lnet/lnet/api-ni.c

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