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, 2016, 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 #include <linux/log2.h>
  35 #include <linux/ktime.h>
  36
  37 #include <lnet/lib-lnet.h>
  38
  39 #define D_LNI D_CONSOLE
  40
  41 lnet_t      the_lnet;                           /* THE state of the network */
  42 EXPORT_SYMBOL(the_lnet);
  43
  44 static char *ip2nets = "";
  45 module_param(ip2nets, charp, 0444);
  46 MODULE_PARM_DESC(ip2nets, "LNET network <- IP table");
  47
  48 static char *networks = "";
  49 module_param(networks, charp, 0444);
  50 MODULE_PARM_DESC(networks, "local networks");
  51
  52 static char *routes = "";
  53 module_param(routes, charp, 0444);
  54 MODULE_PARM_DESC(routes, "routes to non-local networks");
  55
  56 static int rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
  57 module_param(rnet_htable_size, int, 0444);
  58 MODULE_PARM_DESC(rnet_htable_size, "size of remote network hash table");
  59
  60 static int use_tcp_bonding = false;
  61 module_param(use_tcp_bonding, int, 0444);
  62 MODULE_PARM_DESC(use_tcp_bonding,
  63                  "Set to 1 to use socklnd bonding. 0 to use Multi-Rail");
  64
  65 static int lnet_ping(lnet_process_id_t id, signed long timeout,
  66                      lnet_process_id_t __user *ids, int n_ids);
  67
  68 static char *
  69 lnet_get_routes(void)
  70 {
  71         return routes;
  72 }
  73
  74 static char *
  75 lnet_get_networks(void)
  76 {
  77         char   *nets;
  78         int     rc;
  79
  80         if (*networks != 0 && *ip2nets != 0) {
  81                 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
  82                                    "'ip2nets' but not both at once\n");
  83                 return NULL;
  84         }
  85
  86         if (*ip2nets != 0) {
  87                 rc = lnet_parse_ip2nets(&nets, ip2nets);
  88                 return (rc == 0) ? nets : NULL;
  89         }
  90
  91         if (*networks != 0)
  92                 return networks;
  93
  94         return "tcp";
  95 }
  96
  97 static void
  98 lnet_init_locks(void)
  99 {
 100         spin_lock_init(&the_lnet.ln_eq_wait_lock);
 101         init_waitqueue_head(&the_lnet.ln_eq_waitq);
 102         init_waitqueue_head(&the_lnet.ln_rc_waitq);
 103         mutex_init(&the_lnet.ln_lnd_mutex);
 104         mutex_init(&the_lnet.ln_api_mutex);
 105 }
 106
 107 static void
 108 lnet_fini_locks(void)
 109 {
 110 }
 111
 112 struct kmem_cache *lnet_mes_cachep;        /* MEs kmem_cache */
 113 struct kmem_cache *lnet_small_mds_cachep;  /* <= LNET_SMALL_MD_SIZE bytes
 114                                             *  MDs kmem_cache */
 115
 116 static int
 117 lnet_descriptor_setup(void)
 118 {
 119         /* create specific kmem_cache for MEs and small MDs (i.e., originally
 120          * allocated in <size-xxx> kmem_cache).
 121          */
 122         lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(lnet_me_t),
 123                                             0, 0, NULL);
 124         if (!lnet_mes_cachep)
 125                 return -ENOMEM;
 126
 127         lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
 128                                                   LNET_SMALL_MD_SIZE, 0, 0,
 129                                                   NULL);
 130         if (!lnet_small_mds_cachep)
 131                 return -ENOMEM;
 132
 133         return 0;
 134 }
 135
 136 static void
 137 lnet_descriptor_cleanup(void)
 138 {
 139
 140         if (lnet_small_mds_cachep) {
 141                 kmem_cache_destroy(lnet_small_mds_cachep);
 142                 lnet_small_mds_cachep = NULL;
 143         }
 144
 145         if (lnet_mes_cachep) {
 146                 kmem_cache_destroy(lnet_mes_cachep);
 147                 lnet_mes_cachep = NULL;
 148         }
 149 }
 150
 151 static int
 152 lnet_create_remote_nets_table(void)
 153 {
 154         int               i;
 155         struct list_head *hash;
 156
 157         LASSERT(the_lnet.ln_remote_nets_hash == NULL);
 158         LASSERT(the_lnet.ln_remote_nets_hbits > 0);
 159         LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
 160         if (hash == NULL) {
 161                 CERROR("Failed to create remote nets hash table\n");
 162                 return -ENOMEM;
 163         }
 164
 165         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 166                 INIT_LIST_HEAD(&hash[i]);
 167         the_lnet.ln_remote_nets_hash = hash;
 168         return 0;
 169 }
 170
 171 static void
 172 lnet_destroy_remote_nets_table(void)
 173 {
 174         int i;
 175
 176         if (the_lnet.ln_remote_nets_hash == NULL)
 177                 return;
 178
 179         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 180                 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
 181
 182         LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
 183                     LNET_REMOTE_NETS_HASH_SIZE *
 184                     sizeof(the_lnet.ln_remote_nets_hash[0]));
 185         the_lnet.ln_remote_nets_hash = NULL;
 186 }
 187
 188 static void
 189 lnet_destroy_locks(void)
 190 {
 191         if (the_lnet.ln_res_lock != NULL) {
 192                 cfs_percpt_lock_free(the_lnet.ln_res_lock);
 193                 the_lnet.ln_res_lock = NULL;
 194         }
 195
 196         if (the_lnet.ln_net_lock != NULL) {
 197                 cfs_percpt_lock_free(the_lnet.ln_net_lock);
 198                 the_lnet.ln_net_lock = NULL;
 199         }
 200
 201         lnet_fini_locks();
 202 }
 203
 204 static int
 205 lnet_create_locks(void)
 206 {
 207         lnet_init_locks();
 208
 209         the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 210         if (the_lnet.ln_res_lock == NULL)
 211                 goto failed;
 212
 213         the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 214         if (the_lnet.ln_net_lock == NULL)
 215                 goto failed;
 216
 217         return 0;
 218
 219  failed:
 220         lnet_destroy_locks();
 221         return -ENOMEM;
 222 }
 223
 224 static void lnet_assert_wire_constants(void)
 225 {
 226         /* Wire protocol assertions generated by 'wirecheck'
 227          * running on Linux robert.bartonsoftware.com 2.6.8-1.521
 228          * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
 229          * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
 230
 231         /* Constants... */
 232         CLASSERT(LNET_PROTO_TCP_MAGIC == 0xeebc0ded);
 233         CLASSERT(LNET_PROTO_TCP_VERSION_MAJOR == 1);
 234         CLASSERT(LNET_PROTO_TCP_VERSION_MINOR == 0);
 235         CLASSERT(LNET_MSG_ACK == 0);
 236         CLASSERT(LNET_MSG_PUT == 1);
 237         CLASSERT(LNET_MSG_GET == 2);
 238         CLASSERT(LNET_MSG_REPLY == 3);
 239         CLASSERT(LNET_MSG_HELLO == 4);
 240
 241         /* Checks for struct lnet_handle_wire */
 242         CLASSERT((int)sizeof(struct lnet_handle_wire) == 16);
 243         CLASSERT((int)offsetof(struct lnet_handle_wire, wh_interface_cookie) == 0);
 244         CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) == 8);
 245         CLASSERT((int)offsetof(struct lnet_handle_wire, wh_object_cookie) == 8);
 246         CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) == 8);
 247
 248         /* Checks for struct lnet_magicversion_t */
 249         CLASSERT((int)sizeof(lnet_magicversion_t) == 8);
 250         CLASSERT((int)offsetof(lnet_magicversion_t, magic) == 0);
 251         CLASSERT((int)sizeof(((lnet_magicversion_t *)0)->magic) == 4);
 252         CLASSERT((int)offsetof(lnet_magicversion_t, version_major) == 4);
 253         CLASSERT((int)sizeof(((lnet_magicversion_t *)0)->version_major) == 2);
 254         CLASSERT((int)offsetof(lnet_magicversion_t, version_minor) == 6);
 255         CLASSERT((int)sizeof(((lnet_magicversion_t *)0)->version_minor) == 2);
 256
 257         /* Checks for struct lnet_hdr_t */
 258         CLASSERT((int)sizeof(lnet_hdr_t) == 72);
 259         CLASSERT((int)offsetof(lnet_hdr_t, dest_nid) == 0);
 260         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->dest_nid) == 8);
 261         CLASSERT((int)offsetof(lnet_hdr_t, src_nid) == 8);
 262         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->src_nid) == 8);
 263         CLASSERT((int)offsetof(lnet_hdr_t, dest_pid) == 16);
 264         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->dest_pid) == 4);
 265         CLASSERT((int)offsetof(lnet_hdr_t, src_pid) == 20);
 266         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->src_pid) == 4);
 267         CLASSERT((int)offsetof(lnet_hdr_t, type) == 24);
 268         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->type) == 4);
 269         CLASSERT((int)offsetof(lnet_hdr_t, payload_length) == 28);
 270         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->payload_length) == 4);
 271         CLASSERT((int)offsetof(lnet_hdr_t, msg) == 32);
 272         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg) == 40);
 273
 274         /* Ack */
 275         CLASSERT((int)offsetof(lnet_hdr_t, msg.ack.dst_wmd) == 32);
 276         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.ack.dst_wmd) == 16);
 277         CLASSERT((int)offsetof(lnet_hdr_t, msg.ack.match_bits) == 48);
 278         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.ack.match_bits) == 8);
 279         CLASSERT((int)offsetof(lnet_hdr_t, msg.ack.mlength) == 56);
 280         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.ack.mlength) == 4);
 281
 282         /* Put */
 283         CLASSERT((int)offsetof(lnet_hdr_t, msg.put.ack_wmd) == 32);
 284         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.ack_wmd) == 16);
 285         CLASSERT((int)offsetof(lnet_hdr_t, msg.put.match_bits) == 48);
 286         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.match_bits) == 8);
 287         CLASSERT((int)offsetof(lnet_hdr_t, msg.put.hdr_data) == 56);
 288         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.hdr_data) == 8);
 289         CLASSERT((int)offsetof(lnet_hdr_t, msg.put.ptl_index) == 64);
 290         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.ptl_index) == 4);
 291         CLASSERT((int)offsetof(lnet_hdr_t, msg.put.offset) == 68);
 292         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.offset) == 4);
 293
 294         /* Get */
 295         CLASSERT((int)offsetof(lnet_hdr_t, msg.get.return_wmd) == 32);
 296         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.return_wmd) == 16);
 297         CLASSERT((int)offsetof(lnet_hdr_t, msg.get.match_bits) == 48);
 298         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.match_bits) == 8);
 299         CLASSERT((int)offsetof(lnet_hdr_t, msg.get.ptl_index) == 56);
