         lnet_net_lock(LNET_LOCK_EX);
2621         the_lnet.ln_state = LNET_STATE_RUNNING;
2622         lnet_net_unlock(LNET_LOCK_EX);
2623
2624         while (!list_empty(netlist)) {
2625                 net = list_entry(netlist->next, struct lnet_net, net_list);
2626                 list_del_init(&net->net_list);
2627
2628                 rc = lnet_startup_lndnet(net, NULL);
2629
2630                 if (rc < 0)
2631                         goto failed;
2632
2633                 ni_count += rc;
2634         }
2635
2636         return ni_count;
2637 failed:
2638         lnet_shutdown_lndnets();
2639
2640         return rc;
2641 }
2642
2643 static int lnet_genl_parse_list(struct sk_buff *msg,
2644                                 const struct ln_key_list *data[], u16 idx)
2645 {
2646         const struct ln_key_list *list = data[idx];
2647         const struct ln_key_props *props;
2648         struct nlattr *node;
2649         u16 count;
2650
2651         if (!list)
2652                 return 0;
2653
2654         if (!list->lkl_maxattr)
2655                 return -ERANGE;
2656
2657         props = list->lkl_list;
2658         if (!props)
2659                 return -EINVAL;
2660
2661         node = nla_nest_start(msg, LN_SCALAR_ATTR_LIST);
2662         if (!node)
2663                 return -ENOBUFS;
2664
2665         for (count = 1; count <= list->lkl_maxattr; count++) {
2666                 struct nlattr *key = nla_nest_start(msg, count);
2667
2668                 if (count == 1)
2669                         nla_put_u16(msg, LN_SCALAR_ATTR_LIST_SIZE,
2670                                     list->lkl_maxattr);
2671
2672                 nla_put_u16(msg, LN_SCALAR_ATTR_INDEX, count);
2673                 if (props[count].lkp_values)
2674                         nla_put_string(msg, LN_SCALAR_ATTR_VALUE,
2675                                        props[count].lkp_values);
2676                 if (props[count].lkp_key_format)
2677                         nla_put_u16(msg, LN_SCALAR_ATTR_KEY_FORMAT,
2678                                     props[count].lkp_key_format);
2679                 nla_put_u16(msg, LN_SCALAR_ATTR_NLA_TYPE,
2680                             props[count].lkp_data_type);
2681                 if (props[count].lkp_data_type == NLA_NESTED) {
2682                         int rc;
2683
2684                         rc = lnet_genl_parse_list(msg, data, ++idx);
2685                         if (rc < 0)
2686                                 return rc;
2687                 }
2688
2689                 nla_nest_end(msg, key);
2690         }
2691
2692         nla_nest_end(msg, node);
2693         return 0;
2694 }
2695
2696 int lnet_genl_send_scalar_list(struct sk_buff *msg, u32 portid, u32 seq,
2697                                const struct genl_family *family, int flags,
2698                                u8 cmd, const struct ln_key_list *data[])
2699 {
2700         int rc = 0;
2701         void *hdr;
2702
2703         if (!data[0])
2704                 return -EINVAL;
2705
2706         hdr = genlmsg_put(msg, portid, seq, family, flags, cmd);
2707         if (!hdr)
2708                 GOTO(canceled, rc = -EMSGSIZE);
2709
2710         rc = lnet_genl_parse_list(msg, data, 0);
2711         if (rc < 0)
2712                 GOTO(canceled, rc);
2713
2714         genlmsg_end(msg, hdr);
2715 canceled:
2716         if (rc < 0)
2717                 genlmsg_cancel(msg, hdr);
2718         return rc;
2719 }
2720 EXPORT_SYMBOL(lnet_genl_send_scalar_list);
2721
2722 /**
2723  * Initialize LNet library.
2724  *
2725  * Automatically called at module loading time. Caller has to call
2726  * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2727  * latter returned 0. It must be called exactly once.
2728  *
2729  * \retval 0 on success
2730  * \retval -ve on failures.
2731  */
2732 int lnet_lib_init(void)
2733 {
2734         int rc;
2735
2736         lnet_assert_wire_constants();
2737
2738         /* refer to global cfs_cpt_table for now */
2739         the_lnet.ln_cpt_table = cfs_cpt_tab;
2740         the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_tab);
2741
2742         LASSERT(the_lnet.ln_cpt_number > 0);
2743         if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2744                 /* we are under risk of consuming all lh_cookie */
2745                 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2746                        "please change setting of CPT-table and retry\n",
2747                        the_lnet.ln_cpt_number, LNET_CPT_MAX);
2748                 return -E2BIG;
2749         }
2750
2751         while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2752                 the_lnet.ln_cpt_bits++;
2753
2754         rc = lnet_create_locks();
2755         if (rc != 0) {
2756                 CERROR("Can't create LNet global locks: %d\n", rc);
2757                 return rc;
2758         }
2759
2760         the_lnet.ln_refcount = 0;
2761         INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2762         INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2763
2764         /* The hash table size is the number of bits it takes to express the set
2765          * ln_num_routes, minus 1 (better to under estimate than over so we
2766          * don't waste memory). */
2767         if (rnet_htable_size <= 0)
2768                 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2769         else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2770                 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2771         the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2772                                            order_base_2(rnet_htable_size) - 1);
2773
2774         /* All LNDs apart from the LOLND are in separate modules.  They
2775          * register themselves when their module loads, and unregister
2776          * themselves when their module is unloaded. */
2777         lnet_register_lnd(&the_lolnd);
2778         return 0;
2779 }
2780
2781 /**
2782  * Finalize LNet library.
2783  *
2784  * \pre lnet_lib_init() called with success.
2785  * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2786  *
2787  * As this happens at module-unload, all lnds must already be unloaded,
2788  * so they must already be unregistered.
2789  */
2790 void lnet_lib_exit(void)
2791 {
2792         int i;
2793
2794         LASSERT(the_lnet.ln_refcount == 0);
2795         lnet_unregister_lnd(&the_lolnd);
2796         for (i = 0; i < NUM_LNDS; i++)
2797                 LASSERT(!the_lnet.ln_lnds[i]);
2798         lnet_destroy_locks();
2799 }
2800
2801 /**
2802  * Set LNet PID and start LNet interfaces, routing, and forwarding.
2803  *
2804  * Users must call this function at least once before any other functions.
2805  * For each successful call there must be a corresponding call to
2806  * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2807  * ignored.
2808  *
2809  * The PID used by LNet may be different from the one requested.
2810  * See LNetGetId().
2811  *
2812  * \param requested_pid PID requested by the caller.
2813  *
2814  * \return >= 0 on success, and < 0 error code on failures.
2815  */
2816 int
2817 LNetNIInit(lnet_pid_t requested_pid)
2818 {
2819         int                     im_a_router = 0;
2820         int                     rc;
2821         int                     ni_count;
2822         struct lnet_ping_buffer *pbuf;
2823         struct lnet_handle_md   ping_mdh;
2824         LIST_HEAD(net_head);
2825         struct lnet_net         *net;
2826
2827         mutex_lock(&the_lnet.ln_api_mutex);
2828
2829         CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2830
2831         if (the_lnet.ln_refcount > 0) {
2832                 rc = the_lnet.ln_refcount++;
2833                 mutex_unlock(&the_lnet.ln_api_mutex);
2834                 return rc;
2835         }
2836
2837         rc = lnet_prepare(requested_pid);
2838         if (rc != 0) {
2839                 mutex_unlock(&the_lnet.ln_api_mutex);
2840                 return rc;
2841         }
2842
2843         /* create a network for Loopback network */
2844         net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2845         if (net == NULL) {
2846                 rc = -ENOMEM;
2847                 goto err_empty_list;
2848         }
2849
2850         /* Add in the loopback NI */
2851         if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2852                 rc = -ENOMEM;
2853                 goto err_empty_list;
2854         }
2855
2856         if (use_tcp_bonding)
2857                 CWARN("use_tcp_bonding has been removed. Use Multi-Rail and Dynamic Discovery instead, see LU-13641\n");
2858
2859         /* If LNet is being initialized via DLC it is possible
2860          * that the user requests not to load module parameters (ones which
2861          * are supported by DLC) on initialization.  Therefore, make sure not
2862          * to load networks, routes and forwarding from module parameters
2863          * in this case.  On cleanup in case of failure only clean up
2864          * routes if it has been loaded */
2865         if (!the_lnet.ln_nis_from_mod_params) {
2866                 rc = lnet_parse_networks(&net_head, lnet_get_networks());
2867                 if (rc < 0)
2868                         goto err_empty_list;
2869         }
2870
2871         ni_count = lnet_startup_lndnets(&net_head);
2872         if (ni_count < 0) {
2873                 rc = ni_count;
2874                 goto err_empty_list;
2875         }
2876
2877         if (!the_lnet.ln_nis_from_mod_params) {
2878                 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2879                 if (rc != 0)
2880                         goto err_shutdown_lndnis;
2881
2882                 rc = lnet_rtrpools_alloc(im_a_router);
2883                 if (rc != 0)
2884                         goto err_destroy_routes;
2885         }
2886
2887         rc = lnet_acceptor_start();
2888         if (rc != 0)
2889                 goto err_destroy_routes;
2890
2891         the_lnet.ln_refcount = 1;
2892         /* Now I may use my own API functions... */
2893
2894         rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2895         if (rc != 0)
2896                 goto err_acceptor_stop;
2897
2898         lnet_ping_target_update(pbuf, ping_mdh);
2899
2900         the_lnet.ln_mt_handler = lnet_mt_event_handler;
2901
2902         rc = lnet_push_target_init();
2903         if (rc != 0)
2904                 goto err_stop_ping;
2905
2906         rc = lnet_peer_discovery_start();
2907         if (rc != 0)
2908                 goto err_destroy_push_target;
2909
2910         rc = lnet_monitor_thr_start();
2911         if (rc != 0)
2912                 goto err_stop_discovery_thr;
2913
2914         lnet_fault_init();
2915         lnet_router_debugfs_init();
2916
2917         mutex_unlock(&the_lnet.ln_api_mutex);
2918
2919         complete_all(&the_lnet.ln_started);
2920
2921         /* wait for all routers to start */
2922         lnet_wait_router_start();
2923
2924         return 0;
2925
2926 err_stop_discovery_thr:
2927         lnet_peer_discovery_stop();
2928 err_destroy_push_target:
2929         lnet_push_target_fini();
2930 err_stop_ping:
2931         lnet_ping_target_fini();
2932 err_acceptor_stop:
2933         the_lnet.ln_refcount = 0;
2934         lnet_acceptor_stop();
2935 err_destroy_routes:
2936         if (!the_lnet.ln_nis_from_mod_params)
2937                 lnet_destroy_routes();
2938 err_shutdown_lndnis:
2939         lnet_shutdown_lndnets();
2940 err_empty_list:
2941         lnet_unprepare();
2942         LASSERT(rc < 0);
2943         mutex_unlock(&the_lnet.ln_api_mutex);
2944         while (!list_empty(&net_head)) {
2945                 struct lnet_net *net;
2946
2947                 net = list_entry(net_head.next, struct lnet_net, net_list);
2948                 list_del_init(&net->net_list);
2949                 lnet_net_free(net);
2950         }
2951         return rc;
2952 }
2953 EXPORT_SYMBOL(LNetNIInit);
2954
2955 /**
2956  * Stop LNet interfaces, routing, and forwarding.
