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(struct lnet_nid *nid, struct lnet_ni *ni)
1531 {
1532         struct lnet_net *net;
1533         /* FIXME handle long-addr nid */
1534         lnet_nid_t nid4 = lnet_nid_to_nid4(nid);
1535
1536         /* must called with hold of lnet_net_lock */
1537         if (LNET_CPT_NUMBER == 1)
1538                 return 0; /* the only one */
1539
1540         /*
1541          * If NI is provided then use the CPT identified in the NI cpt
1542          * list if one exists. If one doesn't exist, then that NI is
1543          * associated with all CPTs and it follows that the net it belongs
1544          * to is implicitly associated with all CPTs, so just hash the nid
1545          * and return that.
1546          */
1547         if (ni != NULL) {
1548                 if (ni->ni_cpts != NULL)
1549                         return ni->ni_cpts[lnet_nid_cpt_hash(nid4,
1550                                                              ni->ni_ncpts)];
1551                 else
1552                         return lnet_nid_cpt_hash(nid4, LNET_CPT_NUMBER);
1553         }
1554
1555         /* no NI provided so look at the net */
1556         net = lnet_get_net_locked(LNET_NID_NET(nid));
1557
1558         if (net != NULL && net->net_cpts != NULL) {
1559                 return net->net_cpts[lnet_nid_cpt_hash(nid4, net->net_ncpts)];
1560         }
1561
1562         return lnet_nid_cpt_hash(nid4, LNET_CPT_NUMBER);
1563 }
1564
1565 int
1566 lnet_cpt_of_nid(lnet_nid_t nid4, struct lnet_ni *ni)
1567 {
1568         int     cpt;
1569         int     cpt2;
1570         struct lnet_nid nid;
1571
1572         if (LNET_CPT_NUMBER == 1)
1573                 return 0; /* the only one */
1574
1575         lnet_nid4_to_nid(nid4, &nid);
1576         cpt = lnet_net_lock_current();
1577
1578         cpt2 = lnet_cpt_of_nid_locked(&nid, ni);
1579
1580         lnet_net_unlock(cpt);
1581
1582         return cpt2;
1583 }
1584 EXPORT_SYMBOL(lnet_cpt_of_nid);
1585
1586 int
1587 lnet_islocalnet_locked(__u32 net_id)
1588 {
1589         struct lnet_net *net;
1590         bool local;
1591
1592         net = lnet_get_net_locked(net_id);
1593
1594         local = net != NULL;
1595
1596         return local;
1597 }
1598
1599 int
1600 lnet_islocalnet(__u32 net_id)
1601 {
1602         int cpt;
1603         bool local;
1604
1605         cpt = lnet_net_lock_current();
1606
1607         local = lnet_islocalnet_locked(net_id);
1608
1609         lnet_net_unlock(cpt);
1610
1611         return local;
1612 }
1613
1614 struct lnet_ni  *
1615 lnet_nid_to_ni_locked(struct lnet_nid *nid, int cpt)
1616 {
1617         struct lnet_net  *net;
1618         struct lnet_ni *ni;
1619
1620         LASSERT(cpt != LNET_LOCK_EX);
1621
1622         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1623                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1624                         if (nid_same(&ni->ni_nid, nid))
1625                                 return ni;
1626                 }
1627         }
1628
1629         return NULL;
1630 }
1631
1632 struct lnet_ni  *
1633 lnet_nid2ni_locked(lnet_nid_t nid4, int cpt)
1634 {
1635         struct lnet_nid nid;
1636
1637         lnet_nid4_to_nid(nid4, &nid);
1638         return lnet_nid_to_ni_locked(&nid, cpt);
1639 }
1640
1641 struct lnet_ni *
1642 lnet_nid2ni_addref(lnet_nid_t nid4)
1643 {
1644         struct lnet_ni *ni;
1645         struct lnet_nid nid;
1646
1647         lnet_nid4_to_nid(nid4, &nid);
1648
1649         lnet_net_lock(0);
1650         ni = lnet_nid_to_ni_locked(&nid, 0);
1651         if (ni)
1652                 lnet_ni_addref_locked(ni, 0);
1653         lnet_net_unlock(0);
1654
1655         return ni;
1656 }
1657 EXPORT_SYMBOL(lnet_nid2ni_addref);
1658
1659 struct lnet_ni *
1660 lnet_nid_to_ni_addref(struct lnet_nid *nid)
1661 {
1662         struct lnet_ni *ni;
1663
1664         lnet_net_lock(0);
1665         ni = lnet_nid_to_ni_locked(nid, 0);
1666         if (ni)
1667                 lnet_ni_addref_locked(ni, 0);
1668         lnet_net_unlock(0);
1669
1670         return ni;
1671 }
1672 EXPORT_SYMBOL(lnet_nid_to_ni_addref);
1673
1674 int
1675 lnet_islocalnid(lnet_nid_t nid)
1676 {
1677         struct lnet_ni  *ni;
1678         int             cpt;
1679
1680         cpt = lnet_net_lock_current();
1681         ni = lnet_nid2ni_locked(nid, cpt);
1682         lnet_net_unlock(cpt);
1683
1684         return ni != NULL;
1685 }
1686
1687 int
1688 lnet_count_acceptor_nets(void)
1689 {
1690         /* Return the # of NIs that need the acceptor. */
1691         int              count = 0;
1692         struct lnet_net  *net;
1693         int              cpt;
1694
1695         cpt = lnet_net_lock_current();
1696         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1697                 /* all socklnd type networks should have the acceptor
1698                  * thread started */
1699                 if (net->net_lnd->lnd_accept != NULL)
1700                         count++;
1701         }
1702
1703         lnet_net_unlock(cpt);
1704
1705         return count;
1706 }
1707
1708 struct lnet_ping_buffer *
1709 lnet_ping_buffer_alloc(int nnis, gfp_t gfp)
1710 {
1711         struct lnet_ping_buffer *pbuf;
1712
1713         LIBCFS_ALLOC_GFP(pbuf, LNET_PING_BUFFER_SIZE(nnis), gfp);
1714         if (pbuf) {
1715                 pbuf->pb_nnis = nnis;
1716                 pbuf->pb_needs_post = false;
1717                 atomic_set(&pbuf->pb_refcnt, 1);
1718         }
1719
1720         return pbuf;
1721 }
1722
1723 void
1724 lnet_ping_buffer_free(struct lnet_ping_buffer *pbuf)
1725 {
1726         LASSERT(atomic_read(&pbuf->pb_refcnt) == 0);
1727         LIBCFS_FREE(pbuf, LNET_PING_BUFFER_SIZE(pbuf->pb_nnis));
1728 }
1729
1730 static struct lnet_ping_buffer *
1731 lnet_ping_target_create(int nnis)
1732 {
1733         struct lnet_ping_buffer *pbuf;
1734
1735         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
1736         if (pbuf == NULL) {
1737                 CERROR("Can't allocate ping source [%d]\n", nnis);
1738                 return NULL;
1739         }
1740
1741         pbuf->pb_info.pi_nnis = nnis;
1742         pbuf->pb_info.pi_pid = the_lnet.ln_pid;
1743         pbuf->pb_info.pi_magic = LNET_PROTO_PING_MAGIC;
1744         pbuf->pb_info.pi_features =
1745                 LNET_PING_FEAT_NI_STATUS | LNET_PING_FEAT_MULTI_RAIL;
1746
1747         return pbuf;
1748 }
1749
1750 static inline int
1751 lnet_get_net_ni_count_locked(struct lnet_net *net)
1752 {
1753         struct lnet_ni  *ni;
1754         int             count = 0;
1755
1756         list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1757                 count++;
1758
1759         return count;
1760 }
1761
1762 static inline int
1763 lnet_get_net_ni_count_pre(struct lnet_net *net)
1764 {
1765         struct lnet_ni  *ni;
1766         int             count = 0;
1767
1768         list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
1769                 count++;
1770
1771         return count;
1772 }
1773
1774 static inline int
1775 lnet_get_ni_count(void)
1776 {
1777         struct lnet_ni  *ni;
1778         struct lnet_net *net;
1779         int             count = 0;
1780
1781         lnet_net_lock(0);
1782
1783         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1784                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
1785                         count++;
1786         }
1787
1788         lnet_net_unlock(0);
1789
1790         return count;
1791 }
1792
1793 void
1794 lnet_swap_pinginfo(struct lnet_ping_buffer *pbuf)
1795 {
1796         struct lnet_ni_status *stat;
1797         int nnis;
1798         int i;
1799
1800         __swab32s(&pbuf->pb_info.pi_magic);
1801         __swab32s(&pbuf->pb_info.pi_features);
1802         __swab32s(&pbuf->pb_info.pi_pid);
1803         __swab32s(&pbuf->pb_info.pi_nnis);
1804         nnis = pbuf->pb_info.pi_nnis;
1805         if (nnis > pbuf->pb_nnis)
1806                 nnis = pbuf->pb_nnis;
1807         for (i = 0; i < nnis; i++) {
1808                 stat = &pbuf->pb_info.pi_ni[i];
1809                 __swab64s(&stat->ns_nid);
1810                 __swab32s(&stat->ns_status);
1811         }
1812 }
1813
1814 int
1815 lnet_ping_info_validate(struct lnet_ping_info *pinfo)
1816 {
1817         if (!pinfo)
1818                 return -EINVAL;
1819         if (pinfo->pi_magic != LNET_PROTO_PING_MAGIC)
1820                 return -EPROTO;
1821         if (!(pinfo->pi_features & LNET_PING_FEAT_NI_STATUS))
1822                 return -EPROTO;
1823         /* Loopback is guaranteed to be present */
1824         if (pinfo->pi_nnis < 1 || pinfo->pi_nnis > lnet_interfaces_max)
1825                 return -ERANGE;
1826         if (LNET_PING_INFO_LONI(pinfo) != LNET_NID_LO_0)
1827                 return -EPROTO;
1828         return 0;
1829 }
1830
1831 static void
1832 lnet_ping_target_destroy(void)
1833 {
1834         struct lnet_net *net;
1835         struct lnet_ni  *ni;
1836
1837         lnet_net_lock(LNET_LOCK_EX);
1838
1839         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1840                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1841                         lnet_ni_lock(ni);
1842                         ni->ni_status = NULL;
1843                         lnet_ni_unlock(ni);
1844                 }
1845         }
1846
1847         lnet_ping_buffer_decref(the_lnet.ln_ping_target);
1848         the_lnet.ln_ping_target = NULL;
1849
1850         lnet_net_unlock(LNET_LOCK_EX);
1851 }
1852
1853 static void
1854 lnet_ping_target_event_handler(struct lnet_event *event)
1855 {
1856         struct lnet_ping_buffer *pbuf = event->md_user_ptr;
1857
1858         if (event->unlinked)
1859                 lnet_ping_buffer_decref(pbuf);
1860 }
1861
1862 static int
1863 lnet_ping_target_setup(struct lnet_ping_buffer **ppbuf,
1864                        struct lnet_handle_md *ping_mdh,
1865                        int ni_count, bool set_eq)
1866 {
1867         struct lnet_process_id id = {
1868                 .nid = LNET_NID_ANY,
1869                 .pid = LNET_PID_ANY
1870         };
1871         struct lnet_me *me;
1872         struct lnet_md md = { NULL };
1873         int rc;
1874
1875         if (set_eq)
1876                 the_lnet.ln_ping_target_handler =
1877                         lnet_ping_target_event_handler;
1878
1879         *ppbuf = lnet_ping_target_create(ni_count);
1880         if (*ppbuf == NULL) {
1881                 rc = -ENOMEM;
1882                 goto fail_free_eq;
1883         }
1884
1885         /* Ping target ME/MD */
1886         me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1887                           LNET_PROTO_PING_MATCHBITS, 0,
1888                           LNET_UNLINK, LNET_INS_AFTER);
1889         if (IS_ERR(me)) {
1890                 rc = PTR_ERR(me);
1891                 CERROR("Can't create ping target ME: %d\n", rc);
1892                 goto fail_decref_ping_buffer;
1893         }
1894
1895         /* initialize md content */
1896         md.start     = &(*ppbuf)->pb_info;
1897         md.length    = LNET_PING_INFO_SIZE((*ppbuf)->pb_nnis);
1898         md.threshold = LNET_MD_THRESH_INF;
1899         md.max_size  = 0;
1900         md.options   = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1901                        LNET_MD_MANAGE_REMOTE;
1902         md.handler   = the_lnet.ln_ping_target_handler;
1903         md.user_ptr  = *ppbuf;
1904
1905         rc = LNetMDAttach(me, &md, LNET_RETAIN, ping_mdh);
1906         if (rc != 0) {
1907                 CERROR("Can't attach ping target MD: %d\n", rc);
1908                 goto fail_decref_ping_buffer;
1909         }
1910         lnet_ping_buffer_addref(*ppbuf);
1911
1912         return 0;
1913
1914 fail_decref_ping_buffer:
1915         LASSERT(atomic_read(&(*ppbuf)->pb_refcnt) == 1);
1916         lnet_ping_buffer_decref(*ppbuf);
1917         *ppbuf = NULL;
1918 fail_free_eq:
1919         return rc;
1920 }
1921
1922 static void
1923 lnet_ping_md_unlink(struct lnet_ping_buffer *pbuf,
1924                     struct lnet_handle_md *ping_mdh)
1925 {
1926         LNetMDUnlink(*ping_mdh);
1927         LNetInvalidateMDHandle(ping_mdh);
1928
1929         /* NB the MD could be busy; this just starts the unlink */
1930         wait_var_event_warning(&pbuf->pb_refcnt,
1931                                atomic_read(&pbuf->pb_refcnt) <= 1,
1932                                "Still waiting for ping data MD to unlink\n");
1933 }
1934
1935 static void
1936 lnet_ping_target_install_locked(struct lnet_ping_buffer *pbuf)
1937 {
1938         struct lnet_ni          *ni;
1939         struct lnet_net         *net;
1940         struct lnet_ni_status *ns;
1941         int                     i;
1942         int                     rc;
1943
1944         i = 0;
1945         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1946                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1947                         LASSERT(i < pbuf->pb_nnis);
1948
1949                         ns = &pbuf->pb_info.pi_ni[i];
1950
1951                         if (!nid_is_nid4(&ni->ni_nid))
1952                                 continue;
1953                         ns->ns_nid = lnet_nid_to_nid4(&ni->ni_nid);
1954
1955                         lnet_ni_lock(ni);
1956                         ns->ns_status = lnet_ni_get_status_locked(ni);
1957                         ni->ni_status = ns;
1958                         lnet_ni_unlock(ni);
1959
1960                         i++;
1961                 }
1962         }
1963         /*
1964          * We (ab)use the ns_status of the loopback interface to
1965          * transmit the sequence number. The first interface listed
1966          * must be the loopback interface.