 300         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.ptl_index) == 4);
 301         CLASSERT((int)offsetof(lnet_hdr_t, msg.get.src_offset) == 60);
 302         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.src_offset) == 4);
 303         CLASSERT((int)offsetof(lnet_hdr_t, msg.get.sink_length) == 64);
 304         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.sink_length) == 4);
 305
 306         /* Reply */
 307         CLASSERT((int)offsetof(lnet_hdr_t, msg.reply.dst_wmd) == 32);
 308         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.reply.dst_wmd) == 16);
 309
 310         /* Hello */
 311         CLASSERT((int)offsetof(lnet_hdr_t, msg.hello.incarnation) == 32);
 312         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.hello.incarnation) == 8);
 313         CLASSERT((int)offsetof(lnet_hdr_t, msg.hello.type) == 40);
 314         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.hello.type) == 4);
 315 }
 316
 317 static lnd_t *lnet_find_lnd_by_type(__u32 type)
 318 {
 319         lnd_t            *lnd;
 320         struct list_head *tmp;
 321
 322         /* holding lnd mutex */
 323         list_for_each(tmp, &the_lnet.ln_lnds) {
 324                 lnd = list_entry(tmp, lnd_t, lnd_list);
 325
 326                 if (lnd->lnd_type == type)
 327                         return lnd;
 328         }
 329         return NULL;
 330 }
 331
 332 void
 333 lnet_register_lnd (lnd_t *lnd)
 334 {
 335         mutex_lock(&the_lnet.ln_lnd_mutex);
 336
 337         LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
 338         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
 339
 340         list_add_tail(&lnd->lnd_list, &the_lnet.ln_lnds);
 341         lnd->lnd_refcount = 0;
 342
 343         CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
 344
 345         mutex_unlock(&the_lnet.ln_lnd_mutex);
 346 }
 347 EXPORT_SYMBOL(lnet_register_lnd);
 348
 349 void
 350 lnet_unregister_lnd (lnd_t *lnd)
 351 {
 352         mutex_lock(&the_lnet.ln_lnd_mutex);
 353
 354         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
 355         LASSERT(lnd->lnd_refcount == 0);
 356
 357         list_del(&lnd->lnd_list);
 358         CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
 359
 360         mutex_unlock(&the_lnet.ln_lnd_mutex);
 361 }
 362 EXPORT_SYMBOL(lnet_unregister_lnd);
 363
 364 void
 365 lnet_counters_get(lnet_counters_t *counters)
 366 {
 367         lnet_counters_t *ctr;
 368         int             i;
 369
 370         memset(counters, 0, sizeof(*counters));
 371
 372         lnet_net_lock(LNET_LOCK_EX);
 373
 374         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
 375                 counters->msgs_max     += ctr->msgs_max;
 376                 counters->msgs_alloc   += ctr->msgs_alloc;
 377                 counters->errors       += ctr->errors;
 378                 counters->send_count   += ctr->send_count;
 379                 counters->recv_count   += ctr->recv_count;
 380                 counters->route_count  += ctr->route_count;
 381                 counters->drop_count   += ctr->drop_count;
 382                 counters->send_length  += ctr->send_length;
 383                 counters->recv_length  += ctr->recv_length;
 384                 counters->route_length += ctr->route_length;
 385                 counters->drop_length  += ctr->drop_length;
 386
 387         }
 388         lnet_net_unlock(LNET_LOCK_EX);
 389 }
 390 EXPORT_SYMBOL(lnet_counters_get);
 391
 392 void
 393 lnet_counters_reset(void)
 394 {
 395         lnet_counters_t *counters;
 396         int             i;
 397
 398         lnet_net_lock(LNET_LOCK_EX);
 399
 400         cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
 401                 memset(counters, 0, sizeof(lnet_counters_t));
 402
 403         lnet_net_unlock(LNET_LOCK_EX);
 404 }
 405
 406 static char *
 407 lnet_res_type2str(int type)
 408 {
 409         switch (type) {
 410         default:
 411                 LBUG();
 412         case LNET_COOKIE_TYPE_MD:
 413                 return "MD";
 414         case LNET_COOKIE_TYPE_ME:
 415                 return "ME";
 416         case LNET_COOKIE_TYPE_EQ:
 417                 return "EQ";
 418         }
 419 }
 420
 421 static void
 422 lnet_res_container_cleanup(struct lnet_res_container *rec)
 423 {
 424         int     count = 0;
 425
 426         if (rec->rec_type == 0) /* not set yet, it's uninitialized */
 427                 return;
 428
 429         while (!list_empty(&rec->rec_active)) {
 430                 struct list_head *e = rec->rec_active.next;
 431
 432                 list_del_init(e);
 433                 if (rec->rec_type == LNET_COOKIE_TYPE_EQ) {
 434                         lnet_eq_free(list_entry(e, lnet_eq_t, eq_list));
 435
 436                 } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
 437                         lnet_md_free(list_entry(e, lnet_libmd_t, md_list));
 438
 439                 } else { /* NB: Active MEs should be attached on portals */
 440                         LBUG();
 441                 }
 442                 count++;
 443         }
 444
 445         if (count > 0) {
 446                 /* Found alive MD/ME/EQ, user really should unlink/free
 447                  * all of them before finalize LNet, but if someone didn't,
 448                  * we have to recycle garbage for him */
 449                 CERROR("%d active elements on exit of %s container\n",
 450                        count, lnet_res_type2str(rec->rec_type));
 451         }
 452
 453         if (rec->rec_lh_hash != NULL) {
 454                 LIBCFS_FREE(rec->rec_lh_hash,
 455                             LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
 456                 rec->rec_lh_hash = NULL;
 457         }
 458
 459         rec->rec_type = 0; /* mark it as finalized */
 460 }
 461
 462 static int
 463 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
 464 {
 465         int     rc = 0;
 466         int     i;
 467
 468         LASSERT(rec->rec_type == 0);
 469
 470         rec->rec_type = type;
 471         INIT_LIST_HEAD(&rec->rec_active);
 472
 473         rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
 474
 475         /* Arbitrary choice of hash table size */
 476         LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
 477                          LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
 478         if (rec->rec_lh_hash == NULL) {
 479                 rc = -ENOMEM;
 480                 goto out;
 481         }
 482
 483         for (i = 0; i < LNET_LH_HASH_SIZE; i++)
 484                 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
 485
 486         return 0;
 487
 488 out:
 489         CERROR("Failed to setup %s resource container\n",
 490                lnet_res_type2str(type));
 491         lnet_res_container_cleanup(rec);
 492         return rc;
 493 }
 494
 495 static void
 496 lnet_res_containers_destroy(struct lnet_res_container **recs)
 497 {
 498         struct lnet_res_container       *rec;
 499         int                             i;
 500
 501         cfs_percpt_for_each(rec, i, recs)
 502                 lnet_res_container_cleanup(rec);
 503
 504         cfs_percpt_free(recs);
 505 }
 506
 507 static struct lnet_res_container **
 508 lnet_res_containers_create(int type)
 509 {
 510         struct lnet_res_container       **recs;
 511         struct lnet_res_container       *rec;
 512         int                             rc;
 513         int                             i;
 514
 515         recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
 516         if (recs == NULL) {
 517                 CERROR("Failed to allocate %s resource containers\n",
 518                        lnet_res_type2str(type));
 519                 return NULL;
 520         }
 521
 522         cfs_percpt_for_each(rec, i, recs) {
 523                 rc = lnet_res_container_setup(rec, i, type);
 524                 if (rc != 0) {
 525                         lnet_res_containers_destroy(recs);
 526                         return NULL;
 527                 }
 528         }
 529
 530         return recs;
 531 }
 532
 533 lnet_libhandle_t *
 534 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
 535 {
 536         /* ALWAYS called with lnet_res_lock held */
 537         struct list_head        *head;
 538         lnet_libhandle_t        *lh;
 539         unsigned int            hash;
 540
 541         if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
 542                 return NULL;
 543
 544         hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
 545         head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
 546
 547         list_for_each_entry(lh, head, lh_hash_chain) {
 548                 if (lh->lh_cookie == cookie)
 549                         return lh;
 550         }
 551
 552         return NULL;
 553 }
 554
 555 void
 556 lnet_res_lh_initialize(struct lnet_res_container *rec, lnet_libhandle_t *lh)
 557 {
 558         /* ALWAYS called with lnet_res_lock held */
 559         unsigned int    ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
 560         unsigned int    hash;
 561
 562         lh->lh_cookie = rec->rec_lh_cookie;
 563         rec->rec_lh_cookie += 1 << ibits;
 564
 565         hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
 566
 567         list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
 568 }
 569
 570 static int lnet_unprepare(void);
 571
 572 static int
 573 lnet_prepare(lnet_pid_t requested_pid)
 574 {
 575         /* Prepare to bring up the network */
 576         struct lnet_res_container **recs;
 577         int                       rc = 0;
 578
 579         if (requested_pid == LNET_PID_ANY) {
 580                 /* Don't instantiate LNET just for me */
 581                 return -ENETDOWN;
 582         }
 583
 584         LASSERT(the_lnet.ln_refcount == 0);
 585
 586         the_lnet.ln_routing = 0;
 587
 588         LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
 589         the_lnet.ln_pid = requested_pid;
 590
 591         INIT_LIST_HEAD(&the_lnet.ln_test_peers);
 592         INIT_LIST_HEAD(&the_lnet.ln_nets);
 593         INIT_LIST_HEAD(&the_lnet.ln_routers);
 594         INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
 595         INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
 596
 597         rc = lnet_descriptor_setup();
 598         if (rc != 0)
 599                 goto failed;
 600
 601         rc = lnet_create_remote_nets_table();
 602         if (rc != 0)
 603                 goto failed;
 604
 605         /*
 606          * NB the interface cookie in wire handles guards against delayed
 607          * replies and ACKs appearing valid after reboot.
