linux/drivers/staging/lustre/lnet/lnet/api-ni.c
<<
>>
Prefs
   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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19 *
  20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21 * CA 95054 USA or visit www.sun.com if you need additional information or
  22 * have any questions.
  23 *
  24 * GPL HEADER END
  25 */
  26/*
  27 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28 * Use is subject to license terms.
  29 *
  30 * Copyright (c) 2011, 2015, Intel Corporation.
  31 */
  32/*
  33 * This file is part of Lustre, http://www.lustre.org/
  34 * Lustre is a trademark of Sun Microsystems, Inc.
  35 */
  36
  37#define DEBUG_SUBSYSTEM S_LNET
  38#include <linux/log2.h>
  39#include <linux/ktime.h>
  40
  41#include "../../include/linux/lnet/lib-lnet.h"
  42#include "../../include/linux/lnet/lib-dlc.h"
  43
  44#define D_LNI D_CONSOLE
  45
  46lnet_t the_lnet;                           /* THE state of the network */
  47EXPORT_SYMBOL(the_lnet);
  48
  49static char *ip2nets = "";
  50module_param(ip2nets, charp, 0444);
  51MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
  52
  53static char *networks = "";
  54module_param(networks, charp, 0444);
  55MODULE_PARM_DESC(networks, "local networks");
  56
  57static char *routes = "";
  58module_param(routes, charp, 0444);
  59MODULE_PARM_DESC(routes, "routes to non-local networks");
  60
  61static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
  62module_param(rnet_htable_size, int, 0444);
  63MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
  64
  65static int lnet_ping(lnet_process_id_t id, int timeout_ms,
  66                     lnet_process_id_t __user *ids, int n_ids);
  67
  68static char *
  69lnet_get_routes(void)
  70{
  71        return routes;
  72}
  73
  74static char *
  75lnet_get_networks(void)
  76{
  77        char *nets;
  78        int rc;
  79
  80        if (*networks && *ip2nets) {
  81                LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or 'ip2nets' but not both at once\n");
  82                return NULL;
  83        }
  84
  85        if (*ip2nets) {
  86                rc = lnet_parse_ip2nets(&nets, ip2nets);
  87                return !rc ? nets : NULL;
  88        }
  89
  90        if (*networks)
  91                return networks;
  92
  93        return "tcp";
  94}
  95
  96static void
  97lnet_init_locks(void)
  98{
  99        spin_lock_init(&the_lnet.ln_eq_wait_lock);
 100        init_waitqueue_head(&the_lnet.ln_eq_waitq);
 101        init_waitqueue_head(&the_lnet.ln_rc_waitq);
 102        mutex_init(&the_lnet.ln_lnd_mutex);
 103        mutex_init(&the_lnet.ln_api_mutex);
 104}
 105
 106static int
 107lnet_create_remote_nets_table(void)
 108{
 109        int i;
 110        struct list_head *hash;
 111
 112        LASSERT(!the_lnet.ln_remote_nets_hash);
 113        LASSERT(the_lnet.ln_remote_nets_hbits > 0);
 114        LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
 115        if (!hash) {
 116                CERROR("Failed to create remote nets hash table\n");
 117                return -ENOMEM;
 118        }
 119
 120        for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 121                INIT_LIST_HEAD(&hash[i]);
 122        the_lnet.ln_remote_nets_hash = hash;
 123        return 0;
 124}
 125
 126static void
 127lnet_destroy_remote_nets_table(void)
 128{
 129        int i;
 130
 131        if (!the_lnet.ln_remote_nets_hash)
 132                return;
 133
 134        for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 135                LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
 136
 137        LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
 138                    LNET_REMOTE_NETS_HASH_SIZE *
 139                    sizeof(the_lnet.ln_remote_nets_hash[0]));
 140        the_lnet.ln_remote_nets_hash = NULL;
 141}
 142
 143static void
 144lnet_destroy_locks(void)
 145{
 146        if (the_lnet.ln_res_lock) {
 147                cfs_percpt_lock_free(the_lnet.ln_res_lock);
 148                the_lnet.ln_res_lock = NULL;
 149        }
 150
 151        if (the_lnet.ln_net_lock) {
 152                cfs_percpt_lock_free(the_lnet.ln_net_lock);
 153                the_lnet.ln_net_lock = NULL;
 154        }
 155}
 156
 157static int
 158lnet_create_locks(void)
 159{
 160        lnet_init_locks();
 161
 162        the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 163        if (!the_lnet.ln_res_lock)
 164                goto failed;
 165
 166        the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 167        if (!the_lnet.ln_net_lock)
 168                goto failed;
 169
 170        return 0;
 171
 172 failed:
 173        lnet_destroy_locks();
 174        return -ENOMEM;
 175}
 176
 177static void lnet_assert_wire_constants(void)
 178{
 179        /*
 180         * Wire protocol assertions generated by 'wirecheck'
 181         * running on Linux robert.bartonsoftware.com 2.6.8-1.521
 182         * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
 183         * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7)
 184         */
 185
 186        /* Constants... */
 187        CLASSERT(LNET_PROTO_TCP_MAGIC == 0xeebc0ded);
 188        CLASSERT(LNET_PROTO_TCP_VERSION_MAJOR == 1);
 189        CLASSERT(LNET_PROTO_TCP_VERSION_MINOR == 0);
 190        CLASSERT(LNET_MSG_ACK == 0);
 191        CLASSERT(LNET_MSG_PUT == 1);
 192        CLASSERT(LNET_MSG_GET == 2);
 193        CLASSERT(LNET_MSG_REPLY == 3);
 194        CLASSERT(LNET_MSG_HELLO == 4);
 195
 196        /* Checks for struct ptl_handle_wire_t */
 197        CLASSERT((int)sizeof(lnet_handle_wire_t) == 16);
 198        CLASSERT((int)offsetof(lnet_handle_wire_t, wh_interface_cookie) == 0);
 199        CLASSERT((int)sizeof(((lnet_handle_wire_t *)0)->wh_interface_cookie) == 8);
 200        CLASSERT((int)offsetof(lnet_handle_wire_t, wh_object_cookie) == 8);
 201        CLASSERT((int)sizeof(((lnet_handle_wire_t *)0)->wh_object_cookie) == 8);
 202
 203        /* Checks for struct lnet_magicversion_t */
 204        CLASSERT((int)sizeof(lnet_magicversion_t) == 8);
 205        CLASSERT((int)offsetof(lnet_magicversion_t, magic) == 0);
 206        CLASSERT((int)sizeof(((lnet_magicversion_t *)0)->magic) == 4);
 207        CLASSERT((int)offsetof(lnet_magicversion_t, version_major) == 4);
 208        CLASSERT((int)sizeof(((lnet_magicversion_t *)0)->version_major) == 2);
 209        CLASSERT((int)offsetof(lnet_magicversion_t, version_minor) == 6);
 210        CLASSERT((int)sizeof(((lnet_magicversion_t *)0)->version_minor) == 2);
 211
 212        /* Checks for struct lnet_hdr_t */
 213        CLASSERT((int)sizeof(lnet_hdr_t) == 72);
 214        CLASSERT((int)offsetof(lnet_hdr_t, dest_nid) == 0);
 215        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->dest_nid) == 8);
 216        CLASSERT((int)offsetof(lnet_hdr_t, src_nid) == 8);
 217        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->src_nid) == 8);
 218        CLASSERT((int)offsetof(lnet_hdr_t, dest_pid) == 16);
 219        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->dest_pid) == 4);
 220        CLASSERT((int)offsetof(lnet_hdr_t, src_pid) == 20);
 221        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->src_pid) == 4);
 222        CLASSERT((int)offsetof(lnet_hdr_t, type) == 24);
 223        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->type) == 4);
 224        CLASSERT((int)offsetof(lnet_hdr_t, payload_length) == 28);
 225        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->payload_length) == 4);
 226        CLASSERT((int)offsetof(lnet_hdr_t, msg) == 32);
 227        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg) == 40);
 228
 229        /* Ack */
 230        CLASSERT((int)offsetof(lnet_hdr_t, msg.ack.dst_wmd) == 32);
 231        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.ack.dst_wmd) == 16);
 232        CLASSERT((int)offsetof(lnet_hdr_t, msg.ack.match_bits) == 48);
 233        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.ack.match_bits) == 8);
 234        CLASSERT((int)offsetof(lnet_hdr_t, msg.ack.mlength) == 56);
 235        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.ack.mlength) == 4);
 236
 237        /* Put */
 238        CLASSERT((int)offsetof(lnet_hdr_t, msg.put.ack_wmd) == 32);
 239        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.ack_wmd) == 16);
 240        CLASSERT((int)offsetof(lnet_hdr_t, msg.put.match_bits) == 48);
 241        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.match_bits) == 8);
 242        CLASSERT((int)offsetof(lnet_hdr_t, msg.put.hdr_data) == 56);
 243        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.hdr_data) == 8);
 244        CLASSERT((int)offsetof(lnet_hdr_t, msg.put.ptl_index) == 64);
 245        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.ptl_index) == 4);
 246        CLASSERT((int)offsetof(lnet_hdr_t, msg.put.offset) == 68);
 247        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.offset) == 4);
 248
 249        /* Get */
 250        CLASSERT((int)offsetof(lnet_hdr_t, msg.get.return_wmd) == 32);
 251        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.return_wmd) == 16);
 252        CLASSERT((int)offsetof(lnet_hdr_t, msg.get.match_bits) == 48);
 253        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.match_bits) == 8);
 254        CLASSERT((int)offsetof(lnet_hdr_t, msg.get.ptl_index) == 56);
 255        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.ptl_index) == 4);
 256        CLASSERT((int)offsetof(lnet_hdr_t, msg.get.src_offset) == 60);
 257        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.src_offset) == 4);
 258        CLASSERT((int)offsetof(lnet_hdr_t, msg.get.sink_length) == 64);
 259        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.sink_length) == 4);
 260
 261        /* Reply */