1967          */
1968         rc = lnet_ping_info_validate(&pbuf->pb_info);
1969         if (rc) {
1970                 LCONSOLE_EMERG("Invalid ping target: %d\n", rc);
1971                 LBUG();
1972         }
1973         LNET_PING_BUFFER_SEQNO(pbuf) =
1974                 atomic_inc_return(&the_lnet.ln_ping_target_seqno);
1975 }
1976
1977 static void
1978 lnet_ping_target_update(struct lnet_ping_buffer *pbuf,
1979                         struct lnet_handle_md ping_mdh)
1980 {
1981         struct lnet_ping_buffer *old_pbuf = NULL;
1982         struct lnet_handle_md old_ping_md;
1983
1984         /* switch the NIs to point to the new ping info created */
1985         lnet_net_lock(LNET_LOCK_EX);
1986
1987         if (!the_lnet.ln_routing)
1988                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1989         if (!lnet_peer_discovery_disabled)
1990                 pbuf->pb_info.pi_features |= LNET_PING_FEAT_DISCOVERY;
1991
1992         /* Ensure only known feature bits have been set. */
1993         LASSERT(pbuf->pb_info.pi_features & LNET_PING_FEAT_BITS);
1994         LASSERT(!(pbuf->pb_info.pi_features & ~LNET_PING_FEAT_BITS));
1995
1996         lnet_ping_target_install_locked(pbuf);
1997
1998         if (the_lnet.ln_ping_target) {
1999                 old_pbuf = the_lnet.ln_ping_target;
2000                 old_ping_md = the_lnet.ln_ping_target_md;
2001         }
2002         the_lnet.ln_ping_target_md = ping_mdh;
2003         the_lnet.ln_ping_target = pbuf;
2004
2005         lnet_net_unlock(LNET_LOCK_EX);
2006
2007         if (old_pbuf) {
2008                 /* unlink and free the old ping info */
2009                 lnet_ping_md_unlink(old_pbuf, &old_ping_md);
2010                 lnet_ping_buffer_decref(old_pbuf);
2011         }
2012
2013         lnet_push_update_to_peers(0);
2014 }
2015
2016 static void
2017 lnet_ping_target_fini(void)
2018 {
2019         lnet_ping_md_unlink(the_lnet.ln_ping_target,
2020                             &the_lnet.ln_ping_target_md);
2021
2022         lnet_assert_handler_unused(the_lnet.ln_ping_target_handler);
2023         lnet_ping_target_destroy();
2024 }
2025
2026 /* Resize the push target. */
2027 int lnet_push_target_resize(void)
2028 {
2029         struct lnet_handle_md mdh;
2030         struct lnet_handle_md old_mdh;
2031         struct lnet_ping_buffer *pbuf;
2032         struct lnet_ping_buffer *old_pbuf;
2033         int nnis;
2034         int rc;
2035
2036 again:
2037         nnis = the_lnet.ln_push_target_nnis;
2038         if (nnis <= 0) {
2039                 CDEBUG(D_NET, "Invalid nnis %d\n", nnis);
2040                 return -EINVAL;
2041         }
2042
2043         /* NB: lnet_ping_buffer_alloc() sets pbuf refcount to 1. That ref is
2044          * dropped when we need to resize again (see "old_pbuf" below) or when
2045          * LNet is shutdown (see lnet_push_target_fini())
2046          */
2047         pbuf = lnet_ping_buffer_alloc(nnis, GFP_NOFS);
2048         if (!pbuf) {
2049                 CDEBUG(D_NET, "Can't allocate pbuf for nnis %d\n", nnis);
2050                 return -ENOMEM;
2051         }
2052
2053         rc = lnet_push_target_post(pbuf, &mdh);
2054         if (rc) {
2055                 CDEBUG(D_NET, "Failed to post push target: %d\n", rc);
2056                 lnet_ping_buffer_decref(pbuf);
2057                 return rc;
2058         }
2059
2060         lnet_net_lock(LNET_LOCK_EX);
2061         old_pbuf = the_lnet.ln_push_target;
2062         old_mdh = the_lnet.ln_push_target_md;
2063         the_lnet.ln_push_target = pbuf;
2064         the_lnet.ln_push_target_md = mdh;
2065         lnet_net_unlock(LNET_LOCK_EX);
2066
2067         if (old_pbuf) {
2068                 LNetMDUnlink(old_mdh);
2069                 /* Drop ref set by lnet_ping_buffer_alloc() */
2070                 lnet_ping_buffer_decref(old_pbuf);
2071         }
2072
2073         /* Received another push or reply that requires a larger buffer */
2074         if (nnis < the_lnet.ln_push_target_nnis)
2075                 goto again;
2076
2077         CDEBUG(D_NET, "nnis %d success\n", nnis);
2078         return 0;
2079 }
2080
2081 int lnet_push_target_post(struct lnet_ping_buffer *pbuf,
2082                           struct lnet_handle_md *mdhp)
2083 {
2084         struct lnet_process_id id = { LNET_NID_ANY, LNET_PID_ANY };
2085         struct lnet_md md = { NULL };
2086         struct lnet_me *me;
2087         int rc;
2088
2089         me = LNetMEAttach(LNET_RESERVED_PORTAL, id,
2090                           LNET_PROTO_PING_MATCHBITS, 0,
2091                           LNET_UNLINK, LNET_INS_AFTER);
2092         if (IS_ERR(me)) {
2093                 rc = PTR_ERR(me);
2094                 CERROR("Can't create push target ME: %d\n", rc);
2095                 return rc;
2096         }
2097
2098         pbuf->pb_needs_post = false;
2099
2100         /* This reference is dropped by lnet_push_target_event_handler() */
2101         lnet_ping_buffer_addref(pbuf);
2102
2103         /* initialize md content */
2104         md.start     = &pbuf->pb_info;
2105         md.length    = LNET_PING_INFO_SIZE(pbuf->pb_nnis);
2106         md.threshold = 1;
2107         md.max_size  = 0;
2108         md.options   = LNET_MD_OP_PUT | LNET_MD_TRUNCATE;
2109         md.user_ptr  = pbuf;
2110         md.handler   = the_lnet.ln_push_target_handler;
2111
2112         rc = LNetMDAttach(me, &md, LNET_UNLINK, mdhp);
2113         if (rc) {
2114                 CERROR("Can't attach push MD: %d\n", rc);
2115                 lnet_ping_buffer_decref(pbuf);
2116                 pbuf->pb_needs_post = true;
2117                 return rc;
2118         }
2119
2120         CDEBUG(D_NET, "posted push target %p\n", pbuf);
2121
2122         return 0;
2123 }
2124
2125 static void lnet_push_target_event_handler(struct lnet_event *ev)
2126 {
2127         struct lnet_ping_buffer *pbuf = ev->md_user_ptr;
2128
2129         CDEBUG(D_NET, "type %d status %d unlinked %d\n", ev->type, ev->status,
2130                ev->unlinked);
2131
2132         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC))
2133                 lnet_swap_pinginfo(pbuf);
2134
2135         if (ev->type == LNET_EVENT_UNLINK) {
2136                 /* Drop ref added by lnet_push_target_post() */
2137                 lnet_ping_buffer_decref(pbuf);
2138                 return;
2139         }
2140
2141         lnet_peer_push_event(ev);
2142         if (ev->unlinked)
2143                 /* Drop ref added by lnet_push_target_post */
2144                 lnet_ping_buffer_decref(pbuf);
2145 }
2146
2147 /* Initialize the push target. */
2148 static int lnet_push_target_init(void)
2149 {
2150         int rc;
2151
2152         if (the_lnet.ln_push_target)
2153                 return -EALREADY;
2154
2155         the_lnet.ln_push_target_handler =
2156                 lnet_push_target_event_handler;
2157
2158         rc = LNetSetLazyPortal(LNET_RESERVED_PORTAL);
2159         LASSERT(rc == 0);
2160
2161         /* Start at the required minimum, we'll enlarge if required. */
2162         the_lnet.ln_push_target_nnis = LNET_INTERFACES_MIN;
2163
2164         rc = lnet_push_target_resize();
2165
2166         if (rc) {
2167                 LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2168                 the_lnet.ln_push_target_handler = NULL;
2169         }
2170
2171         return rc;
2172 }
2173
2174 /* Clean up the push target. */
2175 static void lnet_push_target_fini(void)
2176 {
2177         if (!the_lnet.ln_push_target)
2178                 return;
2179
2180         /* Unlink and invalidate to prevent new references. */
2181         LNetMDUnlink(the_lnet.ln_push_target_md);
2182         LNetInvalidateMDHandle(&the_lnet.ln_push_target_md);
2183
2184         /* Wait for the unlink to complete. */
2185         wait_var_event_warning(&the_lnet.ln_push_target->pb_refcnt,
2186                                atomic_read(&the_lnet.ln_push_target->pb_refcnt) <= 1,
2187                                "Still waiting for ping data MD to unlink\n");
2188
2189         /* Drop ref set by lnet_ping_buffer_alloc() */
2190         lnet_ping_buffer_decref(the_lnet.ln_push_target);
2191         the_lnet.ln_push_target = NULL;
2192         the_lnet.ln_push_target_nnis = 0;
2193
2194         LNetClearLazyPortal(LNET_RESERVED_PORTAL);
2195         lnet_assert_handler_unused(the_lnet.ln_push_target_handler);
2196         the_lnet.ln_push_target_handler = NULL;
2197 }
2198
2199 static int
2200 lnet_ni_tq_credits(struct lnet_ni *ni)
2201 {
2202         int     credits;
2203
2204         LASSERT(ni->ni_ncpts >= 1);
2205
2206         if (ni->ni_ncpts == 1)
2207                 return ni->ni_net->net_tunables.lct_max_tx_credits;
2208
2209         credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
2210         credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
2211         credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
2212
2213         return credits;
2214 }
2215
2216 static void
2217 lnet_ni_unlink_locked(struct lnet_ni *ni)
2218 {
2219         /* move it to zombie list and nobody can find it anymore */
2220         LASSERT(!list_empty(&ni->ni_netlist));
2221         list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
2222         lnet_ni_decref_locked(ni, 0);
2223 }
2224
2225 static void
2226 lnet_clear_zombies_nis_locked(struct lnet_net *net)
2227 {
2228         int             i;
2229         int             islo;
2230         struct lnet_ni  *ni;
2231         struct list_head *zombie_list = &net->net_ni_zombie;
2232
2233         /*
2234          * Now wait for the NIs I just nuked to show up on the zombie
2235          * list and shut them down in guaranteed thread context
2236          */
2237         i = 2;
2238         while (!list_empty(zombie_list)) {
2239                 int     *ref;
2240                 int     j;
2241
2242                 ni = list_entry(zombie_list->next,
2243                                 struct lnet_ni, ni_netlist);
2244                 list_del_init(&ni->ni_netlist);
2245                 /* the ni should be in deleting state. If it's not it's
2246                  * a bug */
2247                 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
2248                 cfs_percpt_for_each(ref, j, ni->ni_refs) {
2249                         if (*ref == 0)
2250                                 continue;
2251                         /* still busy, add it back to zombie list */
2252                         list_add(&ni->ni_netlist, zombie_list);
2253                         break;
2254                 }
2255
2256                 if (!list_empty(&ni->ni_netlist)) {
2257                         /* Unlock mutex while waiting to allow other
2258                          * threads to read the LNet state and fall through
2259                          * to avoid deadlock
2260                          */
2261                         lnet_net_unlock(LNET_LOCK_EX);
2262                         mutex_unlock(&the_lnet.ln_api_mutex);
2263
2264                         ++i;
2265                         if ((i & (-i)) == i) {
2266                                 CDEBUG(D_WARNING,
2267                                        "Waiting for zombie LNI %s\n",
2268                                        libcfs_nidstr(&ni->ni_nid));
2269                         }
2270                         schedule_timeout_uninterruptible(cfs_time_seconds(1));
2271
2272                         mutex_lock(&the_lnet.ln_api_mutex);
2273                         lnet_net_lock(LNET_LOCK_EX);
2274                         continue;
2275                 }
2276
2277                 lnet_net_unlock(LNET_LOCK_EX);
2278
2279                 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
2280
2281                 LASSERT(!in_interrupt());
2282                 /* Holding the mutex makes it safe for lnd_shutdown
2283                  * to call module_put(). Module unload cannot finish
2284                  * until lnet_unregister_lnd() completes, and that
2285                  * requires the mutex.