2957  *
2958  * Users must call this function once for each successful call to LNetNIInit().
2959  * Once the LNetNIFini() operation has been started, the results of pending
2960  * API operations are undefined.
2961  *
2962  * \return always 0 for current implementation.
2963  */
2964 int
2965 LNetNIFini(void)
2966 {
2967         mutex_lock(&the_lnet.ln_api_mutex);
2968
2969         LASSERT(the_lnet.ln_refcount > 0);
2970
2971         if (the_lnet.ln_refcount != 1) {
2972                 the_lnet.ln_refcount--;
2973         } else {
2974                 LASSERT(!the_lnet.ln_niinit_self);
2975
2976                 lnet_fault_fini();
2977
2978                 lnet_router_debugfs_fini();
2979                 lnet_monitor_thr_stop();
2980                 lnet_peer_discovery_stop();
2981                 lnet_push_target_fini();
2982                 lnet_ping_target_fini();
2983
2984                 /* Teardown fns that use my own API functions BEFORE here */
2985                 the_lnet.ln_refcount = 0;
2986
2987                 lnet_acceptor_stop();
2988                 lnet_destroy_routes();
2989                 lnet_shutdown_lndnets();
2990                 lnet_unprepare();
2991         }
2992
2993         mutex_unlock(&the_lnet.ln_api_mutex);
2994         return 0;
2995 }
2996 EXPORT_SYMBOL(LNetNIFini);
2997
2998 /**
2999  * Grabs the ni data from the ni structure and fills the out
3000  * parameters
3001  *
3002  * \param[in] ni network        interface structure
3003  * \param[out] cfg_ni           NI config information
3004  * \param[out] tun              network and LND tunables
3005  */
3006 static void
3007 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
3008                    struct lnet_ioctl_config_lnd_tunables *tun,
3009                    struct lnet_ioctl_element_stats *stats,
3010                    __u32 tun_size)
3011 {
3012         size_t min_size = 0;
3013         int i;
3014
3015         if (!ni || !cfg_ni || !tun || !nid_is_nid4(&ni->ni_nid))
3016                 return;
3017
3018         if (ni->ni_interface != NULL) {
3019                 strncpy(cfg_ni->lic_ni_intf,
3020                         ni->ni_interface,
3021                         sizeof(cfg_ni->lic_ni_intf));
3022         }
3023
3024         cfg_ni->lic_nid = lnet_nid_to_nid4(&ni->ni_nid);
3025         cfg_ni->lic_status = lnet_ni_get_status_locked(ni);
3026         cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
3027
3028         memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
3029
3030         if (stats) {
3031                 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
3032                                                        LNET_STATS_TYPE_SEND);
3033                 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
3034                                                        LNET_STATS_TYPE_RECV);
3035                 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
3036                                                        LNET_STATS_TYPE_DROP);
3037         }
3038
3039         /*
3040          * tun->lt_tun will always be present, but in order to be
3041          * backwards compatible, we need to deal with the cases when
3042          * tun->lt_tun is smaller than what the kernel has, because it
3043          * comes from an older version of a userspace program, then we'll
3044          * need to copy as much information as we have available space.
3045          */
3046         min_size = tun_size - sizeof(tun->lt_cmn);
3047         memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
3048
3049         /* copy over the cpts */
3050         if (ni->ni_ncpts == LNET_CPT_NUMBER &&
3051             ni->ni_cpts == NULL)  {
3052                 for (i = 0; i < ni->ni_ncpts; i++)
3053                         cfg_ni->lic_cpts[i] = i;
3054         } else {
3055                 for (i = 0;
3056                      ni->ni_cpts != NULL && i < ni->ni_ncpts &&
3057                      i < LNET_MAX_SHOW_NUM_CPT;
3058                      i++)
3059                         cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
3060         }
3061         cfg_ni->lic_ncpts = ni->ni_ncpts;
3062 }
3063
3064 /**
3065  * NOTE: This is a legacy function left in the code to be backwards
3066  * compatible with older userspace programs. It should eventually be
3067  * removed.
3068  *
3069  * Grabs the ni data from the ni structure and fills the out
3070  * parameters
3071  *
3072  * \param[in] ni network        interface structure
3073  * \param[out] config           config information
3074  */
3075 static void
3076 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
3077                          struct lnet_ioctl_config_data *config)
3078 {
3079         struct lnet_ioctl_net_config *net_config;
3080         struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
3081         size_t min_size, tunable_size = 0;
3082         int i;
3083
3084         if (!ni || !config || !nid_is_nid4(&ni->ni_nid))
3085                 return;
3086
3087         net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
3088         if (!net_config)
3089                 return;
3090
3091         if (!ni->ni_interface)
3092                 return;
3093
3094         strncpy(net_config->ni_interface,
3095                 ni->ni_interface,
3096                 sizeof(net_config->ni_interface));
3097
3098         config->cfg_nid = lnet_nid_to_nid4(&ni->ni_nid);
3099         config->cfg_config_u.cfg_net.net_peer_timeout =
3100                 ni->ni_net->net_tunables.lct_peer_timeout;
3101         config->cfg_config_u.cfg_net.net_max_tx_credits =
3102                 ni->ni_net->net_tunables.lct_max_tx_credits;
3103         config->cfg_config_u.cfg_net.net_peer_tx_credits =
3104                 ni->ni_net->net_tunables.lct_peer_tx_credits;
3105         config->cfg_config_u.cfg_net.net_peer_rtr_credits =
3106                 ni->ni_net->net_tunables.lct_peer_rtr_credits;
3107
3108         net_config->ni_status = lnet_ni_get_status_locked(ni);
3109
3110         if (ni->ni_cpts) {
3111                 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
3112
3113                 for (i = 0; i < num_cpts; i++)
3114                         net_config->ni_cpts[i] = ni->ni_cpts[i];
3115
3116                 config->cfg_ncpts = num_cpts;
3117         }
3118
3119         /*
3120          * See if user land tools sent in a newer and larger version
3121          * of struct lnet_tunables than what the kernel uses.
