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 __u32 lnet_numa_range = 0;
  66 module_param(lnet_numa_range, int, 0444);
  67 MODULE_PARM_DESC(lnet_numa_range,
  68                 "NUMA range to consider during Multi-Rail selection");
  69
  70 /*
  71  * This sequence number keeps track of how many times DLC was used to
  72  * update the configuration. It is incremented on any DLC update and
  73  * checked when sending a message to determine if there is a need to
  74  * re-run the selection algorithm to handle configuration change.
  75  * Look at lnet_select_pathway() for more details on its usage.
  76  */
  77 static atomic_t lnet_dlc_seq_no = ATOMIC_INIT(0);
  78
  79 static int lnet_ping(lnet_process_id_t id, signed long timeout,
  80                      lnet_process_id_t __user *ids, int n_ids);
  81
  82 static char *
  83 lnet_get_routes(void)
  84 {
  85         return routes;
  86 }
  87
  88 static char *
  89 lnet_get_networks(void)
  90 {
  91         char   *nets;
  92         int     rc;
  93
  94         if (*networks != 0 && *ip2nets != 0) {
  95                 LCONSOLE_ERROR_MSG(0x101, "Please specify EITHER 'networks' or "
  96                                    "'ip2nets' but not both at once\n");
  97                 return NULL;
  98         }
  99
 100         if (*ip2nets != 0) {
 101                 rc = lnet_parse_ip2nets(&nets, ip2nets);
 102                 return (rc == 0) ? nets : NULL;
 103         }
 104
 105         if (*networks != 0)
 106                 return networks;
 107
 108         return "tcp";
 109 }
 110
 111 static void
 112 lnet_init_locks(void)
 113 {
 114         spin_lock_init(&the_lnet.ln_eq_wait_lock);
 115         init_waitqueue_head(&the_lnet.ln_eq_waitq);
 116         init_waitqueue_head(&the_lnet.ln_rc_waitq);
 117         mutex_init(&the_lnet.ln_lnd_mutex);
 118         mutex_init(&the_lnet.ln_api_mutex);
 119 }
 120
 121 static void
 122 lnet_fini_locks(void)
 123 {
 124 }
 125
 126 struct kmem_cache *lnet_mes_cachep;        /* MEs kmem_cache */
 127 struct kmem_cache *lnet_small_mds_cachep;  /* <= LNET_SMALL_MD_SIZE bytes
 128                                             *  MDs kmem_cache */
 129
 130 static int
 131 lnet_descriptor_setup(void)
 132 {
 133         /* create specific kmem_cache for MEs and small MDs (i.e., originally
 134          * allocated in <size-xxx> kmem_cache).
 135          */
 136         lnet_mes_cachep = kmem_cache_create("lnet_MEs", sizeof(lnet_me_t),
 137                                             0, 0, NULL);
 138         if (!lnet_mes_cachep)
 139                 return -ENOMEM;
 140
 141         lnet_small_mds_cachep = kmem_cache_create("lnet_small_MDs",
 142                                                   LNET_SMALL_MD_SIZE, 0, 0,
 143                                                   NULL);
 144         if (!lnet_small_mds_cachep)
 145                 return -ENOMEM;
 146
 147         return 0;
 148 }
 149
 150 static void
 151 lnet_descriptor_cleanup(void)
 152 {
 153
 154         if (lnet_small_mds_cachep) {
 155                 kmem_cache_destroy(lnet_small_mds_cachep);
 156                 lnet_small_mds_cachep = NULL;
 157         }
 158
 159         if (lnet_mes_cachep) {
 160                 kmem_cache_destroy(lnet_mes_cachep);
 161                 lnet_mes_cachep = NULL;
 162         }
 163 }
 164
 165 static int
 166 lnet_create_remote_nets_table(void)
 167 {
 168         int               i;
 169         struct list_head *hash;
 170
 171         LASSERT(the_lnet.ln_remote_nets_hash == NULL);
 172         LASSERT(the_lnet.ln_remote_nets_hbits > 0);
 173         LIBCFS_ALLOC(hash, LNET_REMOTE_NETS_HASH_SIZE * sizeof(*hash));
 174         if (hash == NULL) {
 175                 CERROR("Failed to create remote nets hash table\n");
 176                 return -ENOMEM;
 177         }
 178
 179         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 180                 INIT_LIST_HEAD(&hash[i]);
 181         the_lnet.ln_remote_nets_hash = hash;
 182         return 0;
 183 }
 184
 185 static void
 186 lnet_destroy_remote_nets_table(void)
 187 {
 188         int i;
 189
 190         if (the_lnet.ln_remote_nets_hash == NULL)
 191                 return;
 192
 193         for (i = 0; i < LNET_REMOTE_NETS_HASH_SIZE; i++)
 194                 LASSERT(list_empty(&the_lnet.ln_remote_nets_hash[i]));
 195
 196         LIBCFS_FREE(the_lnet.ln_remote_nets_hash,
 197                     LNET_REMOTE_NETS_HASH_SIZE *
 198                     sizeof(the_lnet.ln_remote_nets_hash[0]));
 199         the_lnet.ln_remote_nets_hash = NULL;
 200 }
 201
 202 static void
 203 lnet_destroy_locks(void)
 204 {
 205         if (the_lnet.ln_res_lock != NULL) {
 206                 cfs_percpt_lock_free(the_lnet.ln_res_lock);
 207                 the_lnet.ln_res_lock = NULL;
 208         }
 209
 210         if (the_lnet.ln_net_lock != NULL) {
 211                 cfs_percpt_lock_free(the_lnet.ln_net_lock);
 212                 the_lnet.ln_net_lock = NULL;
 213         }
 214
 215         lnet_fini_locks();
 216 }
 217
 218 static int
 219 lnet_create_locks(void)
 220 {
 221         lnet_init_locks();
 222
 223         the_lnet.ln_res_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 224         if (the_lnet.ln_res_lock == NULL)
 225                 goto failed;
 226
 227         the_lnet.ln_net_lock = cfs_percpt_lock_alloc(lnet_cpt_table());
 228         if (the_lnet.ln_net_lock == NULL)
 229                 goto failed;
 230
 231         return 0;
 232
 233  failed:
 234         lnet_destroy_locks();
 235         return -ENOMEM;
 236 }
 237
 238 static void lnet_assert_wire_constants(void)
 239 {
 240         /* Wire protocol assertions generated by 'wirecheck'
 241          * running on Linux robert.bartonsoftware.com 2.6.8-1.521
 242          * #1 Mon Aug 16 09:01:18 EDT 2004 i686 athlon i386 GNU/Linux
 243          * with gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7) */
 244
 245         /* Constants... */
 246         CLASSERT(LNET_PROTO_TCP_MAGIC == 0xeebc0ded);
 247         CLASSERT(LNET_PROTO_TCP_VERSION_MAJOR == 1);
 248         CLASSERT(LNET_PROTO_TCP_VERSION_MINOR == 0);
 249         CLASSERT(LNET_MSG_ACK == 0);
 250         CLASSERT(LNET_MSG_PUT == 1);
 251         CLASSERT(LNET_MSG_GET == 2);
 252         CLASSERT(LNET_MSG_REPLY == 3);
 253         CLASSERT(LNET_MSG_HELLO == 4);
 254
 255         /* Checks for struct lnet_handle_wire */
 256         CLASSERT((int)sizeof(struct lnet_handle_wire) == 16);
 257         CLASSERT((int)offsetof(struct lnet_handle_wire, wh_interface_cookie) == 0);
 258         CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_interface_cookie) == 8);
 259         CLASSERT((int)offsetof(struct lnet_handle_wire, wh_object_cookie) == 8);
 260         CLASSERT((int)sizeof(((struct lnet_handle_wire *)0)->wh_object_cookie) == 8);
 261
 262         /* Checks for struct lnet_magicversion_t */
 263         CLASSERT((int)sizeof(lnet_magicversion_t) == 8);
 264         CLASSERT((int)offsetof(lnet_magicversion_t, magic) == 0);