2286                  */
2287                 mutex_lock(&the_lnet.ln_lnd_mutex);
2288                 (net->net_lnd->lnd_shutdown)(ni);
2289                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2290
2291                 if (!islo)
2292                         CDEBUG(D_LNI, "Removed LNI %s\n",
2293                               libcfs_nidstr(&ni->ni_nid));
2294
2295                 lnet_ni_free(ni);
2296                 i = 2;
2297                 lnet_net_lock(LNET_LOCK_EX);
2298         }
2299 }
2300
2301 /* shutdown down the NI and release refcount */
2302 static void
2303 lnet_shutdown_lndni(struct lnet_ni *ni)
2304 {
2305         int i;
2306         struct lnet_net *net = ni->ni_net;
2307
2308         lnet_net_lock(LNET_LOCK_EX);
2309         lnet_ni_lock(ni);
2310         ni->ni_state = LNET_NI_STATE_DELETING;
2311         lnet_ni_unlock(ni);
2312         lnet_ni_unlink_locked(ni);
2313         lnet_incr_dlc_seq();
2314         lnet_net_unlock(LNET_LOCK_EX);
2315
2316         /* clear messages for this NI on the lazy portal */
2317         for (i = 0; i < the_lnet.ln_nportals; i++)
2318                 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
2319
2320         lnet_net_lock(LNET_LOCK_EX);
2321         lnet_clear_zombies_nis_locked(net);
2322         lnet_net_unlock(LNET_LOCK_EX);
2323 }
2324
2325 static void
2326 lnet_shutdown_lndnet(struct lnet_net *net)
2327 {
2328         struct lnet_ni *ni;
2329
2330         lnet_net_lock(LNET_LOCK_EX);
2331
2332         list_del_init(&net->net_list);
2333
2334         while (!list_empty(&net->net_ni_list)) {
2335                 ni = list_entry(net->net_ni_list.next,
2336                                 struct lnet_ni, ni_netlist);
2337                 lnet_net_unlock(LNET_LOCK_EX);
2338                 lnet_shutdown_lndni(ni);
2339                 lnet_net_lock(LNET_LOCK_EX);
2340         }
2341
2342         lnet_net_unlock(LNET_LOCK_EX);
2343
2344         /* Do peer table cleanup for this net */
2345         lnet_peer_tables_cleanup(net);
2346
2347         lnet_net_free(net);
2348 }
2349
2350 static void
2351 lnet_shutdown_lndnets(void)
2352 {
2353         struct lnet_net *net;
2354         LIST_HEAD(resend);
2355         struct lnet_msg *msg, *tmp;
2356
2357         /* NB called holding the global mutex */
2358
2359         /* All quiet on the API front */
2360         LASSERT(the_lnet.ln_state == LNET_STATE_RUNNING);
2361         LASSERT(the_lnet.ln_refcount == 0);
2362
2363         lnet_net_lock(LNET_LOCK_EX);
2364         the_lnet.ln_state = LNET_STATE_STOPPING;
2365
2366         /*
2367          * move the nets to the zombie list to avoid them being
2368          * picked up for new work. LONET is also included in the
2369          * Nets that will be moved to the zombie list
2370          */
2371         list_splice_init(&the_lnet.ln_nets, &the_lnet.ln_net_zombie);
2372
2373         /* Drop the cached loopback Net. */
2374         if (the_lnet.ln_loni != NULL) {
2375                 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
2376                 the_lnet.ln_loni = NULL;
2377         }
2378         lnet_net_unlock(LNET_LOCK_EX);
2379
2380         /* iterate through the net zombie list and delete each net */
2381         while (!list_empty(&the_lnet.ln_net_zombie)) {
2382                 net = list_entry(the_lnet.ln_net_zombie.next,
2383                                  struct lnet_net, net_list);
2384                 lnet_shutdown_lndnet(net);
2385         }
2386
2387         spin_lock(&the_lnet.ln_msg_resend_lock);
2388         list_splice(&the_lnet.ln_msg_resend, &resend);
2389         spin_unlock(&the_lnet.ln_msg_resend_lock);
2390
2391         list_for_each_entry_safe(msg, tmp, &resend, msg_list) {
2392                 list_del_init(&msg->msg_list);
2393                 msg->msg_no_resend = true;
2394                 lnet_finalize(msg, -ECANCELED);
2395         }
2396
2397         lnet_net_lock(LNET_LOCK_EX);
2398         the_lnet.ln_state = LNET_STATE_SHUTDOWN;
2399         lnet_net_unlock(LNET_LOCK_EX);
2400 }
2401
2402 static int
2403 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
2404 {
2405         int                     rc = -EINVAL;
2406         struct lnet_tx_queue    *tq;
2407         int                     i;
2408         struct lnet_net         *net = ni->ni_net;
2409
2410         mutex_lock(&the_lnet.ln_lnd_mutex);
2411
2412         if (tun) {
2413                 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
2414                 ni->ni_lnd_tunables_set = true;
2415         }
2416
2417         rc = (net->net_lnd->lnd_startup)(ni);
2418
2419         mutex_unlock(&the_lnet.ln_lnd_mutex);
2420
2421         if (rc != 0) {
2422                 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
2423                                    rc, libcfs_lnd2str(net->net_lnd->lnd_type));
2424                 goto failed0;
2425         }
2426
2427         lnet_ni_lock(ni);
2428         ni->ni_state = LNET_NI_STATE_ACTIVE;
2429         lnet_ni_unlock(ni);
2430
2431         /* We keep a reference on the loopback net through the loopback NI */
2432         if (net->net_lnd->lnd_type == LOLND) {
2433                 lnet_ni_addref(ni);
2434                 LASSERT(the_lnet.ln_loni == NULL);
2435                 the_lnet.ln_loni = ni;
2436                 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
2437                 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
2438                 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
2439                 ni->ni_net->net_tunables.lct_peer_timeout = 0;
2440                 return 0;
2441         }
2442
2443         if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
2444             ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
2445                 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
2446                                    libcfs_lnd2str(net->net_lnd->lnd_type),
2447                                    ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
2448                                         "" : "per-peer ");
2449                 /* shutdown the NI since if we get here then it must've already
2450                  * been started
2451                  */
2452                 lnet_shutdown_lndni(ni);
2453                 return -EINVAL;
2454         }
2455
2456         cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
2457                 tq->tq_credits_min =
2458                 tq->tq_credits_max =
2459                 tq->tq_credits = lnet_ni_tq_credits(ni);
2460         }
2461
2462         atomic_set(&ni->ni_tx_credits,
2463                    lnet_ni_tq_credits(ni) * ni->ni_ncpts);
2464         atomic_set(&ni->ni_healthv, LNET_MAX_HEALTH_VALUE);
2465
2466         CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
2467                 libcfs_nidstr(&ni->ni_nid),
2468                 ni->ni_net->net_tunables.lct_peer_tx_credits,
2469                 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
2470                 ni->ni_net->net_tunables.lct_peer_rtr_credits,
2471                 ni->ni_net->net_tunables.lct_peer_timeout);
2472
2473         return 0;
2474 failed0:
2475         lnet_ni_free(ni);
2476         return rc;
2477 }
2478
2479 static int
2480 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
2481 {
2482         struct lnet_ni *ni;
2483         struct lnet_net *net_l = NULL;
2484         LIST_HEAD(local_ni_list);
2485         int rc;
2486         int ni_count = 0;
2487         __u32 lnd_type;
2488         const struct lnet_lnd  *lnd;
2489         int peer_timeout =
2490                 net->net_tunables.lct_peer_timeout;
2491         int maxtxcredits =
2492                 net->net_tunables.lct_max_tx_credits;
2493         int peerrtrcredits =
2494                 net->net_tunables.lct_peer_rtr_credits;
2495
2496         /*
2497          * make sure that this net is unique. If it isn't then
2498          * we are adding interfaces to an already existing network, and
2499          * 'net' is just a convenient way to pass in the list.
2500          * if it is unique we need to find the LND and load it if
2501          * necessary.
2502          */
2503         if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
2504                 lnd_type = LNET_NETTYP(net->net_id);
2505
2506                 mutex_lock(&the_lnet.ln_lnd_mutex);
2507                 lnd = lnet_find_lnd_by_type(lnd_type);
2508
2509                 if (lnd == NULL) {
2510                         mutex_unlock(&the_lnet.ln_lnd_mutex);
2511                         rc = request_module("%s", libcfs_lnd2modname(lnd_type));
2512                         mutex_lock(&the_lnet.ln_lnd_mutex);
2513
2514                         lnd = lnet_find_lnd_by_type(lnd_type);
2515                         if (lnd == NULL) {
2516                                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2517                                 CERROR("Can't load LND %s, module %s, rc=%d\n",
2518                                 libcfs_lnd2str(lnd_type),
2519                                 libcfs_lnd2modname(lnd_type), rc);
2520 #ifndef HAVE_MODULE_LOADING_SUPPORT
2521                                 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
2522                                                 "compiled with kernel module "
2523                                                 "loading support.");
2524 #endif
2525                                 rc = -EINVAL;
2526                                 goto failed0;
2527                         }
2528                 }
2529
2530                 net->net_lnd = lnd;
2531
2532                 mutex_unlock(&the_lnet.ln_lnd_mutex);
2533
2534                 net_l = net;
2535         }
2536
2537         /*
2538          * net_l: if the network being added is unique then net_l
2539          *        will point to that network
2540          *        if the network being added is not unique then
2541          *        net_l points to the existing network.
2542          *
2543          * When we enter the loop below, we'll pick NIs off he
2544          * network beign added and start them up, then add them to
2545          * a local ni list. Once we've successfully started all
2546          * the NIs then we join the local NI list (of started up
2547          * networks) with the net_l->net_ni_list, which should
2548          * point to the correct network to add the new ni list to
2549          *
2550          * If any of the new NIs fail to start up, then we want to
2551          * iterate through the local ni list, which should include
2552          * any NIs which were successfully started up, and shut
2553          * them down.
2554          *
2555          * After than we want to delete the network being added,
2556          * to avoid a memory leak.
2557          */
2558         while (!list_empty(&net->net_ni_added)) {
2559                 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
2560                                 ni_netlist);
2561                 list_del_init(&ni->ni_netlist);
2562
2563                 /* make sure that the the NI we're about to start
2564                  * up is actually unique. if it's not fail. */
2565                 if (!lnet_ni_unique_net(&net_l->net_ni_list,
2566                                         ni->ni_interface)) {
2567                         rc = -EEXIST;
2568                         goto failed1;
2569                 }
2570
2571                 /* adjust the pointer the parent network, just in case it
2572                  * the net is a duplicate */
2573                 ni->ni_net = net_l;
2574
2575                 rc = lnet_startup_lndni(ni, tun);
2576
2577                 if (rc < 0)
2578                         goto failed1;
2579
2580                 lnet_ni_addref(ni);
2581                 list_add_tail(&ni->ni_netlist, &local_ni_list);
2582
2583                 ni_count++;
2584         }
2585
2586         lnet_net_lock(LNET_LOCK_EX);
2587         list_splice_tail(&local_ni_list, &net_l->net_ni_list);
2588         lnet_incr_dlc_seq();
2589         lnet_net_unlock(LNET_LOCK_EX);
2590
2591         /* if the network is not unique then we don't want to keep
2592          * it around after we're done. Free it. Otherwise add that
2593          * net to the global the_lnet.ln_nets */
2594         if (net_l != net && net_l != NULL) {
2595                 /*
2596                  * TODO - note. currently the tunables can not be updated
2597                  * once added
2598                  */
2599                 lnet_net_free(net);
2600         } else {
2601                 /*
2602                  * restore tunables after it has been overwitten by the
2603                  * lnd
2604                  */
2605                 if (peer_timeout != -1)
2606                         net->net_tunables.lct_peer_timeout = peer_timeout;
2607                 if (maxtxcredits != -1)
2608                         net->net_tunables.lct_max_tx_credits = maxtxcredits;
2609                 if (peerrtrcredits != -1)
2610                         net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
2611
2612                 lnet_net_lock(LNET_LOCK_EX);
2613                 list_add_tail(&net->net_list, &the_lnet.ln_nets);
2614                 lnet_net_unlock(LNET_LOCK_EX);
2615         }
2616
2617         return ni_count;
2618
2619 failed1:
2620         /*
2621          * shutdown the new NIs that are being started up
2622          * free the NET being started
2623          */
2624         while (!list_empty(&local_ni_list)) {
2625                 ni = list_entry(local_ni_list.next, struct lnet_ni,
2626                                 ni_netlist);
2627
2628                 lnet_shutdown_lndni(ni);
2629         }
2630
2631 failed0:
2632         lnet_net_free(net);
2633
2634         return rc;
2635 }
2636
2637 static int
2638 lnet_startup_lndnets(struct list_head *netlist)
2639 {
2640         struct lnet_net         *net;
2641         int                     rc;
2642         int                     ni_count = 0;
2643
2644         /*
2645          * Change to running state before bringing up the LNDs. This
2646          * allows lnet_shutdown_lndnets() to assert that we've passed
2647          * through here.
2648          */
2649         lnet_net_lock(LNET_LOCK_EX);
2650         the_lnet.ln_state = LNET_STATE_RUNNING;
2651         lnet_net_unlock(LNET_LOCK_EX);
2652
2653         while (!list_empty(netlist)) {
2654                 net = list_entry(netlist->next, struct lnet_net, net_list);
2655                 list_del_init(&net->net_list);
2656
2657                 rc = lnet_startup_lndnet(net, NULL);
2658
2659                 if (rc < 0)
2660                         goto failed;
2661
2662                 ni_count += rc;
2663         }
2664
2665         return ni_count;
2666 failed:
2667         lnet_shutdown_lndnets();
2668
2669         return rc;
2670 }
2671
2672 static int lnet_genl_parse_list(struct sk_buff *msg,
2673                                 const struct ln_key_list *data[], u16 idx)
2674 {
2675         const struct ln_key_list *list = data[idx];
2676         const struct ln_key_props *props;
2677         struct nlattr *node;
2678         u16 count;
2679
2680         if (!list)
2681                 return 0;
2682
2683         if (!list->lkl_maxattr)
2684                 return -ERANGE;
2685
2686         props = list->lkl_list;
2687         if (!props)
2688                 return -EINVAL;
2689
2690         node = nla_nest_start(msg, LN_SCALAR_ATTR_LIST);
2691         if (!node)
2692                 return -ENOBUFS;
2693
2694         for (count = 1; count <= list->lkl_maxattr; count++) {
2695                 struct nlattr *key = nla_nest_start(msg, count);
2696
2697                 if (count == 1)
2698                         nla_put_u16(msg, LN_SCALAR_ATTR_LIST_SIZE,
2699                                     list->lkl_maxattr);
2700
2701                 nla_put_u16(msg, LN_SCALAR_ATTR_INDEX, count);
2702                 if (props[count].lkp_values)
2703                         nla_put_string(msg, LN_SCALAR_ATTR_VALUE,
2704                                        props[count].lkp_values);
2705                 if (props[count].lkp_key_format)
2706                         nla_put_u16(msg, LN_SCALAR_ATTR_KEY_FORMAT,
2707                                     props[count].lkp_key_format);
2708                 nla_put_u16(msg, LN_SCALAR_ATTR_NLA_TYPE,
2709                             props[count].lkp_data_type);
2710                 if (props[count].lkp_data_type == NLA_NESTED) {
2711                         int rc;
2712
2713                         rc = lnet_genl_parse_list(msg, data, ++idx);
2714                         if (rc < 0)
2715                                 return rc;
2716                 }
2717
2718                 nla_nest_end(msg, key);
2719         }
2720
2721         nla_nest_end(msg, node);
2722         return 0;
2723 }
2724
2725 int lnet_genl_send_scalar_list(struct sk_buff *msg, u32 portid, u32 seq,
2726                                const struct genl_family *family, int flags,
2727                                u8 cmd, const struct ln_key_list *data[])
2728 {
2729         int rc = 0;
2730         void *hdr;
2731
2732         if (!data[0])
2733                 return -EINVAL;
2734
2735         hdr = genlmsg_put(msg, portid, seq, family, flags, cmd);
2736         if (!hdr)
2737                 GOTO(canceled, rc = -EMSGSIZE);
2738
2739         rc = lnet_genl_parse_list(msg, data, 0);
2740         if (rc < 0)
2741                 GOTO(canceled, rc);
2742
2743         genlmsg_end(msg, hdr);
2744 canceled:
2745         if (rc < 0)
2746                 genlmsg_cancel(msg, hdr);
2747         return rc;
2748 }
2749 EXPORT_SYMBOL(lnet_genl_send_scalar_list);
2750
2751 /**
2752  * Initialize LNet library.
2753  *
2754  * Automatically called at module loading time. Caller has to call
2755  * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
2756  * latter returned 0. It must be called exactly once.
2757  *
2758  * \retval 0 on success
2759  * \retval -ve on failures.