 262        CLASSERT((int)offsetof(lnet_hdr_t, msg.reply.dst_wmd) == 32);
 263        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.reply.dst_wmd) == 16);
 264
 265        /* Hello */
 266        CLASSERT((int)offsetof(lnet_hdr_t, msg.hello.incarnation) == 32);
 267        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.hello.incarnation) == 8);
 268        CLASSERT((int)offsetof(lnet_hdr_t, msg.hello.type) == 40);
 269        CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.hello.type) == 4);
 270}
 271
 272static lnd_t *
 273lnet_find_lnd_by_type(__u32 type)
 274{
 275        lnd_t *lnd;
 276        struct list_head *tmp;
 277
 278        /* holding lnd mutex */
 279        list_for_each(tmp, &the_lnet.ln_lnds) {
 280                lnd = list_entry(tmp, lnd_t, lnd_list);
 281
 282                if (lnd->lnd_type == type)
 283                        return lnd;
 284        }
 285
 286        return NULL;
 287}
 288
 289void
 290lnet_register_lnd(lnd_t *lnd)
 291{
 292        mutex_lock(&the_lnet.ln_lnd_mutex);
 293
 294        LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
 295        LASSERT(!lnet_find_lnd_by_type(lnd->lnd_type));
 296
 297        list_add_tail(&lnd->lnd_list, &the_lnet.ln_lnds);
 298        lnd->lnd_refcount = 0;
 299
 300        CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
 301
 302        mutex_unlock(&the_lnet.ln_lnd_mutex);
 303}
 304EXPORT_SYMBOL(lnet_register_lnd);
 305
 306void
 307lnet_unregister_lnd(lnd_t *lnd)
 308{
 309        mutex_lock(&the_lnet.ln_lnd_mutex);
 310
 311        LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
 312        LASSERT(!lnd->lnd_refcount);
 313
 314        list_del(&lnd->lnd_list);
 315        CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
 316
 317        mutex_unlock(&the_lnet.ln_lnd_mutex);
 318}
 319EXPORT_SYMBOL(lnet_unregister_lnd);
 320
 321void
 322lnet_counters_get(lnet_counters_t *counters)
 323{
 324        lnet_counters_t *ctr;
 325        int i;
 326
 327        memset(counters, 0, sizeof(*counters));
 328
 329        lnet_net_lock(LNET_LOCK_EX);
 330
 331        cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
 332                counters->msgs_max     += ctr->msgs_max;
 333                counters->msgs_alloc   += ctr->msgs_alloc;
 334                counters->errors       += ctr->errors;
 335                counters->send_count   += ctr->send_count;
 336                counters->recv_count   += ctr->recv_count;
 337                counters->route_count  += ctr->route_count;
 338                counters->drop_count   += ctr->drop_count;
 339                counters->send_length  += ctr->send_length;
 340                counters->recv_length  += ctr->recv_length;
 341                counters->route_length += ctr->route_length;
 342                counters->drop_length  += ctr->drop_length;
 343        }
 344        lnet_net_unlock(LNET_LOCK_EX);
 345}
 346EXPORT_SYMBOL(lnet_counters_get);
 347
 348void
 349lnet_counters_reset(void)
 350{
 351        lnet_counters_t *counters;
 352        int i;
 353
 354        lnet_net_lock(LNET_LOCK_EX);
 355
 356        cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
 357                memset(counters, 0, sizeof(lnet_counters_t));
 358
 359        lnet_net_unlock(LNET_LOCK_EX);
 360}
 361
 362static char *
 363lnet_res_type2str(int type)
 364{
 365        switch (type) {
 366        default:
 367                LBUG();
 368        case LNET_COOKIE_TYPE_MD:
 369                return "MD";
 370        case LNET_COOKIE_TYPE_ME:
 371                return "ME";
 372        case LNET_COOKIE_TYPE_EQ:
 373                return "EQ";
 374        }
 375}
 376
 377static void
 378lnet_res_container_cleanup(struct lnet_res_container *rec)
 379{
 380        int count = 0;
 381
 382        if (!rec->rec_type) /* not set yet, it's uninitialized */
 383                return;
 384
 385        while (!list_empty(&rec->rec_active)) {
 386                struct list_head *e = rec->rec_active.next;
 387
 388                list_del_init(e);
 389                if (rec->rec_type == LNET_COOKIE_TYPE_EQ) {
 390                        lnet_eq_free(list_entry(e, lnet_eq_t, eq_list));
 391
 392                } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
 393                        lnet_md_free(list_entry(e, lnet_libmd_t, md_list));
 394
 395                } else { /* NB: Active MEs should be attached on portals */
 396                        LBUG();
 397                }
 398                count++;
 399        }
 400
 401        if (count > 0) {
 402                /*
 403                 * Found alive MD/ME/EQ, user really should unlink/free
 404                 * all of them before finalize LNet, but if someone didn't,
 405                 * we have to recycle garbage for him
 406                 */
 407                CERROR("%d active elements on exit of %s container\n",
 408                       count, lnet_res_type2str(rec->rec_type));
 409        }
 410
 411        if (rec->rec_lh_hash) {
 412                LIBCFS_FREE(rec->rec_lh_hash,
 413                            LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
 414                rec->rec_lh_hash = NULL;
 415        }
 416
 417        rec->rec_type = 0; /* mark it as finalized */
 418}
 419
 420static int
 421lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
 422{
 423        int rc = 0;
 424        int i;
 425
 426        LASSERT(!rec->rec_type);
 427
 428        rec->rec_type = type;
 429        INIT_LIST_HEAD(&rec->rec_active);
 430        rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
 431
 432        /* Arbitrary choice of hash table size */
 433        LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
 434                         LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
 435        if (!rec->rec_lh_hash) {
 436                rc = -ENOMEM;
 437                goto out;
 438        }
 439
 440        for (i = 0; i < LNET_LH_HASH_SIZE; i++)
 441                INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
 442
 443        return 0;
 444
 445out:
 446        CERROR("Failed to setup %s resource container\n",
 447               lnet_res_type2str(type));
 448        lnet_res_container_cleanup(rec);
 449        return rc;
 450}
 451
 452static void
 453lnet_res_containers_destroy(struct lnet_res_container **recs)
 454{
 455        struct lnet_res_container *rec;
 456        int i;
 457
 458        cfs_percpt_for_each(rec, i, recs)
 459                lnet_res_container_cleanup(rec);
 460
 461        cfs_percpt_free(recs);
 462}
 463
 464static struct lnet_res_container **
 465lnet_res_containers_create(int type)
 466{
 467        struct lnet_res_container **recs;
 468        struct lnet_res_container *rec;
 469        int rc;
 470        int i;
 471
 472        recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
 473        if (!recs) {
 474                CERROR("Failed to allocate %s resource containers\n",
 475                       lnet_res_type2str(type));
 476                return NULL;
 477        }
 478
 479        cfs_percpt_for_each(rec, i, recs) {
 480                rc = lnet_res_container_setup(rec, i, type);
 481                if (rc) {
 482                        lnet_res_containers_destroy(recs);
 483                        return NULL;
 484                }
 485        }
 486
 487        return recs;
 488}
 489
 490lnet_libhandle_t *
 491lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
 492{
 493        /* ALWAYS called with lnet_res_lock held */
 494        struct list_head *head;
 495        lnet_libhandle_t *lh;
 496        unsigned int hash;
 497
 498        if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
 499                return NULL;
 500
 501        hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
 502        head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
 503
 504        list_for_each_entry(lh, head, lh_hash_chain) {
 505                if (lh->lh_cookie == cookie)
 506                        return lh;
 507        }
 508
 509        return NULL;
 510}
 511
 512void
 513lnet_res_lh_initialize(struct lnet_res_container *rec, lnet_libhandle_t *lh)
 514{
 515        /* ALWAYS called with lnet_res_lock held */
 516        unsigned int ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
 517        unsigned int hash;
 518
 519        lh->lh_cookie = rec->rec_lh_cookie;
 520        rec->rec_lh_cookie += 1 << ibits;
 521
 522        hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
 523
 524        list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
 525}
 526
 527static int lnet_unprepare(void);
 528
 529static int
 530lnet_prepare(lnet_pid_t requested_pid)
 531{
 532        /* Prepare to bring up the network */
 533        struct lnet_res_container **recs;
 534        int rc = 0;
 535
 536        if (requested_pid == LNET_PID_ANY) {
 537                /* Don't instantiate LNET just for me */
 538                return -ENETDOWN;
 539        }
 540
 541        LASSERT(!the_lnet.ln_refcount);
 542
 543        the_lnet.ln_routing = 0;
 544
 545        LASSERT(!(requested_pid & LNET_PID_USERFLAG));
 546        the_lnet.ln_pid = requested_pid;
 547
 548        INIT_LIST_HEAD(&the_lnet.ln_test_peers);
 549        INIT_LIST_HEAD(&the_lnet.ln_nis);
 550        INIT_LIST_HEAD(&the_lnet.ln_nis_cpt);
 551        INIT_LIST_HEAD(&the_lnet.ln_nis_zombie);
 552        INIT_LIST_HEAD(&the_lnet.ln_routers);
 553        INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
 554        INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
 555
 556        rc = lnet_create_remote_nets_table();
 557        if (rc)
 558                goto failed;
 559        /*
 560         * NB the interface cookie in wire handles guards against delayed
 561         * replies and ACKs appearing valid after reboot.