 265         CLASSERT((int)sizeof(((lnet_magicversion_t *)0)->magic) == 4);
 266         CLASSERT((int)offsetof(lnet_magicversion_t, version_major) == 4);
 267         CLASSERT((int)sizeof(((lnet_magicversion_t *)0)->version_major) == 2);
 268         CLASSERT((int)offsetof(lnet_magicversion_t, version_minor) == 6);
 269         CLASSERT((int)sizeof(((lnet_magicversion_t *)0)->version_minor) == 2);
 270
 271         /* Checks for struct lnet_hdr_t */
 272         CLASSERT((int)sizeof(lnet_hdr_t) == 72);
 273         CLASSERT((int)offsetof(lnet_hdr_t, dest_nid) == 0);
 274         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->dest_nid) == 8);
 275         CLASSERT((int)offsetof(lnet_hdr_t, src_nid) == 8);
 276         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->src_nid) == 8);
 277         CLASSERT((int)offsetof(lnet_hdr_t, dest_pid) == 16);
 278         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->dest_pid) == 4);
 279         CLASSERT((int)offsetof(lnet_hdr_t, src_pid) == 20);
 280         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->src_pid) == 4);
 281         CLASSERT((int)offsetof(lnet_hdr_t, type) == 24);
 282         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->type) == 4);
 283         CLASSERT((int)offsetof(lnet_hdr_t, payload_length) == 28);
 284         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->payload_length) == 4);
 285         CLASSERT((int)offsetof(lnet_hdr_t, msg) == 32);
 286         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg) == 40);
 287
 288         /* Ack */
 289         CLASSERT((int)offsetof(lnet_hdr_t, msg.ack.dst_wmd) == 32);
 290         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.ack.dst_wmd) == 16);
 291         CLASSERT((int)offsetof(lnet_hdr_t, msg.ack.match_bits) == 48);
 292         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.ack.match_bits) == 8);
 293         CLASSERT((int)offsetof(lnet_hdr_t, msg.ack.mlength) == 56);
 294         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.ack.mlength) == 4);
 295
 296         /* Put */
 297         CLASSERT((int)offsetof(lnet_hdr_t, msg.put.ack_wmd) == 32);
 298         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.ack_wmd) == 16);
 299         CLASSERT((int)offsetof(lnet_hdr_t, msg.put.match_bits) == 48);
 300         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.match_bits) == 8);
 301         CLASSERT((int)offsetof(lnet_hdr_t, msg.put.hdr_data) == 56);
 302         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.hdr_data) == 8);
 303         CLASSERT((int)offsetof(lnet_hdr_t, msg.put.ptl_index) == 64);
 304         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.ptl_index) == 4);
 305         CLASSERT((int)offsetof(lnet_hdr_t, msg.put.offset) == 68);
 306         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.put.offset) == 4);
 307
 308         /* Get */
 309         CLASSERT((int)offsetof(lnet_hdr_t, msg.get.return_wmd) == 32);
 310         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.return_wmd) == 16);
 311         CLASSERT((int)offsetof(lnet_hdr_t, msg.get.match_bits) == 48);
 312         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.match_bits) == 8);
 313         CLASSERT((int)offsetof(lnet_hdr_t, msg.get.ptl_index) == 56);
 314         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.ptl_index) == 4);
 315         CLASSERT((int)offsetof(lnet_hdr_t, msg.get.src_offset) == 60);
 316         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.src_offset) == 4);
 317         CLASSERT((int)offsetof(lnet_hdr_t, msg.get.sink_length) == 64);
 318         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.get.sink_length) == 4);
 319
 320         /* Reply */
 321         CLASSERT((int)offsetof(lnet_hdr_t, msg.reply.dst_wmd) == 32);
 322         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.reply.dst_wmd) == 16);
 323
 324         /* Hello */
 325         CLASSERT((int)offsetof(lnet_hdr_t, msg.hello.incarnation) == 32);
 326         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.hello.incarnation) == 8);
 327         CLASSERT((int)offsetof(lnet_hdr_t, msg.hello.type) == 40);
 328         CLASSERT((int)sizeof(((lnet_hdr_t *)0)->msg.hello.type) == 4);
 329 }
 330
 331 static lnd_t *lnet_find_lnd_by_type(__u32 type)
 332 {
 333         lnd_t            *lnd;
 334         struct list_head *tmp;
 335
 336         /* holding lnd mutex */
 337         list_for_each(tmp, &the_lnet.ln_lnds) {
 338                 lnd = list_entry(tmp, lnd_t, lnd_list);
 339
 340                 if (lnd->lnd_type == type)
 341                         return lnd;
 342         }
 343         return NULL;
 344 }
 345
 346 void
 347 lnet_register_lnd (lnd_t *lnd)
 348 {
 349         mutex_lock(&the_lnet.ln_lnd_mutex);
 350
 351         LASSERT(libcfs_isknown_lnd(lnd->lnd_type));
 352         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == NULL);
 353
 354         list_add_tail(&lnd->lnd_list, &the_lnet.ln_lnds);
 355         lnd->lnd_refcount = 0;
 356
 357         CDEBUG(D_NET, "%s LND registered\n", libcfs_lnd2str(lnd->lnd_type));
 358
 359         mutex_unlock(&the_lnet.ln_lnd_mutex);
 360 }
 361 EXPORT_SYMBOL(lnet_register_lnd);
 362
 363 void
 364 lnet_unregister_lnd (lnd_t *lnd)
 365 {
 366         mutex_lock(&the_lnet.ln_lnd_mutex);
 367
 368         LASSERT(lnet_find_lnd_by_type(lnd->lnd_type) == lnd);
 369         LASSERT(lnd->lnd_refcount == 0);
 370
 371         list_del(&lnd->lnd_list);
 372         CDEBUG(D_NET, "%s LND unregistered\n", libcfs_lnd2str(lnd->lnd_type));
 373
 374         mutex_unlock(&the_lnet.ln_lnd_mutex);
 375 }
 376 EXPORT_SYMBOL(lnet_unregister_lnd);
 377
 378 void
 379 lnet_counters_get(lnet_counters_t *counters)
 380 {
 381         lnet_counters_t *ctr;
 382         int             i;
 383
 384         memset(counters, 0, sizeof(*counters));
 385
 386         lnet_net_lock(LNET_LOCK_EX);
 387
 388         cfs_percpt_for_each(ctr, i, the_lnet.ln_counters) {
 389                 counters->msgs_max     += ctr->msgs_max;
 390                 counters->msgs_alloc   += ctr->msgs_alloc;
 391                 counters->errors       += ctr->errors;
 392                 counters->send_count   += ctr->send_count;
 393                 counters->recv_count   += ctr->recv_count;
 394                 counters->route_count  += ctr->route_count;
 395                 counters->drop_count   += ctr->drop_count;
 396                 counters->send_length  += ctr->send_length;
 397                 counters->recv_length  += ctr->recv_length;
 398                 counters->route_length += ctr->route_length;
 399                 counters->drop_length  += ctr->drop_length;
 400
 401         }
 402         lnet_net_unlock(LNET_LOCK_EX);
 403 }
 404 EXPORT_SYMBOL(lnet_counters_get);
 405
 406 void
 407 lnet_counters_reset(void)
 408 {
 409         lnet_counters_t *counters;
 410         int             i;
 411
 412         lnet_net_lock(LNET_LOCK_EX);
 413
 414         cfs_percpt_for_each(counters, i, the_lnet.ln_counters)
 415                 memset(counters, 0, sizeof(lnet_counters_t));