2760  */
2761 int lnet_lib_init(void)
2762 {
2763         int rc;
2764
2765         lnet_assert_wire_constants();
2766
2767         /* refer to global cfs_cpt_table for now */
2768         the_lnet.ln_cpt_table = cfs_cpt_tab;
2769         the_lnet.ln_cpt_number = cfs_cpt_number(cfs_cpt_tab);
2770
2771         LASSERT(the_lnet.ln_cpt_number > 0);
2772         if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
2773                 /* we are under risk of consuming all lh_cookie */
2774                 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
2775                        "please change setting of CPT-table and retry\n",
2776                        the_lnet.ln_cpt_number, LNET_CPT_MAX);
2777                 return -E2BIG;
2778         }
2779
2780         while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
2781                 the_lnet.ln_cpt_bits++;
2782
2783         rc = lnet_create_locks();
2784         if (rc != 0) {
2785                 CERROR("Can't create LNet global locks: %d\n", rc);
2786                 return rc;
2787         }
2788
2789         the_lnet.ln_refcount = 0;
2790         INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
2791         INIT_LIST_HEAD(&the_lnet.ln_msg_resend);
2792
2793         /* The hash table size is the number of bits it takes to express the set
2794          * ln_num_routes, minus 1 (better to under estimate than over so we
2795          * don't waste memory). */
2796         if (rnet_htable_size <= 0)
2797                 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
2798         else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
2799                 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
2800         the_lnet.ln_remote_nets_hbits = max_t(int, 1,
2801                                            order_base_2(rnet_htable_size) - 1);
2802
2803         /* All LNDs apart from the LOLND are in separate modules.  They
2804          * register themselves when their module loads, and unregister
2805          * themselves when their module is unloaded. */
2806         lnet_register_lnd(&the_lolnd);
2807         return 0;
2808 }
2809
2810 /**
2811  * Finalize LNet library.
2812  *
2813  * \pre lnet_lib_init() called with success.
2814  * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
2815  *
2816  * As this happens at module-unload, all lnds must already be unloaded,
2817  * so they must already be unregistered.
2818  */
2819 void lnet_lib_exit(void)
2820 {
2821         int i;
2822
2823         LASSERT(the_lnet.ln_refcount == 0);
2824         lnet_unregister_lnd(&the_lolnd);
2825         for (i = 0; i < NUM_LNDS; i++)
2826                 LASSERT(!the_lnet.ln_lnds[i]);
2827         lnet_destroy_locks();
2828 }
2829
2830 /**
2831  * Set LNet PID and start LNet interfaces, routing, and forwarding.
2832  *
2833  * Users must call this function at least once before any other functions.
2834  * For each successful call there must be a corresponding call to
2835  * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
2836  * ignored.
2837  *
2838  * The PID used by LNet may be different from the one requested.
2839  * See LNetGetId().
2840  *
2841  * \param requested_pid PID requested by the caller.
2842  *
2843  * \return >= 0 on success, and < 0 error code on failures.
2844  */
2845 int
2846 LNetNIInit(lnet_pid_t requested_pid)
2847 {
2848         int                     im_a_router = 0;
2849         int                     rc;
2850         int                     ni_count;
2851         struct lnet_ping_buffer *pbuf;
2852         struct lnet_handle_md   ping_mdh;
2853         LIST_HEAD(net_head);
2854         struct lnet_net         *net;
2855
2856         mutex_lock(&the_lnet.ln_api_mutex);
2857
2858         CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
2859
2860         if (the_lnet.ln_refcount > 0) {
2861                 rc = the_lnet.ln_refcount++;
2862                 mutex_unlock(&the_lnet.ln_api_mutex);
2863                 return rc;
2864         }
2865
2866         rc = lnet_prepare(requested_pid);
2867         if (rc != 0) {
2868                 mutex_unlock(&the_lnet.ln_api_mutex);
2869                 return rc;
2870         }
2871
2872         /* create a network for Loopback network */
2873         net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
2874         if (net == NULL) {
2875                 rc = -ENOMEM;
2876                 goto err_empty_list;
2877         }
2878
2879         /* Add in the loopback NI */
2880         if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
2881                 rc = -ENOMEM;
2882                 goto err_empty_list;
2883         }
2884
2885         if (use_tcp_bonding)
2886                 CWARN("use_tcp_bonding has been removed. Use Multi-Rail and Dynamic Discovery instead, see LU-13641\n");
2887
2888         /* If LNet is being initialized via DLC it is possible
2889          * that the user requests not to load module parameters (ones which
2890          * are supported by DLC) on initialization.  Therefore, make sure not
2891          * to load networks, routes and forwarding from module parameters
2892          * in this case.  On cleanup in case of failure only clean up
2893          * routes if it has been loaded */
2894         if (!the_lnet.ln_nis_from_mod_params) {
2895                 rc = lnet_parse_networks(&net_head, lnet_get_networks());
2896                 if (rc < 0)
2897                         goto err_empty_list;
2898         }
2899
2900         ni_count = lnet_startup_lndnets(&net_head);
2901         if (ni_count < 0) {
2902                 rc = ni_count;
2903                 goto err_empty_list;
2904         }
2905
2906         if (!the_lnet.ln_nis_from_mod_params) {
2907                 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
2908                 if (rc != 0)
2909                         goto err_shutdown_lndnis;
2910
2911                 rc = lnet_rtrpools_alloc(im_a_router);
2912                 if (rc != 0)
2913                         goto err_destroy_routes;
2914         }
2915
2916         rc = lnet_acceptor_start();
2917         if (rc != 0)
2918                 goto err_destroy_routes;
2919
2920         the_lnet.ln_refcount = 1;
2921         /* Now I may use my own API functions... */
2922
2923         rc = lnet_ping_target_setup(&pbuf, &ping_mdh, ni_count, true);
2924         if (rc != 0)
2925                 goto err_acceptor_stop;
2926
2927         lnet_ping_target_update(pbuf, ping_mdh);
2928
2929         the_lnet.ln_mt_handler = lnet_mt_event_handler;
2930
2931         rc = lnet_push_target_init();
2932         if (rc != 0)
2933                 goto err_stop_ping;
2934
2935         rc = lnet_peer_discovery_start();
2936         if (rc != 0)
2937                 goto err_destroy_push_target;
2938
2939         rc = lnet_monitor_thr_start();
2940         if (rc != 0)
2941                 goto err_stop_discovery_thr;
2942
2943         lnet_fault_init();
2944         lnet_router_debugfs_init();
2945
2946         mutex_unlock(&the_lnet.ln_api_mutex);
2947
2948         complete_all(&the_lnet.ln_started);
2949
2950         /* wait for all routers to start */
2951         lnet_wait_router_start();
2952
2953         return 0;
2954
2955 err_stop_discovery_thr:
2956         lnet_peer_discovery_stop();
2957 err_destroy_push_target:
2958         lnet_push_target_fini();
2959 err_stop_ping:
2960         lnet_ping_target_fini();
2961 err_acceptor_stop:
2962         the_lnet.ln_refcount = 0;
2963         lnet_acceptor_stop();
2964 err_destroy_routes:
2965         if (!the_lnet.ln_nis_from_mod_params)
2966                 lnet_destroy_routes();
2967 err_shutdown_lndnis:
2968         lnet_shutdown_lndnets();
2969 err_empty_list:
2970         lnet_unprepare();
2971         LASSERT(rc < 0);
2972         mutex_unlock(&the_lnet.ln_api_mutex);
2973         while (!list_empty(&net_head)) {
2974                 struct lnet_net *net;
2975
2976                 net = list_entry(net_head.next, struct lnet_net, net_list);
2977                 list_del_init(&net->net_list);
2978                 lnet_net_free(net);
2979         }
2980         return rc;
2981 }
2982 EXPORT_SYMBOL(LNetNIInit);
2983
2984 /**
2985  * Stop LNet interfaces, routing, and forwarding.
2986  *
2987  * Users must call this function once for each successful call to LNetNIInit().
2988  * Once the LNetNIFini() operation has been started, the results of pending
2989  * API operations are undefined.
2990  *
2991  * \return always 0 for current implementation.
2992  */
2993 int
2994 LNetNIFini(void)
2995 {
2996         mutex_lock(&the_lnet.ln_api_mutex);
2997
2998         LASSERT(the_lnet.ln_refcount > 0);
2999
3000         if (the_lnet.ln_refcount != 1) {
3001                 the_lnet.ln_refcount--;
3002         } else {
3003                 LASSERT(!the_lnet.ln_niinit_self);
3004
3005                 lnet_fault_fini();
3006
3007                 lnet_router_debugfs_fini();
3008                 lnet_monitor_thr_stop();
3009                 lnet_peer_discovery_stop();
3010                 lnet_push_target_fini();
3011                 lnet_ping_target_fini();
3012
3013                 /* Teardown fns that use my own API functions BEFORE here */
3014                 the_lnet.ln_refcount = 0;
3015
3016                 lnet_acceptor_stop();
3017                 lnet_destroy_routes();
3018                 lnet_shutdown_lndnets();
3019                 lnet_unprepare();
3020         }
3021
3022         mutex_unlock(&the_lnet.ln_api_mutex);
3023         return 0;
3024 }
3025 EXPORT_SYMBOL(LNetNIFini);
3026
3027 /**
3028  * Grabs the ni data from the ni structure and fills the out
3029  * parameters
3030  *
3031  * \param[in] ni network        interface structure
3032  * \param[out] cfg_ni           NI config information
3033  * \param[out] tun              network and LND tunables
3034  */
3035 static void
3036 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
3037                    struct lnet_ioctl_config_lnd_tunables *tun,
3038                    struct lnet_ioctl_element_stats *stats,
3039                    __u32 tun_size)
3040 {
3041         size_t min_size = 0;
3042         int i;
3043
3044         if (!ni || !cfg_ni || !tun || !nid_is_nid4(&ni->ni_nid))
3045                 return;
3046
3047         if (ni->ni_interface != NULL) {
3048                 strncpy(cfg_ni->lic_ni_intf,
3049                         ni->ni_interface,
3050                         sizeof(cfg_ni->lic_ni_intf));
3051         }
3052
3053         cfg_ni->lic_nid = lnet_nid_to_nid4(&ni->ni_nid);
3054         cfg_ni->lic_status = lnet_ni_get_status_locked(ni);
3055         cfg_ni->lic_dev_cpt = ni->ni_dev_cpt;
3056
3057         memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
3058
3059         if (stats) {
3060                 stats->iel_send_count = lnet_sum_stats(&ni->ni_stats,
3061                                                        LNET_STATS_TYPE_SEND);
3062                 stats->iel_recv_count = lnet_sum_stats(&ni->ni_stats,
3063                                                        LNET_STATS_TYPE_RECV);
3064                 stats->iel_drop_count = lnet_sum_stats(&ni->ni_stats,
3065                                                        LNET_STATS_TYPE_DROP);
3066         }
3067
3068         /*
3069          * tun->lt_tun will always be present, but in order to be
3070          * backwards compatible, we need to deal with the cases when
3071          * tun->lt_tun is smaller than what the kernel has, because it
3072          * comes from an older version of a userspace program, then we'll
3073          * need to copy as much information as we have available space.
3074          */
3075         min_size = tun_size - sizeof(tun->lt_cmn);
3076         memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
3077
3078         /* copy over the cpts */
3079         if (ni->ni_ncpts == LNET_CPT_NUMBER &&
3080             ni->ni_cpts == NULL)  {
3081                 for (i = 0; i < ni->ni_ncpts; i++)
3082                         cfg_ni->lic_cpts[i] = i;
3083         } else {
3084                 for (i = 0;
3085                      ni->ni_cpts != NULL && i < ni->ni_ncpts &&
3086                      i < LNET_MAX_SHOW_NUM_CPT;
3087                      i++)
3088                         cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
3089         }
3090         cfg_ni->lic_ncpts = ni->ni_ncpts;
3091 }
3092
3093 /**
3094  * NOTE: This is a legacy function left in the code to be backwards
3095  * compatible with older userspace programs. It should eventually be
3096  * removed.
3097  *
3098  * Grabs the ni data from the ni structure and fills the out
3099  * parameters
3100  *
3101  * \param[in] ni network        interface structure
3102  * \param[out] config           config information
3103  */
3104 static void
3105 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
3106                          struct lnet_ioctl_config_data *config)
3107 {
3108         struct lnet_ioctl_net_config *net_config;
3109         struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
3110         size_t min_size, tunable_size = 0;
3111         int i;
3112
3113         if (!ni || !config || !nid_is_nid4(&ni->ni_nid))
3114                 return;
3115
3116         net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
3117         if (!net_config)
3118                 return;
3119
3120         if (!ni->ni_interface)
3121                 return;
3122
3123         strncpy(net_config->ni_interface,
3124                 ni->ni_interface,
3125                 sizeof(net_config->ni_interface));
3126
3127         config->cfg_nid = lnet_nid_to_nid4(&ni->ni_nid);
3128         config->cfg_config_u.cfg_net.net_peer_timeout =
3129                 ni->ni_net->net_tunables.lct_peer_timeout;
3130         config->cfg_config_u.cfg_net.net_max_tx_credits =
3131                 ni->ni_net->net_tunables.lct_max_tx_credits;
3132         config->cfg_config_u.cfg_net.net_peer_tx_credits =
3133                 ni->ni_net->net_tunables.lct_peer_tx_credits;
3134         config->cfg_config_u.cfg_net.net_peer_rtr_credits =
3135                 ni->ni_net->net_tunables.lct_peer_rtr_credits;
3136
3137         net_config->ni_status = lnet_ni_get_status_locked(ni);
3138
3139         if (ni->ni_cpts) {
3140                 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
3141
3142                 for (i = 0; i < num_cpts; i++)
3143                         net_config->ni_cpts[i] = ni->ni_cpts[i];
3144
3145                 config->cfg_ncpts = num_cpts;
3146         }
3147
3148         /*
3149          * See if user land tools sent in a newer and larger version
3150          * of struct lnet_tunables than what the kernel uses.