 608          */
 609         the_lnet.ln_interface_cookie = ktime_get_real_ns();
 610
 611         the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
 612                                                 sizeof(lnet_counters_t));
 613         if (the_lnet.ln_counters == NULL) {
 614                 CERROR("Failed to allocate counters for LNet\n");
 615                 rc = -ENOMEM;
 616                 goto failed;
 617         }
 618
 619         rc = lnet_peer_tables_create();
 620         if (rc != 0)
 621                 goto failed;
 622
 623         rc = lnet_msg_containers_create();
 624         if (rc != 0)
 625                 goto failed;
 626
 627         rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
 628                                       LNET_COOKIE_TYPE_EQ);
 629         if (rc != 0)
 630                 goto failed;
 631
 632         recs = lnet_res_containers_create(LNET_COOKIE_TYPE_ME);
 633         if (recs == NULL) {
 634                 rc = -ENOMEM;
 635                 goto failed;
 636         }
 637
 638         the_lnet.ln_me_containers = recs;
 639
 640         recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
 641         if (recs == NULL) {
 642                 rc = -ENOMEM;
 643                 goto failed;
 644         }
 645
 646         the_lnet.ln_md_containers = recs;
 647
 648         rc = lnet_portals_create();
 649         if (rc != 0) {
 650                 CERROR("Failed to create portals for LNet: %d\n", rc);
 651                 goto failed;
 652         }
 653
 654         return 0;
 655
 656  failed:
 657         lnet_unprepare();
 658         return rc;
 659 }
 660
 661 static int
 662 lnet_unprepare (void)
 663 {
 664         /* NB no LNET_LOCK since this is the last reference.  All LND instances
 665          * have shut down already, so it is safe to unlink and free all
 666          * descriptors, even those that appear committed to a network op (eg MD
 667          * with non-zero pending count) */
 668
 669         lnet_fail_nid(LNET_NID_ANY, 0);
 670
 671         LASSERT(the_lnet.ln_refcount == 0);
 672         LASSERT(list_empty(&the_lnet.ln_test_peers));
 673         LASSERT(list_empty(&the_lnet.ln_nets));
 674
 675         lnet_portals_destroy();
 676
 677         if (the_lnet.ln_md_containers != NULL) {
 678                 lnet_res_containers_destroy(the_lnet.ln_md_containers);
 679                 the_lnet.ln_md_containers = NULL;
 680         }
 681
 682         if (the_lnet.ln_me_containers != NULL) {
 683                 lnet_res_containers_destroy(the_lnet.ln_me_containers);
 684                 the_lnet.ln_me_containers = NULL;
 685         }
 686
 687         lnet_res_container_cleanup(&the_lnet.ln_eq_container);
 688
 689         lnet_msg_containers_destroy();
 690         lnet_peer_tables_destroy();
 691         lnet_rtrpools_free(0);
 692
 693         if (the_lnet.ln_counters != NULL) {
 694                 cfs_percpt_free(the_lnet.ln_counters);
 695                 the_lnet.ln_counters = NULL;
 696         }
 697         lnet_destroy_remote_nets_table();
 698         lnet_descriptor_cleanup();
 699
 700         return 0;
 701 }
 702
 703 lnet_ni_t  *
 704 lnet_net2ni_locked(__u32 net_id, int cpt)
 705 {
 706         struct lnet_ni   *ni;
 707         struct lnet_net  *net;
 708
 709         LASSERT(cpt != LNET_LOCK_EX);
 710
 711         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
 712                 if (net->net_id == net_id) {
 713                         ni = list_entry(net->net_ni_list.next, struct lnet_ni,
 714                                         ni_netlist);
 715                         return ni;
 716                 }
 717         }
 718
 719         return NULL;
 720 }
 721
 722 lnet_ni_t *
 723 lnet_net2ni(__u32 net)
 724 {
 725         lnet_ni_t *ni;
 726
 727         lnet_net_lock(0);
 728         ni = lnet_net2ni_locked(net, 0);
 729         lnet_net_unlock(0);
 730
 731         return ni;
 732 }
 733 EXPORT_SYMBOL(lnet_net2ni);
 734
 735 struct lnet_net *
 736 lnet_get_net_locked(__u32 net_id)
 737 {
 738         struct lnet_net  *net;
 739
 740         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
 741                 if (net->net_id == net_id)
 742                         return net;
 743         }
 744
 745         return NULL;
 746 }
 747
 748 static unsigned int
 749 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
 750 {
 751         __u64           key = nid;
 752         unsigned int    val;
 753
 754         LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
 755
 756         if (number == 1)
 757                 return 0;
 758
 759         val = hash_long(key, LNET_CPT_BITS);
 760         /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
 761         if (val < number)
 762                 return val;
 763
 764         return (unsigned int)(key + val + (val >> 1)) % number;
 765 }
 766
 767 int
 768 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
 769 {
 770         struct lnet_net *net;
 771
 772         /* must called with hold of lnet_net_lock */
 773         if (LNET_CPT_NUMBER == 1)
 774                 return 0; /* the only one */
 775
 776         /*
 777          * If NI is provided then use the CPT identified in the NI cpt
 778          * list if one exists. If one doesn't exist, then that NI is
 779          * associated with all CPTs and it follows that the net it belongs
 780          * to is implicitly associated with all CPTs, so just hash the nid
 781          * and return that.
 782          */
 783         if (ni != NULL) {
 784                 if (ni->ni_cpts != NULL)
 785                         return ni->ni_cpts[lnet_nid_cpt_hash(nid,
 786                                                              ni->ni_ncpts)];
 787                 else
 788                         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
 789         }
 790
 791         /* no NI provided so look at the net */
 792         net = lnet_get_net_locked(LNET_NIDNET(nid));
 793
 794         if (net != NULL && net->net_cpts != NULL) {
 795                 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
 796         }
 797
 798         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
 799 }
 800
 801 int
 802 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
 803 {
 804         int     cpt;
 805         int     cpt2;
 806
 807         if (LNET_CPT_NUMBER == 1)
 808                 return 0; /* the only one */
 809
 810         cpt = lnet_net_lock_current();
 811
 812         cpt2 = lnet_cpt_of_nid_locked(nid, ni);
 813
 814         lnet_net_unlock(cpt);
 815
 816         return cpt2;
 817 }
 818 EXPORT_SYMBOL(lnet_cpt_of_nid);
 819
 820 int
 821 lnet_islocalnet(__u32 net_id)
 822 {
 823         struct lnet_net *net;
 824         int             cpt;
 825         bool            local;
 826
 827         cpt = lnet_net_lock_current();
 828
 829         net = lnet_get_net_locked(net_id);
 830
 831         local = net != NULL;
 832
 833         lnet_net_unlock(cpt);
 834
 835         return local;
 836 }
 837
 838 bool
 839 lnet_is_ni_healthy_locked(struct lnet_ni *ni)
 840 {
 841         if (ni->ni_state == LNET_NI_STATE_ACTIVE ||
 842             ni->ni_state == LNET_NI_STATE_DEGRADED)
 843                 return true;
 844
 845         return false;
 846 }
 847
 848 lnet_ni_t  *
 849 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
 850 {
 851         struct lnet_net  *net;
 852         struct lnet_ni   *ni;
 853
 854         LASSERT(cpt != LNET_LOCK_EX);
 855
 856         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
 857                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
 858                         if (ni->ni_nid == nid)
 859                                 return ni;
 860                 }
 861         }
 862
 863         return NULL;
 864 }
 865
 866 lnet_ni_t *
 867 lnet_nid2ni_addref(lnet_nid_t nid)
 868 {
 869         lnet_ni_t *ni;
 870
 871         lnet_net_lock(0);
 872         ni = lnet_nid2ni_locked(nid, 0);
 873         if (ni)
 874                 lnet_ni_addref_locked(ni, 0);
 875         lnet_net_unlock(0);
 876
 877         return ni;
 878 }
 879 EXPORT_SYMBOL(lnet_nid2ni_addref);
 880
 881 int
 882 lnet_islocalnid(lnet_nid_t nid)
 883 {
 884         struct lnet_ni  *ni;
 885         int             cpt;
 886
 887         cpt = lnet_net_lock_current();
 888         ni = lnet_nid2ni_locked(nid, cpt);
 889         lnet_net_unlock(cpt);
 890
 891         return ni != NULL;
 892 }
 893
 894 int
 895 lnet_count_acceptor_nets(void)
 896 {
 897         /* Return the # of NIs that need the acceptor. */
 898         int              count = 0;
 899         struct lnet_net  *net;
 900         int              cpt;
 901
 902         cpt = lnet_net_lock_current();
 903         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
 904                 /* all socklnd type networks should have the acceptor
 905                  * thread started */
 906                 if (net->net_lnd->lnd_accept != NULL)
 907                         count++;
 908         }
 909
 910         lnet_net_unlock(cpt);
 911
 912         return count;
 913 }
 914
 915 static struct lnet_ping_info *
 916 lnet_ping_info_create(int num_ni)
 917 {
 918         struct lnet_ping_info *ping_info;
 919         unsigned int     infosz;
 920
 921         infosz = offsetof(struct lnet_ping_info, pi_ni[num_ni]);
 922         LIBCFS_ALLOC(ping_info, infosz);
 923         if (ping_info == NULL) {
 924                 CERROR("Can't allocate ping info[%d]\n", num_ni);
 925                 return NULL;
 926         }
 927
 928         ping_info->pi_nnis = num_ni;
 929         ping_info->pi_pid = the_lnet.ln_pid;
 930         ping_info->pi_magic = LNET_PROTO_PING_MAGIC;
 931         ping_info->pi_features = LNET_PING_FEAT_NI_STATUS;
 932
 933         return ping_info;
 934 }
 935
 936 static inline int
 937 lnet_get_net_ni_count_locked(struct lnet_net *net)
 938 {
 939         struct lnet_ni  *ni;
 940         int             count = 0;
 941
 942         list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
 943                 count++;
 944
 945         return count;
 946 }
 947
 948 static inline int
 949 lnet_get_ni_count(void)
 950 {
 951         struct lnet_ni  *ni;
 952         struct lnet_net *net;
 953         int             count = 0;
 954
 955         lnet_net_lock(0);
 956
 957         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
 958                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
 959                         count++;
 960         }
 961
 962         lnet_net_unlock(0);
 963
 964         return count;
 965 }
 966
 967 static inline void
 968 lnet_ping_info_free(struct lnet_ping_info *pinfo)
 969 {
 970         LIBCFS_FREE(pinfo,
 971                     offsetof(struct lnet_ping_info,
 972                              pi_ni[pinfo->pi_nnis]));
 973 }
 974
 975 static void
 976 lnet_ping_info_destroy(void)
 977 {
 978         struct lnet_net *net;
 979         struct lnet_ni  *ni;
 980
 981         lnet_net_lock(LNET_LOCK_EX);
 982
 983         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
 984                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
 985                         lnet_ni_lock(ni);
 986                         ni->ni_status = NULL;
 987                         lnet_ni_unlock(ni);
 988                 }
 989         }
 990
 991         lnet_ping_info_free(the_lnet.ln_ping_info);
 992         the_lnet.ln_ping_info = NULL;
 993
 994         lnet_net_unlock(LNET_LOCK_EX);
 995 }
 996
 997 static void
 998 lnet_ping_event_handler(lnet_event_t *event)
 999 {
1000         struct lnet_ping_info *pinfo = event->md.user_ptr;
1001
1002         if (event->unlinked)
1003                 pinfo->pi_features = LNET_PING_FEAT_INVAL;
1004 }
1005
1006 static int
1007 lnet_ping_info_setup(struct lnet_ping_info **ppinfo, lnet_handle_md_t *md_handle,
1008                      int ni_count, bool set_eq)
1009 {
1010         lnet_handle_me_t  me_handle;
1011         lnet_process_id_t id = {LNET_NID_ANY, LNET_PID_ANY};
1012         lnet_md_t         md = {NULL};
1013         int               rc, rc2;
1014
1015         if (set_eq) {
1016                 rc = LNetEQAlloc(0, lnet_ping_event_handler,
1017                                  &the_lnet.ln_ping_target_eq);
1018                 if (rc != 0) {
1019                         CERROR("Can't allocate ping EQ: %d\n", rc);
1020                         return rc;
1021                 }
1022         }
1023
1024         *ppinfo = lnet_ping_info_create(ni_count);
1025         if (*ppinfo == NULL) {
1026                 rc = -ENOMEM;
1027                 goto failed_0;
1028         }
1029
1030         rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1031                           LNET_PROTO_PING_MATCHBITS, 0,
1032                           LNET_UNLINK, LNET_INS_AFTER,
1033                           &me_handle);
1034         if (rc != 0) {
1035                 CERROR("Can't create ping ME: %d\n", rc);
1036                 goto failed_1;
1037         }
1038
1039         /* initialize md content */
1040         md.start     = *ppinfo;
1041         md.length    = offsetof(struct lnet_ping_info,
1042                                 pi_ni[(*ppinfo)->pi_nnis]);
1043         md.threshold = LNET_MD_THRESH_INF;
1044         md.max_size  = 0;
1045         md.options   = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1046                        LNET_MD_MANAGE_REMOTE;
1047         md.user_ptr  = NULL;
1048         md.eq_handle = the_lnet.ln_ping_target_eq;
1049         md.user_ptr = *ppinfo;
1050
1051         rc = LNetMDAttach(me_handle, md, LNET_RETAIN, md_handle);
1052         if (rc != 0) {
1053                 CERROR("Can't attach ping MD: %d\n", rc);
1054                 goto failed_2;
1055         }
1056
1057         return 0;
1058
1059 failed_2:
1060         rc2 = LNetMEUnlink(me_handle);
1061         LASSERT(rc2 == 0);
1062 failed_1:
1063         lnet_ping_info_free(*ppinfo);
1064         *ppinfo = NULL;
1065 failed_0:
1066         if (set_eq)
1067                 LNetEQFree(the_lnet.ln_ping_target_eq);
1068         return rc;
1069 }
1070
1071 static void
1072 lnet_ping_md_unlink(struct lnet_ping_info *pinfo, lnet_handle_md_t *md_handle)
1073 {
1074         sigset_t        blocked = cfs_block_allsigs();
1075
1076         LNetMDUnlink(*md_handle);
1077         LNetInvalidateHandle(md_handle);
1078
1079         /* NB md could be busy; this just starts the unlink */
1080         while (pinfo->pi_features != LNET_PING_FEAT_INVAL) {
1081                 CDEBUG(D_NET, "Still waiting for ping MD to unlink\n");
1082                 set_current_state(TASK_UNINTERRUPTIBLE);
1083                 schedule_timeout(cfs_time_seconds(1));
1084         }
1085
1086         cfs_restore_sigs(blocked);
1087 }
1088
1089 static void
1090 lnet_ping_info_install_locked(struct lnet_ping_info *ping_info)
1091 {
1092         int                     i;
1093         struct lnet_ni          *ni;
1094         struct lnet_net         *net;
1095         struct lnet_ni_status *ns;
1096
1097         i = 0;
1098         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1099                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1100                         LASSERT(i < ping_info->pi_nnis);
1101
1102                         ns = &ping_info->pi_ni[i];
1103
1104                         ns->ns_nid = ni->ni_nid;
1105
1106                         lnet_ni_lock(ni);
1107                         ns->ns_status = (ni->ni_status != NULL) ?
