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