3151          */
3152         min_size = sizeof(*config) + sizeof(*net_config);
3153
3154         if (config->cfg_hdr.ioc_len > min_size)
3155                 tunable_size = config->cfg_hdr.ioc_len - min_size;
3156
3157         /* Don't copy too much data to user space */
3158         min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
3159         lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
3160
3161         if (lnd_cfg && min_size) {
3162                 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
3163                 config->cfg_config_u.cfg_net.net_interface_count = 1;
3164
3165                 /* Tell user land that kernel side has less data */
3166                 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
3167                         min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
3168                         config->cfg_hdr.ioc_len -= min_size;
3169                 }
3170         }
3171 }
3172
3173 struct lnet_ni *
3174 lnet_get_ni_idx_locked(int idx)
3175 {
3176         struct lnet_ni          *ni;
3177         struct lnet_net         *net;
3178
3179         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3180                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3181                         if (idx-- == 0)
3182                                 return ni;
3183                 }
3184         }
3185
3186         return NULL;
3187 }
3188
3189 int lnet_get_net_healthv_locked(struct lnet_net *net)
3190 {
3191         struct lnet_ni *ni;
3192         int best_healthv = 0;
3193         int healthv, ni_fatal;
3194
3195         list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3196                 healthv = atomic_read(&ni->ni_healthv);
3197                 ni_fatal = atomic_read(&ni->ni_fatal_error_on);
3198                 if (!ni_fatal && healthv > best_healthv)
3199                         best_healthv = healthv;
3200         }
3201
3202         return best_healthv;
3203 }
3204
3205 struct lnet_ni *
3206 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
3207 {
3208         struct lnet_ni          *ni;
3209         struct lnet_net         *net = mynet;
3210
3211         /*
3212          * It is possible that the net has been cleaned out while there is
3213          * a message being sent. This function accessed the net without
3214          * checking if the list is empty
3215          */
3216         if (prev == NULL) {
3217                 if (net == NULL)
3218                         net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
3219                                         net_list);
3220                 if (list_empty(&net->net_ni_list))
3221                         return NULL;
3222                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3223                                 ni_netlist);
3224
3225                 return ni;
3226         }
3227
3228         if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
3229                 /* if you reached the end of the ni list and the net is
3230                  * specified, then there are no more nis in that net */
3231                 if (net != NULL)
3232                         return NULL;
3233
3234                 /* we reached the end of this net ni list. move to the
3235                  * next net */
3236                 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
3237                         /* no more nets and no more NIs. */
3238                         return NULL;
3239
3240                 /* get the next net */
3241                 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
3242                                  net_list);
3243                 if (list_empty(&net->net_ni_list))
3244                         return NULL;
3245                 /* get the ni on it */
3246                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
3247                                 ni_netlist);
3248
3249                 return ni;
3250         }
3251
3252         if (list_empty(&prev->ni_netlist))
3253                 return NULL;
3254
3255         /* there are more nis left */
3256         ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
3257
3258         return ni;
3259 }
3260
3261 int
3262 lnet_get_net_config(struct lnet_ioctl_config_data *config)
3263 {
3264         struct lnet_ni *ni;
3265         int cpt;
3266         int rc = -ENOENT;
3267         int idx = config->cfg_count;
3268
3269         cpt = lnet_net_lock_current();
3270
3271         ni = lnet_get_ni_idx_locked(idx);
3272
3273         if (ni != NULL) {
3274                 rc = 0;
3275                 lnet_ni_lock(ni);
3276                 lnet_fill_ni_info_legacy(ni, config);
3277                 lnet_ni_unlock(ni);
3278         }
3279
3280         lnet_net_unlock(cpt);
3281         return rc;
3282 }
3283
3284 int
3285 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
3286                    struct lnet_ioctl_config_lnd_tunables *tun,
3287                    struct lnet_ioctl_element_stats *stats,
3288                    __u32 tun_size)
3289 {
3290         struct lnet_ni          *ni;
3291         int                     cpt;
3292         int                     rc = -ENOENT;
3293
3294         if (!cfg_ni || !tun || !stats)
3295                 return -EINVAL;
3296
3297         cpt = lnet_net_lock_current();
3298
3299         ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
3300
3301         if (ni) {
3302                 rc = 0;
3303                 lnet_ni_lock(ni);
3304                 lnet_fill_ni_info(ni, cfg_ni, tun, stats, tun_size);
3305                 lnet_ni_unlock(ni);
3306         }
3307
3308         lnet_net_unlock(cpt);
3309         return rc;
3310 }
3311
3312 int lnet_get_ni_stats(struct lnet_ioctl_element_msg_stats *msg_stats)
3313 {
3314         struct lnet_ni *ni;
3315         int cpt;
3316         int rc = -ENOENT;
3317
3318         if (!msg_stats)
3319                 return -EINVAL;
3320
3321         cpt = lnet_net_lock_current();
3322
3323         ni = lnet_get_ni_idx_locked(msg_stats->im_idx);
3324
3325         if (ni) {
3326                 lnet_usr_translate_stats(msg_stats, &ni->ni_stats);
3327                 rc = 0;
3328         }
3329
3330         lnet_net_unlock(cpt);
3331
3332         return rc;
3333 }
3334
3335 static int lnet_add_net_common(struct lnet_net *net,
3336                                struct lnet_ioctl_config_lnd_tunables *tun)
3337 {
3338         struct lnet_handle_md ping_mdh;
3339         struct lnet_ping_buffer *pbuf;
3340         struct lnet_remotenet *rnet;
3341         struct lnet_ni *ni;
3342         int net_ni_count;
3343         __u32 net_id;
3344         int rc;
3345
3346         lnet_net_lock(LNET_LOCK_EX);
3347         rnet = lnet_find_rnet_locked(net->net_id);
3348         lnet_net_unlock(LNET_LOCK_EX);
3349         /*
3350          * make sure that the net added doesn't invalidate the current
3351          * configuration LNet is keeping
3352          */
3353         if (rnet) {
3354                 CERROR("Adding net %s will invalidate routing configuration\n",
3355                        libcfs_net2str(net->net_id));
3356                 lnet_net_free(net);
3357                 return -EUSERS;
3358         }
3359
3360         /*
3361          * make sure you calculate the correct number of slots in the ping
3362          * buffer. Since the ping info is a flattened list of all the NIs,
3363          * we should allocate enough slots to accomodate the number of NIs
3364          * which will be added.
3365          *
3366          * since ni hasn't been configured yet, use
3367          * lnet_get_net_ni_count_pre() which checks the net_ni_added list
3368          */
3369         net_ni_count = lnet_get_net_ni_count_pre(net);
3370
3371         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3372                                     net_ni_count + lnet_get_ni_count(),
3373                                     false);
3374         if (rc < 0) {
3375                 lnet_net_free(net);
3376                 return rc;
3377         }
3378
3379         if (tun)
3380                 memcpy(&net->net_tunables,
3381                        &tun->lt_cmn, sizeof(net->net_tunables));
3382         else
3383                 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
3384
3385         net_id = net->net_id;
3386
3387         rc = lnet_startup_lndnet(net,
3388                                  (tun) ? &tun->lt_tun : NULL);
3389         if (rc < 0)
3390                 goto failed;
3391
3392         lnet_net_lock(LNET_LOCK_EX);
3393         net = lnet_get_net_locked(net_id);
3394         LASSERT(net);
3395
3396         /* apply the UDSPs */
3397         rc = lnet_udsp_apply_policies_on_net(net);
3398         if (rc)
3399                 CERROR("Failed to apply UDSPs on local net %s\n",
3400                        libcfs_net2str(net->net_id));
3401
3402         /* At this point we lost track of which NI was just added, so we
3403          * just re-apply the policies on all of the NIs on this net
3404          */
3405         list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3406                 rc = lnet_udsp_apply_policies_on_ni(ni);
3407                 if (rc)
3408                         CERROR("Failed to apply UDSPs on ni %s\n",
3409                                libcfs_nidstr(&ni->ni_nid));
3410         }
3411         lnet_net_unlock(LNET_LOCK_EX);
3412
3413         /*
3414          * Start the acceptor thread if this is the first network
3415          * being added that requires the thread.
3416          */
3417         if (net->net_lnd->lnd_accept) {
3418                 rc = lnet_acceptor_start();
3419                 if (rc < 0) {
3420                         /* shutdown the net that we just started */
3421                         CERROR("Failed to start up acceptor thread\n");
3422                         lnet_shutdown_lndnet(net);
3423                         goto failed;
3424                 }
3425         }
3426
3427         lnet_net_lock(LNET_LOCK_EX);
3428         lnet_peer_net_added(net);
3429         lnet_net_unlock(LNET_LOCK_EX);
3430
3431         lnet_ping_target_update(pbuf, ping_mdh);
3432
3433         return 0;
3434
3435 failed:
3436         lnet_ping_md_unlink(pbuf, &ping_mdh);
3437         lnet_ping_buffer_decref(pbuf);
3438         return rc;
3439 }
3440
3441 static void
3442 lnet_set_tune_defaults(struct lnet_ioctl_config_lnd_tunables *tun)
3443 {
3444         if (tun) {
3445                 if (!tun->lt_cmn.lct_peer_timeout)
3446                         tun->lt_cmn.lct_peer_timeout = DEFAULT_PEER_TIMEOUT;
3447                 if (!tun->lt_cmn.lct_peer_tx_credits)
3448                         tun->lt_cmn.lct_peer_tx_credits = DEFAULT_PEER_CREDITS;
3449                 if (!tun->lt_cmn.lct_max_tx_credits)
3450                         tun->lt_cmn.lct_max_tx_credits = DEFAULT_CREDITS;
3451         }
3452 }
3453
3454 static int lnet_handle_legacy_ip2nets(char *ip2nets,
3455                                       struct lnet_ioctl_config_lnd_tunables *tun)
3456 {
3457         struct lnet_net *net;
3458         const char *nets;
3459         int rc;
3460         LIST_HEAD(net_head);
3461
3462         rc = lnet_parse_ip2nets(&nets, ip2nets);
3463         if (rc < 0)
3464                 return rc;
3465
3466         rc = lnet_parse_networks(&net_head, nets);
3467         if (rc < 0)
3468                 return rc;
3469
3470         lnet_set_tune_defaults(tun);
3471
3472         mutex_lock(&the_lnet.ln_api_mutex);
3473         while (!list_empty(&net_head)) {
3474                 net = list_entry(net_head.next, struct lnet_net, net_list);
3475                 list_del_init(&net->net_list);
3476                 rc = lnet_add_net_common(net, tun);
3477                 if (rc < 0)
3478                         goto out;
3479         }
3480
3481 out:
3482         mutex_unlock(&the_lnet.ln_api_mutex);
3483
3484         while (!list_empty(&net_head)) {
3485                 net = list_entry(net_head.next, struct lnet_net, net_list);
3486                 list_del_init(&net->net_list);
3487                 lnet_net_free(net);
3488         }
3489         return rc;
3490 }
3491
3492 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
3493 {
3494         struct lnet_net *net;
3495         struct lnet_ni *ni;
3496         struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3497         int rc, i;
3498         __u32 net_id, lnd_type;
3499
3500         /* get the tunables if they are available */
3501         if (conf->lic_cfg_hdr.ioc_len >=
3502             sizeof(*conf) + sizeof(*tun))
3503                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3504                         conf->lic_bulk;
3505
3506         /* handle legacy ip2nets from DLC */
3507         if (conf->lic_legacy_ip2nets[0] != '\0')
3508                 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
3509                                                   tun);
3510
3511         net_id = LNET_NIDNET(conf->lic_nid);
3512         lnd_type = LNET_NETTYP(net_id);
3513
3514         if (!libcfs_isknown_lnd(lnd_type)) {
3515                 CERROR("No valid net and lnd information provided\n");
3516                 return -EINVAL;
3517         }
3518
3519         net = lnet_net_alloc(net_id, NULL);
3520         if (!net)
3521                 return -ENOMEM;
3522
3523         for (i = 0; i < conf->lic_ncpts; i++) {
3524                 if (conf->lic_cpts[i] >= LNET_CPT_NUMBER)
3525                         return -EINVAL;
3526         }
3527
3528         ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
3529                                        conf->lic_ni_intf);
3530         if (!ni)
3531                 return -ENOMEM;
3532
3533         lnet_set_tune_defaults(tun);
3534
3535         mutex_lock(&the_lnet.ln_api_mutex);
3536
3537         rc = lnet_add_net_common(net, tun);
3538
3539         mutex_unlock(&the_lnet.ln_api_mutex);
3540
3541         return rc;
3542 }
3543
3544 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
3545 {
3546         struct lnet_net  *net;
3547         struct lnet_ni *ni;
3548         __u32 net_id = LNET_NIDNET(conf->lic_nid);
3549         struct lnet_ping_buffer *pbuf;
3550         struct lnet_handle_md  ping_mdh;
3551         int               rc;
3552         int               net_count;
3553         __u32             addr;
3554
3555         /* don't allow userspace to shutdown the LOLND */
3556         if (LNET_NETTYP(net_id) == LOLND)
3557                 return -EINVAL;
3558
3559         mutex_lock(&the_lnet.ln_api_mutex);
3560
3561         lnet_net_lock(0);
3562
3563         net = lnet_get_net_locked(net_id);
3564         if (!net) {
3565                 CERROR("net %s not found\n",
3566                        libcfs_net2str(net_id));
3567                 rc = -ENOENT;
3568                 goto unlock_net;
3569         }
3570
3571         addr = LNET_NIDADDR(conf->lic_nid);
3572         if (addr == 0) {
3573                 /* remove the entire net */
3574                 net_count = lnet_get_net_ni_count_locked(net);
3575
3576                 lnet_net_unlock(0);
3577
3578                 /* create and link a new ping info, before removing the old one */
3579                 rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3580                                         lnet_get_ni_count() - net_count,
3581                                         false);
3582                 if (rc != 0)
3583                         goto unlock_api_mutex;
3584
3585                 lnet_shutdown_lndnet(net);
3586
3587                 lnet_acceptor_stop();
3588
3589                 lnet_ping_target_update(pbuf, ping_mdh);
3590
3591                 goto unlock_api_mutex;
3592         }
3593
3594         ni = lnet_nid2ni_locked(conf->lic_nid, 0);
3595         if (!ni) {
3596                 CERROR("nid %s not found\n",
3597                        libcfs_nid2str(conf->lic_nid));
3598                 rc = -ENOENT;
3599                 goto unlock_net;
3600         }
3601
3602         net_count = lnet_get_net_ni_count_locked(net);
3603
3604         lnet_net_unlock(0);
3605
3606         /* create and link a new ping info, before removing the old one */
3607         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3608                                   lnet_get_ni_count() - 1, false);
3609         if (rc != 0)
3610                 goto unlock_api_mutex;
3611
3612         lnet_shutdown_lndni(ni);
3613
3614         lnet_acceptor_stop();
3615
3616         lnet_ping_target_update(pbuf, ping_mdh);
3617
3618         /* check if the net is empty and remove it if it is */
3619         if (net_count == 1)
3620                 lnet_shutdown_lndnet(net);
3621
3622         goto unlock_api_mutex;
3623
3624 unlock_net:
3625         lnet_net_unlock(0);
3626 unlock_api_mutex:
3627         mutex_unlock(&the_lnet.ln_api_mutex);
3628
3629         return rc;
3630 }
3631
3632 /*
3633  * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
3634  * They are only expected to be called for unique networks.
3635  * That can be as a result of older DLC library
3636  * calls. Multi-Rail DLC and beyond no longer uses these APIs.