1108                                         ni->ni_status->ns_status :
1109                                                 LNET_NI_STATUS_UP;
1110                         ni->ni_status = ns;
1111                         lnet_ni_unlock(ni);
1112
1113                         i++;
1114                 }
1115
1116         }
1117 }
1118
1119 static void
1120 lnet_ping_target_update(struct lnet_ping_info *pinfo, lnet_handle_md_t md_handle)
1121 {
1122         struct lnet_ping_info *old_pinfo = NULL;
1123         lnet_handle_md_t old_md;
1124
1125         /* switch the NIs to point to the new ping info created */
1126         lnet_net_lock(LNET_LOCK_EX);
1127
1128         if (!the_lnet.ln_routing)
1129                 pinfo->pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1130         lnet_ping_info_install_locked(pinfo);
1131
1132         if (the_lnet.ln_ping_info != NULL) {
1133                 old_pinfo = the_lnet.ln_ping_info;
1134                 old_md = the_lnet.ln_ping_target_md;
1135         }
1136         the_lnet.ln_ping_target_md = md_handle;
1137         the_lnet.ln_ping_info = pinfo;
1138
1139         lnet_net_unlock(LNET_LOCK_EX);
1140
1141         if (old_pinfo != NULL) {
1142                 /* unlink the old ping info */
1143                 lnet_ping_md_unlink(old_pinfo, &old_md);
1144                 lnet_ping_info_free(old_pinfo);
1145         }
1146 }
1147
1148 static void
1149 lnet_ping_target_fini(void)
1150 {
1151         int             rc;
1152
1153         lnet_ping_md_unlink(the_lnet.ln_ping_info,
1154                             &the_lnet.ln_ping_target_md);
1155
1156         rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1157         LASSERT(rc == 0);
1158
1159         lnet_ping_info_destroy();
1160 }
1161
1162 static int
1163 lnet_ni_tq_credits(lnet_ni_t *ni)
1164 {
1165         int     credits;
1166
1167         LASSERT(ni->ni_ncpts >= 1);
1168
1169         if (ni->ni_ncpts == 1)
1170                 return ni->ni_net->net_tunables.lct_max_tx_credits;
1171
1172         credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
1173         credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
1174         credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
1175
1176         return credits;
1177 }
1178
1179 static void
1180 lnet_ni_unlink_locked(lnet_ni_t *ni)
1181 {
1182         if (!list_empty(&ni->ni_cptlist)) {
1183                 list_del_init(&ni->ni_cptlist);
1184                 lnet_ni_decref_locked(ni, 0);
1185         }
1186
1187         /* move it to zombie list and nobody can find it anymore */
1188         LASSERT(!list_empty(&ni->ni_netlist));
1189         list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
1190         lnet_ni_decref_locked(ni, 0);
1191 }
1192
1193 static void
1194 lnet_clear_zombies_nis_locked(struct lnet_net *net)
1195 {
1196         int             i;
1197         int             islo;
1198         lnet_ni_t       *ni;
1199         struct list_head *zombie_list = &net->net_ni_zombie;
1200
1201         /*
1202          * Now wait for the NIs I just nuked to show up on the zombie
1203          * list and shut them down in guaranteed thread context
1204          */
1205         i = 2;
1206         while (!list_empty(zombie_list)) {
1207                 int     *ref;
1208                 int     j;
1209
1210                 ni = list_entry(zombie_list->next,
1211                                 lnet_ni_t, ni_netlist);
1212                 list_del_init(&ni->ni_netlist);
1213                 /* the ni should be in deleting state. If it's not it's
1214                  * a bug */
1215                 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
1216                 cfs_percpt_for_each(ref, j, ni->ni_refs) {
1217                         if (*ref == 0)
1218                                 continue;
1219                         /* still busy, add it back to zombie list */
1220                         list_add(&ni->ni_netlist, zombie_list);
1221                         break;
1222                 }
1223
1224                 if (!list_empty(&ni->ni_netlist)) {
1225                         lnet_net_unlock(LNET_LOCK_EX);
1226                         ++i;
1227                         if ((i & (-i)) == i) {
1228                                 CDEBUG(D_WARNING,
1229                                        "Waiting for zombie LNI %s\n",
1230                                        libcfs_nid2str(ni->ni_nid));
1231                         }
1232                         set_current_state(TASK_UNINTERRUPTIBLE);
1233                         schedule_timeout(cfs_time_seconds(1));
1234                         lnet_net_lock(LNET_LOCK_EX);
1235                         continue;
1236                 }
1237
1238                 lnet_net_unlock(LNET_LOCK_EX);
1239
1240                 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
1241
1242                 LASSERT(!in_interrupt());
1243                 (net->net_lnd->lnd_shutdown)(ni);
1244
1245                 if (!islo)
1246                         CDEBUG(D_LNI, "Removed LNI %s\n",
1247                               libcfs_nid2str(ni->ni_nid));
1248
1249                 lnet_ni_free(ni);
1250                 i = 2;
1251                 lnet_net_lock(LNET_LOCK_EX);
1252         }
1253 }
1254
1255 /* shutdown down the NI and release refcount */
1256 static void
1257 lnet_shutdown_lndni(struct lnet_ni *ni)
1258 {
1259         int i;
1260         struct lnet_net *net = ni->ni_net;
1261
1262         lnet_net_lock(LNET_LOCK_EX);
1263         ni->ni_state = LNET_NI_STATE_DELETING;
1264         lnet_ni_unlink_locked(ni);
1265         lnet_net_unlock(LNET_LOCK_EX);
1266
1267         /* clear messages for this NI on the lazy portal */
1268         for (i = 0; i < the_lnet.ln_nportals; i++)
1269                 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
1270
1271         /* Do peer table cleanup for this ni */
1272         lnet_peer_tables_cleanup(ni);
1273
1274         lnet_net_lock(LNET_LOCK_EX);
1275         lnet_clear_zombies_nis_locked(net);
1276         lnet_net_unlock(LNET_LOCK_EX);
1277 }
1278
1279 static void
1280 lnet_shutdown_lndnet(struct lnet_net *net)
1281 {
1282         struct lnet_ni *ni;
1283
1284         lnet_net_lock(LNET_LOCK_EX);
1285
1286         net->net_state = LNET_NET_STATE_DELETING;
1287
1288         list_del_init(&net->net_list);
1289
1290         while (!list_empty(&net->net_ni_list)) {
1291                 ni = list_entry(net->net_ni_list.next,
1292                                 lnet_ni_t, ni_netlist);
1293                 lnet_net_unlock(LNET_LOCK_EX);
1294                 lnet_shutdown_lndni(ni);
1295                 lnet_net_lock(LNET_LOCK_EX);
1296         }
1297
1298         /*
1299          * decrement ref count on lnd only when the entire network goes
1300          * away
1301          */
1302         net->net_lnd->lnd_refcount--;
1303
1304         lnet_net_unlock(LNET_LOCK_EX);
1305
1306         lnet_net_free(net);
1307 }
1308
1309 static void
1310 lnet_shutdown_lndnets(void)
1311 {
1312         struct lnet_net *net;
1313
1314         /* NB called holding the global mutex */
1315
1316         /* All quiet on the API front */
1317         LASSERT(!the_lnet.ln_shutdown);
1318         LASSERT(the_lnet.ln_refcount == 0);
1319
1320         lnet_net_lock(LNET_LOCK_EX);
1321         the_lnet.ln_shutdown = 1;       /* flag shutdown */
1322
1323         while (!list_empty(&the_lnet.ln_nets)) {
1324                 /*
1325                  * move the nets to the zombie list to avoid them being
1326                  * picked up for new work. LONET is also included in the
1327                  * Nets that will be moved to the zombie list
1328                  */
1329                 net = list_entry(the_lnet.ln_nets.next,
1330                                  struct lnet_net, net_list);
1331                 list_move(&net->net_list, &the_lnet.ln_net_zombie);
1332         }
1333
1334         /* Drop the cached loopback Net. */
1335         if (the_lnet.ln_loni != NULL) {
1336                 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
1337                 the_lnet.ln_loni = NULL;
1338         }
1339         lnet_net_unlock(LNET_LOCK_EX);
1340
1341         /* iterate through the net zombie list and delete each net */
1342         while (!list_empty(&the_lnet.ln_net_zombie)) {
1343                 net = list_entry(the_lnet.ln_net_zombie.next,
1344                                  struct lnet_net, net_list);
1345                 lnet_shutdown_lndnet(net);
1346         }
1347
1348         lnet_net_lock(LNET_LOCK_EX);
1349         the_lnet.ln_shutdown = 0;
1350         lnet_net_unlock(LNET_LOCK_EX);
1351 }
1352
1353 static int
1354 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
1355 {
1356         int                     rc = -EINVAL;
1357         struct lnet_tx_queue    *tq;
1358         int                     i;
1359         struct lnet_net         *net = ni->ni_net;
1360
1361         mutex_lock(&the_lnet.ln_lnd_mutex);
1362
1363         if (tun) {
1364                 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
1365                 ni->ni_lnd_tunables_set = true;
1366         }
1367
1368         rc = (net->net_lnd->lnd_startup)(ni);
1369
1370         mutex_unlock(&the_lnet.ln_lnd_mutex);
1371
1372         if (rc != 0) {
1373                 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
1374                                    rc, libcfs_lnd2str(net->net_lnd->lnd_type));
1375                 lnet_net_lock(LNET_LOCK_EX);
1376                 net->net_lnd->lnd_refcount--;
1377                 lnet_net_unlock(LNET_LOCK_EX);
1378                 goto failed0;
1379         }
1380
1381         ni->ni_state = LNET_NI_STATE_ACTIVE;
1382
1383         /* We keep a reference on the loopback net through the loopback NI */
1384         if (net->net_lnd->lnd_type == LOLND) {
1385                 lnet_ni_addref(ni);
1386                 LASSERT(the_lnet.ln_loni == NULL);
1387                 the_lnet.ln_loni = ni;
1388                 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
1389                 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
1390                 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
1391                 ni->ni_net->net_tunables.lct_peer_timeout = 0;
1392                 return 0;
1393         }
1394
1395         if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
1396             ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
1397                 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
1398                                    libcfs_lnd2str(net->net_lnd->lnd_type),
1399                                    ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
1400                                         "" : "per-peer ");
1401                 /* shutdown the NI since if we get here then it must've already
1402                  * been started
1403                  */
1404                 lnet_shutdown_lndni(ni);
1405                 return -EINVAL;
1406         }
1407
1408         cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
1409                 tq->tq_credits_min =
1410                 tq->tq_credits_max =
1411                 tq->tq_credits = lnet_ni_tq_credits(ni);
1412         }
1413
1414         CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
1415                 libcfs_nid2str(ni->ni_nid),
1416                 ni->ni_net->net_tunables.lct_peer_tx_credits,
1417                 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
1418                 ni->ni_net->net_tunables.lct_peer_rtr_credits,
1419                 ni->ni_net->net_tunables.lct_peer_timeout);
1420
1421         return 0;
1422 failed0:
1423         lnet_ni_free(ni);
1424         return rc;
1425 }
1426
1427 static int
1428 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
1429 {
1430         struct lnet_ni          *ni;
1431         struct lnet_net         *net_l = NULL;
1432         struct list_head        local_ni_list;
1433         int                     rc;
1434         int                     ni_count = 0;
1435         __u32                   lnd_type;
1436         lnd_t                   *lnd;
1437         int                     peer_timeout =
1438                 net->net_tunables.lct_peer_timeout;
1439         int                     maxtxcredits =
1440                 net->net_tunables.lct_max_tx_credits;
1441         int                     peerrtrcredits =
1442                 net->net_tunables.lct_peer_rtr_credits;
1443
1444         INIT_LIST_HEAD(&local_ni_list);
1445
1446         /*
1447          * make sure that this net is unique. If it isn't then
1448          * we are adding interfaces to an already existing network, and
1449          * 'net' is just a convenient way to pass in the list.