 416
 417         lnet_net_unlock(LNET_LOCK_EX);
 418 }
 419
 420 static char *
 421 lnet_res_type2str(int type)
 422 {
 423         switch (type) {
 424         default:
 425                 LBUG();
 426         case LNET_COOKIE_TYPE_MD:
 427                 return "MD";
 428         case LNET_COOKIE_TYPE_ME:
 429                 return "ME";
 430         case LNET_COOKIE_TYPE_EQ:
 431                 return "EQ";
 432         }
 433 }
 434
 435 static void
 436 lnet_res_container_cleanup(struct lnet_res_container *rec)
 437 {
 438         int     count = 0;
 439
 440         if (rec->rec_type == 0) /* not set yet, it's uninitialized */
 441                 return;
 442
 443         while (!list_empty(&rec->rec_active)) {
 444                 struct list_head *e = rec->rec_active.next;
 445
 446                 list_del_init(e);
 447                 if (rec->rec_type == LNET_COOKIE_TYPE_EQ) {
 448                         lnet_eq_free(list_entry(e, lnet_eq_t, eq_list));
 449
 450                 } else if (rec->rec_type == LNET_COOKIE_TYPE_MD) {
 451                         lnet_md_free(list_entry(e, lnet_libmd_t, md_list));
 452
 453                 } else { /* NB: Active MEs should be attached on portals */
 454                         LBUG();
 455                 }
 456                 count++;
 457         }
 458
 459         if (count > 0) {
 460                 /* Found alive MD/ME/EQ, user really should unlink/free
 461                  * all of them before finalize LNet, but if someone didn't,
 462                  * we have to recycle garbage for him */
 463                 CERROR("%d active elements on exit of %s container\n",
 464                        count, lnet_res_type2str(rec->rec_type));
 465         }
 466
 467         if (rec->rec_lh_hash != NULL) {
 468                 LIBCFS_FREE(rec->rec_lh_hash,
 469                             LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
 470                 rec->rec_lh_hash = NULL;
 471         }
 472
 473         rec->rec_type = 0; /* mark it as finalized */
 474 }
 475
 476 static int
 477 lnet_res_container_setup(struct lnet_res_container *rec, int cpt, int type)
 478 {
 479         int     rc = 0;
 480         int     i;
 481
 482         LASSERT(rec->rec_type == 0);
 483
 484         rec->rec_type = type;
 485         INIT_LIST_HEAD(&rec->rec_active);
 486
 487         rec->rec_lh_cookie = (cpt << LNET_COOKIE_TYPE_BITS) | type;
 488
 489         /* Arbitrary choice of hash table size */
 490         LIBCFS_CPT_ALLOC(rec->rec_lh_hash, lnet_cpt_table(), cpt,
 491                          LNET_LH_HASH_SIZE * sizeof(rec->rec_lh_hash[0]));
 492         if (rec->rec_lh_hash == NULL) {
 493                 rc = -ENOMEM;
 494                 goto out;
 495         }
 496
 497         for (i = 0; i < LNET_LH_HASH_SIZE; i++)
 498                 INIT_LIST_HEAD(&rec->rec_lh_hash[i]);
 499
 500         return 0;
 501
 502 out:
 503         CERROR("Failed to setup %s resource container\n",
 504                lnet_res_type2str(type));
 505         lnet_res_container_cleanup(rec);
 506         return rc;
 507 }
 508
 509 static void
 510 lnet_res_containers_destroy(struct lnet_res_container **recs)
 511 {
 512         struct lnet_res_container       *rec;
 513         int                             i;
 514
 515         cfs_percpt_for_each(rec, i, recs)
 516                 lnet_res_container_cleanup(rec);
 517
 518         cfs_percpt_free(recs);
 519 }
 520
 521 static struct lnet_res_container **
 522 lnet_res_containers_create(int type)
 523 {
 524         struct lnet_res_container       **recs;
 525         struct lnet_res_container       *rec;
 526         int                             rc;
 527         int                             i;
 528
 529         recs = cfs_percpt_alloc(lnet_cpt_table(), sizeof(*rec));
 530         if (recs == NULL) {
 531                 CERROR("Failed to allocate %s resource containers\n",
 532                        lnet_res_type2str(type));
 533                 return NULL;
 534         }
 535
 536         cfs_percpt_for_each(rec, i, recs) {
 537                 rc = lnet_res_container_setup(rec, i, type);
 538                 if (rc != 0) {
 539                         lnet_res_containers_destroy(recs);
 540                         return NULL;
 541                 }
 542         }
 543
 544         return recs;
 545 }
 546
 547 lnet_libhandle_t *
 548 lnet_res_lh_lookup(struct lnet_res_container *rec, __u64 cookie)
 549 {
 550         /* ALWAYS called with lnet_res_lock held */
 551         struct list_head        *head;
 552         lnet_libhandle_t        *lh;
 553         unsigned int            hash;
 554
 555         if ((cookie & LNET_COOKIE_MASK) != rec->rec_type)
 556                 return NULL;
 557
 558         hash = cookie >> (LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS);
 559         head = &rec->rec_lh_hash[hash & LNET_LH_HASH_MASK];
 560
 561         list_for_each_entry(lh, head, lh_hash_chain) {
 562                 if (lh->lh_cookie == cookie)
 563                         return lh;
 564         }
 565
 566         return NULL;
 567 }
 568
 569 void
 570 lnet_res_lh_initialize(struct lnet_res_container *rec, lnet_libhandle_t *lh)
 571 {
 572         /* ALWAYS called with lnet_res_lock held */
 573         unsigned int    ibits = LNET_COOKIE_TYPE_BITS + LNET_CPT_BITS;
 574         unsigned int    hash;
 575
 576         lh->lh_cookie = rec->rec_lh_cookie;
 577         rec->rec_lh_cookie += 1 << ibits;
 578
 579         hash = (lh->lh_cookie >> ibits) & LNET_LH_HASH_MASK;
 580
 581         list_add(&lh->lh_hash_chain, &rec->rec_lh_hash[hash]);
 582 }
 583
 584 static int lnet_unprepare(void);
 585
 586 static int
 587 lnet_prepare(lnet_pid_t requested_pid)
 588 {
 589         /* Prepare to bring up the network */
 590         struct lnet_res_container **recs;
 591         int                       rc = 0;
 592
 593         if (requested_pid == LNET_PID_ANY) {
 594                 /* Don't instantiate LNET just for me */
 595                 return -ENETDOWN;
 596         }
 597
 598         LASSERT(the_lnet.ln_refcount == 0);
 599
 600         the_lnet.ln_routing = 0;
 601
 602         LASSERT((requested_pid & LNET_PID_USERFLAG) == 0);
 603         the_lnet.ln_pid = requested_pid;
 604
 605         INIT_LIST_HEAD(&the_lnet.ln_test_peers);
 606         INIT_LIST_HEAD(&the_lnet.ln_peers);
 607         INIT_LIST_HEAD(&the_lnet.ln_remote_peer_ni_list);
 608         INIT_LIST_HEAD(&the_lnet.ln_nets);
 609         INIT_LIST_HEAD(&the_lnet.ln_routers);
 610         INIT_LIST_HEAD(&the_lnet.ln_drop_rules);
 611         INIT_LIST_HEAD(&the_lnet.ln_delay_rules);
 612
 613         rc = lnet_descriptor_setup();
 614         if (rc != 0)
 615                 goto failed;
 616
 617         rc = lnet_create_remote_nets_table();
 618         if (rc != 0)
 619                 goto failed;
 620
 621         /*
 622          * NB the interface cookie in wire handles guards against delayed
 623          * replies and ACKs appearing valid after reboot.