3637  */
3638 int
3639 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
3640 {
3641         struct lnet_net *net;
3642         LIST_HEAD(net_head);
3643         int rc;
3644         struct lnet_ioctl_config_lnd_tunables tun;
3645         const char *nets = conf->cfg_config_u.cfg_net.net_intf;
3646
3647         /* Create a net/ni structures for the network string */
3648         rc = lnet_parse_networks(&net_head, nets);
3649         if (rc <= 0)
3650                 return rc == 0 ? -EINVAL : rc;
3651
3652         mutex_lock(&the_lnet.ln_api_mutex);
3653
3654         if (rc > 1) {
3655                 rc = -EINVAL; /* only add one network per call */
3656                 goto out_unlock_clean;
3657         }
3658
3659         net = list_entry(net_head.next, struct lnet_net, net_list);
3660         list_del_init(&net->net_list);
3661
3662         LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
3663
3664         memset(&tun, 0, sizeof(tun));
3665
3666         tun.lt_cmn.lct_peer_timeout =
3667           (!conf->cfg_config_u.cfg_net.net_peer_timeout) ? DEFAULT_PEER_TIMEOUT :
3668                 conf->cfg_config_u.cfg_net.net_peer_timeout;
3669         tun.lt_cmn.lct_peer_tx_credits =
3670           (!conf->cfg_config_u.cfg_net.net_peer_tx_credits) ? DEFAULT_PEER_CREDITS :
3671                 conf->cfg_config_u.cfg_net.net_peer_tx_credits;
3672         tun.lt_cmn.lct_peer_rtr_credits =
3673           conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
3674         tun.lt_cmn.lct_max_tx_credits =
3675           (!conf->cfg_config_u.cfg_net.net_max_tx_credits) ? DEFAULT_CREDITS :
3676                 conf->cfg_config_u.cfg_net.net_max_tx_credits;
3677
3678         rc = lnet_add_net_common(net, &tun);
3679
3680 out_unlock_clean:
3681         mutex_unlock(&the_lnet.ln_api_mutex);
3682         while (!list_empty(&net_head)) {
3683                 /* net_head list is empty in success case */
3684                 net = list_entry(net_head.next, struct lnet_net, net_list);
3685                 list_del_init(&net->net_list);
3686                 lnet_net_free(net);
3687         }
3688         return rc;
3689 }
3690
3691 int
3692 lnet_dyn_del_net(__u32 net_id)
3693 {
3694         struct lnet_net  *net;
3695         struct lnet_ping_buffer *pbuf;
3696         struct lnet_handle_md ping_mdh;
3697         int               rc;
3698         int               net_ni_count;
3699
3700         /* don't allow userspace to shutdown the LOLND */
3701         if (LNET_NETTYP(net_id) == LOLND)
3702                 return -EINVAL;
3703
3704         mutex_lock(&the_lnet.ln_api_mutex);
3705
3706         lnet_net_lock(0);
3707
3708         net = lnet_get_net_locked(net_id);
3709         if (net == NULL) {
3710                 lnet_net_unlock(0);
3711                 rc = -EINVAL;
3712                 goto out;
3713         }
3714
3715         net_ni_count = lnet_get_net_ni_count_locked(net);
3716
3717         lnet_net_unlock(0);
3718
3719         /* create and link a new ping info, before removing the old one */
3720         rc = lnet_ping_target_setup(&pbuf, &ping_mdh,
3721                                     lnet_get_ni_count() - net_ni_count, false);
3722         if (rc != 0)
3723                 goto out;
3724
3725         lnet_shutdown_lndnet(net);
3726
3727         lnet_acceptor_stop();
3728
3729         lnet_ping_target_update(pbuf, ping_mdh);
3730
3731 out:
3732         mutex_unlock(&the_lnet.ln_api_mutex);
3733
3734         return rc;
3735 }
3736
3737 void lnet_incr_dlc_seq(void)
3738 {
3739         atomic_inc(&lnet_dlc_seq_no);
3740 }
3741
3742 __u32 lnet_get_dlc_seq_locked(void)
3743 {
3744         return atomic_read(&lnet_dlc_seq_no);
3745 }
3746
3747 static void
3748 lnet_ni_set_healthv(lnet_nid_t nid, int value, bool all)
3749 {
3750         struct lnet_net *net;
3751         struct lnet_ni *ni;
3752
3753         lnet_net_lock(LNET_LOCK_EX);
3754         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3755                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3756                         if (all || (nid_is_nid4(&ni->ni_nid) &&
3757                                     lnet_nid_to_nid4(&ni->ni_nid) == nid)) {
3758                                 atomic_set(&ni->ni_healthv, value);
3759                                 if (list_empty(&ni->ni_recovery) &&
3760                                     value < LNET_MAX_HEALTH_VALUE) {
3761                                         CERROR("manually adding local NI %s to recovery\n",
3762                                                libcfs_nidstr(&ni->ni_nid));
3763                                         list_add_tail(&ni->ni_recovery,
3764                                                       &the_lnet.ln_mt_localNIRecovq);
3765                                         lnet_ni_addref_locked(ni, 0);
3766                                 }
3767                                 if (!all) {
3768                                         lnet_net_unlock(LNET_LOCK_EX);
3769                                         return;
3770                                 }
3771                         }
3772                 }
3773         }
3774         lnet_net_unlock(LNET_LOCK_EX);
3775 }
3776
3777 static void
3778 lnet_ni_set_conns_per_peer(lnet_nid_t nid, int value, bool all)
3779 {
3780         struct lnet_net *net;
3781         struct lnet_ni *ni;
3782
3783         lnet_net_lock(LNET_LOCK_EX);
3784         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
3785                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
3786                         if (lnet_nid_to_nid4(&ni->ni_nid) != nid && !all)
3787                                 continue;
3788                         if (LNET_NETTYP(net->net_id) == SOCKLND)
3789                                 ni->ni_lnd_tunables.lnd_tun_u.lnd_sock.lnd_conns_per_peer = value;
3790                         else if (LNET_NETTYP(net->net_id) == O2IBLND)
3791                                 ni->ni_lnd_tunables.lnd_tun_u.lnd_o2ib.lnd_conns_per_peer = value;
3792                         if (!all) {
3793                                 lnet_net_unlock(LNET_LOCK_EX);
3794                                 return;
3795                         }
3796                 }
3797         }
3798         lnet_net_unlock(LNET_LOCK_EX);
3799 }
3800
3801 static int
3802 lnet_get_local_ni_hstats(struct lnet_ioctl_local_ni_hstats *stats)
3803 {
3804         int cpt, rc = 0;
3805         struct lnet_ni *ni;
3806         lnet_nid_t nid = stats->hlni_nid;
3807
3808         cpt = lnet_net_lock_current();
3809         ni = lnet_nid2ni_locked(nid, cpt);
3810
3811         if (!ni) {
3812                 rc = -ENOENT;
3813                 goto unlock;
3814         }
3815
3816         stats->hlni_local_interrupt = atomic_read(&ni->ni_hstats.hlt_local_interrupt);
3817         stats->hlni_local_dropped = atomic_read(&ni->ni_hstats.hlt_local_dropped);
3818         stats->hlni_local_aborted = atomic_read(&ni->ni_hstats.hlt_local_aborted);
3819         stats->hlni_local_no_route = atomic_read(&ni->ni_hstats.hlt_local_no_route);
3820         stats->hlni_local_timeout = atomic_read(&ni->ni_hstats.hlt_local_timeout);
3821         stats->hlni_local_error = atomic_read(&ni->ni_hstats.hlt_local_error);
3822         stats->hlni_fatal_error = atomic_read(&ni->ni_fatal_error_on);
3823         stats->hlni_health_value = atomic_read(&ni->ni_healthv);
3824         stats->hlni_ping_count = ni->ni_ping_count;
3825         stats->hlni_next_ping = ni->ni_next_ping;
3826
3827 unlock:
3828         lnet_net_unlock(cpt);
3829
3830         return rc;
3831 }
3832
3833 static int
3834 lnet_get_local_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3835 {
3836         struct lnet_ni *ni;
3837         int i = 0;
3838
3839         lnet_net_lock(LNET_LOCK_EX);
3840         list_for_each_entry(ni, &the_lnet.ln_mt_localNIRecovq, ni_recovery) {
3841                 if (!nid_is_nid4(&ni->ni_nid))
3842                         continue;
3843                 list->rlst_nid_array[i] = lnet_nid_to_nid4(&ni->ni_nid);
3844                 i++;
3845                 if (i >= LNET_MAX_SHOW_NUM_NID)
3846                         break;
3847         }
3848         lnet_net_unlock(LNET_LOCK_EX);
3849         list->rlst_num_nids = i;
3850
3851         return 0;
3852 }
3853
3854 static int
3855 lnet_get_peer_ni_recovery_list(struct lnet_ioctl_recovery_list *list)
3856 {
3857         struct lnet_peer_ni *lpni;
3858         int i = 0;
3859
3860         lnet_net_lock(LNET_LOCK_EX);
3861         list_for_each_entry(lpni, &the_lnet.ln_mt_peerNIRecovq, lpni_recovery) {
3862                 list->rlst_nid_array[i] = lnet_nid_to_nid4(&lpni->lpni_nid);
3863                 i++;
3864                 if (i >= LNET_MAX_SHOW_NUM_NID)
3865                         break;
3866         }
3867         lnet_net_unlock(LNET_LOCK_EX);
3868         list->rlst_num_nids = i;
3869
3870         return 0;
3871 }
3872
3873 /**
3874  * LNet ioctl handler.
3875  *
3876  */
3877 int
3878 LNetCtl(unsigned int cmd, void *arg)
3879 {
3880         struct libcfs_ioctl_data *data = arg;
3881         struct lnet_ioctl_config_data *config;
3882         struct lnet_process_id    id = {0};
3883         struct lnet_ni           *ni;
3884         int                       rc;
3885
3886         BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
3887                      sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
3888
3889         switch (cmd) {
3890         case IOC_LIBCFS_GET_NI:
3891                 rc = LNetGetId(data->ioc_count, &id);
3892                 data->ioc_nid = id.nid;
3893                 return rc;
3894
3895         case IOC_LIBCFS_FAIL_NID:
3896                 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
3897
3898         case IOC_LIBCFS_ADD_ROUTE: {
3899                 /* default router sensitivity to 1 */
3900                 unsigned int sensitivity = 1;
3901                 config = arg;
3902
3903                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3904                         return -EINVAL;
3905
3906                 if (config->cfg_config_u.cfg_route.rtr_sensitivity) {
3907                         sensitivity =
3908                           config->cfg_config_u.cfg_route.rtr_sensitivity;
3909                 }
3910
3911                 mutex_lock(&the_lnet.ln_api_mutex);
3912                 rc = lnet_add_route(config->cfg_net,
3913                                     config->cfg_config_u.cfg_route.rtr_hop,
3914                                     config->cfg_nid,
3915                                     config->cfg_config_u.cfg_route.
3916                                         rtr_priority, sensitivity);
3917                 mutex_unlock(&the_lnet.ln_api_mutex);
3918                 return rc;
3919         }
3920
3921         case IOC_LIBCFS_DEL_ROUTE:
3922                 config = arg;
3923
3924                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3925                         return -EINVAL;
3926
3927                 mutex_lock(&the_lnet.ln_api_mutex);
3928                 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
3929                 mutex_unlock(&the_lnet.ln_api_mutex);
3930                 return rc;
3931
3932         case IOC_LIBCFS_GET_ROUTE:
3933                 config = arg;
3934
3935                 if (config->cfg_hdr.ioc_len < sizeof(*config))
3936                         return -EINVAL;
3937
3938                 mutex_lock(&the_lnet.ln_api_mutex);
3939                 rc = lnet_get_route(config->cfg_count,
3940                                     &config->cfg_net,
3941                                     &config->cfg_config_u.cfg_route.rtr_hop,
3942                                     &config->cfg_nid,
3943                                     &config->cfg_config_u.cfg_route.rtr_flags,
3944                                     &config->cfg_config_u.cfg_route.
3945                                         rtr_priority,
3946                                     &config->cfg_config_u.cfg_route.
3947                                         rtr_sensitivity);
3948                 mutex_unlock(&the_lnet.ln_api_mutex);
3949                 return rc;
3950
3951         case IOC_LIBCFS_GET_LOCAL_NI: {
3952                 struct lnet_ioctl_config_ni *cfg_ni;
3953                 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
3954                 struct lnet_ioctl_element_stats *stats;
3955                 __u32 tun_size;
3956
3957                 cfg_ni = arg;
3958
3959                 /* get the tunables if they are available */
3960                 if (cfg_ni->lic_cfg_hdr.ioc_len <
3961                     sizeof(*cfg_ni) + sizeof(*stats) + sizeof(*tun))
3962                         return -EINVAL;
3963
3964                 stats = (struct lnet_ioctl_element_stats *)
3965                         cfg_ni->lic_bulk;
3966                 tun = (struct lnet_ioctl_config_lnd_tunables *)
3967                                 (cfg_ni->lic_bulk + sizeof(*stats));
3968
3969                 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni) -
3970                         sizeof(*stats);
3971
3972                 mutex_lock(&the_lnet.ln_api_mutex);
3973                 rc = lnet_get_ni_config(cfg_ni, tun, stats, tun_size);
3974                 mutex_unlock(&the_lnet.ln_api_mutex);
3975                 return rc;
3976         }
3977
3978         case IOC_LIBCFS_GET_LOCAL_NI_MSG_STATS: {
3979                 struct lnet_ioctl_element_msg_stats *msg_stats = arg;
3980
3981                 if (msg_stats->im_hdr.ioc_len != sizeof(*msg_stats))
3982                         return -EINVAL;
3983
3984                 mutex_lock(&the_lnet.ln_api_mutex);
3985                 rc = lnet_get_ni_stats(msg_stats);
3986                 mutex_unlock(&the_lnet.ln_api_mutex);
3987
3988                 return rc;
3989         }
3990
3991         case IOC_LIBCFS_GET_NET: {
3992                 size_t total = sizeof(*config) +
3993                                sizeof(struct lnet_ioctl_net_config);
3994                 config = arg;
3995
3996                 if (config->cfg_hdr.ioc_len < total)
3997                         return -EINVAL;
3998
3999                 mutex_lock(&the_lnet.ln_api_mutex);
4000                 rc = lnet_get_net_config(config);
4001                 mutex_unlock(&the_lnet.ln_api_mutex);
4002                 return rc;
4003         }
4004
4005         case IOC_LIBCFS_GET_LNET_STATS:
4006         {
4007                 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
4008
4009                 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
4010                         return -EINVAL;
4011
4012                 mutex_lock(&the_lnet.ln_api_mutex);
4013                 rc = lnet_counters_get(&lnet_stats->st_cntrs);
4014                 mutex_unlock(&the_lnet.ln_api_mutex);
4015                 return rc;
4016         }
4017
4018         case IOC_LIBCFS_RESET_LNET_STATS:
4019         {
4020                 mutex_lock(&the_lnet.ln_api_mutex);
4021                 lnet_counters_reset();
4022                 mutex_unlock(&the_lnet.ln_api_mutex);
4023                 return 0;
4024         }
4025
4026         case IOC_LIBCFS_CONFIG_RTR:
4027                 config = arg;
4028
4029                 if (config->cfg_hdr.ioc_len < sizeof(*config))
4030                         return -EINVAL;
4031
4032                 mutex_lock(&the_lnet.ln_api_mutex);
4033                 if (config->cfg_config_u.cfg_buffers.buf_enable) {
4034                         rc = lnet_rtrpools_enable();
4035                         mutex_unlock(&the_lnet.ln_api_mutex);
4036                         return rc;
4037                 }
4038                 lnet_rtrpools_disable();
4039                 mutex_unlock(&the_lnet.ln_api_mutex);
4040                 return 0;
4041
4042         case IOC_LIBCFS_ADD_BUF:
4043                 config = arg;
4044
4045                 if (config->cfg_hdr.ioc_len < sizeof(*config))
4046                         return -EINVAL;
4047
4048                 mutex_lock(&the_lnet.ln_api_mutex);
4049                 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
4050                                                 buf_tiny,
4051                                           config->cfg_config_u.cfg_buffers.