1450          * if it is unique we need to find the LND and load it if
1451          * necessary.
1452          */
1453         if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
1454                 lnd_type = LNET_NETTYP(net->net_id);
1455
1456                 LASSERT(libcfs_isknown_lnd(lnd_type));
1457
1458                 if (lnd_type == CIBLND || lnd_type == OPENIBLND ||
1459                     lnd_type == IIBLND || lnd_type == VIBLND) {
1460                         CERROR("LND %s obsoleted\n", libcfs_lnd2str(lnd_type));
1461                         rc = -EINVAL;
1462                         goto failed0;
1463                 }
1464
1465                 mutex_lock(&the_lnet.ln_lnd_mutex);
1466                 lnd = lnet_find_lnd_by_type(lnd_type);
1467
1468                 if (lnd == NULL) {
1469                         mutex_unlock(&the_lnet.ln_lnd_mutex);
1470                         rc = request_module("%s", libcfs_lnd2modname(lnd_type));
1471                         mutex_lock(&the_lnet.ln_lnd_mutex);
1472
1473                         lnd = lnet_find_lnd_by_type(lnd_type);
1474                         if (lnd == NULL) {
1475                                 mutex_unlock(&the_lnet.ln_lnd_mutex);
1476                                 CERROR("Can't load LND %s, module %s, rc=%d\n",
1477                                 libcfs_lnd2str(lnd_type),
1478                                 libcfs_lnd2modname(lnd_type), rc);
1479 #ifndef HAVE_MODULE_LOADING_SUPPORT
1480                                 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
1481                                                 "compiled with kernel module "
1482                                                 "loading support.");
1483 #endif
1484                                 rc = -EINVAL;
1485                                 goto failed0;
1486                         }
1487                 }
1488
1489                 lnet_net_lock(LNET_LOCK_EX);
1490                 lnd->lnd_refcount++;
1491                 lnet_net_unlock(LNET_LOCK_EX);
1492
1493                 net->net_lnd = lnd;
1494
1495                 mutex_unlock(&the_lnet.ln_lnd_mutex);
1496
1497                 net_l = net;
1498         }
1499
1500         /*
1501          * net_l: if the network being added is unique then net_l
1502          *        will point to that network
1503          *        if the network being added is not unique then
1504          *        net_l points to the existing network.
1505          *
1506          * When we enter the loop below, we'll pick NIs off he
1507          * network beign added and start them up, then add them to
1508          * a local ni list. Once we've successfully started all
1509          * the NIs then we join the local NI list (of started up
1510          * networks) with the net_l->net_ni_list, which should
1511          * point to the correct network to add the new ni list to
1512          *
1513          * If any of the new NIs fail to start up, then we want to
1514          * iterate through the local ni list, which should include
1515          * any NIs which were successfully started up, and shut
1516          * them down.
1517          *
1518          * After than we want to delete the network being added,
1519          * to avoid a memory leak.
1520          */
1521
1522         /*
1523          * When a network uses TCP bonding then all its interfaces
1524          * must be specified when the network is first defined: the
1525          * TCP bonding code doesn't allow for interfaces to be added
1526          * or removed.
1527          */
1528         if (net_l != net && net_l != NULL && use_tcp_bonding &&
1529             LNET_NETTYP(net_l->net_id) == SOCKLND) {
1530                 rc = -EINVAL;
1531                 goto failed0;
1532         }
1533
1534         while (!list_empty(&net->net_ni_added)) {
1535                 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
1536                                 ni_netlist);
1537                 list_del_init(&ni->ni_netlist);
1538
1539                 /* make sure that the the NI we're about to start
1540                  * up is actually unique. if it's not fail. */
1541                 if (!lnet_ni_unique_net(&net_l->net_ni_list,
1542                                         ni->ni_interfaces[0])) {
1543                         rc = -EINVAL;
1544                         goto failed1;
1545                 }
1546
1547                 /* adjust the pointer the parent network, just in case it
1548                  * the net is a duplicate */
1549                 ni->ni_net = net_l;
1550
1551                 rc = lnet_startup_lndni(ni, tun);
1552
1553                 LASSERT(ni->ni_net->net_tunables.lct_peer_timeout <= 0 ||
1554                         ni->ni_net->net_lnd->lnd_query != NULL);
1555
1556                 if (rc < 0)
1557                         goto failed1;
1558
1559                 lnet_ni_addref(ni);
1560                 list_add_tail(&ni->ni_netlist, &local_ni_list);
1561
1562                 ni_count++;
1563         }
1564
1565         lnet_net_lock(LNET_LOCK_EX);
1566         list_splice_tail(&local_ni_list, &net_l->net_ni_list);
1567         lnet_net_unlock(LNET_LOCK_EX);
1568
1569         /* if the network is not unique then we don't want to keep
1570          * it around after we're done. Free it. Otherwise add that
1571          * net to the global the_lnet.ln_nets */
1572         if (net_l != net && net_l != NULL) {
1573                 /*
1574                  * TODO - note. currently the tunables can not be updated
1575                  * once added
1576                  */
1577                 lnet_net_free(net);
1578         } else {
1579                 net->net_state = LNET_NET_STATE_ACTIVE;
1580                 /*
1581                  * restore tunables after it has been overwitten by the
1582                  * lnd
1583                  */
1584                 if (peer_timeout != -1)
1585                         net->net_tunables.lct_peer_timeout = peer_timeout;
1586                 if (maxtxcredits != -1)
1587                         net->net_tunables.lct_max_tx_credits = maxtxcredits;
1588                 if (peerrtrcredits != -1)
1589                         net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
1590
1591                 lnet_net_lock(LNET_LOCK_EX);
1592                 list_add_tail(&net->net_list, &the_lnet.ln_nets);
1593                 lnet_net_unlock(LNET_LOCK_EX);
1594         }
1595
1596         return ni_count;
1597
1598 failed1:
1599         /*
1600          * shutdown the new NIs that are being started up
1601          * free the NET being started
1602          */
1603         while (!list_empty(&local_ni_list)) {
1604                 ni = list_entry(local_ni_list.next, struct lnet_ni,
1605                                 ni_netlist);
1606
1607                 lnet_shutdown_lndni(ni);
1608         }
1609
1610 failed0:
1611         lnet_net_free(net);
1612
1613         return rc;
1614 }
1615
1616 static int
1617 lnet_startup_lndnets(struct list_head *netlist)
1618 {
1619         struct lnet_net         *net;
1620         int                     rc;
1621         int                     ni_count = 0;
1622
1623         while (!list_empty(netlist)) {
1624                 net = list_entry(netlist->next, struct lnet_net, net_list);
1625                 list_del_init(&net->net_list);
1626
1627                 rc = lnet_startup_lndnet(net, NULL);
1628
1629                 if (rc < 0)
1630                         goto failed;
1631
1632                 ni_count += rc;
1633         }
1634
1635         return ni_count;
1636 failed:
1637         lnet_shutdown_lndnets();
1638
1639         return rc;
1640 }
1641
1642 /**
1643  * Initialize LNet library.
1644  *
1645  * Automatically called at module loading time. Caller has to call
1646  * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
1647  * latter returned 0. It must be called exactly once.
1648  *
1649  * \retval 0 on success
1650  * \retval -ve on failures.
1651  */
1652 int lnet_lib_init(void)
1653 {
1654         int rc;
1655
1656         lnet_assert_wire_constants();
1657
1658         memset(&the_lnet, 0, sizeof(the_lnet));
1659
1660         /* refer to global cfs_cpt_table for now */
1661         the_lnet.ln_cpt_table   = cfs_cpt_table;
1662         the_lnet.ln_cpt_number  = cfs_cpt_number(cfs_cpt_table);
1663
1664         LASSERT(the_lnet.ln_cpt_number > 0);
1665         if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
1666                 /* we are under risk of consuming all lh_cookie */
1667                 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
1668                        "please change setting of CPT-table and retry\n",
1669                        the_lnet.ln_cpt_number, LNET_CPT_MAX);
1670                 return -E2BIG;
1671         }
1672
1673         while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
1674                 the_lnet.ln_cpt_bits++;
1675
1676         rc = lnet_create_locks();
1677         if (rc != 0) {
1678                 CERROR("Can't create LNet global locks: %d\n", rc);
1679                 return rc;
1680         }
1681
1682         the_lnet.ln_refcount = 0;
1683         LNetInvalidateHandle(&the_lnet.ln_rc_eqh);
1684         INIT_LIST_HEAD(&the_lnet.ln_lnds);
1685         INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
1686         INIT_LIST_HEAD(&the_lnet.ln_rcd_zombie);
1687         INIT_LIST_HEAD(&the_lnet.ln_rcd_deathrow);
1688
1689         /* The hash table size is the number of bits it takes to express the set
1690          * ln_num_routes, minus 1 (better to under estimate than over so we
1691          * don't waste memory). */
1692         if (rnet_htable_size <= 0)
1693                 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
1694         else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
1695                 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
1696         the_lnet.ln_remote_nets_hbits = max_t(int, 1,
1697                                            order_base_2(rnet_htable_size) - 1);
1698
1699         /* All LNDs apart from the LOLND are in separate modules.  They
1700          * register themselves when their module loads, and unregister
1701          * themselves when their module is unloaded. */
1702         lnet_register_lnd(&the_lolnd);
1703         return 0;
1704 }
1705
1706 /**
1707  * Finalize LNet library.