 562         */
 563        the_lnet.ln_interface_cookie = ktime_get_ns();
 564
 565        the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
 566                                                sizeof(lnet_counters_t));
 567        if (!the_lnet.ln_counters) {
 568                CERROR("Failed to allocate counters for LNet\n");
 569                rc = -ENOMEM;
 570                goto failed;
 571        }
 572
 573        rc = lnet_peer_tables_create();
 574        if (rc)
 575                goto failed;
 576
 577        rc = lnet_msg_containers_create();
 578        if (rc)
 579                goto failed;
 580
 581        rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
 582                                      LNET_COOKIE_TYPE_EQ);
 583        if (rc)
 584                goto failed;
 585
 586        recs = lnet_res_containers_create(LNET_COOKIE_TYPE_ME);
 587        if (!recs) {
 588                rc = -ENOMEM;
 589                goto failed;
 590        }
 591
 592        the_lnet.ln_me_containers = recs;
 593
 594        recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
 595        if (!recs) {
 596                rc = -ENOMEM;
 597                goto failed;
 598        }
 599
 600        the_lnet.ln_md_containers = recs;
 601
 602        rc = lnet_portals_create();
 603        if (rc) {
 604                CERROR("Failed to create portals for LNet: %d\n", rc);
 605                goto failed;
 606        }
 607
 608        return 0;
 609
 610 failed:
 611        lnet_unprepare();
 612        return rc;
 613}
 614
 615static int
 616lnet_unprepare(void)
 617{
 618        /*
 619         * NB no LNET_LOCK since this is the last reference.  All LND instances
 620         * have shut down already, so it is safe to unlink and free all
 621         * descriptors, even those that appear committed to a network op (eg MD
 622         * with non-zero pending count)
 623         */
 624        lnet_fail_nid(LNET_NID_ANY, 0);
 625
 626        LASSERT(!the_lnet.ln_refcount);
 627        LASSERT(list_empty(&the_lnet.ln_test_peers));
 628        LASSERT(list_empty(&the_lnet.ln_nis));
 629        LASSERT(list_empty(&the_lnet.ln_nis_cpt));
 630        LASSERT(list_empty(&the_lnet.ln_nis_zombie));
 631
 632        lnet_portals_destroy();
 633
 634        if (the_lnet.ln_md_containers) {
 635                lnet_res_containers_destroy(the_lnet.ln_md_containers);
 636                the_lnet.ln_md_containers = NULL;
 637        }
 638
 639        if (the_lnet.ln_me_containers) {
 640                lnet_res_containers_destroy(the_lnet.ln_me_containers);
 641                the_lnet.ln_me_containers = NULL;
 642        }
 643
 644        lnet_res_container_cleanup(&the_lnet.ln_eq_container);
 645
 646        lnet_msg_containers_destroy();
 647        lnet_peer_tables_destroy();
 648        lnet_rtrpools_free(0);
 649
 650        if (the_lnet.ln_counters) {
 651                cfs_percpt_free(the_lnet.ln_counters);
 652                the_lnet.ln_counters = NULL;
 653        }
 654        lnet_destroy_remote_nets_table();
 655
 656        return 0;
 657}
 658
 659lnet_ni_t  *
 660lnet_net2ni_locked(__u32 net, int cpt)
 661{
 662        struct list_head *tmp;
 663        lnet_ni_t *ni;
 664
 665        LASSERT(cpt != LNET_LOCK_EX);
 666
 667        list_for_each(tmp, &the_lnet.ln_nis) {
 668                ni = list_entry(tmp, lnet_ni_t, ni_list);
 669
 670                if (LNET_NIDNET(ni->ni_nid) == net) {
 671                        lnet_ni_addref_locked(ni, cpt);
 672                        return ni;
 673                }
 674        }
 675
 676        return NULL;
 677}
 678
 679lnet_ni_t *
 680lnet_net2ni(__u32 net)
 681{
 682        lnet_ni_t *ni;
 683
 684        lnet_net_lock(0);
 685        ni = lnet_net2ni_locked(net, 0);
 686        lnet_net_unlock(0);
 687
 688        return ni;
 689}
 690EXPORT_SYMBOL(lnet_net2ni);
 691
 692static unsigned int
 693lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
 694{
 695        __u64 key = nid;
 696        unsigned int val;
 697
 698        LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
 699
 700        if (number == 1)
 701                return 0;
 702
 703        val = hash_long(key, LNET_CPT_BITS);
 704        /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
 705        if (val < number)
 706                return val;
 707
 708        return (unsigned int)(key + val + (val >> 1)) % number;
 709}
 710
 711int
 712lnet_cpt_of_nid_locked(lnet_nid_t nid)
 713{
 714        struct lnet_ni *ni;
 715
 716        /* must called with hold of lnet_net_lock */
 717        if (LNET_CPT_NUMBER == 1)
 718                return 0; /* the only one */
 719
 720        /* take lnet_net_lock(any) would be OK */
 721        if (!list_empty(&the_lnet.ln_nis_cpt)) {
 722                list_for_each_entry(ni, &the_lnet.ln_nis_cpt, ni_cptlist) {
 723                        if (LNET_NIDNET(ni->ni_nid) != LNET_NIDNET(nid))
 724                                continue;
 725
 726                        LASSERT(ni->ni_cpts);
 727                        return ni->ni_cpts[lnet_nid_cpt_hash
 728                                           (nid, ni->ni_ncpts)];
 729                }
 730        }
 731
 732        return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
 733}
 734
 735int
 736lnet_cpt_of_nid(lnet_nid_t nid)
 737{
 738        int cpt;
 739        int cpt2;
 740
 741        if (LNET_CPT_NUMBER == 1)
 742                return 0; /* the only one */
 743
 744        if (list_empty(&the_lnet.ln_nis_cpt))
 745                return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
 746
 747        cpt = lnet_net_lock_current();
 748        cpt2 = lnet_cpt_of_nid_locked(nid);
 749        lnet_net_unlock(cpt);
 750
 751        return cpt2;
 752}
 753EXPORT_SYMBOL(lnet_cpt_of_nid);
 754
 755int
 756lnet_islocalnet(__u32 net)
 757{
 758        struct lnet_ni *ni;
 759        int cpt;
 760
 761        cpt = lnet_net_lock_current();
 762
 763        ni = lnet_net2ni_locked(net, cpt);
 764        if (ni)
 765                lnet_ni_decref_locked(ni, cpt);
 766
 767        lnet_net_unlock(cpt);
 768
 769        return !!ni;
 770}
 771
 772lnet_ni_t  *
 773lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
 774{
 775        struct lnet_ni *ni;
 776        struct list_head *tmp;
 777
 778        LASSERT(cpt != LNET_LOCK_EX);
 779
 780        list_for_each(tmp, &the_lnet.ln_nis) {
 781                ni = list_entry(tmp, lnet_ni_t, ni_list);
 782
 783                if (ni->ni_nid == nid) {
 784                        lnet_ni_addref_locked(ni, cpt);
 785                        return ni;
 786                }
 787        }
 788
 789        return NULL;
 790}
 791
 792int
 793lnet_islocalnid(lnet_nid_t nid)
 794{
 795        struct lnet_ni *ni;
 796        int cpt;
 797
 798        cpt = lnet_net_lock_current();
 799        ni = lnet_nid2ni_locked(nid, cpt);
 800        if (ni)
 801                lnet_ni_decref_locked(ni, cpt);
 802        lnet_net_unlock(cpt);
 803
 804        return !!ni;
 805}
 806
 807int
 808lnet_count_acceptor_nis(void)
 809{
 810        /* Return the # of NIs that need the acceptor. */
 811        int count = 0;
 812        struct list_head *tmp;
 813        struct lnet_ni *ni;
 814        int cpt;
 815
 816        cpt = lnet_net_lock_current();
 817        list_for_each(tmp, &the_lnet.ln_nis) {
 818                ni = list_entry(tmp, lnet_ni_t, ni_list);
 819
 820                if (ni->ni_lnd->lnd_accept)
 821                        count++;
 822        }
 823
 824        lnet_net_unlock(cpt);
 825
 826        return count;
 827}
 828
 829static lnet_ping_info_t *
 830lnet_ping_info_create(int num_ni)
 831{
 832        lnet_ping_info_t *ping_info;
 833        unsigned int infosz;
 834
 835        infosz = offsetof(lnet_ping_info_t, pi_ni[num_ni]);
 836        LIBCFS_ALLOC(ping_info, infosz);
 837        if (!ping_info) {
 838                CERROR("Can't allocate ping info[%d]\n", num_ni);
 839                return NULL;
 840        }
 841
 842        ping_info->pi_nnis = num_ni;
 843        ping_info->pi_pid = the_lnet.ln_pid;
 844        ping_info->pi_magic = LNET_PROTO_PING_MAGIC;
 845        ping_info->pi_features = LNET_PING_FEAT_NI_STATUS;
 846
 847        return ping_info;
 848}
 849
 850static inline int
 851lnet_get_ni_count(void)
 852{
 853        struct lnet_ni *ni;
 854        int count = 0;
 855
 856        lnet_net_lock(0);
 857
 858        list_for_each_entry(ni, &the_lnet.ln_nis, ni_list)
 859                count++;
 860
 861        lnet_net_unlock(0);
 862
 863        return count;
 864}
 865
 866static inline void
 867lnet_ping_info_free(lnet_ping_info_t *pinfo)
 868{