3122          */
3123         min_size = sizeof(*config) + sizeof(*net_config);
3124
3125         if (config->cfg_hdr.ioc_len > min_size)
3126                 tunable_size = config->cfg_hdr.ioc_len - min_size;
3127
3128         /* Don't copy too much data to user space */
3129         min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
3130         lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
3131
3132         if (lnd_cfg && min_size) {
3133                 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
3134                 config->cfg_config_u.cfg_net.net_interface_count = 1;
3135
3136                 /* Tell user land that kernel side has less data */
3137                 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
3138                         min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
3139                         config->cfg_hdr.ioc_len -= min_size;
3140                 }
3141         }
3142 }
3143
3144 struct lnet_ni *
3145 lnet_get_ni_idx_locked(int idx)
3146 {
3147         struct lnet_ni          *ni;
3148         struct lnet_net         *net;
3149
3150         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3151                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3152                         if (idx-- == 0)
3153                                 return ni;
3154                 }
3155         }
3156
3157         return NULL;
3158 }
3159
3160 int lnet_get_net_healthv_locked(struct lnet_net *net)
3161 {
3162         struct lnet_ni *ni;
3163         int best_healthv = 0;
3164         int healthv, ni_fatal;
3165
3166         list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3167                 healthv = atomic_read(&ni->ni_healthv);
3168                 ni_fatal = atomic_read(&ni->ni_fatal_error_on);
3169                 if (!ni_fatal && healthv > best_healthv)
3170                         best_healthv = healthv;
3171         }
3172
3173         return best_healthv;
3174 }
3175
3176 struct lnet_ni *
3177 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
3178 {
3179         struct lnet_ni          *ni;
3180         struct lnet_net         *net = mynet;
3181
3182         /*
3183          * It is possible that the net has been cleaned out while there is
3184          * a message being sent. This function accessed the net without
3185          * checking if the list is empty
3186          */
3187         if (prev == NULL) {
3188                 if (net == NULL)
3189                         net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
3190                                         net_list);
3191                 if (list_empty(&net->net_ni_list))
3192                         return NULL;
3193                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3194                                 ni_netlist);
3195
3196                 return ni;
3197         }
3198
3199         if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
3200                 /* if you reached the end of the ni list and the net is
3201                  * specified, then there are no more nis in that net */
3202                 if (net != NULL)
3203                         return NULL;
3204
3205                 /* we reached the end of this net ni list. move to the
3206                  * next net */
3207                 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
3208                         /* no more nets and no more NIs. */
3209                         return NULL;
3210
3211                 /* get the next net */
3212                 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
3213                                  net_list);
3214                 if (list_empty(&net->net_ni_list))
3215                         return NULL;
3216                 /* get the ni on it */
3217                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3218                                 ni_netlist);
3219
3220                 return ni;
3221         }
3222
3223         if (list_empty(&prev->ni_netlist))
3224                 return NULL;
3225
3226         /* there are more nis left */
3227         ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
3228
3229         return ni;
3230 }
3231
3232 int
3233 lnet_get_net_config(struct lnet_ioctl_config_data *config)
3234 {
3235         struct lnet_ni *ni;
3236         int cpt;
3237         int rc = -ENOENT;
3238         int idx = config->cfg_count;
3239
3240         cpt = lnet_net_lock_current();
3241
3242         ni = lnet_get_ni_idx_locked(idx);
3243
3244         if (ni != NULL) {
3245                 rc = 0;
3246                 lnet_ni_lock(ni);
3247                 lnet_fill_ni_info_legacy(ni, config);
3248                 lnet_ni_unlock(ni);
3249         }
3250
3251         lnet_net_unlock(cpt);
3252         return rc;
3253 }
3254
3255 int
3256 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3257                    struct lnet_ioctl_config_lnd_tunables *tun,
3258                    struct lnet_ioctl_element_stats *stats,
3259                    __u32 tun_size)
3260 {
3261         struct lnet_ni          *ni;
3262         int                     cpt;
3263         int                     rc = -ENOENT;
3264
3265         if (!cfg_ni || !tun || !stats)
3266                 return -EINVAL;
3267
3268         cpt = lnet_net_lock_current();
3269
3270         ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3271
3272         if (ni) {
3273                 rc = 0;
3274                 lnet_ni_lock(ni);
3275                 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3276                 lnet_ni_unlock(ni);
3277         }
3278
3279         lnet_net_unlock(cpt);
3280         return rc;
3281 }
3282
3283 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3284 {
3285         struct lnet_ni *ni;
3286         int cpt;
3287         int rc = -ENOENT;
3288
3289         if (!msg_stats)
3290                 return -EINVAL;
3291
3292         cpt = lnet_net_lock_current();
3293
3294         ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3295
3296         if (ni) {
3297                 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3298                 rc = 0;
3299         }
3300
3301         lnet_net_unlock(cpt);
3302
3303         return rc;
3304 }
3305
3306 static int lnet_add_net_common(struct lnet_net *net,
3307                                struct lnet_ioctl_config_lnd_tunables *tun)
3308 {
3309         struct lnet_handle_md ping_mdh;
3310         struct lnet_ping_buffer *pbuf;
3311         struct lnet_remotenet *rnet;
3312         struct lnet_ni *ni;
3313         int net_ni_count;
3314         __u32 net_id;
3315         int rc;
3316
3317         lnet_net_lock(LNET_LOCK_EX);
3318         rnet = lnet_find_rnet_locked(net->net_id);
3319         lnet_net_unlock(LNET_LOCK_EX);
3320         /*
3321          * make sure that the net added doesn't invalidate the current
3322          * configuration LNet is keeping
3323          */
3324         if (rnet) {
3325                 CERROR("Adding net %s will invalidate routing configuration\n",
3326                        libcfs_net2str(net->net_id));
3327                 lnet_net_free(net);
3328                 return -EUSERS;
3329         }
3330
3331         /*
3332          * make sure you calculate the correct number of slots in the ping
3333          * buffer. Since the ping info is a flattened list of all the NIs,
3334          * we should allocate enough slots to accomodate the number of NIs
3335          * which will be added.
3336          *
3337          * since ni hasn't been configured yet, use
3338          * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3339          */
3340         net_ni_count = lnet_get_net_ni_count_pre(net);
3341
3342         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3343                                     net_ni_count + lnet_get_ni_count(),
3344                                     false);
3345         if (rc < 0) {
3346                 lnet_net_free(net);
3347                 return rc;
3348         }
3349
3350         if (tun)
3351                 memcpy(&net->net_tunables,
3352                        &tun->lt_cmn, sizeof(net->net_tunables));
3353         else
3354                 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3355
3356         net_id = net->net_id;
3357
3358         rc = lnet_startup_lndnet(net,
3359                                  (tun) ? &tun->lt_tun : NULL);
3360         if (rc < 0)
3361                 goto failed;
3362
3363         lnet_net_lock(LNET_LOCK_EX);
3364         net = lnet_get_net_locked(net_id);
3365         LASSERT(net);
3366
3367         /* apply the UDSPs */
3368         rc = lnet_udsp_apply_policies_on_net(net);
3369         if (rc)
3370                 CERROR("Failed to apply UDSPs on local net %s\n",
3371                        libcfs_net2str(net->net_id));
3372
3373         /* At this point we lost track of which NI was just added, so we
3374          * just re-apply the policies on all of the NIs on this net
3375          */
3376         list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3377                 rc = lnet_udsp_apply_policies_on_ni(ni);
3378                 if (rc)
3379                         CERROR("Failed to apply UDSPs on ni %s\n",
3380                                libcfs_nidstr(&ni->ni_nid));
3381         }
3382         lnet_net_unlock(LNET_LOCK_EX);
3383
3384         /*
3385          * Start the acceptor thread if this is the first network
3386          * being added that requires the thread.