1708  *
1709  * \pre lnet_lib_init() called with success.
1710  * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
1711  */
1712 void lnet_lib_exit(void)
1713 {
1714         LASSERT(the_lnet.ln_refcount == 0);
1715
1716         while (!list_empty(&the_lnet.ln_lnds))
1717                 lnet_unregister_lnd(list_entry(the_lnet.ln_lnds.next,
1718                                                lnd_t, lnd_list));
1719         lnet_destroy_locks();
1720 }
1721
1722 /**
1723  * Set LNet PID and start LNet interfaces, routing, and forwarding.
1724  *
1725  * Users must call this function at least once before any other functions.
1726  * For each successful call there must be a corresponding call to
1727  * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
1728  * ignored.
1729  *
1730  * The PID used by LNet may be different from the one requested.
1731  * See LNetGetId().
1732  *
1733  * \param requested_pid PID requested by the caller.
1734  *
1735  * \return >= 0 on success, and < 0 error code on failures.
1736  */
1737 int
1738 LNetNIInit(lnet_pid_t requested_pid)
1739 {
1740         int                     im_a_router = 0;
1741         int                     rc;
1742         int                     ni_count;
1743         struct lnet_ping_info   *pinfo;
1744         lnet_handle_md_t        md_handle;
1745         struct list_head        net_head;
1746         struct lnet_net         *net;
1747
1748         INIT_LIST_HEAD(&net_head);
1749
1750         mutex_lock(&the_lnet.ln_api_mutex);
1751
1752         CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
1753
1754         if (the_lnet.ln_refcount > 0) {
1755                 rc = the_lnet.ln_refcount++;
1756                 mutex_unlock(&the_lnet.ln_api_mutex);
1757                 return rc;
1758         }
1759
1760         rc = lnet_prepare(requested_pid);
1761         if (rc != 0) {
1762                 mutex_unlock(&the_lnet.ln_api_mutex);
1763                 return rc;
1764         }
1765
1766         /* create a network for Loopback network */
1767         net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
1768         if (net == NULL) {
1769                 rc = -ENOMEM;
1770                 goto err_empty_list;
1771         }
1772
1773         /* Add in the loopback NI */
1774         if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
1775                 rc = -ENOMEM;
1776                 goto err_empty_list;
1777         }
1778
1779         /* If LNet is being initialized via DLC it is possible
1780          * that the user requests not to load module parameters (ones which
1781          * are supported by DLC) on initialization.  Therefore, make sure not
1782          * to load networks, routes and forwarding from module parameters
1783          * in this case.  On cleanup in case of failure only clean up
1784          * routes if it has been loaded */
1785         if (!the_lnet.ln_nis_from_mod_params) {
1786                 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
1787                                          use_tcp_bonding);
1788                 if (rc < 0)
1789                         goto err_empty_list;
1790         }
1791
1792         ni_count = lnet_startup_lndnets(&net_head);
1793         if (ni_count < 0) {
1794                 rc = ni_count;
1795                 goto err_empty_list;
1796         }
1797
1798         if (!the_lnet.ln_nis_from_mod_params) {
1799                 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
1800                 if (rc != 0)
1801                         goto err_shutdown_lndnis;
1802
1803                 rc = lnet_check_routes();
1804                 if (rc != 0)
1805                         goto err_destroy_routes;
1806
1807                 rc = lnet_rtrpools_alloc(im_a_router);
1808                 if (rc != 0)
1809                         goto err_destroy_routes;
1810         }
1811
1812         rc = lnet_acceptor_start();
1813         if (rc != 0)
1814                 goto err_destroy_routes;
1815
1816         the_lnet.ln_refcount = 1;
1817         /* Now I may use my own API functions... */
1818
1819         rc = lnet_ping_info_setup(&pinfo, &md_handle, ni_count, true);
1820         if (rc != 0)
1821                 goto err_acceptor_stop;
1822
1823         lnet_ping_target_update(pinfo, md_handle);
1824
1825         rc = lnet_router_checker_start();
1826         if (rc != 0)
1827                 goto err_stop_ping;
1828
1829         lnet_fault_init();
1830         lnet_proc_init();
1831
1832         mutex_unlock(&the_lnet.ln_api_mutex);
1833
1834         return 0;
1835
1836 err_stop_ping:
1837         lnet_ping_target_fini();
1838 err_acceptor_stop:
1839         the_lnet.ln_refcount = 0;
1840         lnet_acceptor_stop();
1841 err_destroy_routes:
1842         if (!the_lnet.ln_nis_from_mod_params)
1843                 lnet_destroy_routes();
1844 err_shutdown_lndnis:
1845         lnet_shutdown_lndnets();
1846 err_empty_list:
1847         lnet_unprepare();
1848         LASSERT(rc < 0);
1849         mutex_unlock(&the_lnet.ln_api_mutex);
1850         while (!list_empty(&net_head)) {
1851                 struct lnet_net *net;
1852
1853                 net = list_entry(net_head.next, struct lnet_net, net_list);
1854                 list_del_init(&net->net_list);
1855                 lnet_net_free(net);
1856         }
1857         return rc;
1858 }
1859 EXPORT_SYMBOL(LNetNIInit);
1860
1861 /**
1862  * Stop LNet interfaces, routing, and forwarding.
1863  *
1864  * Users must call this function once for each successful call to LNetNIInit().
1865  * Once the LNetNIFini() operation has been started, the results of pending
1866  * API operations are undefined.
1867  *
1868  * \return always 0 for current implementation.
1869  */
1870 int
1871 LNetNIFini()
1872 {
1873         mutex_lock(&the_lnet.ln_api_mutex);
1874
1875         LASSERT(the_lnet.ln_refcount > 0);
1876
1877         if (the_lnet.ln_refcount != 1) {
1878                 the_lnet.ln_refcount--;
1879         } else {
1880                 LASSERT(!the_lnet.ln_niinit_self);
1881
1882                 lnet_fault_fini();
1883
1884                 lnet_proc_fini();
1885                 lnet_router_checker_stop();
1886                 lnet_ping_target_fini();
1887
1888                 /* Teardown fns that use my own API functions BEFORE here */
1889                 the_lnet.ln_refcount = 0;
1890
1891                 lnet_acceptor_stop();
1892                 lnet_destroy_routes();
1893                 lnet_shutdown_lndnets();
1894                 lnet_unprepare();
1895         }
1896
1897         mutex_unlock(&the_lnet.ln_api_mutex);
1898         return 0;
1899 }
1900 EXPORT_SYMBOL(LNetNIFini);
1901
1902 /**
1903  * Grabs the ni data from the ni structure and fills the out
1904  * parameters
1905  *
1906  * \param[in] ni network        interface structure
1907  * \param[out] cpt_count        the number of cpts the ni is on
1908  * \param[out] nid              Network Interface ID
1909  * \param[out] peer_timeout     NI peer timeout
1910  * \param[out] peer_tx_crdits   NI peer transmit credits
1911  * \param[out] peer_rtr_credits NI peer router credits
1912  * \param[out] max_tx_credits   NI max transmit credit
1913  * \param[out] net_config       Network configuration
1914  */
1915 static void
1916 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_data *config)
1917 {
1918         struct lnet_ioctl_net_config *net_config;
1919         struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
1920         size_t min_size, tunable_size = 0;
1921         int i;
1922
1923         if (!ni || !config)
1924                 return;
1925
1926         net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
1927         if (!net_config)
1928                 return;
1929
1930         BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
1931                      ARRAY_SIZE(net_config->ni_interfaces));
1932
1933         for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
1934                 if (!ni->ni_interfaces[i])
1935                         break;
1936
1937                 strncpy(net_config->ni_interfaces[i],
1938                         ni->ni_interfaces[i],
1939                         sizeof(net_config->ni_interfaces[i]));
1940         }
1941
1942         config->cfg_nid = ni->ni_nid;
1943         config->cfg_config_u.cfg_net.net_peer_timeout =
1944                 ni->ni_net->net_tunables.lct_peer_timeout;
1945         config->cfg_config_u.cfg_net.net_max_tx_credits =
1946                 ni->ni_net->net_tunables.lct_max_tx_credits;
1947         config->cfg_config_u.cfg_net.net_peer_tx_credits =
1948                 ni->ni_net->net_tunables.lct_peer_tx_credits;
1949         config->cfg_config_u.cfg_net.net_peer_rtr_credits =
1950                 ni->ni_net->net_tunables.lct_peer_rtr_credits;
1951
1952         net_config->ni_status = ni->ni_status->ns_status;
1953
1954         if (ni->ni_cpts) {
1955                 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
1956
1957                 for (i = 0; i < num_cpts; i++)
1958                         net_config->ni_cpts[i] = ni->ni_cpts[i];
1959
1960                 config->cfg_ncpts = num_cpts;
1961         }
1962
1963         /*
1964          * See if user land tools sent in a newer and larger version
1965          * of struct lnet_tunables than what the kernel uses.
1966          */
1967         min_size = sizeof(*config) + sizeof(*net_config);
1968
1969         if (config->cfg_hdr.ioc_len > min_size)
1970                 tunable_size = config->cfg_hdr.ioc_len - min_size;
1971
1972         /* Don't copy too much data to user space */
1973         min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
1974         lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
1975
1976         if (lnd_cfg && min_size) {
1977                 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
1978                 config->cfg_config_u.cfg_net.net_interface_count = 1;
1979
1980                 /* Tell user land that kernel side has less data */
1981                 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
1982                         min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
1983                         config->cfg_hdr.ioc_len -= min_size;
1984                 }
1985         }
1986 }
1987
1988 struct lnet_ni *
1989 lnet_get_ni_idx_locked(int idx)
1990 {
1991         struct lnet_ni          *ni;
1992         struct lnet_net         *net;
1993
1994         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1995                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1996                         if (idx-- == 0)
1997                                 return ni;
1998                 }
1999         }
2000
2001         return NULL;
2002 }
2003
2004 struct lnet_ni *
2005 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2006 {
2007         struct lnet_ni          *ni;
2008         struct lnet_net         *net = mynet;
2009
2010         if (prev == NULL) {
2011                 if (net == NULL)
2012                         net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2013                                         net_list);
2014                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2015                                 ni_netlist);
2016
2017                 return ni;
2018         }
2019
2020         if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
2021                 /* if you reached the end of the ni list and the net is
2022                  * specified, then there are no more nis in that net */
2023                 if (net != NULL)
2024                         return NULL;
2025
2026                 /* we reached the end of this net ni list. move to the
2027                  * next net */
2028                 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
2029                         /* no more nets and no more NIs. */
2030                         return NULL;
2031
2032                 /* get the next net */
2033                 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
2034                                  net_list);
2035                 /* get the ni on it */
2036                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2037                                 ni_netlist);
2038
2039                 return ni;
2040         }
2041
2042         /* there are more nis left */
2043         ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
2044
2045         return ni;
2046 }
2047
2048 int
2049 lnet_get_net_config(struct lnet_ioctl_config_data *config)
2050 {
2051         struct lnet_ni *ni;
2052         int cpt;
2053         int rc = -ENOENT;
2054         int idx = config->cfg_count;
2055
2056         cpt = lnet_net_lock_current();
2057
2058         ni = lnet_get_ni_idx_locked(idx);
2059
2060         if (ni != NULL) {
2061                 rc = 0;
2062                 lnet_ni_lock(ni);
2063                 lnet_fill_ni_info(ni, config);
2064                 lnet_ni_unlock(ni);
2065         }
2066
2067         lnet_net_unlock(cpt);
2068         return rc;
2069 }
2070
2071 int
2072 lnet_dyn_add_ni(lnet_pid_t requested_pid, struct lnet_ioctl_config_data *conf)
2073 {
2074         char                    *nets = conf->cfg_config_u.cfg_net.net_intf;
2075         struct lnet_ping_info   *pinfo;
2076         lnet_handle_md_t        md_handle;
2077         struct lnet_net         *net;
2078         struct list_head        net_head;
2079         int                     rc;
2080         lnet_remotenet_t        *rnet;
2081         int                     net_ni_count;
2082         int                     num_acceptor_nets;
2083         __u32                   net_type;
2084         struct lnet_ioctl_config_lnd_tunables *lnd_tunables = NULL;
2085
2086         INIT_LIST_HEAD(&net_head);
2087
2088         if (conf && conf->cfg_hdr.ioc_len > sizeof(*conf))
2089                 lnd_tunables = (struct lnet_ioctl_config_lnd_tunables *)conf->cfg_bulk;
2090
2091         /* Create a net/ni structures for the network string */
2092         rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
2093         if (rc <= 0)
2094                 return rc == 0 ? -EINVAL : rc;
2095
2096         mutex_lock(&the_lnet.ln_api_mutex);
2097
2098         if (rc > 1) {
2099                 rc = -EINVAL; /* only add one network per call */
2100                 goto failed0;
2101         }
2102
2103         net = list_entry(net_head.next, struct lnet_net, net_list);
2104
2105         lnet_net_lock(LNET_LOCK_EX);
2106         rnet = lnet_find_rnet_locked(net->net_id);
2107         lnet_net_unlock(LNET_LOCK_EX);
2108         /* make sure that the net added doesn't invalidate the current
2109          * configuration LNet is keeping */
2110         if (rnet != NULL) {
2111                 CERROR("Adding net %s will invalidate routing configuration\n",
2112                        nets);
2113                 rc = -EUSERS;
2114                 goto failed0;
2115         }
2116
2117         /*
2118          * make sure you calculate the correct number of slots in the ping
2119          * info. Since the ping info is a flattened list of all the NIs,
2120          * we should allocate enough slots to accomodate the number of NIs
2121          * which will be added.