4052                                                 buf_small,
4053                                           config->cfg_config_u.cfg_buffers.
4054                                                 buf_large);
4055                 mutex_unlock(&the_lnet.ln_api_mutex);
4056                 return rc;
4057
4058         case IOC_LIBCFS_SET_NUMA_RANGE: {
4059                 struct lnet_ioctl_set_value *numa;
4060                 numa = arg;
4061                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
4062                         return -EINVAL;
4063                 lnet_net_lock(LNET_LOCK_EX);
4064                 lnet_numa_range = numa->sv_value;
4065                 lnet_net_unlock(LNET_LOCK_EX);
4066                 return 0;
4067         }
4068
4069         case IOC_LIBCFS_GET_NUMA_RANGE: {
4070                 struct lnet_ioctl_set_value *numa;
4071                 numa = arg;
4072                 if (numa->sv_hdr.ioc_len != sizeof(*numa))
4073                         return -EINVAL;
4074                 numa->sv_value = lnet_numa_range;
4075                 return 0;
4076         }
4077
4078         case IOC_LIBCFS_GET_BUF: {
4079                 struct lnet_ioctl_pool_cfg *pool_cfg;
4080                 size_t total = sizeof(*config) + sizeof(*pool_cfg);
4081
4082                 config = arg;
4083
4084                 if (config->cfg_hdr.ioc_len < total)
4085                         return -EINVAL;
4086
4087                 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
4088
4089                 mutex_lock(&the_lnet.ln_api_mutex);
4090                 rc = lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
4091                 mutex_unlock(&the_lnet.ln_api_mutex);
4092                 return rc;
4093         }
4094
4095         case IOC_LIBCFS_GET_LOCAL_HSTATS: {
4096                 struct lnet_ioctl_local_ni_hstats *stats = arg;
4097
4098                 if (stats->hlni_hdr.ioc_len < sizeof(*stats))
4099                         return -EINVAL;
4100
4101                 mutex_lock(&the_lnet.ln_api_mutex);
4102                 rc = lnet_get_local_ni_hstats(stats);
4103                 mutex_unlock(&the_lnet.ln_api_mutex);
4104
4105                 return rc;
4106         }
4107
4108         case IOC_LIBCFS_GET_RECOVERY_QUEUE: {
4109                 struct lnet_ioctl_recovery_list *list = arg;
4110                 if (list->rlst_hdr.ioc_len < sizeof(*list))
4111                         return -EINVAL;
4112
4113                 mutex_lock(&the_lnet.ln_api_mutex);
4114                 if (list->rlst_type == LNET_HEALTH_TYPE_LOCAL_NI)
4115                         rc = lnet_get_local_ni_recovery_list(list);
4116                 else
4117                         rc = lnet_get_peer_ni_recovery_list(list);
4118                 mutex_unlock(&the_lnet.ln_api_mutex);
4119                 return rc;
4120         }
4121
4122         case IOC_LIBCFS_ADD_PEER_NI: {
4123                 struct lnet_ioctl_peer_cfg *cfg = arg;
4124
4125                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4126                         return -EINVAL;
4127
4128                 mutex_lock(&the_lnet.ln_api_mutex);
4129                 rc = lnet_add_peer_ni(cfg->prcfg_prim_nid,
4130                                       cfg->prcfg_cfg_nid,
4131                                       cfg->prcfg_mr, false);
4132                 mutex_unlock(&the_lnet.ln_api_mutex);
4133                 return rc;
4134         }
4135
4136         case IOC_LIBCFS_DEL_PEER_NI: {
4137                 struct lnet_ioctl_peer_cfg *cfg = arg;
4138
4139                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4140                         return -EINVAL;
4141
4142                 mutex_lock(&the_lnet.ln_api_mutex);
4143                 rc = lnet_del_peer_ni(cfg->prcfg_prim_nid,
4144                                       cfg->prcfg_cfg_nid);
4145                 mutex_unlock(&the_lnet.ln_api_mutex);
4146                 return rc;
4147         }
4148
4149         case IOC_LIBCFS_GET_PEER_INFO: {
4150                 struct lnet_ioctl_peer *peer_info = arg;
4151
4152                 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
4153                         return -EINVAL;
4154
4155                 mutex_lock(&the_lnet.ln_api_mutex);
4156                 rc = lnet_get_peer_ni_info(
4157                    peer_info->pr_count,
4158                    &peer_info->pr_nid,
4159                    peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
4160                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
4161                    &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
4162                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
4163                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
4164                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
4165                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_tx_credits,
4166                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
4167                 mutex_unlock(&the_lnet.ln_api_mutex);
4168                 return rc;
4169         }
4170
4171         case IOC_LIBCFS_GET_PEER_NI: {
4172                 struct lnet_ioctl_peer_cfg *cfg = arg;
4173
4174                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4175                         return -EINVAL;
4176
4177                 mutex_lock(&the_lnet.ln_api_mutex);
4178                 rc = lnet_get_peer_info(cfg,
4179                                         (void __user *)cfg->prcfg_bulk);
4180                 mutex_unlock(&the_lnet.ln_api_mutex);
4181                 return rc;
4182         }
4183
4184         case IOC_LIBCFS_GET_PEER_LIST: {
4185                 struct lnet_ioctl_peer_cfg *cfg = arg;
4186
4187                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
4188                         return -EINVAL;
4189
4190                 mutex_lock(&the_lnet.ln_api_mutex);
4191                 rc = lnet_get_peer_list(&cfg->prcfg_count, &cfg->prcfg_size,
4192                                 (struct lnet_process_id __user *)cfg->prcfg_bulk);
4193                 mutex_unlock(&the_lnet.ln_api_mutex);
4194                 return rc;
4195         }
4196
4197         case IOC_LIBCFS_SET_HEALHV: {
4198                 struct lnet_ioctl_reset_health_cfg *cfg = arg;
4199                 int value;
4200                 if (cfg->rh_hdr.ioc_len < sizeof(*cfg))
4201                         return -EINVAL;
4202                 if (cfg->rh_value < 0 ||
4203                     cfg->rh_value > LNET_MAX_HEALTH_VALUE)
4204                         value = LNET_MAX_HEALTH_VALUE;
4205                 else
4206                         value = cfg->rh_value;
4207                 CDEBUG(D_NET, "Manually setting healthv to %d for %s:%s. all = %d\n",
4208                        value, (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI) ?
4209                        "local" : "peer", libcfs_nid2str(cfg->rh_nid), cfg->rh_all);
4210                 mutex_lock(&the_lnet.ln_api_mutex);
4211                 if (cfg->rh_type == LNET_HEALTH_TYPE_LOCAL_NI)
4212                         lnet_ni_set_healthv(cfg->rh_nid, value,
4213                                              cfg->rh_all);
4214                 else
4215                         lnet_peer_ni_set_healthv(cfg->rh_nid, value,
4216                                                   cfg->rh_all);
4217                 mutex_unlock(&the_lnet.ln_api_mutex);
4218                 return 0;
4219         }
4220
4221         case IOC_LIBCFS_SET_CONNS_PER_PEER: {
4222                 struct lnet_ioctl_reset_conns_per_peer_cfg *cfg = arg;
4223                 int value;
4224
4225                 if (cfg->rcpp_hdr.ioc_len < sizeof(*cfg))
4226                         return -EINVAL;
4227                 if (cfg->rcpp_value < 0)
4228                         value = 1;
4229                 else
4230                         value = cfg->rcpp_value;
4231                 CDEBUG(D_NET,
4232                        "Setting conns_per_peer to %d for %s. all = %d\n",
4233                        value, libcfs_nid2str(cfg->rcpp_nid), cfg->rcpp_all);
4234                 mutex_lock(&the_lnet.ln_api_mutex);
4235                 lnet_ni_set_conns_per_peer(cfg->rcpp_nid, value, cfg->rcpp_all);
4236                 mutex_unlock(&the_lnet.ln_api_mutex);
4237                 return 0;
4238         }
4239
4240         case IOC_LIBCFS_NOTIFY_ROUTER: {
4241                 time64_t deadline = ktime_get_real_seconds() - data->ioc_u64[0];
4242
4243                 /* The deadline passed in by the user should be some time in
4244                  * seconds in the future since the UNIX epoch. We have to map
4245                  * that deadline to the wall clock.
4246                  */
4247                 deadline += ktime_get_seconds();
4248                 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags, false,
4249                                    deadline);
4250         }
4251
4252         case IOC_LIBCFS_LNET_DIST:
4253                 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
4254                 if (rc < 0 && rc != -EHOSTUNREACH)
4255                         return rc;
4256
4257                 data->ioc_u32[0] = rc;
4258                 return 0;
4259
4260         case IOC_LIBCFS_TESTPROTOCOMPAT:
4261                 the_lnet.ln_testprotocompat = data->ioc_flags;
4262                 return 0;
4263
4264         case IOC_LIBCFS_LNET_FAULT:
4265                 return lnet_fault_ctl(data->ioc_flags, data);
4266
4267         case IOC_LIBCFS_PING: {
4268                 signed long timeout;
4269
4270                 id.nid = data->ioc_nid;
4271                 id.pid = data->ioc_u32[0];
4272
4273                 /* If timeout is negative then set default of 3 minutes */
4274                 if (((s32)data->ioc_u32[1] <= 0) ||
4275                     data->ioc_u32[1] > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
4276                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
4277                 else
4278                         timeout = nsecs_to_jiffies(data->ioc_u32[1] * NSEC_PER_MSEC);
4279
4280                 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
4281                                data->ioc_plen1 / sizeof(struct lnet_process_id));
4282
4283                 if (rc < 0)
4284                         return rc;
4285
4286                 data->ioc_count = rc;
4287                 return 0;
4288         }
4289
4290         case IOC_LIBCFS_PING_PEER: {
4291                 struct lnet_ioctl_ping_data *ping = arg;
4292                 struct lnet_peer *lp;
4293                 signed long timeout;
4294
4295                 /* If timeout is negative then set default of 3 minutes */
4296                 if (((s32)ping->op_param) <= 0 ||
4297                     ping->op_param > (DEFAULT_PEER_TIMEOUT * MSEC_PER_SEC))
4298                         timeout = cfs_time_seconds(DEFAULT_PEER_TIMEOUT);
4299                 else
4300                         timeout = nsecs_to_jiffies(ping->op_param * NSEC_PER_MSEC);
4301
4302                 rc = lnet_ping(ping->ping_id, timeout,
4303                                ping->ping_buf,
4304                                ping->ping_count);
4305                 if (rc < 0)
4306                         return rc;
4307
4308                 mutex_lock(&the_lnet.ln_api_mutex);
4309                 lp = lnet_find_peer(ping->ping_id.nid);
4310                 if (lp) {
4311                         ping->ping_id.nid = lp->lp_primary_nid;
4312                         ping->mr_info = lnet_peer_is_multi_rail(lp);
4313                         lnet_peer_decref_locked(lp);
4314                 }
4315                 mutex_unlock(&the_lnet.ln_api_mutex);
4316
4317                 ping->ping_count = rc;
4318                 return 0;
4319         }
4320
4321         case IOC_LIBCFS_DISCOVER: {
4322                 struct lnet_ioctl_ping_data *discover = arg;
4323                 struct lnet_peer *lp;
4324
4325                 rc = lnet_discover(discover->ping_id, discover->op_param,
4326                                    discover->ping_buf,
4327                                    discover->ping_count);
4328                 if (rc < 0)
4329                         return rc;
4330
4331                 mutex_lock(&the_lnet.ln_api_mutex);
4332                 lp = lnet_find_peer(discover->ping_id.nid);
4333                 if (lp) {
4334                         discover->ping_id.nid = lp->lp_primary_nid;
4335                         discover->mr_info = lnet_peer_is_multi_rail(lp);
4336                         lnet_peer_decref_locked(lp);
4337                 }
4338                 mutex_unlock(&the_lnet.ln_api_mutex);
4339
4340                 discover->ping_count = rc;
4341                 return 0;
4342         }
4343
4344         case IOC_LIBCFS_ADD_UDSP: {
4345                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4346                 __u32 bulk_size = ioc_udsp->iou_hdr.ioc_len;
4347
4348                 mutex_lock(&the_lnet.ln_api_mutex);
4349                 rc = lnet_udsp_demarshal_add(arg, bulk_size);
4350                 if (!rc) {
4351                         rc = lnet_udsp_apply_policies(NULL, false);
4352                         CDEBUG(D_NET, "policy application returned %d\n", rc);
4353                         rc = 0;
4354                 }
4355                 mutex_unlock(&the_lnet.ln_api_mutex);
4356
4357                 return rc;
4358         }
4359
4360         case IOC_LIBCFS_DEL_UDSP: {
4361                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4362                 int idx = ioc_udsp->iou_idx;
4363
4364                 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4365                         return -EINVAL;
4366
4367                 mutex_lock(&the_lnet.ln_api_mutex);
4368                 rc = lnet_udsp_del_policy(idx);
4369                 if (!rc) {
4370                         rc = lnet_udsp_apply_policies(NULL, false);
4371                         CDEBUG(D_NET, "policy re-application returned %d\n",
4372                                rc);
4373                         rc = 0;
4374                 }
4375                 mutex_unlock(&the_lnet.ln_api_mutex);
4376
4377                 return rc;
4378         }
4379
4380         case IOC_LIBCFS_GET_UDSP_SIZE: {
4381                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4382                 struct lnet_udsp *udsp;
4383
4384                 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4385                         return -EINVAL;
4386
4387                 rc = 0;
4388
4389                 mutex_lock(&the_lnet.ln_api_mutex);
4390                 udsp = lnet_udsp_get_policy(ioc_udsp->iou_idx);
4391                 if (!udsp) {
4392                         rc = -ENOENT;
4393                 } else {
4394                         /* coming in iou_idx will hold the idx of the udsp
4395                          * to get the size of. going out the iou_idx will
4396                          * hold the size of the UDSP found at the passed
4397                          * in index.