3387          */
3388         if (net->net_lnd->lnd_accept) {
3389                 rc = lnet_acceptor_start();
3390                 if (rc < 0) {
3391                         /* shutdown the net that we just started */
3392                         CERROR("Failed to start up acceptor thread\n");
3393                         lnet_shutdown_lndnet(net);
3394                         goto failed;
3395                 }
3396         }
3397
3398         lnet_net_lock(LNET_LOCK_EX);
3399         lnet_peer_net_added(net);
3400         lnet_net_unlock(LNET_LOCK_EX);
3401
3402         lnet_ping_target_update(pbuf, ping_mdh);
3403
3404         return 0;
3405
3406 failed:
3407         lnet_ping_md_unlink(pbuf, &ping_mdh);
3408         lnet_ping_buffer_decref(pbuf);
3409         return rc;
3410 }
3411
3412 static void
3413 lnet_set_tune_defaults(struct lnet_ioctl_config_lnd_tunables *tun)
3414 {
3415         if (tun) {
3416                 if (!tun->lt_cmn.lct_peer_timeout)
3417                         tun->lt_cmn.lct_peer_timeout = DEFAULT_PEER_TIMEOUT;
3418                 if (!tun->lt_cmn.lct_peer_tx_credits)
3419                         tun->lt_cmn.lct_peer_tx_credits = DEFAULT_PEER_CREDITS;
3420                 if (!tun->lt_cmn.lct_max_tx_credits)
3421                         tun->lt_cmn.lct_max_tx_credits = DEFAULT_CREDITS;
3422         }
3423 }
3424
3425 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3426                                       struct lnet_ioctl_config_lnd_tunables *tun)
3427 {
3428         struct lnet_net *net;
3429         const char *nets;
3430         int rc;
3431         LIST_HEAD(net_head);
3432
3433         rc = lnet_parse_ip2nets(&nets, ip2nets);
3434         if (rc < 0)
3435                 return rc;
3436
3437         rc = lnet_parse_networks(&net_head, nets);
3438         if (rc < 0)
3439                 return rc;
3440
3441         lnet_set_tune_defaults(tun);
3442
3443         mutex_lock(&the_lnet.ln_api_mutex);
3444         while (!list_empty(&net_head)) {
3445                 net = list_entry(net_head.next, struct lnet_net, net_list);
3446                 list_del_init(&net->net_list);
3447                 rc = lnet_add_net_common(net, tun);
3448                 if (rc < 0)
3449                         goto out;
3450         }
3451
3452 out:
3453         mutex_unlock(&the_lnet.ln_api_mutex);
3454
3455         while (!list_empty(&net_head)) {
3456                 net = list_entry(net_head.next, struct lnet_net, net_list);
3457                 list_del_init(&net->net_list);
3458                 lnet_net_free(net);
3459         }
3460         return rc;
3461 }
3462
3463 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3464 {
3465         struct lnet_net *net;
3466         struct lnet_ni *ni;
3467         struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3468         int rc, i;
3469         __u32 net_id, lnd_type;
3470
3471         /* get the tunables if they are available */
3472         if (conf->lic_cfg_hdr.ioc_len >=
3473             sizeof(*conf) + sizeof(*tun))
3474                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3475                         conf->lic_bulk;
3476
3477         /* handle legacy ip2nets from DLC */
3478         if (conf->lic_legacy_ip2nets[0] != '\0')
3479                 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3480                                                   tun);
3481
3482         net_id = LNET_NIDNET(conf->lic_nid);
3483         lnd_type = LNET_NETTYP(net_id);
3484
3485         if (!libcfs_isknown_lnd(lnd_type)) {
3486                 CERROR("No valid net and lnd information provided\n");
3487                 return -EINVAL;
3488         }
3489
3490         net = lnet_net_alloc(net_id, NULL);
3491         if (!net)
3492                 return -ENOMEM;
3493
3494         for (i = 0; i < conf->lic_ncpts; i++) {
3495                 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3496                         return -EINVAL;
3497         }
3498
3499         ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3500                                        conf->lic_ni_intf);
3501         if (!ni)
3502                 return -ENOMEM;
3503
3504         lnet_set_tune_defaults(tun);
3505
3506         mutex_lock(&the_lnet.ln_api_mutex);
3507
3508         rc = lnet_add_net_common(net, tun);
3509
3510         mutex_unlock(&the_lnet.ln_api_mutex);
3511
3512         return rc;
3513 }
3514
3515 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3516 {
3517         struct lnet_net  *net;
3518         struct lnet_ni *ni;
3519         __u32 net_id = LNET_NIDNET(conf->lic_nid);
3520         struct lnet_ping_buffer *pbuf;
3521         struct lnet_handle_md  ping_mdh;
3522         int               rc;
3523         int               net_count;
3524         __u32             addr;
3525
3526         /* don't allow userspace to shutdown the LOLND */
3527         if (LNET_NETTYP(net_id) == LOLND)
3528                 return -EINVAL;
3529
3530         mutex_lock(&the_lnet.ln_api_mutex);
3531
3532         lnet_net_lock(0);
3533
3534         net = lnet_get_net_locked(net_id);
3535         if (!net) {
3536                 CERROR("net %s not found\n",
3537                        libcfs_net2str(net_id));
3538                 rc = -ENOENT;
3539                 goto unlock_net;
3540         }
3541
3542         addr = LNET_NIDADDR(conf->lic_nid);
3543         if (addr == 0) {
3544                 /* remove the entire net */
3545                 net_count = lnet_get_net_ni_count_locked(net);
3546
3547                 lnet_net_unlock(0);
3548
3549                 /* create and link a new ping info, before removing the old one */
3550                 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3551                                         lnet_get_ni_count() - net_count,
3552                                         false);
3553                 if (rc != 0)
3554                         goto unlock_api_mutex;
3555
3556                 lnet_shutdown_lndnet(net);
3557
3558                 lnet_acceptor_stop();
3559
3560                 lnet_ping_target_update(pbuf, ping_mdh);
3561
3562                 goto unlock_api_mutex;
3563         }
3564
3565         ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3566         if (!ni) {
3567                 CERROR("nid %s not found\n",
3568                        libcfs_nid2str(conf->lic_nid));
3569                 rc = -ENOENT;
3570                 goto unlock_net;
3571         }
3572
3573         net_count = lnet_get_net_ni_count_locked(net);
3574
3575         lnet_net_unlock(0);
3576
3577         /* create and link a new ping info, before removing the old one */
3578         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3579                                   lnet_get_ni_count() - 1, false);
3580         if (rc != 0)
3581                 goto unlock_api_mutex;
3582
3583         lnet_shutdown_lndni(ni);
3584
3585         lnet_acceptor_stop();
3586
3587         lnet_ping_target_update(pbuf, ping_mdh);
3588
3589         /* check if the net is empty and remove it if it is */
3590         if (net_count == 1)
3591                 lnet_shutdown_lndnet(net);
3592
3593         goto unlock_api_mutex;
3594
3595 unlock_net:
3596         lnet_net_unlock(0);
3597 unlock_api_mutex:
3598         mutex_unlock(&the_lnet.ln_api_mutex);
3599
3600         return rc;
3601 }
3602
3603 /*
3604  * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3605  * They are only expected to be called for unique networks.
3606  * That can be as a result of older DLC library
3607  * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3608  */
3609 int
3610 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3611 {
3612         struct lnet_net *net;
3613         LIST_HEAD(net_head);
3614         int rc;
3615         struct lnet_ioctl_config_lnd_tunables tun;
3616         const char *nets = conf->cfg_config_u.cfg_net.net_intf;
3617
3618         /* Create a net/ni structures for the network string */
3619         rc = lnet_parse_networks(&net_head, nets);
3620         if (rc <= 0)
3621                 return rc == 0 ? -EINVAL : rc;
3622
3623         mutex_lock(&the_lnet.ln_api_mutex);
3624
3625         if (rc > 1) {
3626                 rc = -EINVAL; /* only add one network per call */
3627                 goto out_unlock_clean;
3628         }
3629
3630         net = list_entry(net_head.next, struct lnet_net, net_list);
3631         list_del_init(&net->net_list);
3632
3633         LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3634
3635         memset(&tun, 0, sizeof(tun));
3636
3637         tun.lt_cmn.lct_peer_timeout =
3638           (!conf->cfg_config_u.cfg_net.net_peer_timeout) ? DEFAULT_PEER_TIMEOUT :
3639                 conf->cfg_config_u.cfg_net.net_peer_timeout;
3640         tun.lt_cmn.lct_peer_tx_credits =
3641           (!conf->cfg_config_u.cfg_net.net_peer_tx_credits) ? DEFAULT_PEER_CREDITS :
3642                 conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3643         tun.lt_cmn.lct_peer_rtr_credits =
3644           conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3645         tun.lt_cmn.lct_max_tx_credits =
3646           (!conf->cfg_config_u.cfg_net.net_max_tx_credits) ? DEFAULT_CREDITS :
3647                 conf->cfg_config_u.cfg_net.net_max_tx_credits;
3648
3649         rc = lnet_add_net_common(net, &tun);
3650
3651 out_unlock_clean:
3652         mutex_unlock(&the_lnet.ln_api_mutex);
3653         while (!list_empty(&net_head)) {
3654                 /* net_head list is empty in success case */
3655                 net = list_entry(net_head.next, struct lnet_net, net_list);
3656                 list_del_init(&net->net_list);
3657                 lnet_net_free(net);
3658         }
3659         return rc;
3660 }
3661
3662 int
3663 lnet_dyn_del_net(__u32 net_id)
3664 {
3665         struct lnet_net  *net;
3666         struct lnet_ping_buffer *pbuf;
3667         struct lnet_handle_md ping_mdh;
3668         int               rc;
3669         int               net_ni_count;
3670
3671         /* don't allow userspace to shutdown the LOLND */
3672         if (LNET_NETTYP(net_id) == LOLND)
3673                 return -EINVAL;
3674
3675         mutex_lock(&the_lnet.ln_api_mutex);
3676
3677         lnet_net_lock(0);
3678
3679         net = lnet_get_net_locked(net_id);
3680         if (net == NULL) {
3681                 lnet_net_unlock(0);
3682                 rc = -EINVAL;
3683                 goto out;
3684         }
3685
3686         net_ni_count = lnet_get_net_ni_count_locked(net);
3687
3688         lnet_net_unlock(0);
3689
3690         /* create and link a new ping info, before removing the old one */
3691         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3692                                     lnet_get_ni_count() - net_ni_count, false);
3693         if (rc != 0)
3694                 goto out;
3695
3696         lnet_shutdown_lndnet(net);
3697
3698         lnet_acceptor_stop();
3699
3700         lnet_ping_target_update(pbuf, ping_mdh);
3701
3702 out:
3703         mutex_unlock(&the_lnet.ln_api_mutex);
3704
3705         return rc;
3706 }
3707
3708 void lnet_incr_dlc_seq(void)
3709 {
3710         atomic_inc(&lnet_dlc_seq_no);
3711 }
3712
3713 __u32 lnet_get_dlc_seq_locked(void)
3714 {
3715         return atomic_read(&lnet_dlc_seq_no);
3716 }
3717
3718 static void
3719 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3720 {
3721         struct lnet_net *net;
3722         struct lnet_ni *ni;
3723
3724         lnet_net_lock(LNET_LOCK_EX);
3725         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3726                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3727                         if (all || (nid_is_nid4(&ni->ni_nid) &&
3728                                     lnet_nid_to_nid4(&ni->ni_nid) == nid)) {
3729                                 atomic_set(&ni->ni_healthv, value);