 869        LIBCFS_FREE(pinfo,
 870                    offsetof(lnet_ping_info_t,
 871                             pi_ni[pinfo->pi_nnis]));
 872}
 873
 874static void
 875lnet_ping_info_destroy(void)
 876{
 877        struct lnet_ni *ni;
 878
 879        lnet_net_lock(LNET_LOCK_EX);
 880
 881        list_for_each_entry(ni, &the_lnet.ln_nis, ni_list) {
 882                lnet_ni_lock(ni);
 883                ni->ni_status = NULL;
 884                lnet_ni_unlock(ni);
 885        }
 886
 887        lnet_ping_info_free(the_lnet.ln_ping_info);
 888        the_lnet.ln_ping_info = NULL;
 889
 890        lnet_net_unlock(LNET_LOCK_EX);
 891}
 892
 893static void
 894lnet_ping_event_handler(lnet_event_t *event)
 895{
 896        lnet_ping_info_t *pinfo = event->md.user_ptr;
 897
 898        if (event->unlinked)
 899                pinfo->pi_features = LNET_PING_FEAT_INVAL;
 900}
 901
 902static int
 903lnet_ping_info_setup(lnet_ping_info_t **ppinfo, lnet_handle_md_t *md_handle,
 904                     int ni_count, bool set_eq)
 905{
 906        lnet_process_id_t id = {LNET_NID_ANY, LNET_PID_ANY};
 907        lnet_handle_me_t me_handle;
 908        lnet_md_t md = { NULL };
 909        int rc, rc2;
 910
 911        if (set_eq) {
 912                rc = LNetEQAlloc(0, lnet_ping_event_handler,
 913                                 &the_lnet.ln_ping_target_eq);
 914                if (rc) {
 915                        CERROR("Can't allocate ping EQ: %d\n", rc);
 916                        return rc;
 917                }
 918        }
 919
 920        *ppinfo = lnet_ping_info_create(ni_count);
 921        if (!*ppinfo) {
 922                rc = -ENOMEM;
 923                goto failed_0;
 924        }
 925
 926        rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
 927                          LNET_PROTO_PING_MATCHBITS, 0,
 928                          LNET_UNLINK, LNET_INS_AFTER,
 929                          &me_handle);
 930        if (rc) {
 931                CERROR("Can't create ping ME: %d\n", rc);
 932                goto failed_1;
 933        }
 934
 935        /* initialize md content */
 936        md.start = *ppinfo;
 937        md.length = offsetof(lnet_ping_info_t,
 938                             pi_ni[(*ppinfo)->pi_nnis]);
 939        md.threshold = LNET_MD_THRESH_INF;
 940        md.max_size = 0;
 941        md.options = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
 942                     LNET_MD_MANAGE_REMOTE;
 943        md.user_ptr  = NULL;
 944        md.eq_handle = the_lnet.ln_ping_target_eq;
 945        md.user_ptr = *ppinfo;
 946
 947        rc = LNetMDAttach(me_handle, md, LNET_RETAIN, md_handle);
 948        if (rc) {
 949                CERROR("Can't attach ping MD: %d\n", rc);
 950                goto failed_2;
 951        }
 952
 953        return 0;
 954
 955failed_2:
 956        rc2 = LNetMEUnlink(me_handle);
 957        LASSERT(!rc2);
 958failed_1:
 959        lnet_ping_info_free(*ppinfo);
 960        *ppinfo = NULL;
 961failed_0:
 962        if (set_eq)
 963                LNetEQFree(the_lnet.ln_ping_target_eq);
 964        return rc;
 965}
 966
 967static void
 968lnet_ping_md_unlink(lnet_ping_info_t *pinfo, lnet_handle_md_t *md_handle)
 969{
 970        sigset_t blocked = cfs_block_allsigs();
 971
 972        LNetMDUnlink(*md_handle);
 973        LNetInvalidateHandle(md_handle);
 974
 975        /* NB md could be busy; this just starts the unlink */
 976        while (pinfo->pi_features != LNET_PING_FEAT_INVAL) {
 977                CDEBUG(D_NET, "Still waiting for ping MD to unlink\n");
 978                set_current_state(TASK_UNINTERRUPTIBLE);
 979                schedule_timeout(cfs_time_seconds(1));
 980        }
 981
 982        cfs_restore_sigs(blocked);
 983}
 984
 985static void
 986lnet_ping_info_install_locked(lnet_ping_info_t *ping_info)
 987{
 988        lnet_ni_status_t *ns;
 989        lnet_ni_t *ni;
 990        int i = 0;
 991
 992        list_for_each_entry(ni, &the_lnet.ln_nis, ni_list) {
 993                LASSERT(i < ping_info->pi_nnis);
 994
 995                ns = &ping_info->pi_ni[i];
 996
 997                ns->ns_nid = ni->ni_nid;
 998
 999                lnet_ni_lock(ni);
1000                ns->ns_status = (ni->ni_status) ?
1001                                 ni->ni_status->ns_status : LNET_NI_STATUS_UP;
1002                ni->ni_status = ns;
1003                lnet_ni_unlock(ni);
1004
1005                i++;
1006        }
1007}
1008
1009static void
1010lnet_ping_target_update(lnet_ping_info_t *pinfo, lnet_handle_md_t md_handle)
1011{
1012        lnet_ping_info_t *old_pinfo = NULL;
1013        lnet_handle_md_t old_md;
1014
1015        /* switch the NIs to point to the new ping info created */
1016        lnet_net_lock(LNET_LOCK_EX);
1017
1018        if (!the_lnet.ln_routing)
1019                pinfo->pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1020        lnet_ping_info_install_locked(pinfo);
1021
1022        if (the_lnet.ln_ping_info) {
1023                old_pinfo = the_lnet.ln_ping_info;
1024                old_md = the_lnet.ln_ping_target_md;
1025        }
1026        the_lnet.ln_ping_target_md = md_handle;
1027        the_lnet.ln_ping_info = pinfo;
1028
1029        lnet_net_unlock(LNET_LOCK_EX);
1030
1031        if (old_pinfo) {
1032                /* unlink the old ping info */
1033                lnet_ping_md_unlink(old_pinfo, &old_md);
1034                lnet_ping_info_free(old_pinfo);
1035        }
1036}
1037
1038static void
1039lnet_ping_target_fini(void)
1040{
1041        int rc;
1042
1043        lnet_ping_md_unlink(the_lnet.ln_ping_info,
1044                            &the_lnet.ln_ping_target_md);
1045
1046        rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1047        LASSERT(!rc);
1048
1049        lnet_ping_info_destroy();
1050}
1051
1052static int
1053lnet_ni_tq_credits(lnet_ni_t *ni)
1054{
1055        int credits;
1056
1057        LASSERT(ni->ni_ncpts >= 1);
1058
1059        if (ni->ni_ncpts == 1)
1060                return ni->ni_maxtxcredits;
1061
1062        credits = ni->ni_maxtxcredits / ni->ni_ncpts;
1063        credits = max(credits, 8 * ni->ni_peertxcredits);
1064        credits = min(credits, ni->ni_maxtxcredits);
1065
1066        return credits;
1067}
1068
1069static void
1070lnet_ni_unlink_locked(lnet_ni_t *ni)
1071{
1072        if (!list_empty(&ni->ni_cptlist)) {
1073                list_del_init(&ni->ni_cptlist);
1074                lnet_ni_decref_locked(ni, 0);
1075        }
1076
1077        /* move it to zombie list and nobody can find it anymore */
1078        LASSERT(!list_empty(&ni->ni_list));
1079        list_move(&ni->ni_list, &the_lnet.ln_nis_zombie);
1080        lnet_ni_decref_locked(ni, 0);   /* drop ln_nis' ref */
1081}
1082
1083static void
1084lnet_clear_zombies_nis_locked(void)
1085{
1086        int i;
1087        int islo;
1088        lnet_ni_t *ni;
1089        lnet_ni_t *temp;
1090
1091        /*
1092         * Now wait for the NI's I just nuked to show up on ln_zombie_nis
1093         * and shut them down in guaranteed thread context
1094         */
1095        i = 2;
1096        list_for_each_entry_safe(ni, temp, &the_lnet.ln_nis_zombie, ni_list) {
1097                int *ref;
1098                int j;
1099
1100                list_del_init(&ni->ni_list);
1101                cfs_percpt_for_each(ref, j, ni->ni_refs) {
1102                        if (!*ref)
1103                                continue;
1104                        /* still busy, add it back to zombie list */
1105                        list_add(&ni->ni_list, &the_lnet.ln_nis_zombie);
1106                        break;
1107                }
1108
1109                if (!list_empty(&ni->ni_list)) {
1110                        lnet_net_unlock(LNET_LOCK_EX);
1111                        ++i;
1112                        if ((i & (-i)) == i) {
1113                                CDEBUG(D_WARNING, "Waiting for zombie LNI %s\n",
1114                                       libcfs_nid2str(ni->ni_nid));
1115                        }
1116                        set_current_state(TASK_UNINTERRUPTIBLE);
1117                        schedule_timeout(cfs_time_seconds(1));
1118                        lnet_net_lock(LNET_LOCK_EX);
1119                        continue;
1120                }
1121
1122                ni->ni_lnd->lnd_refcount--;
1123                lnet_net_unlock(LNET_LOCK_EX);
1124
1125                islo = ni->ni_lnd->lnd_type == LOLND;
1126
1127                LASSERT(!in_interrupt());
1128                ni->ni_lnd->lnd_shutdown(ni);
1129
1130                /*
1131                 * can't deref lnd anymore now; it might have unregistered
1132                 * itself...