 624          */
 625         the_lnet.ln_interface_cookie = ktime_get_real_ns();
 626
 627         the_lnet.ln_counters = cfs_percpt_alloc(lnet_cpt_table(),
 628                                                 sizeof(lnet_counters_t));
 629         if (the_lnet.ln_counters == NULL) {
 630                 CERROR("Failed to allocate counters for LNet\n");
 631                 rc = -ENOMEM;
 632                 goto failed;
 633         }
 634
 635         rc = lnet_peer_tables_create();
 636         if (rc != 0)
 637                 goto failed;
 638
 639         rc = lnet_msg_containers_create();
 640         if (rc != 0)
 641                 goto failed;
 642
 643         rc = lnet_res_container_setup(&the_lnet.ln_eq_container, 0,
 644                                       LNET_COOKIE_TYPE_EQ);
 645         if (rc != 0)
 646                 goto failed;
 647
 648         recs = lnet_res_containers_create(LNET_COOKIE_TYPE_ME);
 649         if (recs == NULL) {
 650                 rc = -ENOMEM;
 651                 goto failed;
 652         }
 653
 654         the_lnet.ln_me_containers = recs;
 655
 656         recs = lnet_res_containers_create(LNET_COOKIE_TYPE_MD);
 657         if (recs == NULL) {
 658                 rc = -ENOMEM;
 659                 goto failed;
 660         }
 661
 662         the_lnet.ln_md_containers = recs;
 663
 664         rc = lnet_portals_create();
 665         if (rc != 0) {
 666                 CERROR("Failed to create portals for LNet: %d\n", rc);
 667                 goto failed;
 668         }
 669
 670         return 0;
 671
 672  failed:
 673         lnet_unprepare();
 674         return rc;
 675 }
 676
 677 static int
 678 lnet_unprepare (void)
 679 {
 680         /* NB no LNET_LOCK since this is the last reference.  All LND instances
 681          * have shut down already, so it is safe to unlink and free all
 682          * descriptors, even those that appear committed to a network op (eg MD
 683          * with non-zero pending count) */
 684
 685         lnet_fail_nid(LNET_NID_ANY, 0);
 686
 687         LASSERT(the_lnet.ln_refcount == 0);
 688         LASSERT(list_empty(&the_lnet.ln_test_peers));
 689         LASSERT(list_empty(&the_lnet.ln_nets));
 690
 691         lnet_portals_destroy();
 692
 693         if (the_lnet.ln_md_containers != NULL) {
 694                 lnet_res_containers_destroy(the_lnet.ln_md_containers);
 695                 the_lnet.ln_md_containers = NULL;
 696         }
 697
 698         if (the_lnet.ln_me_containers != NULL) {
 699                 lnet_res_containers_destroy(the_lnet.ln_me_containers);
 700                 the_lnet.ln_me_containers = NULL;
 701         }
 702
 703         lnet_res_container_cleanup(&the_lnet.ln_eq_container);
 704
 705         lnet_msg_containers_destroy();
 706         lnet_peer_uninit();
 707         lnet_rtrpools_free(0);
 708
 709         if (the_lnet.ln_counters != NULL) {
 710                 cfs_percpt_free(the_lnet.ln_counters);
 711                 the_lnet.ln_counters = NULL;
 712         }
 713         lnet_destroy_remote_nets_table();
 714         lnet_descriptor_cleanup();
 715
 716         return 0;
 717 }
 718
 719 lnet_ni_t  *
 720 lnet_net2ni_locked(__u32 net_id, int cpt)
 721 {
 722         struct lnet_ni   *ni;
 723         struct lnet_net  *net;
 724
 725         LASSERT(cpt != LNET_LOCK_EX);
 726
 727         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
 728                 if (net->net_id == net_id) {
 729                         ni = list_entry(net->net_ni_list.next, struct lnet_ni,
 730                                         ni_netlist);
 731                         return ni;
 732                 }
 733         }
 734
 735         return NULL;
 736 }
 737
 738 lnet_ni_t *
 739 lnet_net2ni_addref(__u32 net)
 740 {
 741         lnet_ni_t *ni;
 742
 743         lnet_net_lock(0);
 744         ni = lnet_net2ni_locked(net, 0);
 745         if (ni)
 746                 lnet_ni_addref_locked(ni, 0);
 747         lnet_net_unlock(0);
 748
 749         return ni;
 750 }
 751 EXPORT_SYMBOL(lnet_net2ni_addref);
 752
 753 struct lnet_net *
 754 lnet_get_net_locked(__u32 net_id)
 755 {
 756         struct lnet_net  *net;
 757
 758         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
 759                 if (net->net_id == net_id)
 760                         return net;
 761         }
 762
 763         return NULL;
 764 }
 765
 766 unsigned int
 767 lnet_nid_cpt_hash(lnet_nid_t nid, unsigned int number)
 768 {
 769         __u64           key = nid;
 770         unsigned int    val;
 771
 772         LASSERT(number >= 1 && number <= LNET_CPT_NUMBER);
 773
 774         if (number == 1)
 775                 return 0;
 776
 777         val = hash_long(key, LNET_CPT_BITS);
 778         /* NB: LNET_CP_NUMBER doesn't have to be PO2 */
 779         if (val < number)
 780                 return val;
 781
 782         return (unsigned int)(key + val + (val >> 1)) % number;
 783 }
 784
 785 int
 786 lnet_cpt_of_nid_locked(lnet_nid_t nid, struct lnet_ni *ni)
 787 {
 788         struct lnet_net *net;
 789
 790         /* must called with hold of lnet_net_lock */
 791         if (LNET_CPT_NUMBER == 1)
 792                 return 0; /* the only one */
 793
 794         /*
 795          * If NI is provided then use the CPT identified in the NI cpt
 796          * list if one exists. If one doesn't exist, then that NI is
 797          * associated with all CPTs and it follows that the net it belongs
 798          * to is implicitly associated with all CPTs, so just hash the nid
 799          * and return that.