2122          *
2123          * We can use lnet_get_net_ni_count_locked() since the net is not
2124          * on a public list yet, so locking is not a problem
2125          */
2126         net_ni_count = lnet_get_net_ni_count_locked(net);
2127
2128         rc = lnet_ping_info_setup(&pinfo, &md_handle,
2129                                   net_ni_count + lnet_get_ni_count(),
2130                                   false);
2131         if (rc != 0)
2132                 goto failed0;
2133
2134         list_del_init(&net->net_list);
2135
2136         if (lnd_tunables)
2137                 memcpy(&net->net_tunables,
2138                        &lnd_tunables->lt_cmn, sizeof(lnd_tunables->lt_cmn));
2139
2140         /*
2141          * before starting this network get a count of the current TCP
2142          * networks which require the acceptor thread running. If that
2143          * count is == 0 before we start up this network, then we'd want to
2144          * start up the acceptor thread after starting up this network
2145          */
2146         num_acceptor_nets = lnet_count_acceptor_nets();
2147
2148         /*
2149          * lnd_startup_lndnet() can deallocate 'net' even if it it returns
2150          * success, because we endded up adding interfaces to an existing
2151          * network. So grab the net_type now
2152          */
2153         net_type = LNET_NETTYP(net->net_id);
2154
2155         rc = lnet_startup_lndnet(net,
2156                                  (lnd_tunables) ? &lnd_tunables->lt_tun : NULL);
2157         if (rc < 0)
2158                 goto failed1;
2159
2160         /*
2161          * Start the acceptor thread if this is the first network
2162          * being added that requires the thread.
2163          */
2164         if (net_type == SOCKLND && num_acceptor_nets == 0)
2165         {
2166                 rc = lnet_acceptor_start();
2167                 if (rc < 0) {
2168                         /* shutdown the net that we just started */
2169                         CERROR("Failed to start up acceptor thread\n");
2170                         /*
2171                          * Note that if we needed to start the acceptor
2172                          * thread, then 'net' must have been the first TCP
2173                          * network, therefore was unique, and therefore
2174                          * wasn't deallocated by lnet_startup_lndnet()
2175                          */
2176                         lnet_shutdown_lndnet(net);
2177                         goto failed1;
2178                 }
2179         }
2180
2181         lnet_ping_target_update(pinfo, md_handle);
2182         mutex_unlock(&the_lnet.ln_api_mutex);
2183
2184         return 0;
2185
2186 failed1:
2187         lnet_ping_md_unlink(pinfo, &md_handle);
2188         lnet_ping_info_free(pinfo);
2189 failed0:
2190         mutex_unlock(&the_lnet.ln_api_mutex);
2191         while (!list_empty(&net_head)) {
2192                 net = list_entry(net_head.next, struct lnet_net, net_list);
2193                 list_del_init(&net->net_list);
2194                 lnet_net_free(net);
2195         }
2196         return rc;
2197 }
2198
2199 int
2200 lnet_dyn_del_ni(__u32 net_id)
2201 {
2202         struct lnet_net  *net;
2203         struct lnet_ping_info *pinfo;
2204         lnet_handle_md_t  md_handle;
2205         int               rc;
2206         int               net_ni_count;
2207
2208         /* don't allow userspace to shutdown the LOLND */
2209         if (LNET_NETTYP(net_id) == LOLND)
2210                 return -EINVAL;
2211
2212         mutex_lock(&the_lnet.ln_api_mutex);
2213
2214         lnet_net_lock(0);
2215
2216         net = lnet_get_net_locked(net_id);
2217         if (net == NULL) {
2218                 rc = -EINVAL;
2219                 goto out;
2220         }
2221
2222         net_ni_count = lnet_get_net_ni_count_locked(net);
2223
2224         lnet_net_unlock(0);
2225
2226         /* create and link a new ping info, before removing the old one */
2227         rc = lnet_ping_info_setup(&pinfo, &md_handle,
2228                                   lnet_get_ni_count() - net_ni_count, false);
2229         if (rc != 0)
2230                 goto out;
2231
2232         lnet_shutdown_lndnet(net);
2233
2234         if (lnet_count_acceptor_nets() == 0)
2235                 lnet_acceptor_stop();
2236
2237         lnet_ping_target_update(pinfo, md_handle);
2238
2239 out:
2240         mutex_unlock(&the_lnet.ln_api_mutex);
2241
2242         return rc;
2243 }
2244
2245 /**
2246  * LNet ioctl handler.
2247  *
2248  */
2249 int
2250 LNetCtl(unsigned int cmd, void *arg)
2251 {
2252         struct libcfs_ioctl_data *data = arg;
2253         struct lnet_ioctl_config_data *config;
2254         lnet_process_id_t         id = {0};
2255         lnet_ni_t                *ni;
2256         int                       rc;
2257
2258         BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
2259                      sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
2260
2261         switch (cmd) {
2262         case IOC_LIBCFS_GET_NI:
2263                 rc = LNetGetId(data->ioc_count, &id);
2264                 data->ioc_nid = id.nid;
2265                 return rc;
2266
2267         case IOC_LIBCFS_FAIL_NID:
2268                 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
2269
2270         case IOC_LIBCFS_ADD_ROUTE:
2271                 config = arg;
2272
2273                 if (config->cfg_hdr.ioc_len < sizeof(*config))
2274                         return -EINVAL;
2275
2276                 mutex_lock(&the_lnet.ln_api_mutex);
2277                 rc = lnet_add_route(config->cfg_net,
2278                                     config->cfg_config_u.cfg_route.rtr_hop,
2279                                     config->cfg_nid,
2280                                     config->cfg_config_u.cfg_route.
2281                                         rtr_priority);
2282                 if (rc == 0) {
2283                         rc = lnet_check_routes();
2284                         if (rc != 0)
2285                                 lnet_del_route(config->cfg_net,
2286                                                config->cfg_nid);
2287                 }
2288                 mutex_unlock(&the_lnet.ln_api_mutex);
2289                 return rc;
2290
2291         case IOC_LIBCFS_DEL_ROUTE:
2292                 config = arg;
2293
2294                 if (config->cfg_hdr.ioc_len < sizeof(*config))
2295                         return -EINVAL;
2296
2297                 mutex_lock(&the_lnet.ln_api_mutex);
2298                 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
2299                 mutex_unlock(&the_lnet.ln_api_mutex);
2300                 return rc;
2301
2302         case IOC_LIBCFS_GET_ROUTE:
2303                 config = arg;
2304
2305                 if (config->cfg_hdr.ioc_len < sizeof(*config))
2306                         return -EINVAL;
2307
2308                 return lnet_get_route(config->cfg_count,
2309                                       &config->cfg_net,
2310                                       &config->cfg_config_u.cfg_route.rtr_hop,
2311                                       &config->cfg_nid,
2312                                       &config->cfg_config_u.cfg_route.rtr_flags,
2313                                       &config->cfg_config_u.cfg_route.
2314                                         rtr_priority);
2315
2316         case IOC_LIBCFS_GET_NET: {
2317                 size_t total = sizeof(*config) +
2318                                sizeof(struct lnet_ioctl_net_config);
2319                 config = arg;
2320
2321                 if (config->cfg_hdr.ioc_len < total)
2322                         return -EINVAL;
2323
2324                 return lnet_get_net_config(config);
2325         }
2326
2327         case IOC_LIBCFS_GET_LNET_STATS:
2328         {
2329                 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
2330
2331                 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
2332                         return -EINVAL;
2333
2334                 lnet_counters_get(&lnet_stats->st_cntrs);
2335                 return 0;
2336         }
2337
2338         case IOC_LIBCFS_CONFIG_RTR:
2339                 config = arg;
2340
2341                 if (config->cfg_hdr.ioc_len < sizeof(*config))
2342                         return -EINVAL;
2343
2344                 mutex_lock(&the_lnet.ln_api_mutex);
2345                 if (config->cfg_config_u.cfg_buffers.buf_enable) {
2346                         rc = lnet_rtrpools_enable();
2347                         mutex_unlock(&the_lnet.ln_api_mutex);
2348                         return rc;
2349                 }
2350                 lnet_rtrpools_disable();
2351                 mutex_unlock(&the_lnet.ln_api_mutex);
2352                 return 0;
2353
2354         case IOC_LIBCFS_ADD_BUF:
2355                 config = arg;
2356
2357                 if (config->cfg_hdr.ioc_len < sizeof(*config))
2358                         return -EINVAL;
2359
2360                 mutex_lock(&the_lnet.ln_api_mutex);
2361                 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
2362                                                 buf_tiny,
2363                                           config->cfg_config_u.cfg_buffers.
2364                                                 buf_small,
2365                                           config->cfg_config_u.cfg_buffers.