3730                                 if (list_empty(&ni->ni_recovery) &&
3731                                     value < LNET_MAX_HEALTH_VALUE) {
3732                                         CERROR("manually adding local NI %s to recovery\n",
3733                                                libcfs_nidstr(&ni->ni_nid));
3734                                         list_add_tail(&ni->ni_recovery,
3735                                                       &the_lnet.ln_mt_localNIRecovq);
3736                                         lnet_ni_addref_locked(ni, 0);
3737                                 }
3738                                 if (!all) {
3739                                         lnet_net_unlock(LNET_LOCK_EX);
3740                                         return;
3741                                 }
3742                         }
3743                 }
3744         }
3745         lnet_net_unlock(LNET_LOCK_EX);
3746 }
3747
3748 static void
3749 lnet_ni_set_conns_per_peer(lnet_nid_t nid, int value, bool all)
3750 {
3751         struct lnet_net *net;
3752         struct lnet_ni *ni;
3753
3754         lnet_net_lock(LNET_LOCK_EX);
3755         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3756                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3757                         if (lnet_nid_to_nid4(&ni->ni_nid) != nid && !all)
3758                                 continue;
3759                         if (LNET_NETTYP(net->net_id) == SOCKLND)
3760                                 ni->ni_lnd_tunables.lnd_tun_u.lnd_sock.lnd_conns_per_peer = value;
3761                         else if (LNET_NETTYP(net->net_id) == O2IBLND)
3762                                 ni->ni_lnd_tunables.lnd_tun_u.lnd_o2ib.lnd_conns_per_peer = value;
3763                         if (!all) {
3764                                 lnet_net_unlock(LNET_LOCK_EX);
3765                                 return;
3766                         }
3767                 }
3768         }
3769         lnet_net_unlock(LNET_LOCK_EX);
3770 }
3771
3772 static int
3773 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3774 {
3775         int cpt, rc = 0;
3776         struct lnet_ni *ni;
3777         lnet_nid_t nid = stats->hlni_nid;
3778
3779         cpt = lnet_net_lock_current();
3780         ni = lnet_nid2ni_locked(nid, cpt);
3781
3782         if (!ni) {
3783                 rc = -ENOENT;
3784                 goto unlock;
3785         }
3786
3787         stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3788         stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3789         stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3790         stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3791         stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3792         stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3793         stats->hlni_fatal_error = atomic_read(&ni->ni_fatal_error_on);
3794         stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3795         stats->hlni_ping_count = ni->ni_ping_count;
3796         stats->hlni_next_ping = ni->ni_next_ping;
3797
3798 unlock:
3799         lnet_net_unlock(cpt);
3800
3801         return rc;
3802 }
3803
3804 static int
3805 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3806 {
3807         struct lnet_ni *ni;
3808         int i = 0;
3809
3810         lnet_net_lock(LNET_LOCK_EX);
3811         list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3812                 if (!nid_is_nid4(&ni->ni_nid))
3813                         continue;
3814                 list->rlst_nid_array[i] = lnet_nid_to_nid4(&ni->ni_nid);
3815                 i++;
3816                 if (i >= LNET_MAX_SHOW_NUM_NID)
3817                         break;
3818         }
3819         lnet_net_unlock(LNET_LOCK_EX);
3820         list->rlst_num_nids = i;
3821
3822         return 0;
3823 }
3824
3825 static int
3826 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3827 {
3828         struct lnet_peer_ni *lpni;
3829         int i = 0;
3830
3831         lnet_net_lock(LNET_LOCK_EX);
3832         list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3833                 list->rlst_nid_array[i] = lpni->lpni_nid;
3834                 i++;
3835                 if (i >= LNET_MAX_SHOW_NUM_NID)
3836                         break;
3837         }
3838         lnet_net_unlock(LNET_LOCK_EX);
3839         list->rlst_num_nids = i;
3840
3841         return 0;
3842 }
3843
3844 /**
3845  * LNet ioctl handler.
3846  *
3847  */
3848 int
3849 LNetCtl(unsigned int cmd, void *arg)
3850 {
3851         struct libcfs_ioctl_data *data = arg;
3852         struct lnet_ioctl_config_data *config;
3853         struct lnet_process_id    id = {0};
3854         struct lnet_ni           *ni;
3855         int                       rc;
3856
3857         BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3858                      sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3859
3860         switch (cmd) {
3861         case IOC_LIBCFS_GET_NI:
3862                 rc = LNetGetId(data->ioc_count, &id);
3863                 data->ioc_nid = id.nid;
3864                 return rc;
3865
3866         case IOC_LIBCFS_FAIL_NID:
3867                 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3868
3869         case IOC_LIBCFS_ADD_ROUTE: {
3870                 /* default router sensitivity to 1 */
3871                 unsigned int sensitivity = 1;
3872                 config = arg;
3873
3874                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3875                         return -EINVAL;
3876
3877                 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3878                         sensitivity =
3879                           config->cfg_config_u.cfg_route.rtr_sensitivity;
3880                 }
3881
3882                 mutex_lock(&the_lnet.ln_api_mutex);
3883                 rc = lnet_add_route(config->cfg_net,
3884                                     config->cfg_config_u.cfg_route.rtr_hop,
3885                                     config->cfg_nid,
3886                                     config->cfg_config_u.cfg_route.
3887                                         rtr_priority, sensitivity);
3888                 mutex_unlock(&the_lnet.ln_api_mutex);
3889                 return rc;
3890         }
3891
3892         case IOC_LIBCFS_DEL_ROUTE:
3893                 config = arg;
3894
3895                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3896                         return -EINVAL;
3897
3898                 mutex_lock(&the_lnet.ln_api_mutex);
3899                 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3900                 mutex_unlock(&the_lnet.ln_api_mutex);
3901                 return rc;
3902
3903         case IOC_LIBCFS_GET_ROUTE:
3904                 config = arg;
3905
3906                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3907                         return -EINVAL;
3908
3909                 mutex_lock(&the_lnet.ln_api_mutex);
3910                 rc = lnet_get_route(config->cfg_count,
3911                                     &config->cfg_net,
3912                                     &config->cfg_config_u.cfg_route.rtr_hop,
3913                                     &config->cfg_nid,
3914                                     &config->cfg_config_u.cfg_route.rtr_flags,
3915                                     &config->cfg_config_u.cfg_route.
3916                                         rtr_priority,
3917                                     &config->cfg_config_u.cfg_route.
3918                                         rtr_sensitivity);
3919                 mutex_unlock(&the_lnet.ln_api_mutex);
3920                 return rc;
3921
3922         case IOC_LIBCFS_GET_LOCAL_NI: {
3923                 struct lnet_ioctl_config_ni *cfg_ni;
3924                 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3925                 struct lnet_ioctl_element_stats *stats;
3926                 __u32 tun_size;
3927
3928                 cfg_ni = arg;
3929
3930                 /* get the tunables if they are available */
3931                 if (cfg_ni->lic_cfg_hdr.ioc_len <
3932                     sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3933                         return -EINVAL;
3934
3935                 stats = (struct lnet_ioctl_element_stats *)
3936                         cfg_ni->lic_bulk;
3937                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3938                                 (cfg_ni->lic_bulk + sizeof(*stats));
3939
3940                 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3941                         sizeof(*stats);
3942
3943                 mutex_lock(&the_lnet.ln_api_mutex);
3944                 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3945                 mutex_unlock(&the_lnet.ln_api_mutex);
3946                 return rc;
3947         }
3948
3949         case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3950                 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3951
3952                 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3953                         return -EINVAL;
3954
3955                 mutex_lock(&the_lnet.ln_api_mutex);
3956                 rc = lnet_get_ni_stats(msg_stats);
3957                 mutex_unlock(&the_lnet.ln_api_mutex);
3958
3959                 return rc;
3960         }
3961
3962         case IOC_LIBCFS_GET_NET: {
3963                 size_t total = sizeof(*config) +
3964                                sizeof(struct lnet_ioctl_net_config);
3965                 config = arg;
3966
3967                 if (config->cfg_hdr.ioc_len < total)
3968                         return -EINVAL;
3969
3970                 mutex_lock(&the_lnet.ln_api_mutex);
3971                 rc = lnet_get_net_config(config);
3972                 mutex_unlock(&the_lnet.ln_api_mutex);
3973                 return rc;
3974         }
3975
3976         case IOC_LIBCFS_GET_LNET_STATS:
3977         {
3978                 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
3979
3980                 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
3981                         return -EINVAL;
3982
3983                 mutex_lock(&the_lnet.ln_api_mutex);
3984                 rc = lnet_counters_get(&lnet_stats->st_cntrs);
3985                 mutex_unlock(&the_lnet.ln_api_mutex);
3986                 return rc;
3987         }
3988
3989         case IOC_LIBCFS_RESET_LNET_STATS:
3990         {
3991                 mutex_lock(&the_lnet.ln_api_mutex);
3992                 lnet_counters_reset();
3993                 mutex_unlock(&the_lnet.ln_api_mutex);
3994                 return 0;
3995         }
3996
3997         case IOC_LIBCFS_CONFIG_RTR:
3998                 config = arg;
3999
4000                 if (config->cfg_hdr.ioc_len < sizeof(*config))
4001                         return -EINVAL;
4002
4003                 mutex_lock(&the_lnet.ln_api_mutex);
4004                 if (config->cfg_config_u.cfg_buffers.buf_enable) {
4005                         rc = lnet_rtrpools_enable();
4006                         mutex_unlock(&the_lnet.ln_api_mutex);
4007                         return rc;
4008                 }
4009                 lnet_rtrpools_disable();
4010                 mutex_unlock(&the_lnet.ln_api_mutex);
4011                 return 0;
4012
4013         case IOC_LIBCFS_ADD_BUF:
4014                 config = arg;
4015
4016                 if (config->cfg_hdr.ioc_len < sizeof(*config))
4017                         return -EINVAL;
4018
4019                 mutex_lock(&the_lnet.ln_api_mutex);
4020                 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
4021                                                 buf_tiny,
4022                                           config->cfg_config_u.cfg_buffers.