1133                 */
1134                if (!islo)
1135                        CDEBUG(D_LNI, "Removed LNI %s\n",
1136                               libcfs_nid2str(ni->ni_nid));
1137
1138                lnet_ni_free(ni);
1139                i = 2;
1140
1141                lnet_net_lock(LNET_LOCK_EX);
1142        }
1143}
1144
1145static void
1146lnet_shutdown_lndnis(void)
1147{
1148        lnet_ni_t *ni;
1149        lnet_ni_t *temp;
1150        int i;
1151
1152        /* NB called holding the global mutex */
1153
1154        /* All quiet on the API front */
1155        LASSERT(!the_lnet.ln_shutdown);
1156        LASSERT(!the_lnet.ln_refcount);
1157        LASSERT(list_empty(&the_lnet.ln_nis_zombie));
1158
1159        lnet_net_lock(LNET_LOCK_EX);
1160        the_lnet.ln_shutdown = 1;       /* flag shutdown */
1161
1162        /* Unlink NIs from the global table */
1163        list_for_each_entry_safe(ni, temp, &the_lnet.ln_nis, ni_list) {
1164                lnet_ni_unlink_locked(ni);
1165        }
1166
1167        /* Drop the cached loopback NI. */
1168        if (the_lnet.ln_loni) {
1169                lnet_ni_decref_locked(the_lnet.ln_loni, 0);
1170                the_lnet.ln_loni = NULL;
1171        }
1172
1173        lnet_net_unlock(LNET_LOCK_EX);
1174
1175        /*
1176         * Clear lazy portals and drop delayed messages which hold refs
1177         * on their lnet_msg_t::msg_rxpeer
1178         */
1179        for (i = 0; i < the_lnet.ln_nportals; i++)
1180                LNetClearLazyPortal(i);
1181
1182        /*
1183         * Clear the peer table and wait for all peers to go (they hold refs on
1184         * their NIs)
1185         */
1186        lnet_peer_tables_cleanup(NULL);
1187
1188        lnet_net_lock(LNET_LOCK_EX);
1189
1190        lnet_clear_zombies_nis_locked();
1191        the_lnet.ln_shutdown = 0;
1192        lnet_net_unlock(LNET_LOCK_EX);
1193}
1194
1195/* shutdown down the NI and release refcount */
1196static void
1197lnet_shutdown_lndni(struct lnet_ni *ni)
1198{
1199        int i;
1200
1201        lnet_net_lock(LNET_LOCK_EX);
1202        lnet_ni_unlink_locked(ni);
1203        lnet_net_unlock(LNET_LOCK_EX);
1204
1205        /* clear messages for this NI on the lazy portal */
1206        for (i = 0; i < the_lnet.ln_nportals; i++)
1207                lnet_clear_lazy_portal(ni, i, "Shutting down NI");
1208
1209        /* Do peer table cleanup for this ni */
1210        lnet_peer_tables_cleanup(ni);
1211
1212        lnet_net_lock(LNET_LOCK_EX);
1213        lnet_clear_zombies_nis_locked();
1214        lnet_net_unlock(LNET_LOCK_EX);
1215}
1216
1217static int
1218lnet_startup_lndni(struct lnet_ni *ni, __s32 peer_timeout,
1219                   __s32 peer_cr, __s32 peer_buf_cr, __s32 credits)
1220{
1221        int rc = -EINVAL;
1222        int lnd_type;
1223        lnd_t *lnd;
1224        struct lnet_tx_queue *tq;
1225        int i;
1226
1227        lnd_type = LNET_NETTYP(LNET_NIDNET(ni->ni_nid));
1228
1229        LASSERT(libcfs_isknown_lnd(lnd_type));
1230
1231        if (lnd_type == CIBLND || lnd_type == OPENIBLND ||
1232            lnd_type == IIBLND || lnd_type == VIBLND) {
1233                CERROR("LND %s obsoleted\n", libcfs_lnd2str(lnd_type));
1234                goto failed0;
1235        }
1236
1237        /* Make sure this new NI is unique. */
1238        lnet_net_lock(LNET_LOCK_EX);
1239        rc = lnet_net_unique(LNET_NIDNET(ni->ni_nid), &the_lnet.ln_nis);
1240        lnet_net_unlock(LNET_LOCK_EX);
1241        if (!rc) {
1242                if (lnd_type == LOLND) {
1243                        lnet_ni_free(ni);
1244                        return 0;
1245                }
1246
1247                CERROR("Net %s is not unique\n",
1248                       libcfs_net2str(LNET_NIDNET(ni->ni_nid)));
1249                rc = -EEXIST;
1250                goto failed0;
1251        }
1252
1253        mutex_lock(&the_lnet.ln_lnd_mutex);
1254        lnd = lnet_find_lnd_by_type(lnd_type);
1255
1256        if (!lnd) {
1257                mutex_unlock(&the_lnet.ln_lnd_mutex);
1258                rc = request_module("%s", libcfs_lnd2modname(lnd_type));
1259                mutex_lock(&the_lnet.ln_lnd_mutex);
1260
1261                lnd = lnet_find_lnd_by_type(lnd_type);
1262                if (!lnd) {
1263                        mutex_unlock(&the_lnet.ln_lnd_mutex);
1264                        CERROR("Can't load LND %s, module %s, rc=%d\n",
1265                               libcfs_lnd2str(lnd_type),
1266                               libcfs_lnd2modname(lnd_type), rc);
1267                        rc = -EINVAL;
1268                        goto failed0;
1269                }
1270        }
1271
1272        lnet_net_lock(LNET_LOCK_EX);
1273        lnd->lnd_refcount++;
1274        lnet_net_unlock(LNET_LOCK_EX);
1275
1276        ni->ni_lnd = lnd;
1277
1278        rc = lnd->lnd_startup(ni);
1279
1280        mutex_unlock(&the_lnet.ln_lnd_mutex);
1281
1282        if (rc) {
1283                LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
1284                                   rc, libcfs_lnd2str(lnd->lnd_type));
1285                lnet_net_lock(LNET_LOCK_EX);
1286                lnd->lnd_refcount--;
1287                lnet_net_unlock(LNET_LOCK_EX);
1288                goto failed0;
1289        }
1290
1291        /*
1292         * If given some LND tunable parameters, parse those now to
1293         * override the values in the NI structure.
1294         */
1295        if (peer_buf_cr >= 0)
1296                ni->ni_peerrtrcredits = peer_buf_cr;
1297        if (peer_timeout >= 0)
1298                ni->ni_peertimeout = peer_timeout;
1299        /*
1300         * TODO
1301         * Note: For now, don't allow the user to change
1302         * peertxcredits as this number is used in the
1303         * IB LND to control queue depth.
1304         * if (peer_cr != -1)
1305         *      ni->ni_peertxcredits = peer_cr;
1306         */
1307        if (credits >= 0)
1308                ni->ni_maxtxcredits = credits;
1309
1310        LASSERT(ni->ni_peertimeout <= 0 || lnd->lnd_query);
1311
1312        lnet_net_lock(LNET_LOCK_EX);
1313        /* refcount for ln_nis */
1314        lnet_ni_addref_locked(ni, 0);
1315        list_add_tail(&ni->ni_list, &the_lnet.ln_nis);
1316        if (ni->ni_cpts) {
1317                lnet_ni_addref_locked(ni, 0);
1318                list_add_tail(&ni->ni_cptlist, &the_lnet.ln_nis_cpt);
1319        }
1320
1321        lnet_net_unlock(LNET_LOCK_EX);
1322
1323        if (lnd->lnd_type == LOLND) {
1324                lnet_ni_addref(ni);
1325                LASSERT(!the_lnet.ln_loni);
1326                the_lnet.ln_loni = ni;
1327                return 0;
1328        }
1329
1330        if (!ni->ni_peertxcredits || !ni->ni_maxtxcredits) {
1331                LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
1332                                   libcfs_lnd2str(lnd->lnd_type),
1333                                   !ni->ni_peertxcredits ?
1334                                   "" : "per-peer ");
1335                /*
1336                 * shutdown the NI since if we get here then it must've already
1337                 * been started
1338                 */
1339                lnet_shutdown_lndni(ni);
1340                return -EINVAL;
1341        }
1342
1343        cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
1344                tq->tq_credits_min =
1345                tq->tq_credits_max =
1346                tq->tq_credits = lnet_ni_tq_credits(ni);
1347        }
1348
1349        CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
1350               libcfs_nid2str(ni->ni_nid), ni->ni_peertxcredits,
1351               lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
1352               ni->ni_peerrtrcredits, ni->ni_peertimeout);
1353
1354        return 0;
1355failed0:
1356        lnet_ni_free(ni);
1357        return rc;
1358}
1359
1360static int
1361lnet_startup_lndnis(struct list_head *nilist)
1362{
1363        struct lnet_ni *ni;
1364        int rc;
1365        int ni_count = 0;
1366
1367        while (!list_empty(nilist)) {
1368                ni = list_entry(nilist->next, lnet_ni_t, ni_list);
1369                list_del(&ni->ni_list);
1370                rc = lnet_startup_lndni(ni, -1, -1, -1, -1);
1371
1372                if (rc < 0)
1373                        goto failed;
1374
1375                ni_count++;
1376        }
1377
1378        return ni_count;
1379failed:
1380        lnet_shutdown_lndnis();
1381
1382        return rc;
1383}
1384
1385/**
1386 * Initialize LNet library.
1387 *
1388 * Automatically called at module loading time. Caller has to call
1389 * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
1390 * latter returned 0. It must be called exactly once.
1391 *
1392 * \retval 0 on success
1393 * \retval -ve on failures.
1394 */
1395int lnet_lib_init(void)
1396{
1397        int rc;
1398
1399        lnet_assert_wire_constants();
1400
1401        memset(&the_lnet, 0, sizeof(the_lnet));
1402
1403        /* refer to global cfs_cpt_table for now */
1404        the_lnet.ln_cpt_table   = cfs_cpt_table;
1405        the_lnet.ln_cpt_number  = cfs_cpt_number(cfs_cpt_table);
1406
1407        LASSERT(the_lnet.ln_cpt_number > 0);
1408        if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
1409                /* we are under risk of consuming all lh_cookie */
1410                CERROR("Can't have %d CPTs for LNet (max allowed is %d), please change setting of CPT-table and retry\n",
1411                       the_lnet.ln_cpt_number, LNET_CPT_MAX);
1412                return -E2BIG;
1413        }
1414
1415        while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
1416                the_lnet.ln_cpt_bits++;
1417
1418        rc = lnet_create_locks();
1419        if (rc) {
1420                CERROR("Can't create LNet global locks: %d\n", rc);
1421                return rc;
1422        }
1423
1424        the_lnet.ln_refcount = 0;
1425        LNetInvalidateHandle(&the_lnet.ln_rc_eqh);
1426        INIT_LIST_HEAD(&the_lnet.ln_lnds);
1427        INIT_LIST_HEAD(&the_lnet.ln_rcd_zombie);
1428        INIT_LIST_HEAD(&the_lnet.ln_rcd_deathrow);
1429
1430        /*
1431         * The hash table size is the number of bits it takes to express the set
1432         * ln_num_routes, minus 1 (better to under estimate than over so we
1433         * don't waste memory).
1434         */
1435        if (rnet_htable_size <= 0)
1436                rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
1437        else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
1438                rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
1439        the_lnet.ln_remote_nets_hbits = max_t(int, 1,
1440                                           order_base_2(rnet_htable_size) - 1);
1441
1442        /*
1443         * All LNDs apart from the LOLND are in separate modules.  They
1444         * register themselves when their module loads, and unregister
1445         * themselves when their module is unloaded.
1446         */
1447        lnet_register_lnd(&the_lolnd);
1448        return 0;
1449}
1450
1451/**
1452 * Finalize LNet library.
1453 *
1454 * \pre lnet_lib_init() called with success.
1455 * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
1456 */
1457void lnet_lib_exit(void)
1458{
1459        LASSERT(!the_lnet.ln_refcount);
1460
1461        while (!list_empty(&the_lnet.ln_lnds))
1462                lnet_unregister_lnd(list_entry(the_lnet.ln_lnds.next,
1463                                               lnd_t, lnd_list));
1464        lnet_destroy_locks();
1465}
1466
1467/**
1468 * Set LNet PID and start LNet interfaces, routing, and forwarding.