 800          */
 801         if (ni != NULL) {
 802                 if (ni->ni_cpts != NULL)
 803                         return ni->ni_cpts[lnet_nid_cpt_hash(nid,
 804                                                              ni->ni_ncpts)];
 805                 else
 806                         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
 807         }
 808
 809         /* no NI provided so look at the net */
 810         net = lnet_get_net_locked(LNET_NIDNET(nid));
 811
 812         if (net != NULL && net->net_cpts != NULL) {
 813                 return net->net_cpts[lnet_nid_cpt_hash(nid, net->net_ncpts)];
 814         }
 815
 816         return lnet_nid_cpt_hash(nid, LNET_CPT_NUMBER);
 817 }
 818
 819 int
 820 lnet_cpt_of_nid(lnet_nid_t nid, struct lnet_ni *ni)
 821 {
 822         int     cpt;
 823         int     cpt2;
 824
 825         if (LNET_CPT_NUMBER == 1)
 826                 return 0; /* the only one */
 827
 828         cpt = lnet_net_lock_current();
 829
 830         cpt2 = lnet_cpt_of_nid_locked(nid, ni);
 831
 832         lnet_net_unlock(cpt);
 833
 834         return cpt2;
 835 }
 836 EXPORT_SYMBOL(lnet_cpt_of_nid);
 837
 838 int
 839 lnet_islocalnet(__u32 net_id)
 840 {
 841         struct lnet_net *net;
 842         int             cpt;
 843         bool            local;
 844
 845         cpt = lnet_net_lock_current();
 846
 847         net = lnet_get_net_locked(net_id);
 848
 849         local = net != NULL;
 850
 851         lnet_net_unlock(cpt);
 852
 853         return local;
 854 }
 855
 856 bool
 857 lnet_is_ni_healthy_locked(struct lnet_ni *ni)
 858 {
 859         if (ni->ni_state == LNET_NI_STATE_ACTIVE ||
 860             ni->ni_state == LNET_NI_STATE_DEGRADED)
 861                 return true;
 862
 863         return false;
 864 }
 865
 866 lnet_ni_t  *
 867 lnet_nid2ni_locked(lnet_nid_t nid, int cpt)
 868 {
 869         struct lnet_net  *net;
 870         struct lnet_ni   *ni;
 871
 872         LASSERT(cpt != LNET_LOCK_EX);
 873
 874         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
 875                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
 876                         if (ni->ni_nid == nid)
 877                                 return ni;
 878                 }
 879         }
 880
 881         return NULL;
 882 }
 883
 884 lnet_ni_t *
 885 lnet_nid2ni_addref(lnet_nid_t nid)
 886 {
 887         lnet_ni_t *ni;
 888
 889         lnet_net_lock(0);
 890         ni = lnet_nid2ni_locked(nid, 0);
 891         if (ni)
 892                 lnet_ni_addref_locked(ni, 0);
 893         lnet_net_unlock(0);
 894
 895         return ni;
 896 }
 897 EXPORT_SYMBOL(lnet_nid2ni_addref);
 898
 899 int
 900 lnet_islocalnid(lnet_nid_t nid)
 901 {
 902         struct lnet_ni  *ni;
 903         int             cpt;
 904
 905         cpt = lnet_net_lock_current();
 906         ni = lnet_nid2ni_locked(nid, cpt);
 907         lnet_net_unlock(cpt);
 908
 909         return ni != NULL;
 910 }
 911
 912 int
 913 lnet_count_acceptor_nets(void)
 914 {
 915         /* Return the # of NIs that need the acceptor. */
 916         int              count = 0;
 917         struct lnet_net  *net;
 918         int              cpt;
 919
 920         cpt = lnet_net_lock_current();
 921         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
 922                 /* all socklnd type networks should have the acceptor
 923                  * thread started */
 924                 if (net->net_lnd->lnd_accept != NULL)
 925                         count++;
 926         }
 927
 928         lnet_net_unlock(cpt);
 929
 930         return count;
 931 }
 932
 933 static struct lnet_ping_info *
 934 lnet_ping_info_create(int num_ni)
 935 {
 936         struct lnet_ping_info *ping_info;
 937         unsigned int     infosz;
 938
 939         infosz = offsetof(struct lnet_ping_info, pi_ni[num_ni]);
 940         LIBCFS_ALLOC(ping_info, infosz);
 941         if (ping_info == NULL) {
 942                 CERROR("Can't allocate ping info[%d]\n", num_ni);
 943                 return NULL;
 944         }
 945
 946         ping_info->pi_nnis = num_ni;
 947         ping_info->pi_pid = the_lnet.ln_pid;
 948         ping_info->pi_magic = LNET_PROTO_PING_MAGIC;
 949         ping_info->pi_features = LNET_PING_FEAT_NI_STATUS;
 950
 951         return ping_info;
 952 }
 953
 954 static inline int
 955 lnet_get_net_ni_count_locked(struct lnet_net *net)
 956 {
 957         struct lnet_ni  *ni;
 958         int             count = 0;
 959
 960         list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
 961                 count++;
 962
 963         return count;
 964 }
 965
 966 static inline int
 967 lnet_get_net_ni_count_pre(struct lnet_net *net)
 968 {
 969         struct lnet_ni  *ni;
 970         int             count = 0;
 971
 972         list_for_each_entry(ni, &net->net_ni_added, ni_netlist)
 973                 count++;
 974
 975         return count;
 976 }
 977
 978 static inline int
 979 lnet_get_ni_count(void)
 980 {
 981         struct lnet_ni  *ni;
 982         struct lnet_net *net;
 983         int             count = 0;
 984
 985         lnet_net_lock(0);
 986
 987         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
 988                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist)
 989                         count++;
 990         }
 991
 992         lnet_net_unlock(0);
 993
 994         return count;
 995 }
 996
 997 static inline void
 998 lnet_ping_info_free(struct lnet_ping_info *pinfo)
 999 {
1000         LIBCFS_FREE(pinfo,
1001                     offsetof(struct lnet_ping_info,
1002                              pi_ni[pinfo->pi_nnis]));
1003 }
1004
1005 static void
1006 lnet_ping_info_destroy(void)
1007 {
1008         struct lnet_net *net;
1009         struct lnet_ni  *ni;
1010
1011         lnet_net_lock(LNET_LOCK_EX);
1012
1013         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1014                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1015                         lnet_ni_lock(ni);
1016                         ni->ni_status = NULL;
1017                         lnet_ni_unlock(ni);
1018                 }
1019         }
1020
1021         lnet_ping_info_free(the_lnet.ln_ping_info);
1022         the_lnet.ln_ping_info = NULL;
1023
1024         lnet_net_unlock(LNET_LOCK_EX);
1025 }
1026
1027 static void
1028 lnet_ping_event_handler(lnet_event_t *event)
1029 {
1030         struct lnet_ping_info *pinfo = event->md.user_ptr;
1031
1032         if (event->unlinked)
1033                 pinfo->pi_features = LNET_PING_FEAT_INVAL;
1034 }
1035
1036 static int
1037 lnet_ping_info_setup(struct lnet_ping_info **ppinfo, lnet_handle_md_t *md_handle,
1038                      int ni_count, bool set_eq)
1039 {
1040         lnet_handle_me_t  me_handle;
1041         lnet_process_id_t id = {LNET_NID_ANY, LNET_PID_ANY};
1042         lnet_md_t         md = {NULL};
1043         int               rc, rc2;
1044
1045         if (set_eq) {
1046                 rc = LNetEQAlloc(0, lnet_ping_event_handler,
1047                                  &the_lnet.ln_ping_target_eq);
1048                 if (rc != 0) {
1049                         CERROR("Can't allocate ping EQ: %d\n", rc);
1050                         return rc;
1051                 }
1052         }
1053
1054         *ppinfo = lnet_ping_info_create(ni_count);
1055         if (*ppinfo == NULL) {
1056                 rc = -ENOMEM;
1057                 goto failed_0;
1058         }
1059
1060         rc = LNetMEAttach(LNET_RESERVED_PORTAL, id,
1061                           LNET_PROTO_PING_MATCHBITS, 0,
1062                           LNET_UNLINK, LNET_INS_AFTER,
1063                           &me_handle);
1064         if (rc != 0) {
1065                 CERROR("Can't create ping ME: %d\n", rc);
1066                 goto failed_1;
1067         }
1068
1069         /* initialize md content */
1070         md.start     = *ppinfo;
1071         md.length    = offsetof(struct lnet_ping_info,
1072                                 pi_ni[(*ppinfo)->pi_nnis]);
1073         md.threshold = LNET_MD_THRESH_INF;
1074         md.max_size  = 0;
1075         md.options   = LNET_MD_OP_GET | LNET_MD_TRUNCATE |
1076                        LNET_MD_MANAGE_REMOTE;
1077         md.user_ptr  = NULL;
1078         md.eq_handle = the_lnet.ln_ping_target_eq;
1079         md.user_ptr = *ppinfo;
1080
1081         rc = LNetMDAttach(me_handle, md, LNET_RETAIN, md_handle);
1082         if (rc != 0) {
1083                 CERROR("Can't attach ping MD: %d\n", rc);
1084                 goto failed_2;
1085         }
1086
1087         return 0;
1088
1089 failed_2:
1090         rc2 = LNetMEUnlink(me_handle);
1091         LASSERT(rc2 == 0);
1092 failed_1:
1093         lnet_ping_info_free(*ppinfo);
1094         *ppinfo = NULL;
1095 failed_0:
1096         if (set_eq)
1097                 LNetEQFree(the_lnet.ln_ping_target_eq);
1098         return rc;
1099 }
1100
1101 static void
1102 lnet_ping_md_unlink(struct lnet_ping_info *pinfo, lnet_handle_md_t *md_handle)
1103 {
1104         sigset_t        blocked = cfs_block_allsigs();
1105
1106         LNetMDUnlink(*md_handle);
1107         LNetInvalidateHandle(md_handle);
1108
1109         /* NB md could be busy; this just starts the unlink */
1110         while (pinfo->pi_features != LNET_PING_FEAT_INVAL) {
1111                 CDEBUG(D_NET, "Still waiting for ping MD to unlink\n");
1112                 set_current_state(TASK_UNINTERRUPTIBLE);
1113                 schedule_timeout(cfs_time_seconds(1));
1114         }
1115
1116         cfs_restore_sigs(blocked);
1117 }
1118
1119 static void
1120 lnet_ping_info_install_locked(struct lnet_ping_info *ping_info)
1121 {
1122         int                     i;
1123         struct lnet_ni          *ni;
1124         struct lnet_net         *net;
1125         struct lnet_ni_status *ns;
1126
1127         i = 0;
1128         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
1129                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
1130                         LASSERT(i < ping_info->pi_nnis);
1131
1132                         ns = &ping_info->pi_ni[i];
1133
1134                         ns->ns_nid = ni->ni_nid;
1135
1136                         lnet_ni_lock(ni);
1137                         ns->ns_status = (ni->ni_status != NULL) ?