4023                                                 buf_small,
4024                                           config->cfg_config_u.cfg_buffers.
4025                                                 buf_large);
4026                 mutex_unlock(&the_lnet.ln_api_mutex);
4027                 return rc;
4028
4029         case IOC_LIBCFS_SET_NUMA_RANGE: {
4030                 struct lnet_ioctl_set_value *numa;
4031                 numa = arg;
4032                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
4033                         return -EINVAL;
4034                 lnet_net_lock(LNET_LOCK_EX);
4035                 lnet_numa_range = numa->sv_value;
4036                 lnet_net_unlock(LNET_LOCK_EX);
4037                 return 0;
4038         }
4039
4040         case IOC_LIBCFS_GET_NUMA_RANGE: {
4041                 struct lnet_ioctl_set_value *numa;
4042                 numa = arg;
4043                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
4044                         return -EINVAL;
4045                 numa->sv_value = lnet_numa_range;
4046                 return 0;
4047         }
4048
4049         case IOC_LIBCFS_GET_BUF: {
4050                 struct lnet_ioctl_pool_cfg *pool_cfg;
4051                 size_t total = sizeof(*config) + sizeof(*pool_cfg);
4052
4053                 config = arg;
4054
4055                 if (config->cfg_hdr.ioc_len < total)
4056                         return -EINVAL;
4057
4058                 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
4059
4060                 mutex_lock(&the_lnet.ln_api_mutex);
4061                 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
4062                 mutex_unlock(&the_lnet.ln_api_mutex);
4063                 return rc;
4064         }
4065
4066         case IOC_LIBCFS_GET_LOCAL_HSTATS: {
4067                 struct lnet_ioctl_local_ni_hstats *stats = arg;
4068
4069                 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
4070                         return -EINVAL;
4071
4072                 mutex_lock(&the_lnet.ln_api_mutex);
4073                 rc = lnet_get_local_ni_hstats(stats);
4074                 mutex_unlock(&the_lnet.ln_api_mutex);
4075
4076                 return rc;
4077         }
4078
4079         case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
4080                 struct lnet_ioctl_recovery_list *list = arg;
4081                 if (list->rlst_hdr.ioc_len < sizeof(*list))
4082                         return -EINVAL;
4083
4084                 mutex_lock(&the_lnet.ln_api_mutex);
4085                 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
4086                         rc = lnet_get_local_ni_recovery_list(list);
4087                 else
4088                         rc = lnet_get_peer_ni_recovery_list(list);
4089                 mutex_unlock(&the_lnet.ln_api_mutex);
4090                 return rc;
4091         }
4092
4093         case IOC_LIBCFS_ADD_PEER_NI: {
4094                 struct lnet_ioctl_peer_cfg *cfg = arg;
4095
4096                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4097                         return -EINVAL;
4098
4099                 mutex_lock(&the_lnet.ln_api_mutex);
4100                 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
4101                                       cfg->prcfg_cfg_nid,
4102                                       cfg->prcfg_mr, false);
4103                 mutex_unlock(&the_lnet.ln_api_mutex);
4104                 return rc;
4105         }
4106
4107         case IOC_LIBCFS_DEL_PEER_NI: {
4108                 struct lnet_ioctl_peer_cfg *cfg = arg;
4109
4110                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4111                         return -EINVAL;
4112
4113                 mutex_lock(&the_lnet.ln_api_mutex);
4114                 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
4115                                       cfg->prcfg_cfg_nid);
4116                 mutex_unlock(&the_lnet.ln_api_mutex);
4117                 return rc;
4118         }
4119
4120         case IOC_LIBCFS_GET_PEER_INFO: {
4121                 struct lnet_ioctl_peer *peer_info = arg;
4122
4123                 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
4124                         return -EINVAL;
4125
4126                 mutex_lock(&the_lnet.ln_api_mutex);
4127                 rc = lnet_get_peer_ni_info(
4128                    peer_info->pr_count,
4129                    &peer_info->pr_nid,
4130                    peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
4131                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
4132                    &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
4133                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
4134                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
4135                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
4136                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
4137                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
4138                 mutex_unlock(&the_lnet.ln_api_mutex);
4139                 return rc;
4140         }
4141
4142         case IOC_LIBCFS_GET_PEER_NI: {
4143                 struct lnet_ioctl_peer_cfg *cfg = arg;
4144
4145                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4146                         return -EINVAL;
4147
4148                 mutex_lock(&the_lnet.ln_api_mutex);
4149                 rc = lnet_get_peer_info(cfg,
4150                                         (void __user *)cfg->prcfg_bulk);
4151                 mutex_unlock(&the_lnet.ln_api_mutex);
4152                 return rc;
4153         }
4154
4155         case IOC_LIBCFS_GET_PEER_LIST: {
4156                 struct lnet_ioctl_peer_cfg *cfg = arg;
4157
4158                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4159                         return -EINVAL;
4160
4161                 mutex_lock(&the_lnet.ln_api_mutex);
4162                 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
4163                                 (struct lnet_process_id __user *)cfg->prcfg_bulk);
4164                 mutex_unlock(&the_lnet.ln_api_mutex);
4165                 return rc;
4166         }
4167
4168         case IOC_LIBCFS_SET_HEALHV: {
4169                 struct lnet_ioctl_reset_health_cfg *cfg = arg;
4170                 int value;
4171                 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
4172                         return -EINVAL;
4173                 if (cfg->rh_value < 0 ||
4174                     cfg->rh_value > LNET_MAX_HEALTH_VALUE)
4175                         value = LNET_MAX_HEALTH_VALUE;
4176                 else
4177                         value = cfg->rh_value;
4178                 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
4179                        value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
4180                        "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
4181                 mutex_lock(&the_lnet.ln_api_mutex);
4182                 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
4183                         lnet_ni_set_healthv(cfg->rh_nid, value,
4184                                              cfg->rh_all);
4185                 else
4186                         lnet_peer_ni_set_healthv(cfg->rh_nid, value,
4187                                                   cfg->rh_all);
4188                 mutex_unlock(&the_lnet.ln_api_mutex);
4189                 return 0;
4190         }
4191
4192         case IOC_LIBCFS_SET_CONNS_PER_PEER: {
4193                 struct lnet_ioctl_reset_conns_per_peer_cfg *cfg = arg;
4194                 int value;
4195
4196                 if (cfg->rcpp_hdr.ioc_len < sizeof(*cfg))
4197                         return -EINVAL;
4198                 if (cfg->rcpp_value < 0)
4199                         value = 1;
4200                 else
4201                         value = cfg->rcpp_value;
4202                 CDEBUG(D_NET,
4203                        "Setting conns_per_peer to %d for %s. all = %d\n",
4204                        value, libcfs_nid2str(cfg->rcpp_nid), cfg->rcpp_all);
4205                 mutex_lock(&the_lnet.ln_api_mutex);
4206                 lnet_ni_set_conns_per_peer(cfg->rcpp_nid, value, cfg->rcpp_all);
4207                 mutex_unlock(&the_lnet.ln_api_mutex);
4208                 return 0;
4209         }
4210
4211         case IOC_LIBCFS_NOTIFY_ROUTER: {
4212                 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
4213
4214                 /* The deadline passed in by the user should be some time in
4215                  * seconds in the future since the UNIX epoch. We have to map
4216                  * that deadline to the wall clock.