4398                          */
4399                         ioc_udsp->iou_idx = lnet_get_udsp_size(udsp);
4400                         if (ioc_udsp->iou_idx < 0)
4401                                 rc = -EINVAL;
4402                 }
4403                 mutex_unlock(&the_lnet.ln_api_mutex);
4404
4405                 return rc;
4406         }
4407
4408         case IOC_LIBCFS_GET_UDSP: {
4409                 struct lnet_ioctl_udsp *ioc_udsp = arg;
4410                 struct lnet_udsp *udsp;
4411
4412                 if (ioc_udsp->iou_hdr.ioc_len < sizeof(*ioc_udsp))
4413                         return -EINVAL;
4414
4415                 rc = 0;
4416
4417                 mutex_lock(&the_lnet.ln_api_mutex);
4418                 udsp = lnet_udsp_get_policy(ioc_udsp->iou_idx);
4419                 if (!udsp)
4420                         rc = -ENOENT;
4421                 else
4422                         rc = lnet_udsp_marshal(udsp, ioc_udsp);
4423                 mutex_unlock(&the_lnet.ln_api_mutex);
4424
4425                 return rc;
4426         }
4427
4428         case IOC_LIBCFS_GET_CONST_UDSP_INFO: {
4429                 struct lnet_ioctl_construct_udsp_info *info = arg;
4430
4431                 if (info->cud_hdr.ioc_len < sizeof(*info))
4432                         return -EINVAL;
4433
4434                 CDEBUG(D_NET, "GET_UDSP_INFO for %s\n",
4435                        libcfs_nid2str(info->cud_nid));
4436
4437                 mutex_lock(&the_lnet.ln_api_mutex);
4438                 lnet_udsp_get_construct_info(info);
4439                 mutex_unlock(&the_lnet.ln_api_mutex);
4440
4441                 return 0;
4442         }
4443
4444         default:
4445                 ni = lnet_net2ni_addref(data->ioc_net);
4446                 if (ni == NULL)
4447                         return -EINVAL;
4448
4449                 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
4450                         rc = -EINVAL;
4451                 else
4452                         rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
4453
4454                 lnet_ni_decref(ni);
4455                 return rc;
4456         }
4457         /* not reached */
4458 }
4459 EXPORT_SYMBOL(LNetCtl);
4460
4461 void LNetDebugPeer(struct lnet_process_id id)
4462 {
4463         lnet_debug_peer(id.nid);
4464 }
4465 EXPORT_SYMBOL(LNetDebugPeer);
4466
4467 /**
4468  * Determine if the specified peer \a nid is on the local node.
4469  *
4470  * \param nid   peer nid to check
4471  *
4472  * \retval true         If peer NID is on the local node.
4473  * \retval false        If peer NID is not on the local node.
4474  */
4475 bool LNetIsPeerLocal(lnet_nid_t nid)
4476 {
4477         struct lnet_net *net;
4478         struct lnet_ni *ni;
4479         int cpt;
4480
4481         cpt = lnet_net_lock_current();
4482         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4483                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4484                         if (lnet_nid_to_nid4(&ni->ni_nid) == nid) {
4485                                 lnet_net_unlock(cpt);
4486                                 return true;
4487                         }
4488                 }
4489         }
4490         lnet_net_unlock(cpt);
4491
4492         return false;
4493 }
4494 EXPORT_SYMBOL(LNetIsPeerLocal);
4495
4496 /**
4497  * Retrieve the struct lnet_process_id ID of LNet interface at \a index.
4498  * Note that all interfaces share a same PID, as requested by LNetNIInit().
4499  *
4500  * \param index Index of the interface to look up.
4501  * \param id On successful return, this location will hold the
4502  * struct lnet_process_id ID of the interface.
4503  *
4504  * \retval 0 If an interface exists at \a index.
4505  * \retval -ENOENT If no interface has been found.
4506  */
4507 int
4508 LNetGetId(unsigned int index, struct lnet_process_id *id)
4509 {
4510         struct lnet_ni   *ni;
4511         struct lnet_net  *net;
4512         int               cpt;
4513         int               rc = -ENOENT;
4514
4515         LASSERT(the_lnet.ln_refcount > 0);
4516
4517         cpt = lnet_net_lock_current();
4518
4519         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
4520                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
4521                         if (!nid_is_nid4(&ni->ni_nid))
4522                                 /* FIXME this needs to be handled */
4523                                 continue;
4524                         if (index-- != 0)
4525                                 continue;
4526
4527                         id->nid = lnet_nid_to_nid4(&ni->ni_nid);
4528                         id->pid = the_lnet.ln_pid;
4529                         rc = 0;
4530                         break;
4531                 }
4532         }
4533
4534         lnet_net_unlock(cpt);
4535         return rc;
4536 }
4537 EXPORT_SYMBOL(LNetGetId);
4538
4539 struct ping_data {
4540         int rc;
4541         int replied;
4542         struct lnet_handle_md mdh;
4543         struct completion completion;
4544 };
4545
4546 static void
4547 lnet_ping_event_handler(struct lnet_event *event)
4548 {
4549         struct ping_data *pd = event->md_user_ptr;
4550
4551         CDEBUG(D_NET, "ping event (%d %d)%s\n",
4552                event->type, event->status,
4553                event->unlinked ? " unlinked" : "");
4554
4555         if (event->status) {
4556                 if (!pd->rc)
4557                         pd->rc = event->status;
4558         } else if (event->type == LNET_EVENT_REPLY) {
4559                 pd->replied = 1;
4560                 pd->rc = event->mlength;
4561         }
4562         if (event->unlinked)
4563                 complete(&pd->completion);
4564 }
4565
4566 static int lnet_ping(struct lnet_process_id id, signed long timeout,
4567                      struct lnet_process_id __user *ids, int n_ids)
4568 {
4569         struct lnet_md md = { NULL };
4570         struct ping_data pd = { 0 };
4571         struct lnet_ping_buffer *pbuf;
4572         struct lnet_process_id tmpid;
4573         int i;
4574         int nob;
4575         int rc;
4576         int rc2;
4577
4578         /* n_ids limit is arbitrary */
4579         if (n_ids <= 0 || id.nid == LNET_NID_ANY)
4580                 return -EINVAL;
4581
4582         /*
4583          * if the user buffer has more space than the lnet_interfaces_max
4584          * then only fill it up to lnet_interfaces_max
4585          */
4586         if (n_ids > lnet_interfaces_max)
4587                 n_ids = lnet_interfaces_max;
4588
4589         if (id.pid == LNET_PID_ANY)
4590                 id.pid = LNET_PID_LUSTRE;
4591
4592         pbuf = lnet_ping_buffer_alloc(n_ids, GFP_NOFS);
4593         if (!pbuf)
4594                 return -ENOMEM;
4595
4596         /* initialize md content */
4597         md.start     = &pbuf->pb_info;
4598         md.length    = LNET_PING_INFO_SIZE(n_ids);
4599         md.threshold = 2; /* GET/REPLY */
4600         md.max_size  = 0;
4601         md.options   = LNET_MD_TRUNCATE;
4602         md.user_ptr  = &pd;
4603         md.handler   = lnet_ping_event_handler;
4604
4605         init_completion(&pd.completion);
4606
4607         rc = LNetMDBind(&md, LNET_UNLINK, &pd.mdh);
4608         if (rc != 0) {
4609                 CERROR("Can't bind MD: %d\n", rc);
4610                 goto fail_ping_buffer_decref;
4611         }
4612
4613         rc = LNetGet(LNET_NID_ANY, pd.mdh, id,
4614                      LNET_RESERVED_PORTAL,
4615                      LNET_PROTO_PING_MATCHBITS, 0, false);
4616
4617         if (rc != 0) {
4618                 /* Don't CERROR; this could be deliberate! */
4619                 rc2 = LNetMDUnlink(pd.mdh);
4620                 LASSERT(rc2 == 0);
4621
4622                 /* NB must wait for the UNLINK event below... */
4623         }
4624
4625         if (wait_for_completion_timeout(&pd.completion, timeout) == 0) {
4626                 /* Ensure completion in finite time... */
4627                 LNetMDUnlink(pd.mdh);
4628                 wait_for_completion(&pd.completion);
4629         }
4630         if (!pd.replied) {
4631                 rc = -EIO;
4632                 goto fail_ping_buffer_decref;
4633         }
4634
4635         nob = pd.rc;
4636         LASSERT(nob >= 0 && nob <= LNET_PING_INFO_SIZE(n_ids));
4637
4638         rc = -EPROTO;           /* if I can't parse... */
4639
4640         if (nob < 8) {
4641                 CERROR("%s: ping info too short %d\n",
4642                        libcfs_id2str(id), nob);
4643                 goto fail_ping_buffer_decref;
4644         }
4645
4646         if (pbuf->pb_info.pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
4647                 lnet_swap_pinginfo(pbuf);
4648         } else if (pbuf->pb_info.pi_magic != LNET_PROTO_PING_MAGIC) {
4649                 CERROR("%s: Unexpected magic %08x\n",
4650                        libcfs_id2str(id), pbuf->pb_info.pi_magic);
4651                 goto fail_ping_buffer_decref;
4652         }
4653
4654         if ((pbuf->pb_info.pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
4655                 CERROR("%s: ping w/o NI status: 0x%x\n",
4656                        libcfs_id2str(id), pbuf->pb_info.pi_features);
4657                 goto fail_ping_buffer_decref;
4658         }
4659
4660         if (nob < LNET_PING_INFO_SIZE(0)) {
4661                 CERROR("%s: Short reply %d(%d min)\n",
4662                        libcfs_id2str(id),
4663                        nob, (int)LNET_PING_INFO_SIZE(0));
4664                 goto fail_ping_buffer_decref;
4665         }
4666
4667         if (pbuf->pb_info.pi_nnis < n_ids)
4668                 n_ids = pbuf->pb_info.pi_nnis;
4669
4670         if (nob < LNET_PING_INFO_SIZE(n_ids)) {
4671                 CERROR("%s: Short reply %d(%d expected)\n",
4672                        libcfs_id2str(id),
4673                        nob, (int)LNET_PING_INFO_SIZE(n_ids));
4674                 goto fail_ping_buffer_decref;
4675         }
4676
4677         rc = -EFAULT;           /* if I segv in copy_to_user()... */
4678
4679         memset(&tmpid, 0, sizeof(tmpid));
4680         for (i = 0; i < n_ids; i++) {
4681                 tmpid.pid = pbuf->pb_info.pi_pid;
4682                 tmpid.nid = pbuf->pb_info.pi_ni[i].ns_nid;
4683                 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
4684                         goto fail_ping_buffer_decref;
4685         }
4686         rc = pbuf->pb_info.pi_nnis;
4687
4688  fail_ping_buffer_decref:
4689         lnet_ping_buffer_decref(pbuf);
4690         return rc;
4691 }
4692
4693 static int
4694 lnet_discover(struct lnet_process_id id, __u32 force,
4695               struct lnet_process_id __user *ids, int n_ids)
4696 {
4697         struct lnet_peer_ni *lpni;
4698         struct lnet_peer_ni *p;
4699         struct lnet_peer *lp;
4700         struct lnet_process_id *buf;
4701         int cpt;
4702         int i;
4703         int rc;
4704
4705         if (n_ids <= 0 ||
4706             id.nid == LNET_NID_ANY)
4707                 return -EINVAL;
4708
4709         if (id.pid == LNET_PID_ANY)
4710                 id.pid = LNET_PID_LUSTRE;
4711
4712         /*
4713          * If the user buffer has more space than the lnet_interfaces_max,
4714          * then only fill it up to lnet_interfaces_max.
4715          */
4716         if (n_ids > lnet_interfaces_max)
4717                 n_ids = lnet_interfaces_max;
4718
4719         CFS_ALLOC_PTR_ARRAY(buf, n_ids);
4720         if (!buf)
4721                 return -ENOMEM;
4722
4723         cpt = lnet_net_lock_current();
4724         lpni = lnet_nid2peerni_locked(id.nid, LNET_NID_ANY, cpt);
4725         if (IS_ERR(lpni)) {
4726                 rc = PTR_ERR(lpni);
4727                 goto out;
4728         }
4729
4730         /*
4731          * Clearing the NIDS_UPTODATE flag ensures the peer will
4732          * be discovered, provided discovery has not been disabled.
4733          */
4734         lp = lpni->lpni_peer_net->lpn_peer;
4735         spin_lock(&lp->lp_lock);
4736         lp->lp_state &= ~LNET_PEER_NIDS_UPTODATE;
4737         /* If the force flag is set, force a PING and PUSH as well. */
4738         if (force)
4739                 lp->lp_state |= LNET_PEER_FORCE_PING | LNET_PEER_FORCE_PUSH;
4740         spin_unlock(&lp->lp_lock);
4741         rc = lnet_discover_peer_locked(lpni, cpt, true);
4742         if (rc)
4743                 goto out_decref;
4744
4745         /* The lpni (or lp) for this NID may have changed and our ref is
4746          * the only thing keeping the old one around. Release the ref
4747          * and lookup the lpni again
4748          */
4749         lnet_peer_ni_decref_locked(lpni);
4750         lpni = lnet_find_peer_ni_locked(id.nid);
4751         if (!lpni) {
4752                 rc = -ENOENT;
4753                 goto out;
4754         }
4755         lp = lpni->lpni_peer_net->lpn_peer;
4756
4757         i = 0;
4758         p = NULL;
4759         while ((p = lnet_get_next_peer_ni_locked(lp, NULL, p)) != NULL) {
4760                 buf[i].pid = id.pid;
4761                 buf[i].nid = lnet_nid_to_nid4(&p->lpni_nid);
4762                 if (++i >= n_ids)
4763                         break;
4764         }
4765         rc = i;
4766
4767 out_decref:
4768         lnet_peer_ni_decref_locked(lpni);
4769 out:
4770         lnet_net_unlock(cpt);
4771
4772         if (rc >= 0)
4773                 if (copy_to_user(ids, buf, rc * sizeof(*buf)))
4774                         rc = -EFAULT;
4775         CFS_FREE_PTR_ARRAY(buf, n_ids);
4776
4777         return rc;
4778 }
4779
4780 /**
4781  * Retrieve peer discovery status.
4782  *
4783  * \retval 1 if lnet_peer_discovery_disabled is 0
4784  * \retval 0 if lnet_peer_discovery_disabled is 1
4785  */
4786 int
4787 LNetGetPeerDiscoveryStatus(void)
4788 {
4789         return !lnet_peer_discovery_disabled;
4790 }
4791 EXPORT_SYMBOL(LNetGetPeerDiscoveryStatus);