2366                                                 buf_large);
2367                 mutex_unlock(&the_lnet.ln_api_mutex);
2368                 return rc;
2369
2370         case IOC_LIBCFS_GET_BUF: {
2371                 struct lnet_ioctl_pool_cfg *pool_cfg;
2372                 size_t total = sizeof(*config) + sizeof(*pool_cfg);
2373
2374                 config = arg;
2375
2376                 if (config->cfg_hdr.ioc_len < total)
2377                         return -EINVAL;
2378
2379                 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
2380                 return lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
2381         }
2382
2383         case IOC_LIBCFS_GET_PEER_INFO: {
2384                 struct lnet_ioctl_peer *peer_info = arg;
2385
2386                 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
2387                         return -EINVAL;
2388
2389                 return lnet_get_peer_info(
2390                    peer_info->pr_count,
2391                    &peer_info->pr_nid,
2392                    peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
2393                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
2394                    &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
2395                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
2396                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
2397                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
2398                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_rtr_credits,
2399                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
2400         }
2401
2402         case IOC_LIBCFS_NOTIFY_ROUTER: {
2403                 unsigned long jiffies_passed;
2404
2405                 jiffies_passed = ktime_get_real_seconds() - data->ioc_u64[0];
2406                 jiffies_passed = cfs_time_seconds(jiffies_passed);
2407
2408                 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags,
2409                                    jiffies - jiffies_passed);
2410         }
2411
2412         case IOC_LIBCFS_LNET_DIST:
2413                 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
2414                 if (rc < 0 && rc != -EHOSTUNREACH)
2415                         return rc;
2416
2417                 data->ioc_u32[0] = rc;
2418                 return 0;
2419
2420         case IOC_LIBCFS_TESTPROTOCOMPAT:
2421                 lnet_net_lock(LNET_LOCK_EX);
2422                 the_lnet.ln_testprotocompat = data->ioc_flags;
2423                 lnet_net_unlock(LNET_LOCK_EX);
2424                 return 0;
2425
2426         case IOC_LIBCFS_LNET_FAULT:
2427                 return lnet_fault_ctl(data->ioc_flags, data);
2428
2429         case IOC_LIBCFS_PING: {
2430                 signed long timeout;
2431
2432                 id.nid = data->ioc_nid;
2433                 id.pid = data->ioc_u32[0];
2434
2435                 /* Don't block longer than 2 minutes */
2436                 if (data->ioc_u32[1] > 120 * MSEC_PER_SEC)
2437                         return -EINVAL;
2438
2439                 /* If timestamp is negative then disable timeout */
2440                 if ((s32)data->ioc_u32[1] < 0)
2441                         timeout = MAX_SCHEDULE_TIMEOUT;
2442                 else
2443                         timeout = msecs_to_jiffies(data->ioc_u32[1]);
2444
2445                 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
2446                                data->ioc_plen1 / sizeof(lnet_process_id_t));
2447                 if (rc < 0)
2448                         return rc;
2449                 data->ioc_count = rc;
2450                 return 0;
2451         }
2452         default:
2453                 ni = lnet_net2ni(data->ioc_net);
2454                 if (ni == NULL)
2455                         return -EINVAL;
2456
2457                 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
2458                         rc = -EINVAL;
2459                 else
2460                         rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
2461
2462                 return rc;
2463         }
2464         /* not reached */
2465 }
2466 EXPORT_SYMBOL(LNetCtl);
2467
2468 void LNetDebugPeer(lnet_process_id_t id)
2469 {
2470         lnet_debug_peer(id.nid);
2471 }
2472 EXPORT_SYMBOL(LNetDebugPeer);
2473
2474 /**
2475  * Retrieve the lnet_process_id_t ID of LNet interface at \a index. Note that
2476  * all interfaces share a same PID, as requested by LNetNIInit().
2477  *
2478  * \param index Index of the interface to look up.
2479  * \param id On successful return, this location will hold the
2480  * lnet_process_id_t ID of the interface.
2481  *
2482  * \retval 0 If an interface exists at \a index.
2483  * \retval -ENOENT If no interface has been found.
2484  */
2485 int
2486 LNetGetId(unsigned int index, lnet_process_id_t *id)
2487 {
2488         struct lnet_ni   *ni;
2489         struct lnet_net  *net;
2490         int               cpt;
2491         int               rc = -ENOENT;
2492
2493         LASSERT(the_lnet.ln_refcount > 0);
2494
2495         cpt = lnet_net_lock_current();
2496
2497         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2498                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2499                         if (index-- != 0)
2500                                 continue;
2501
2502                         id->nid = ni->ni_nid;
2503                         id->pid = the_lnet.ln_pid;
2504                         rc = 0;
2505                         break;
2506                 }
2507         }
2508
2509         lnet_net_unlock(cpt);
2510         return rc;
2511 }
2512 EXPORT_SYMBOL(LNetGetId);
2513
2514 /**
2515  * Print a string representation of handle \a h into buffer \a str of
2516  * \a len bytes.
2517  */
2518 void
2519 LNetSnprintHandle(char *str, int len, lnet_handle_any_t h)
2520 {
2521         snprintf(str, len, "%#llx", h.cookie);
2522 }
2523 EXPORT_SYMBOL(LNetSnprintHandle);
2524
2525 static int lnet_ping(lnet_process_id_t id, signed long timeout,
2526                      lnet_process_id_t __user *ids, int n_ids)
2527 {
2528         lnet_handle_eq_t     eqh;
2529         lnet_handle_md_t     mdh;
2530         lnet_event_t         event;
2531         lnet_md_t            md = { NULL };
2532         int                  which;
2533         int                  unlinked = 0;
2534         int                  replied = 0;
2535         const signed long a_long_time = msecs_to_jiffies(60 * MSEC_PER_SEC);
2536         int                  infosz;
2537         struct lnet_ping_info    *info;
2538         lnet_process_id_t    tmpid;
2539         int                  i;
2540         int                  nob;
2541         int                  rc;
2542         int                  rc2;
2543         sigset_t         blocked;
2544
2545         infosz = offsetof(struct lnet_ping_info, pi_ni[n_ids]);
2546
2547         /* n_ids limit is arbitrary */
2548         if (n_ids <= 0 || n_ids > 20 || id.nid == LNET_NID_ANY)
2549                 return -EINVAL;
2550
2551         if (id.pid == LNET_PID_ANY)
2552                 id.pid = LNET_PID_LUSTRE;
2553
2554         LIBCFS_ALLOC(info, infosz);
2555         if (info == NULL)
2556                 return -ENOMEM;
2557
2558         /* NB 2 events max (including any unlink event) */
2559         rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &eqh);
2560         if (rc != 0) {
2561                 CERROR("Can't allocate EQ: %d\n", rc);
2562                 goto out_0;
2563         }
2564
2565         /* initialize md content */
2566         md.start     = info;
2567         md.length    = infosz;
2568         md.threshold = 2; /*GET/REPLY*/
2569         md.max_size  = 0;
2570         md.options   = LNET_MD_TRUNCATE;
2571         md.user_ptr  = NULL;
2572         md.eq_handle = eqh;
2573
2574         rc = LNetMDBind(md, LNET_UNLINK, &mdh);
2575         if (rc != 0) {
2576                 CERROR("Can't bind MD: %d\n", rc);
2577                 goto out_1;
2578         }
2579
2580         rc = LNetGet(LNET_NID_ANY, mdh, id,
2581                      LNET_RESERVED_PORTAL,
2582                      LNET_PROTO_PING_MATCHBITS, 0);
2583
2584         if (rc != 0) {
2585                 /* Don't CERROR; this could be deliberate! */
2586
2587                 rc2 = LNetMDUnlink(mdh);
2588                 LASSERT(rc2 == 0);
2589
2590                 /* NB must wait for the UNLINK event below... */
2591                 unlinked = 1;
2592                 timeout = a_long_time;
2593         }
2594
2595         do {
2596                 /* MUST block for unlink to complete */
2597                 if (unlinked)
2598                         blocked = cfs_block_allsigs();
2599
2600                 rc2 = LNetEQPoll(&eqh, 1, timeout, &event, &which);
2601
2602                 if (unlinked)
2603                         cfs_restore_sigs(blocked);
2604
2605                 CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
2606                        (rc2 <= 0) ? -1 : event.type,
2607                        (rc2 <= 0) ? -1 : event.status,
2608                        (rc2 > 0 && event.unlinked) ? " unlinked" : "");
2609
2610                 LASSERT(rc2 != -EOVERFLOW);     /* can't miss anything */
2611
2612                 if (rc2 <= 0 || event.status != 0) {
2613                         /* timeout or error */
2614                         if (!replied && rc == 0)
2615                                 rc = (rc2 < 0) ? rc2 :
2616                                      (rc2 == 0) ? -ETIMEDOUT :
2617                                      event.status;
2618
2619                         if (!unlinked) {
2620                                 /* Ensure completion in finite time... */
2621                                 LNetMDUnlink(mdh);
2622                                 /* No assertion (racing with network) */
2623                                 unlinked = 1;
2624                                 timeout = a_long_time;
2625                         } else if (rc2 == 0) {
2626                                 /* timed out waiting for unlink */
2627                                 CWARN("ping %s: late network completion\n",
2628                                       libcfs_id2str(id));
2629                         }
2630                 } else if (event.type == LNET_EVENT_REPLY) {
2631                         replied = 1;
2632                         rc = event.mlength;
2633                 }
2634
2635         } while (rc2 <= 0 || !event.unlinked);
2636
2637         if (!replied) {
2638                 if (rc >= 0)
2639                         CWARN("%s: Unexpected rc >= 0 but no reply!\n",
2640                               libcfs_id2str(id));
2641                 rc = -EIO;
2642                 goto out_1;
2643         }
2644
2645         nob = rc;
2646         LASSERT(nob >= 0 && nob <= infosz);
2647
2648         rc = -EPROTO;                           /* if I can't parse... */
2649
2650         if (nob < 8) {
2651                 /* can't check magic/version */
2652                 CERROR("%s: ping info too short %d\n",
2653                        libcfs_id2str(id), nob);
2654                 goto out_1;
2655         }
2656
2657         if (info->pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
2658                 lnet_swap_pinginfo(info);
2659         } else if (info->pi_magic != LNET_PROTO_PING_MAGIC) {
2660                 CERROR("%s: Unexpected magic %08x\n",
2661                        libcfs_id2str(id), info->pi_magic);
2662                 goto out_1;
2663         }
2664
2665         if ((info->pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
2666                 CERROR("%s: ping w/o NI status: 0x%x\n",
2667                        libcfs_id2str(id), info->pi_features);
2668                 goto out_1;
2669         }
2670
2671         if (nob < offsetof(struct lnet_ping_info, pi_ni[0])) {
2672                 CERROR("%s: Short reply %d(%d min)\n", libcfs_id2str(id),
2673                        nob, (int)offsetof(struct lnet_ping_info, pi_ni[0]));
2674                 goto out_1;
2675         }
2676
2677         if (info->pi_nnis < n_ids)
2678                 n_ids = info->pi_nnis;
2679
2680         if (nob < offsetof(struct lnet_ping_info, pi_ni[n_ids])) {
2681                 CERROR("%s: Short reply %d(%d expected)\n", libcfs_id2str(id),
2682                        nob, (int)offsetof(struct lnet_ping_info, pi_ni[n_ids]));
2683                 goto out_1;
2684         }
2685
2686         rc = -EFAULT;                           /* If I SEGV... */
2687
2688         memset(&tmpid, 0, sizeof(tmpid));
2689         for (i = 0; i < n_ids; i++) {
2690                 tmpid.pid = info->pi_pid;
2691                 tmpid.nid = info->pi_ni[i].ns_nid;
2692                 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
2693                         goto out_1;
2694         }
2695         rc = info->pi_nnis;
2696
2697  out_1:
2698         rc2 = LNetEQFree(eqh);
2699         if (rc2 != 0)
2700                 CERROR("rc2 %d\n", rc2);
2701         LASSERT(rc2 == 0);
2702
2703  out_0:
2704         LIBCFS_FREE(info, infosz);
2705         return rc;
2706 }