4217                  */
4218                 deadline += ktime_get_seconds();
4219                 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
4220                                    deadline);
4221         }
4222
4223         case IOC_LIBCFS_LNET_DIST:
4224                 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
4225                 if (rc < 0 && rc != -EHOSTUNREACH)
4226                         return rc;
4227
4228                 data->ioc_u32[0] = rc;
4229                 return 0;
4230
4231         case IOC_LIBCFS_TESTPROTOCOMPAT:
4232                 the_lnet.ln_testprotocompat = data->ioc_flags;
4233                 return 0;
4234
4235         case IOC_LIBCFS_LNET_FAULT:
4236                 return lnet_fault_ctl(data->ioc_flags, data);
4237
4238         case IOC_LIBCFS_PING: {
4239                 signed long timeout;
4240
4241                 id.nid = data->ioc_nid;
4242                 id.pid = data->ioc_u32[0];
4243
4244                 /* If timeout is negative then set default of 3 minutes */
4245                 if (((s32)data->ioc_u32[1] <= 0) ||
4246                     data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
4247                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
4248                 else
4249                         timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
4250
4251                 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
4252                                data->ioc_plen1 / sizeof(struct lnet_process_id));
4253
4254                 if (rc < 0)
4255                         return rc;
4256
4257                 data->ioc_count = rc;
4258                 return 0;
4259         }
4260
4261         case IOC_LIBCFS_PING_PEER: {
4262                 struct lnet_ioctl_ping_data *ping = arg;
4263                 struct lnet_peer *lp;
4264                 signed long timeout;
4265
4266                 /* If timeout is negative then set default of 3 minutes */
4267                 if (((s32)ping->op_param) <= 0 ||
4268                     ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
4269                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
4270                 else
4271                         timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
4272
4273                 rc = lnet_ping(ping->ping_id, timeout,
4274                                ping->ping_buf,
4275                                ping->ping_count);
4276                 if (rc < 0)
4277                         return rc;
4278
4279                 mutex_lock(&the_lnet.ln_api_mutex);
4280                 lp = lnet_find_peer(ping->ping_id.nid);
4281                 if (lp) {
4282                         ping->ping_id.nid = lp->lp_primary_nid;
4283                         ping->mr_info = lnet_peer_is_multi_rail(lp);
4284                         lnet_peer_decref_locked(lp);
4285                 }
4286                 mutex_unlock(&the_lnet.ln_api_mutex);
4287
4288                 ping->ping_count = rc;
4289                 return 0;
4290         }
4291
4292         case IOC_LIBCFS_DISCOVER: {
4293                 struct lnet_ioctl_ping_data *discover = arg;
4294                 struct lnet_peer *lp;
4295
4296                 rc = lnet_discover(discover->ping_id, discover->op_param,
4297                                    discover->ping_buf,
4298                                    discover->ping_count);
4299                 if (rc < 0)
4300                         return rc;
4301
4302                 mutex_lock(&the_lnet.ln_api_mutex);
4303                 lp = lnet_find_peer(discover->ping_id.nid);
4304                 if (lp) {
4305                         discover->ping_id.nid = lp->lp_primary_nid;
4306                         discover->mr_info = lnet_peer_is_multi_rail(lp);
4307                         lnet_peer_decref_locked(lp);
4308                 }
4309                 mutex_unlock(&the_lnet.ln_api_mutex);
4310
4311                 discover->ping_count = rc;
4312                 return 0;
4313         }
4314
4315         case IOC_LIBCFS_ADD_UDSP: {
4316                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4317                 __u32 bulk_size = ioc_udsp->iou_hdr.ioc_len;
4318
4319                 mutex_lock(&the_lnet.ln_api_mutex);
4320                 rc = lnet_udsp_demarshal_add(arg, bulk_size);
4321                 if (!rc) {
4322                         rc = lnet_udsp_apply_policies(NULL, false);
4323                         CDEBUG(D_NET, "policy application returned %d\n", rc);
4324                         rc = 0;
4325                 }
4326                 mutex_unlock(&the_lnet.ln_api_mutex);
4327
4328                 return rc;
4329         }
4330
4331         case IOC_LIBCFS_DEL_UDSP: {
4332                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4333                 int idx = ioc_udsp->iou_idx;
4334
4335                 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4336                         return -EINVAL;
4337
4338                 mutex_lock(&the_lnet.ln_api_mutex);
4339                 rc = lnet_udsp_del_policy(idx);
4340                 if (!rc) {
4341                         rc = lnet_udsp_apply_policies(NULL, false);
4342                         CDEBUG(D_NET, "policy re-application returned %d\n",
4343                                rc);
4344                         rc = 0;
4345                 }
4346                 mutex_unlock(&the_lnet.ln_api_mutex);
4347
4348                 return rc;
4349         }
4350
4351         case IOC_LIBCFS_GET_UDSP_SIZE: {
4352                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4353                 struct lnet_udsp *udsp;
4354
4355                 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4356                         return -EINVAL;
4357
4358                 rc = 0;
4359
4360                 mutex_lock(&the_lnet.ln_api_mutex);
4361                 udsp = lnet_udsp_get_policy(ioc_udsp->iou_idx);
4362                 if (!udsp) {
4363                         rc = -ENOENT;
4364                 } else {
4365                         /* coming in iou_idx will hold the idx of the udsp
4366                          * to get the size of. going out the iou_idx will
4367                          * hold the size of the UDSP found at the passed
4368                          * in index.
4369                          */
4370                         ioc_udsp->iou_idx = lnet_get_udsp_size(udsp);
4371                         if (ioc_udsp->iou_idx < 0)
4372                                 rc = -EINVAL;
4373                 }
4374                 mutex_unlock(&the_lnet.ln_api_mutex);
4375
4376                 return rc;
4377         }
4378
4379         case IOC_LIBCFS_GET_UDSP: {
4380                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4381                 struct lnet_udsp *udsp;
4382
4383                 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4384                         return -EINVAL;
4385
4386                 rc = 0;
4387
4388                 mutex_lock(&the_lnet.ln_api_mutex);
4389                 udsp = lnet_udsp_get_policy(ioc_udsp->iou_idx);
4390                 if (!udsp)
4391                         rc = -ENOENT;
4392                 else
4393                         rc = lnet_udsp_marshal(udsp, ioc_udsp);
4394                 mutex_unlock(&the_lnet.ln_api_mutex);
4395
4396                 return rc;
4397         }
4398
4399         case IOC_LIBCFS_GET_CONST_UDSP_INFO: {
4400                 struct lnet_ioctl_construct_udsp_info *info = arg;
4401
4402                 if (info->cud_hdr.ioc_len < sizeof(*info))
4403                         return -EINVAL;
4404
4405                 CDEBUG(D_NET, "GET_UDSP_INFO for %s\n",
4406                        libcfs_nid2str(info->cud_nid));
4407
4408                 mutex_lock(&the_lnet.ln_api_mutex);
4409                 lnet_udsp_get_construct_info(info);
4410                 mutex_unlock(&the_lnet.ln_api_mutex);
4411
4412                 return 0;
4413         }
4414
4415         default:
4416                 ni = lnet_net2ni_addref(data->ioc_net);
4417                 if (ni == NULL)
4418                         return -EINVAL;
4419
4420                 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
4421                         rc = -EINVAL;
4422                 else
4423                         rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
4424
4425                 lnet_ni_decref(ni);
4426                 return rc;
4427         }
4428         /* not reached */
4429 }
4430 EXPORT_SYMBOL(LNetCtl);
4431
4432 void LNetDebugPeer(struct lnet_process_id id)
4433 {
4434         lnet_debug_peer(id.nid);
4435 }
4436 EXPORT_SYMBOL(LNetDebugPeer);
4437
4438 /**
4439  * Determine if the specified peer \a nid is on the local node.
4440  *
4441  * \param nid   peer nid to check
4442  *
4443  * \retval true         If peer NID is on the local node.
4444  * \retval false        If peer NID is not on the local node.
4445  */
4446 bool LNetIsPeerLocal(lnet_nid_t nid)
4447 {
4448         struct lnet_net *net;
4449         struct lnet_ni *ni;
4450         int cpt;
4451
4452         cpt = lnet_net_lock_current();
4453         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4454                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4455                         if (lnet_nid_to_nid4(&ni->ni_nid) == nid) {
4456                                 lnet_net_unlock(cpt);
4457                                 return true;
4458                         }
4459                 }
4460         }
4461         lnet_net_unlock(cpt);
4462
4463         return false;
4464 }
4465 EXPORT_SYMBOL(LNetIsPeerLocal);
4466
4467 /**
4468  * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4469  * Note that all interfaces share a same PID, as requested by LNetNIInit().
4470  *
4471  * \param index Index of the interface to look up.
4472  * \param id On successful return, this location will hold the
4473  * struct lnet_process_id ID of the interface.
4474  *
4475  * \retval 0 If an interface exists at \a index.
4476  * \retval -ENOENT If no interface has been found.