1138                                         ni->ni_status->ns_status :
1139                                                 LNET_NI_STATUS_UP;
1140                         ni->ni_status = ns;
1141                         lnet_ni_unlock(ni);
1142
1143                         i++;
1144                 }
1145
1146         }
1147 }
1148
1149 static void
1150 lnet_ping_target_update(struct lnet_ping_info *pinfo, lnet_handle_md_t md_handle)
1151 {
1152         struct lnet_ping_info *old_pinfo = NULL;
1153         lnet_handle_md_t old_md;
1154
1155         /* switch the NIs to point to the new ping info created */
1156         lnet_net_lock(LNET_LOCK_EX);
1157
1158         if (!the_lnet.ln_routing)
1159                 pinfo->pi_features |= LNET_PING_FEAT_RTE_DISABLED;
1160         lnet_ping_info_install_locked(pinfo);
1161
1162         if (the_lnet.ln_ping_info != NULL) {
1163                 old_pinfo = the_lnet.ln_ping_info;
1164                 old_md = the_lnet.ln_ping_target_md;
1165         }
1166         the_lnet.ln_ping_target_md = md_handle;
1167         the_lnet.ln_ping_info = pinfo;
1168
1169         lnet_net_unlock(LNET_LOCK_EX);
1170
1171         if (old_pinfo != NULL) {
1172                 /* unlink the old ping info */
1173                 lnet_ping_md_unlink(old_pinfo, &old_md);
1174                 lnet_ping_info_free(old_pinfo);
1175         }
1176 }
1177
1178 static void
1179 lnet_ping_target_fini(void)
1180 {
1181         int             rc;
1182
1183         lnet_ping_md_unlink(the_lnet.ln_ping_info,
1184                             &the_lnet.ln_ping_target_md);
1185
1186         rc = LNetEQFree(the_lnet.ln_ping_target_eq);
1187         LASSERT(rc == 0);
1188
1189         lnet_ping_info_destroy();
1190 }
1191
1192 static int
1193 lnet_ni_tq_credits(lnet_ni_t *ni)
1194 {
1195         int     credits;
1196
1197         LASSERT(ni->ni_ncpts >= 1);
1198
1199         if (ni->ni_ncpts == 1)
1200                 return ni->ni_net->net_tunables.lct_max_tx_credits;
1201
1202         credits = ni->ni_net->net_tunables.lct_max_tx_credits / ni->ni_ncpts;
1203         credits = max(credits, 8 * ni->ni_net->net_tunables.lct_peer_tx_credits);
1204         credits = min(credits, ni->ni_net->net_tunables.lct_max_tx_credits);
1205
1206         return credits;
1207 }
1208
1209 static void
1210 lnet_ni_unlink_locked(lnet_ni_t *ni)
1211 {
1212         if (!list_empty(&ni->ni_cptlist)) {
1213                 list_del_init(&ni->ni_cptlist);
1214                 lnet_ni_decref_locked(ni, 0);
1215         }
1216
1217         /* move it to zombie list and nobody can find it anymore */
1218         LASSERT(!list_empty(&ni->ni_netlist));
1219         list_move(&ni->ni_netlist, &ni->ni_net->net_ni_zombie);
1220         lnet_ni_decref_locked(ni, 0);
1221 }
1222
1223 static void
1224 lnet_clear_zombies_nis_locked(struct lnet_net *net)
1225 {
1226         int             i;
1227         int             islo;
1228         lnet_ni_t       *ni;
1229         struct list_head *zombie_list = &net->net_ni_zombie;
1230
1231         /*
1232          * Now wait for the NIs I just nuked to show up on the zombie
1233          * list and shut them down in guaranteed thread context
1234          */
1235         i = 2;
1236         while (!list_empty(zombie_list)) {
1237                 int     *ref;
1238                 int     j;
1239
1240                 ni = list_entry(zombie_list->next,
1241                                 lnet_ni_t, ni_netlist);
1242                 list_del_init(&ni->ni_netlist);
1243                 /* the ni should be in deleting state. If it's not it's
1244                  * a bug */
1245                 LASSERT(ni->ni_state == LNET_NI_STATE_DELETING);
1246                 cfs_percpt_for_each(ref, j, ni->ni_refs) {
1247                         if (*ref == 0)
1248                                 continue;
1249                         /* still busy, add it back to zombie list */
1250                         list_add(&ni->ni_netlist, zombie_list);
1251                         break;
1252                 }
1253
1254                 if (!list_empty(&ni->ni_netlist)) {
1255                         lnet_net_unlock(LNET_LOCK_EX);
1256                         ++i;
1257                         if ((i & (-i)) == i) {
1258                                 CDEBUG(D_WARNING,
1259                                        "Waiting for zombie LNI %s\n",
1260                                        libcfs_nid2str(ni->ni_nid));
1261                         }
1262                         set_current_state(TASK_UNINTERRUPTIBLE);
1263                         schedule_timeout(cfs_time_seconds(1));
1264                         lnet_net_lock(LNET_LOCK_EX);
1265                         continue;
1266                 }
1267
1268                 lnet_net_unlock(LNET_LOCK_EX);
1269
1270                 islo = ni->ni_net->net_lnd->lnd_type == LOLND;
1271
1272                 LASSERT(!in_interrupt());
1273                 (net->net_lnd->lnd_shutdown)(ni);
1274
1275                 if (!islo)
1276                         CDEBUG(D_LNI, "Removed LNI %s\n",
1277                               libcfs_nid2str(ni->ni_nid));
1278
1279                 lnet_ni_free(ni);
1280                 i = 2;
1281                 lnet_net_lock(LNET_LOCK_EX);
1282         }
1283 }
1284
1285 /* shutdown down the NI and release refcount */
1286 static void
1287 lnet_shutdown_lndni(struct lnet_ni *ni)
1288 {
1289         int i;
1290         struct lnet_net *net = ni->ni_net;
1291
1292         lnet_net_lock(LNET_LOCK_EX);
1293         ni->ni_state = LNET_NI_STATE_DELETING;
1294         lnet_ni_unlink_locked(ni);
1295         lnet_net_unlock(LNET_LOCK_EX);
1296
1297         /* clear messages for this NI on the lazy portal */
1298         for (i = 0; i < the_lnet.ln_nportals; i++)
1299                 lnet_clear_lazy_portal(ni, i, "Shutting down NI");
1300
1301         /* Do peer table cleanup for this ni */
1302         lnet_peer_tables_cleanup(ni);
1303
1304         lnet_net_lock(LNET_LOCK_EX);
1305         lnet_clear_zombies_nis_locked(net);
1306         lnet_net_unlock(LNET_LOCK_EX);
1307 }
1308
1309 static void
1310 lnet_shutdown_lndnet(struct lnet_net *net)
1311 {
1312         struct lnet_ni *ni;
1313
1314         lnet_net_lock(LNET_LOCK_EX);
1315
1316         net->net_state = LNET_NET_STATE_DELETING;
1317
1318         list_del_init(&net->net_list);
1319