1469 *
1470 * Users must call this function at least once before any other functions.
1471 * For each successful call there must be a corresponding call to
1472 * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
1473 * ignored.
1474 *
1475 * The PID used by LNet may be different from the one requested.
1476 * See LNetGetId().
1477 *
1478 * \param requested_pid PID requested by the caller.
1479 *
1480 * \return >= 0 on success, and < 0 error code on failures.
1481 */
1482int
1483LNetNIInit(lnet_pid_t requested_pid)
1484{
1485        int im_a_router = 0;
1486        int rc;
1487        int ni_count;
1488        lnet_ping_info_t *pinfo;
1489        lnet_handle_md_t md_handle;
1490        struct list_head net_head;
1491
1492        INIT_LIST_HEAD(&net_head);
1493
1494        mutex_lock(&the_lnet.ln_api_mutex);
1495
1496        CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
1497
1498        if (the_lnet.ln_refcount > 0) {
1499                rc = the_lnet.ln_refcount++;
1500                mutex_unlock(&the_lnet.ln_api_mutex);
1501                return rc;
1502        }
1503
1504        rc = lnet_prepare(requested_pid);
1505        if (rc) {
1506                mutex_unlock(&the_lnet.ln_api_mutex);
1507                return rc;
1508        }
1509
1510        /* Add in the loopback network */
1511        if (!lnet_ni_alloc(LNET_MKNET(LOLND, 0), NULL, &net_head)) {
1512                rc = -ENOMEM;
1513                goto err_empty_list;
1514        }
1515
1516        /*
1517         * If LNet is being initialized via DLC it is possible
1518         * that the user requests not to load module parameters (ones which
1519         * are supported by DLC) on initialization.  Therefore, make sure not
1520         * to load networks, routes and forwarding from module parameters
1521         * in this case. On cleanup in case of failure only clean up
1522         * routes if it has been loaded
1523         */
1524        if (!the_lnet.ln_nis_from_mod_params) {
1525                rc = lnet_parse_networks(&net_head, lnet_get_networks());
1526                if (rc < 0)
1527                        goto err_empty_list;
1528        }
1529
1530        ni_count = lnet_startup_lndnis(&net_head);
1531        if (ni_count < 0) {
1532                rc = ni_count;
1533                goto err_empty_list;
1534        }
1535
1536        if (!the_lnet.ln_nis_from_mod_params) {
1537                rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
1538                if (rc)
1539                        goto err_shutdown_lndnis;
1540
1541                rc = lnet_check_routes();
1542                if (rc)
1543                        goto err_destory_routes;
1544
1545                rc = lnet_rtrpools_alloc(im_a_router);
1546                if (rc)
1547                        goto err_destory_routes;
1548        }
1549
1550        rc = lnet_acceptor_start();
1551        if (rc)
1552                goto err_destory_routes;
1553
1554        the_lnet.ln_refcount = 1;
1555        /* Now I may use my own API functions... */
1556
1557        rc = lnet_ping_info_setup(&pinfo, &md_handle, ni_count, true);
1558        if (rc)
1559                goto err_acceptor_stop;
1560
1561        lnet_ping_target_update(pinfo, md_handle);
1562
1563        rc = lnet_router_checker_start();
1564        if (rc)
1565                goto err_stop_ping;
1566
1567        lnet_fault_init();
1568        lnet_router_debugfs_init();
1569
1570        mutex_unlock(&the_lnet.ln_api_mutex);
1571
1572        return 0;
1573
1574err_stop_ping:
1575        lnet_ping_target_fini();
1576err_acceptor_stop:
1577        the_lnet.ln_refcount = 0;
1578        lnet_acceptor_stop();
1579err_destory_routes:
1580        if (!the_lnet.ln_nis_from_mod_params)
1581                lnet_destroy_routes();
1582err_shutdown_lndnis:
1583        lnet_shutdown_lndnis();
1584err_empty_list:
1585        lnet_unprepare();
1586        LASSERT(rc < 0);
1587        mutex_unlock(&the_lnet.ln_api_mutex);
1588        while (!list_empty(&net_head)) {
1589                struct lnet_ni *ni;
1590
1591                ni = list_entry(net_head.next, struct lnet_ni, ni_list);
1592                list_del_init(&ni->ni_list);
1593                lnet_ni_free(ni);
1594        }
1595        return rc;
1596}
1597EXPORT_SYMBOL(LNetNIInit);
1598
1599/**
1600 * Stop LNet interfaces, routing, and forwarding.
1601 *
1602 * Users must call this function once for each successful call to LNetNIInit().
1603 * Once the LNetNIFini() operation has been started, the results of pending
1604 * API operations are undefined.
1605 *
1606 * \return always 0 for current implementation.
1607 */
1608int
1609LNetNIFini(void)
1610{
1611        mutex_lock(&the_lnet.ln_api_mutex);
1612
1613        LASSERT(the_lnet.ln_refcount > 0);
1614
1615        if (the_lnet.ln_refcount != 1) {
1616                the_lnet.ln_refcount--;
1617        } else {
1618                LASSERT(!the_lnet.ln_niinit_self);
1619
1620                lnet_fault_fini();
1621                lnet_router_debugfs_fini();
1622                lnet_router_checker_stop();
1623                lnet_ping_target_fini();
1624
1625                /* Teardown fns that use my own API functions BEFORE here */
1626                the_lnet.ln_refcount = 0;
1627
1628                lnet_acceptor_stop();
1629                lnet_destroy_routes();
1630                lnet_shutdown_lndnis();
1631                lnet_unprepare();
1632        }
1633
1634        mutex_unlock(&the_lnet.ln_api_mutex);
1635        return 0;
1636}
1637EXPORT_SYMBOL(LNetNIFini);
1638
1639/**
1640 * Grabs the ni data from the ni structure and fills the out
1641 * parameters
1642 *
1643 * \param[in] ni network       interface structure
1644 * \param[out] cpt_count       the number of cpts the ni is on
1645 * \param[out] nid             Network Interface ID
1646 * \param[out] peer_timeout    NI peer timeout
1647 * \param[out] peer_tx_crdits  NI peer transmit credits
1648 * \param[out] peer_rtr_credits NI peer router credits
1649 * \param[out] max_tx_credits  NI max transmit credit
1650 * \param[out] net_config      Network configuration
1651 */
1652static void
1653lnet_fill_ni_info(struct lnet_ni *ni, __u32 *cpt_count, __u64 *nid,
1654                  int *peer_timeout, int *peer_tx_credits,
1655                  int *peer_rtr_credits, int *max_tx_credits,
1656                  struct lnet_ioctl_net_config *net_config)
1657{
1658        int i;
1659
1660        if (!ni)
1661                return;
1662
1663        if (!net_config)
1664                return;
1665
1666        BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
1667                     ARRAY_SIZE(net_config->ni_interfaces));
1668
1669        for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
1670                if (!ni->ni_interfaces[i])
1671                        break;
1672
1673                strncpy(net_config->ni_interfaces[i],
1674                        ni->ni_interfaces[i],
1675                        sizeof(net_config->ni_interfaces[i]));
1676        }
1677
1678        *nid = ni->ni_nid;
1679        *peer_timeout = ni->ni_peertimeout;
1680        *peer_tx_credits = ni->ni_peertxcredits;
1681        *peer_rtr_credits = ni->ni_peerrtrcredits;
1682        *max_tx_credits = ni->ni_maxtxcredits;
1683
1684        net_config->ni_status = ni->ni_status->ns_status;
1685
1686        if (ni->ni_cpts) {
1687                int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
1688
1689                for (i = 0; i < num_cpts; i++)
1690                        net_config->ni_cpts[i] = ni->ni_cpts[i];
1691
1692                *cpt_count = num_cpts;
1693        }
1694}
1695
1696int
1697lnet_get_net_config(int idx, __u32 *cpt_count, __u64 *nid, int *peer_timeout,
1698                    int *peer_tx_credits, int *peer_rtr_credits,
1699                    int *max_tx_credits,
1700                    struct lnet_ioctl_net_config *net_config)
1701{
1702        struct lnet_ni *ni;
1703        struct list_head *tmp;
1704        int cpt, i = 0;
1705        int rc = -ENOENT;
1706
1707        cpt = lnet_net_lock_current();
1708
1709        list_for_each(tmp, &the_lnet.ln_nis) {
1710                if (i++ != idx)
1711                        continue;
1712
1713                ni = list_entry(tmp, lnet_ni_t, ni_list);
1714                lnet_ni_lock(ni);
1715                lnet_fill_ni_info(ni, cpt_count, nid, peer_timeout,
1716                                  peer_tx_credits, peer_rtr_credits,
1717                                  max_tx_credits, net_config);
1718                lnet_ni_unlock(ni);
1719                rc = 0;
1720                break;
1721        }
1722
1723        lnet_net_unlock(cpt);
1724        return rc;
1725}
1726
1727int
1728lnet_dyn_add_ni(lnet_pid_t requested_pid, char *nets,
1729                __s32 peer_timeout, __s32 peer_cr, __s32 peer_buf_cr,
1730                __s32 credits)
1731{
1732        lnet_ping_info_t *pinfo;
1733        lnet_handle_md_t md_handle;
1734        struct lnet_ni *ni;
1735        struct list_head net_head;
1736        lnet_remotenet_t *rnet;
1737        int rc;
1738
1739        INIT_LIST_HEAD(&net_head);
1740
1741        /* Create a ni structure for the network string */
1742        rc = lnet_parse_networks(&net_head, nets);
1743        if (rc <= 0)
1744                return !rc ? -EINVAL : rc;
1745
1746        mutex_lock(&the_lnet.ln_api_mutex);
1747
1748        if (rc > 1) {
1749                rc = -EINVAL; /* only add one interface per call */
1750                goto failed0;
1751        }
1752
1753        ni = list_entry(net_head.next, struct lnet_ni, ni_list);
1754
1755        lnet_net_lock(LNET_LOCK_EX);
1756        rnet = lnet_find_net_locked(LNET_NIDNET(ni->ni_nid));
1757        lnet_net_unlock(LNET_LOCK_EX);
1758        /*
1759         * make sure that the net added doesn't invalidate the current
1760         * configuration LNet is keeping
1761         */
1762        if (rnet) {
1763                CERROR("Adding net %s will invalidate routing configuration\n",
1764                       nets);
1765                rc = -EUSERS;
1766                goto failed0;
1767        }
1768
1769        rc = lnet_ping_info_setup(&pinfo, &md_handle, 1 + lnet_get_ni_count(),
1770                                  false);
1771        if (rc)
1772                goto failed0;
1773
1774        list_del_init(&ni->ni_list);
1775
1776        rc = lnet_startup_lndni(ni, peer_timeout, peer_cr,
1777                                peer_buf_cr, credits);
1778        if (rc)
1779                goto failed1;
1780
1781        if (ni->ni_lnd->lnd_accept) {
1782                rc = lnet_acceptor_start();
1783                if (rc < 0) {
1784                        /* shutdown the ni that we just started */
1785                        CERROR("Failed to start up acceptor thread\n");
1786                        lnet_shutdown_lndni(ni);
1787                        goto failed1;
1788                }
1789        }
1790