4477  */
4478 int
4479 LNetGetId(unsigned int index, struct lnet_process_id *id)
4480 {
4481         struct lnet_ni   *ni;
4482         struct lnet_net  *net;
4483         int               cpt;
4484         int               rc = -ENOENT;
4485
4486         LASSERT(the_lnet.ln_refcount > 0);
4487
4488         cpt = lnet_net_lock_current();
4489
4490         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4491                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4492                         if (!nid_is_nid4(&ni->ni_nid))
4493                                 /* FIXME this needs to be handled */
4494                                 continue;
4495                         if (index-- != 0)
4496                                 continue;
4497
4498                         id->nid = lnet_nid_to_nid4(&ni->ni_nid);
4499                         id->pid = the_lnet.ln_pid;
4500                         rc = 0;
4501                         break;
4502                 }
4503         }
4504
4505         lnet_net_unlock(cpt);
4506         return rc;
4507 }
4508 EXPORT_SYMBOL(LNetGetId);
4509
4510 struct ping_data {
4511         int rc;
4512         int replied;
4513         struct lnet_handle_md mdh;
4514         struct completion completion;
4515 };
4516
4517 static void
4518 lnet_ping_event_handler(struct lnet_event *event)
4519 {
4520         struct ping_data *pd = event->md_user_ptr;
4521
4522         CDEBUG(D_NET, "ping event (%d %d)%s\n",
4523                event->type, event->status,
4524                event->unlinked ? " unlinked" : "");
4525
4526         if (event->status) {
4527                 if (!pd->rc)
4528                         pd->rc = event->status;
4529         } else if (event->type == LNET_EVENT_REPLY) {
4530                 pd->replied = 1;
4531                 pd->rc = event->mlength;
4532         }
4533         if (event->unlinked)
4534                 complete(&pd->completion);
4535 }
4536
4537 static int lnet_ping(struct lnet_process_id id, signed long timeout,
4538                      struct lnet_process_id __user *ids, int n_ids)
4539 {
4540         struct lnet_md md = { NULL };
4541         struct ping_data pd = { 0 };
4542         struct lnet_ping_buffer *pbuf;
4543         struct lnet_process_id tmpid;
4544         int i;
4545         int nob;
4546         int rc;
4547         int rc2;
4548
4549         /* n_ids limit is arbitrary */
4550         if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4551                 return -EINVAL;
4552
4553         /*
4554          * if the user buffer has more space than the lnet_interfaces_max
4555          * then only fill it up to lnet_interfaces_max
4556          */
4557         if (n_ids > lnet_interfaces_max)
4558                 n_ids = lnet_interfaces_max;
4559
4560         if (id.pid == LNET_PID_ANY)
4561                 id.pid = LNET_PID_LUSTRE;
4562
4563         pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4564         if (!pbuf)
4565                 return -ENOMEM;
4566
4567         /* initialize md content */
4568         md.start     = &pbuf->pb_info;
4569         md.length    = LNET_PING_INFO_SIZE(n_ids);
4570         md.threshold = 2; /* GET/REPLY */
4571         md.max_size  = 0;
4572         md.options   = LNET_MD_TRUNCATE;
4573         md.user_ptr  = &pd;
4574         md.handler   = lnet_ping_event_handler;
4575
4576         init_completion(&pd.completion);
4577
4578         rc = LNetMDBind(&md, LNET_UNLINK, &pd.mdh);
4579         if (rc != 0) {
4580                 CERROR("Can't bind MD: %d\n", rc);
4581                 goto fail_ping_buffer_decref;
4582         }
4583
4584         rc = LNetGet(LNET_NID_ANY, pd.mdh, id,
4585                      LNET_RESERVED_PORTAL,
4586                      LNET_PROTO_PING_MATCHBITS, 0, false);
4587
4588         if (rc != 0) {
4589                 /* Don't CERROR; this could be deliberate! */
4590                 rc2 = LNetMDUnlink(pd.mdh);
4591                 LASSERT(rc2 == 0);
4592
4593                 /* NB must wait for the UNLINK event below... */
4594         }
4595
4596         if (wait_for_completion_timeout(&pd.completion, timeout) == 0) {
4597                 /* Ensure completion in finite time... */
4598                 LNetMDUnlink(pd.mdh);
4599                 wait_for_completion(&pd.completion);
4600         }
4601         if (!pd.replied) {
4602                 rc = -EIO;
4603                 goto fail_ping_buffer_decref;
4604         }
4605
4606         nob = pd.rc;
4607         LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4608
4609         rc = -EPROTO;           /* if I can't parse... */
4610
4611         if (nob < 8) {
4612                 CERROR("%s: ping info too short %d\n",
4613                        libcfs_id2str(id), nob);
4614                 goto fail_ping_buffer_decref;
4615         }
4616
4617         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4618                 lnet_swap_pinginfo(pbuf);
4619         } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4620                 CERROR("%s: Unexpected magic %08x\n",
4621                        libcfs_id2str(id), pbuf->pb_info.pi_magic);
4622                 goto fail_ping_buffer_decref;
4623         }
4624
4625         if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4626                 CERROR("%s: ping w/o NI status: 0x%x\n",
4627                        libcfs_id2str(id), pbuf->pb_info.pi_features);
4628                 goto fail_ping_buffer_decref;
4629         }
4630
4631         if (nob < LNET_PING_INFO_SIZE(0)) {
4632                 CERROR("%s: Short reply %d(%d min)\n",
4633                        libcfs_id2str(id),
4634                        nob, (int)LNET_PING_INFO_SIZE(0));
4635                 goto fail_ping_buffer_decref;
4636         }
4637
4638         if (pbuf->pb_info.pi_nnis < n_ids)
4639                 n_ids = pbuf->pb_info.pi_nnis;
4640
4641         if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4642                 CERROR("%s: Short reply %d(%d expected)\n",
4643                        libcfs_id2str(id),
4644                        nob, (int)LNET_PING_INFO_SIZE(n_ids));
4645                 goto fail_ping_buffer_decref;
4646         }
4647
4648         rc = -EFAULT;           /* if I segv in copy_to_user()... */
4649
4650         memset(&tmpid, 0, sizeof(tmpid));
4651         for (i = 0; i < n_ids; i++) {
4652                 tmpid.pid = pbuf->pb_info.pi_pid;
4653                 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4654                 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4655                         goto fail_ping_buffer_decref;
4656         }
4657         rc = pbuf->pb_info.pi_nnis;
4658
4659  fail_ping_buffer_decref:
4660         lnet_ping_buffer_decref(pbuf);
4661         return rc;
4662 }
4663
4664 static int
4665 lnet_discover(struct lnet_process_id id, __u32 force,
4666               struct lnet_process_id __user *ids, int n_ids)
4667 {
4668         struct lnet_peer_ni *lpni;
4669         struct lnet_peer_ni *p;
4670         struct lnet_peer *lp;
4671         struct lnet_process_id *buf;
4672         int cpt;
4673         int i;
4674         int rc;
4675
4676         if (n_ids <= 0 ||
4677             id.nid == LNET_NID_ANY)
4678                 return -EINVAL;
4679
4680         if (id.pid == LNET_PID_ANY)
4681                 id.pid = LNET_PID_LUSTRE;
4682
4683         /*
4684          * If the user buffer has more space than the lnet_interfaces_max,
4685          * then only fill it up to lnet_interfaces_max.
4686          */
4687         if (n_ids > lnet_interfaces_max)
4688                 n_ids = lnet_interfaces_max;
4689
4690         CFS_ALLOC_PTR_ARRAY(buf, n_ids);
4691         if (!buf)
4692                 return -ENOMEM;
4693
4694         cpt = lnet_net_lock_current();
4695         lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4696         if (IS_ERR(lpni)) {
4697                 rc = PTR_ERR(lpni);
4698                 goto out;
4699         }
4700
4701         /*
4702          * Clearing the NIDS_UPTODATE flag ensures the peer will
4703          * be discovered, provided discovery has not been disabled.
4704          */
4705         lp = lpni->lpni_peer_net->lpn_peer;
4706         spin_lock(&lp->lp_lock);
4707         lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4708         /* If the force flag is set, force a PING and PUSH as well. */
4709         if (force)
4710                 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4711         spin_unlock(&lp->lp_lock);
4712         rc = lnet_discover_peer_locked(lpni, cpt, true);
4713         if (rc)
4714                 goto out_decref;
4715
4716         /* The lpni (or lp) for this NID may have changed and our ref is
4717          * the only thing keeping the old one around. Release the ref
4718          * and lookup the lpni again
4719          */
4720         lnet_peer_ni_decref_locked(lpni);
4721         lpni = lnet_find_peer_ni_locked(id.nid);
4722         if (!lpni) {
4723                 rc = -ENOENT;
4724                 goto out;
4725         }
4726         lp = lpni->lpni_peer_net->lpn_peer;
4727
4728         i = 0;
4729         p = NULL;
4730         while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4731                 buf[i].pid = id.pid;
4732                 buf[i].nid = p->lpni_nid;
4733                 if (++i >= n_ids)
4734                         break;
4735         }
4736         rc = i;
4737
4738 out_decref:
4739         lnet_peer_ni_decref_locked(lpni);
4740 out:
4741         lnet_net_unlock(cpt);
4742
4743         if (rc >= 0)
4744                 if (copy_to_user(ids, buf, rc * sizeof(*buf)))
4745                         rc = -EFAULT;
4746         CFS_FREE_PTR_ARRAY(buf, n_ids);
4747
4748         return rc;
4749 }
4750
4751 /**
4752  * Retrieve peer discovery status.
4753  *
4754  * \retval 1 if lnet_peer_discovery_disabled is 0
4755  * \retval 0 if lnet_peer_discovery_disabled is 1
4756  */
4757 int
4758 LNetGetPeerDiscoveryStatus(void)
4759 {
4760         return !lnet_peer_discovery_disabled;
4761 }
4762 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);