1320         while (!list_empty(&net->net_ni_list)) {
1321                 ni = list_entry(net->net_ni_list.next,
1322                                 lnet_ni_t, ni_netlist);
1323                 lnet_net_unlock(LNET_LOCK_EX);
1324                 lnet_shutdown_lndni(ni);
1325                 lnet_net_lock(LNET_LOCK_EX);
1326         }
1327
1328         /*
1329          * decrement ref count on lnd only when the entire network goes
1330          * away
1331          */
1332         net->net_lnd->lnd_refcount--;
1333
1334         lnet_net_unlock(LNET_LOCK_EX);
1335
1336         lnet_net_free(net);
1337 }
1338
1339 static void
1340 lnet_shutdown_lndnets(void)
1341 {
1342         struct lnet_net *net;
1343
1344         /* NB called holding the global mutex */
1345
1346         /* All quiet on the API front */
1347         LASSERT(!the_lnet.ln_shutdown);
1348         LASSERT(the_lnet.ln_refcount == 0);
1349
1350         lnet_net_lock(LNET_LOCK_EX);
1351         the_lnet.ln_shutdown = 1;       /* flag shutdown */
1352
1353         while (!list_empty(&the_lnet.ln_nets)) {
1354                 /*
1355                  * move the nets to the zombie list to avoid them being
1356                  * picked up for new work. LONET is also included in the
1357                  * Nets that will be moved to the zombie list
1358                  */
1359                 net = list_entry(the_lnet.ln_nets.next,
1360                                  struct lnet_net, net_list);
1361                 list_move(&net->net_list, &the_lnet.ln_net_zombie);
1362         }
1363
1364         /* Drop the cached loopback Net. */
1365         if (the_lnet.ln_loni != NULL) {
1366                 lnet_ni_decref_locked(the_lnet.ln_loni, 0);
1367                 the_lnet.ln_loni = NULL;
1368         }
1369         lnet_net_unlock(LNET_LOCK_EX);
1370
1371         /* iterate through the net zombie list and delete each net */
1372         while (!list_empty(&the_lnet.ln_net_zombie)) {
1373                 net = list_entry(the_lnet.ln_net_zombie.next,
1374                                  struct lnet_net, net_list);
1375                 lnet_shutdown_lndnet(net);
1376         }
1377
1378         lnet_net_lock(LNET_LOCK_EX);
1379         the_lnet.ln_shutdown = 0;
1380         lnet_net_unlock(LNET_LOCK_EX);
1381 }
1382
1383 static int
1384 lnet_startup_lndni(struct lnet_ni *ni, struct lnet_lnd_tunables *tun)
1385 {
1386         int                     rc = -EINVAL;
1387         struct lnet_tx_queue    *tq;
1388         int                     i;
1389         struct lnet_net         *net = ni->ni_net;
1390
1391         mutex_lock(&the_lnet.ln_lnd_mutex);
1392
1393         if (tun) {
1394                 memcpy(&ni->ni_lnd_tunables, tun, sizeof(*tun));
1395                 ni->ni_lnd_tunables_set = true;
1396         }
1397
1398         rc = (net->net_lnd->lnd_startup)(ni);
1399
1400         mutex_unlock(&the_lnet.ln_lnd_mutex);
1401
1402         if (rc != 0) {
1403                 LCONSOLE_ERROR_MSG(0x105, "Error %d starting up LNI %s\n",
1404                                    rc, libcfs_lnd2str(net->net_lnd->lnd_type));
1405                 lnet_net_lock(LNET_LOCK_EX);
1406                 net->net_lnd->lnd_refcount--;
1407                 lnet_net_unlock(LNET_LOCK_EX);
1408                 goto failed0;
1409         }
1410
1411         ni->ni_state = LNET_NI_STATE_ACTIVE;
1412
1413         /* We keep a reference on the loopback net through the loopback NI */
1414         if (net->net_lnd->lnd_type == LOLND) {
1415                 lnet_ni_addref(ni);
1416                 LASSERT(the_lnet.ln_loni == NULL);
1417                 the_lnet.ln_loni = ni;
1418                 ni->ni_net->net_tunables.lct_peer_tx_credits = 0;
1419                 ni->ni_net->net_tunables.lct_peer_rtr_credits = 0;
1420                 ni->ni_net->net_tunables.lct_max_tx_credits = 0;
1421                 ni->ni_net->net_tunables.lct_peer_timeout = 0;
1422                 return 0;
1423         }
1424
1425         if (ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ||
1426             ni->ni_net->net_tunables.lct_max_tx_credits == 0) {
1427                 LCONSOLE_ERROR_MSG(0x107, "LNI %s has no %scredits\n",
1428                                    libcfs_lnd2str(net->net_lnd->lnd_type),
1429                                    ni->ni_net->net_tunables.lct_peer_tx_credits == 0 ?
1430                                         "" : "per-peer ");
1431                 /* shutdown the NI since if we get here then it must've already
1432                  * been started
1433                  */
1434                 lnet_shutdown_lndni(ni);
1435                 return -EINVAL;
1436         }
1437
1438         cfs_percpt_for_each(tq, i, ni->ni_tx_queues) {
1439                 tq->tq_credits_min =
1440                 tq->tq_credits_max =
1441                 tq->tq_credits = lnet_ni_tq_credits(ni);
1442         }
1443
1444         CDEBUG(D_LNI, "Added LNI %s [%d/%d/%d/%d]\n",
1445                 libcfs_nid2str(ni->ni_nid),
1446                 ni->ni_net->net_tunables.lct_peer_tx_credits,
1447                 lnet_ni_tq_credits(ni) * LNET_CPT_NUMBER,
1448                 ni->ni_net->net_tunables.lct_peer_rtr_credits,
1449                 ni->ni_net->net_tunables.lct_peer_timeout);
1450
1451         return 0;
1452 failed0:
1453         lnet_ni_free(ni);
1454         return rc;
1455 }
1456
1457 static int
1458 lnet_startup_lndnet(struct lnet_net *net, struct lnet_lnd_tunables *tun)
1459 {
1460         struct lnet_ni          *ni;
1461         struct lnet_net         *net_l = NULL;
1462         struct list_head        local_ni_list;
1463         int                     rc;
1464         int                     ni_count = 0;
1465         __u32                   lnd_type;
1466         lnd_t                   *lnd;
1467         int                     peer_timeout =
1468                 net->net_tunables.lct_peer_timeout;
1469         int                     maxtxcredits =
1470                 net->net_tunables.lct_max_tx_credits;
1471         int                     peerrtrcredits =
1472                 net->net_tunables.lct_peer_rtr_credits;
1473
1474         INIT_LIST_HEAD(&local_ni_list);
1475
1476         /*
1477          * make sure that this net is unique. If it isn't then
1478          * we are adding interfaces to an already existing network, and
1479          * 'net' is just a convenient way to pass in the list.