1791        lnet_ping_target_update(pinfo, md_handle);
1792        mutex_unlock(&the_lnet.ln_api_mutex);
1793
1794        return 0;
1795
1796failed1:
1797        lnet_ping_md_unlink(pinfo, &md_handle);
1798        lnet_ping_info_free(pinfo);
1799failed0:
1800        mutex_unlock(&the_lnet.ln_api_mutex);
1801        while (!list_empty(&net_head)) {
1802                ni = list_entry(net_head.next, struct lnet_ni, ni_list);
1803                list_del_init(&ni->ni_list);
1804                lnet_ni_free(ni);
1805        }
1806        return rc;
1807}
1808
1809int
1810lnet_dyn_del_ni(__u32 net)
1811{
1812        lnet_ni_t *ni;
1813        lnet_ping_info_t *pinfo;
1814        lnet_handle_md_t md_handle;
1815        int rc;
1816
1817        /* don't allow userspace to shutdown the LOLND */
1818        if (LNET_NETTYP(net) == LOLND)
1819                return -EINVAL;
1820
1821        mutex_lock(&the_lnet.ln_api_mutex);
1822        /* create and link a new ping info, before removing the old one */
1823        rc = lnet_ping_info_setup(&pinfo, &md_handle,
1824                                  lnet_get_ni_count() - 1, false);
1825        if (rc)
1826                goto out;
1827
1828        ni = lnet_net2ni(net);
1829        if (!ni) {
1830                rc = -EINVAL;
1831                goto failed;
1832        }
1833
1834        /* decrement the reference counter taken by lnet_net2ni() */
1835        lnet_ni_decref_locked(ni, 0);
1836
1837        lnet_shutdown_lndni(ni);
1838
1839        if (!lnet_count_acceptor_nis())
1840                lnet_acceptor_stop();
1841
1842        lnet_ping_target_update(pinfo, md_handle);
1843        goto out;
1844failed:
1845        lnet_ping_md_unlink(pinfo, &md_handle);
1846        lnet_ping_info_free(pinfo);
1847out:
1848        mutex_unlock(&the_lnet.ln_api_mutex);
1849
1850        return rc;
1851}
1852
1853/**
1854 * LNet ioctl handler.
1855 *
1856 */
1857int
1858LNetCtl(unsigned int cmd, void *arg)
1859{
1860        struct libcfs_ioctl_data *data = arg;
1861        struct lnet_ioctl_config_data *config;
1862        lnet_process_id_t id = {0};
1863        lnet_ni_t *ni;
1864        int rc;
1865        unsigned long secs_passed;
1866
1867        switch (cmd) {
1868        case IOC_LIBCFS_GET_NI:
1869                rc = LNetGetId(data->ioc_count, &id);
1870                data->ioc_nid = id.nid;
1871                return rc;
1872
1873        case IOC_LIBCFS_FAIL_NID:
1874                return lnet_fail_nid(data->ioc_nid, data->ioc_count);
1875
1876        case IOC_LIBCFS_ADD_ROUTE:
1877                config = arg;
1878
1879                if (config->cfg_hdr.ioc_len < sizeof(*config))
1880                        return -EINVAL;
1881
1882                mutex_lock(&the_lnet.ln_api_mutex);
1883                rc = lnet_add_route(config->cfg_net,
1884                                    config->cfg_config_u.cfg_route.rtr_hop,
1885                                    config->cfg_nid,
1886                                    config->cfg_config_u.cfg_route.rtr_priority);
1887                if (!rc) {
1888                        rc = lnet_check_routes();
1889                        if (rc)
1890                                lnet_del_route(config->cfg_net,
1891                                               config->cfg_nid);
1892                }
1893                mutex_unlock(&the_lnet.ln_api_mutex);
1894                return rc;
1895
1896        case IOC_LIBCFS_DEL_ROUTE:
1897                config = arg;
1898
1899                if (config->cfg_hdr.ioc_len < sizeof(*config))
1900                        return -EINVAL;
1901
1902                mutex_lock(&the_lnet.ln_api_mutex);
1903                rc = lnet_del_route(config->cfg_net, config->cfg_nid);
1904                mutex_unlock(&the_lnet.ln_api_mutex);
1905                return rc;
1906
1907        case IOC_LIBCFS_GET_ROUTE:
1908                config = arg;
1909
1910                if (config->cfg_hdr.ioc_len < sizeof(*config))
1911                        return -EINVAL;
1912
1913                return lnet_get_route(config->cfg_count,
1914                                      &config->cfg_net,
1915                                      &config->cfg_config_u.cfg_route.rtr_hop,
1916                                      &config->cfg_nid,
1917                                      &config->cfg_config_u.cfg_route.rtr_flags,
1918                                      &config->cfg_config_u.cfg_route.rtr_priority);
1919
1920        case IOC_LIBCFS_GET_NET: {
1921                struct lnet_ioctl_net_config *net_config;
1922                size_t total = sizeof(*config) + sizeof(*net_config);
1923
1924                config = arg;
1925
1926                if (config->cfg_hdr.ioc_len < total)
1927                        return -EINVAL;
1928
1929                net_config = (struct lnet_ioctl_net_config *)
1930                                config->cfg_bulk;
1931                if (!net_config)
1932                        return -EINVAL;
1933
1934                return lnet_get_net_config(config->cfg_count,
1935                                           &config->cfg_ncpts,
1936                                           &config->cfg_nid,
1937                                           &config->cfg_config_u.cfg_net.net_peer_timeout,
1938                                           &config->cfg_config_u.cfg_net.net_peer_tx_credits,
1939                                           &config->cfg_config_u.cfg_net.net_peer_rtr_credits,
1940                                           &config->cfg_config_u.cfg_net.net_max_tx_credits,
1941                                           net_config);
1942        }
1943
1944        case IOC_LIBCFS_GET_LNET_STATS: {
1945                struct lnet_ioctl_lnet_stats *lnet_stats = arg;
1946
1947                if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
1948                        return -EINVAL;
1949
1950                lnet_counters_get(&lnet_stats->st_cntrs);
1951                return 0;
1952        }
1953
1954        case IOC_LIBCFS_CONFIG_RTR:
1955                config = arg;
1956
1957                if (config->cfg_hdr.ioc_len < sizeof(*config))
1958                        return -EINVAL;
1959
1960                mutex_lock(&the_lnet.ln_api_mutex);
1961                if (config->cfg_config_u.cfg_buffers.buf_enable) {
1962                        rc = lnet_rtrpools_enable();
1963                        mutex_unlock(&the_lnet.ln_api_mutex);
1964                        return rc;
1965                }
1966                lnet_rtrpools_disable();
1967                mutex_unlock(&the_lnet.ln_api_mutex);
1968                return 0;
1969
1970        case IOC_LIBCFS_ADD_BUF:
1971                config = arg;
1972
1973                if (config->cfg_hdr.ioc_len < sizeof(*config))
1974                        return -EINVAL;
1975
1976                mutex_lock(&the_lnet.ln_api_mutex);
1977                rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.buf_tiny,
1978                                          config->cfg_config_u.cfg_buffers.buf_small,
1979                                          config->cfg_config_u.cfg_buffers.buf_large);
1980                mutex_unlock(&the_lnet.ln_api_mutex);
1981                return rc;
1982
1983        case IOC_LIBCFS_GET_BUF: {
1984                struct lnet_ioctl_pool_cfg *pool_cfg;
1985                size_t total = sizeof(*config) + sizeof(*pool_cfg);
1986
1987                config = arg;
1988
1989                if (config->cfg_hdr.ioc_len < total)
1990                        return -EINVAL;
1991
1992                pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
1993                return lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
1994        }
1995
1996        case IOC_LIBCFS_GET_PEER_INFO: {
1997                struct lnet_ioctl_peer *peer_info = arg;
1998
1999                if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
2000                        return -EINVAL;
2001
2002                return lnet_get_peer_info(peer_info->pr_count,
2003                        &peer_info->pr_nid,
2004                        peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
2005                        &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
2006                        &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
2007                        &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
2008                        &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
2009                        &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
2010                        &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_rtr_credits,
2011                        &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
2012        }
2013
2014        case IOC_LIBCFS_NOTIFY_ROUTER:
2015                secs_passed = (ktime_get_real_seconds() - data->ioc_u64[0]);
2016                secs_passed *= msecs_to_jiffies(MSEC_PER_SEC);
2017
2018                return lnet_notify(NULL, data->ioc_nid, data->ioc_flags,
2019                                   jiffies - secs_passed);
2020
2021        case IOC_LIBCFS_LNET_DIST:
2022                rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
2023                if (rc < 0 && rc != -EHOSTUNREACH)
2024                        return rc;
2025
2026                data->ioc_u32[0] = rc;
2027                return 0;
2028
2029        case IOC_LIBCFS_TESTPROTOCOMPAT:
2030                lnet_net_lock(LNET_LOCK_EX);
2031                the_lnet.ln_testprotocompat = data->ioc_flags;
2032                lnet_net_unlock(LNET_LOCK_EX);
2033                return 0;
2034
2035        case IOC_LIBCFS_LNET_FAULT:
2036                return lnet_fault_ctl(data->ioc_flags, data);
2037
2038        case IOC_LIBCFS_PING:
2039                id.nid = data->ioc_nid;
2040                id.pid = data->ioc_u32[0];
2041                rc = lnet_ping(id, data->ioc_u32[1], /* timeout */
2042                               data->ioc_pbuf1,
2043                               data->ioc_plen1 / sizeof(lnet_process_id_t));
2044                if (rc < 0)
2045                        return rc;
2046                data->ioc_count = rc;
2047                return 0;
2048
2049        default:
2050                ni = lnet_net2ni(data->ioc_net);
2051                if (!ni)
2052                        return -EINVAL;
2053
2054                if (!ni->ni_lnd->lnd_ctl)
2055                        rc = -EINVAL;
2056                else
2057                        rc = ni->ni_lnd->lnd_ctl(ni, cmd, arg);
2058
2059                lnet_ni_decref(ni);
2060                return rc;
2061        }
2062        /* not reached */
2063}
2064EXPORT_SYMBOL(LNetCtl);
2065
2066void LNetDebugPeer(lnet_process_id_t id)
2067{
2068        lnet_debug_peer(id.nid);
2069}
2070EXPORT_SYMBOL(LNetDebugPeer);
2071
2072/**
2073 * Retrieve the lnet_process_id_t ID of LNet interface at \a index. Note that
2074 * all interfaces share a same PID, as requested by LNetNIInit().
2075 *
2076 * \param index Index of the interface to look up.
2077 * \param id On successful return, this location will hold the
2078 * lnet_process_id_t ID of the interface.
2079 *
2080 * \retval 0 If an interface exists at \a index.
2081 * \retval -ENOENT If no interface has been found.
2082 */
2083int
2084LNetGetId(unsigned int index, lnet_process_id_t *id)
2085{
2086        struct lnet_ni *ni;
2087        struct list_head *tmp;
2088        int cpt;
2089        int rc = -ENOENT;
2090
2091        LASSERT(the_lnet.ln_refcount > 0);
2092
2093        cpt = lnet_net_lock_current();
2094
2095        list_for_each(tmp, &the_lnet.ln_nis) {
2096                if (index--)
2097                        continue;
2098
2099                ni = list_entry(tmp, lnet_ni_t, ni_list);
2100
2101                id->nid = ni->ni_nid;
2102                id->pid = the_lnet.ln_pid;
2103                rc = 0;
2104                break;
2105        }
2106
2107        lnet_net_unlock(cpt);
2108        return rc;
2109}
2110EXPORT_SYMBOL(LNetGetId);
2111
2112/**
2113 * Print a string representation of handle \a h into buffer \a str of
2114 * \a len bytes.
2115 */
2116void
2117LNetSnprintHandle(char *str, int len, lnet_handle_any_t h)
2118{
2119        snprintf(str, len, "%#llx", h.cookie);
2120}
2121EXPORT_SYMBOL(LNetSnprintHandle);
2122
2123static int lnet_ping(lnet_process_id_t id, int timeout_ms,
2124                     lnet_process_id_t __user *ids, int n_ids)
2125{
2126        lnet_handle_eq_t eqh;
2127        lnet_handle_md_t mdh;
2128        lnet_event_t event;
2129        lnet_md_t md = { NULL };
2130        int which;
2131        int unlinked = 0;
2132        int replied = 0;
2133        const int a_long_time = 60000; /* mS */
2134        int infosz;
2135        lnet_ping_info_t *info;
2136        lnet_process_id_t tmpid;
2137        int i;
2138        int nob;
2139        int rc;
2140        int rc2;
2141        sigset_t blocked;
2142
2143        infosz = offsetof(lnet_ping_info_t, pi_ni[n_ids]);
2144
2145        if (n_ids <= 0 ||
2146            id.nid == LNET_NID_ANY ||
2147            timeout_ms > 500000 ||            /* arbitrary limit! */
2148            n_ids > 20)                  /* arbitrary limit! */
2149                return -EINVAL;
2150
2151        if (id.pid == LNET_PID_ANY)
2152                id.pid = LNET_PID_LUSTRE;
2153
2154        LIBCFS_ALLOC(info, infosz);
2155        if (!info)
2156                return -ENOMEM;
2157
2158        /* NB 2 events max (including any unlink event) */
2159        rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &eqh);
2160        if (rc) {
2161                CERROR("Can't allocate EQ: %d\n", rc);
2162                goto out_0;
2163        }
2164
2165        /* initialize md content */
2166        md.start     = info;
2167        md.length    = infosz;
2168        md.threshold = 2; /*GET/REPLY*/
2169        md.max_size  = 0;
2170        md.options   = LNET_MD_TRUNCATE;
2171        md.user_ptr  = NULL;
2172        md.eq_handle = eqh;
2173
2174        rc = LNetMDBind(md, LNET_UNLINK, &mdh);
2175        if (rc) {
2176                CERROR("Can't bind MD: %d\n", rc);
2177                goto out_1;
2178        }
2179
2180        rc = LNetGet(LNET_NID_ANY, mdh, id,
2181                     LNET_RESERVED_PORTAL,
2182                     LNET_PROTO_PING_MATCHBITS, 0);
2183
2184        if (rc) {
2185                /* Don't CERROR; this could be deliberate! */
2186
2187                rc2 = LNetMDUnlink(mdh);
2188                LASSERT(!rc2);
2189
2190                /* NB must wait for the UNLINK event below... */
2191                unlinked = 1;
2192                timeout_ms = a_long_time;
2193        }
2194
2195        do {
2196                /* MUST block for unlink to complete */
2197                if (unlinked)
2198                        blocked = cfs_block_allsigs();
2199
2200                rc2 = LNetEQPoll(&eqh, 1, timeout_ms, &event, &which);
2201
2202                if (unlinked)
2203                        cfs_restore_sigs(blocked);
2204
2205                CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
2206                       (rc2 <= 0) ? -1 : event.type,
2207                       (rc2 <= 0) ? -1 : event.status,
2208                       (rc2 > 0 && event.unlinked) ? " unlinked" : "");
2209
2210                LASSERT(rc2 != -EOVERFLOW);     /* can't miss anything */
2211
2212                if (rc2 <= 0 || event.status) {
2213                        /* timeout or error */
2214                        if (!replied && !rc)
2215                                rc = (rc2 < 0) ? rc2 :
2216                                     !rc2 ? -ETIMEDOUT :
2217                                     event.status;
2218
2219                        if (!unlinked) {
2220                                /* Ensure completion in finite time... */
2221                                LNetMDUnlink(mdh);
2222                                /* No assertion (racing with network) */
2223                                unlinked = 1;
2224                                timeout_ms = a_long_time;
2225                        } else if (!rc2) {
2226                                /* timed out waiting for unlink */
2227                                CWARN("ping %s: late network completion\n",
2228                                      libcfs_id2str(id));
2229                        }
2230                } else if (event.type == LNET_EVENT_REPLY) {
2231                        replied = 1;
2232                        rc = event.mlength;
2233                }
2234
2235        } while (rc2 <= 0 || !event.unlinked);
2236
2237        if (!replied) {
2238                if (rc >= 0)
2239                        CWARN("%s: Unexpected rc >= 0 but no reply!\n",
2240                              libcfs_id2str(id));
2241                rc = -EIO;
2242                goto out_1;
2243        }
2244
2245        nob = rc;
2246        LASSERT(nob >= 0 && nob <= infosz);
2247
2248        rc = -EPROTO;                      /* if I can't parse... */
2249
2250        if (nob < 8) {
2251                /* can't check magic/version */
2252                CERROR("%s: ping info too short %d\n",
2253                       libcfs_id2str(id), nob);
2254                goto out_1;
2255        }
2256
2257        if (info->pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
2258                lnet_swap_pinginfo(info);
2259        } else if (info->pi_magic != LNET_PROTO_PING_MAGIC) {
2260                CERROR("%s: Unexpected magic %08x\n",
2261                       libcfs_id2str(id), info->pi_magic);
2262                goto out_1;
2263        }
2264
2265        if (!(info->pi_features & LNET_PING_FEAT_NI_STATUS)) {
2266                CERROR("%s: ping w/o NI status: 0x%x\n",
2267                       libcfs_id2str(id), info->pi_features);
2268                goto out_1;
2269        }
2270
2271        if (nob < offsetof(lnet_ping_info_t, pi_ni[0])) {
2272                CERROR("%s: Short reply %d(%d min)\n", libcfs_id2str(id),
2273                       nob, (int)offsetof(lnet_ping_info_t, pi_ni[0]));
2274                goto out_1;
2275        }
2276
2277        if (info->pi_nnis < n_ids)
2278                n_ids = info->pi_nnis;
2279
2280        if (nob < offsetof(lnet_ping_info_t, pi_ni[n_ids])) {
2281                CERROR("%s: Short reply %d(%d expected)\n", libcfs_id2str(id),
2282                       nob, (int)offsetof(lnet_ping_info_t, pi_ni[n_ids]));
2283                goto out_1;
2284        }
2285
2286        rc = -EFAULT;                      /* If I SEGV... */
2287
2288        memset(&tmpid, 0, sizeof(tmpid));
2289        for (i = 0; i < n_ids; i++) {
2290                tmpid.pid = info->pi_pid;
2291                tmpid.nid = info->pi_ni[i].ns_nid;
2292                if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
2293                        goto out_1;
2294        }
2295        rc = info->pi_nnis;
2296
2297 out_1:
2298        rc2 = LNetEQFree(eqh);
2299        if (rc2)
2300                CERROR("rc2 %d\n", rc2);
2301        LASSERT(!rc2);
2302
2303 out_0:
2304        LIBCFS_FREE(info, infosz);
2305        return rc;
2306}
2307