1480          * if it is unique we need to find the LND and load it if
1481          * necessary.
1482          */
1483         if (lnet_net_unique(net->net_id, &the_lnet.ln_nets, &net_l)) {
1484                 lnd_type = LNET_NETTYP(net->net_id);
1485
1486                 LASSERT(libcfs_isknown_lnd(lnd_type));
1487
1488                 if (lnd_type == CIBLND || lnd_type == OPENIBLND ||
1489                     lnd_type == IIBLND || lnd_type == VIBLND) {
1490                         CERROR("LND %s obsoleted\n", libcfs_lnd2str(lnd_type));
1491                         rc = -EINVAL;
1492                         goto failed0;
1493                 }
1494
1495                 mutex_lock(&the_lnet.ln_lnd_mutex);
1496                 lnd = lnet_find_lnd_by_type(lnd_type);
1497
1498                 if (lnd == NULL) {
1499                         mutex_unlock(&the_lnet.ln_lnd_mutex);
1500                         rc = request_module("%s", libcfs_lnd2modname(lnd_type));
1501                         mutex_lock(&the_lnet.ln_lnd_mutex);
1502
1503                         lnd = lnet_find_lnd_by_type(lnd_type);
1504                         if (lnd == NULL) {
1505                                 mutex_unlock(&the_lnet.ln_lnd_mutex);
1506                                 CERROR("Can't load LND %s, module %s, rc=%d\n",
1507                                 libcfs_lnd2str(lnd_type),
1508                                 libcfs_lnd2modname(lnd_type), rc);
1509 #ifndef HAVE_MODULE_LOADING_SUPPORT
1510                                 LCONSOLE_ERROR_MSG(0x104, "Your kernel must be "
1511                                                 "compiled with kernel module "
1512                                                 "loading support.");
1513 #endif
1514                                 rc = -EINVAL;
1515                                 goto failed0;
1516                         }
1517                 }
1518
1519                 lnet_net_lock(LNET_LOCK_EX);
1520                 lnd->lnd_refcount++;
1521                 lnet_net_unlock(LNET_LOCK_EX);
1522
1523                 net->net_lnd = lnd;
1524
1525                 mutex_unlock(&the_lnet.ln_lnd_mutex);
1526
1527                 net_l = net;
1528         }
1529
1530         /*
1531          * net_l: if the network being added is unique then net_l
1532          *        will point to that network
1533          *        if the network being added is not unique then
1534          *        net_l points to the existing network.
1535          *
1536          * When we enter the loop below, we'll pick NIs off he
1537          * network beign added and start them up, then add them to
1538          * a local ni list. Once we've successfully started all
1539          * the NIs then we join the local NI list (of started up
1540          * networks) with the net_l->net_ni_list, which should
1541          * point to the correct network to add the new ni list to
1542          *
1543          * If any of the new NIs fail to start up, then we want to
1544          * iterate through the local ni list, which should include
1545          * any NIs which were successfully started up, and shut
1546          * them down.
1547          *
1548          * After than we want to delete the network being added,
1549          * to avoid a memory leak.
1550          */
1551
1552         /*
1553          * When a network uses TCP bonding then all its interfaces
1554          * must be specified when the network is first defined: the
1555          * TCP bonding code doesn't allow for interfaces to be added
1556          * or removed.
1557          */
1558         if (net_l != net && net_l != NULL && use_tcp_bonding &&
1559             LNET_NETTYP(net_l->net_id) == SOCKLND) {
1560                 rc = -EINVAL;
1561                 goto failed0;
1562         }
1563
1564         while (!list_empty(&net->net_ni_added)) {
1565                 ni = list_entry(net->net_ni_added.next, struct lnet_ni,
1566                                 ni_netlist);
1567                 list_del_init(&ni->ni_netlist);
1568
1569                 /* make sure that the the NI we're about to start
1570                  * up is actually unique. if it's not fail. */
1571                 if (!lnet_ni_unique_net(&net_l->net_ni_list,
1572                                         ni->ni_interfaces[0])) {
1573                         rc = -EINVAL;
1574                         goto failed1;
1575                 }
1576
1577                 /* adjust the pointer the parent network, just in case it
1578                  * the net is a duplicate */
1579                 ni->ni_net = net_l;
1580
1581                 rc = lnet_startup_lndni(ni, tun);
1582
1583                 LASSERT(ni->ni_net->net_tunables.lct_peer_timeout <= 0 ||
1584                         ni->ni_net->net_lnd->lnd_query != NULL);
1585
1586                 if (rc < 0)
1587                         goto failed1;
1588
1589                 lnet_ni_addref(ni);
1590                 list_add_tail(&ni->ni_netlist, &local_ni_list);
1591
1592                 ni_count++;
1593         }
1594
1595         lnet_net_lock(LNET_LOCK_EX);
1596         list_splice_tail(&local_ni_list, &net_l->net_ni_list);
1597         lnet_incr_dlc_seq();
1598         lnet_net_unlock(LNET_LOCK_EX);
1599
1600         /* if the network is not unique then we don't want to keep
1601          * it around after we're done. Free it. Otherwise add that
1602          * net to the global the_lnet.ln_nets */
1603         if (net_l != net && net_l != NULL) {
1604                 /*
1605                  * TODO - note. currently the tunables can not be updated
1606                  * once added
1607                  */
1608                 lnet_net_free(net);
1609         } else {
1610                 net->net_state = LNET_NET_STATE_ACTIVE;
1611                 /*
1612                  * restore tunables after it has been overwitten by the
1613                  * lnd
1614                  */
1615                 if (peer_timeout != -1)
1616                         net->net_tunables.lct_peer_timeout = peer_timeout;
1617                 if (maxtxcredits != -1)
1618                         net->net_tunables.lct_max_tx_credits = maxtxcredits;
1619                 if (peerrtrcredits != -1)
1620                         net->net_tunables.lct_peer_rtr_credits = peerrtrcredits;
1621
1622                 lnet_net_lock(LNET_LOCK_EX);
1623                 list_add_tail(&net->net_list, &the_lnet.ln_nets);
1624                 lnet_net_unlock(LNET_LOCK_EX);
1625         }
1626
1627         return ni_count;
1628
1629 failed1:
1630         /*
1631          * shutdown the new NIs that are being started up
1632          * free the NET being started
1633          */
1634         while (!list_empty(&local_ni_list)) {
1635                 ni = list_entry(local_ni_list.next, struct lnet_ni,
1636                                 ni_netlist);
1637
1638                 lnet_shutdown_lndni(ni);
1639         }
1640
1641 failed0:
1642         lnet_net_free(net);
1643
1644         return rc;
1645 }
1646
1647 static int
1648 lnet_startup_lndnets(struct list_head *netlist)
1649 {
1650         struct lnet_net         *net;
1651         int                     rc;
1652         int                     ni_count = 0;
1653
1654         while (!list_empty(netlist)) {
1655                 net = list_entry(netlist->next, struct lnet_net, net_list);
1656                 list_del_init(&net->net_list);
1657
1658                 rc = lnet_startup_lndnet(net, NULL);
1659
1660                 if (rc < 0)
1661                         goto failed;
1662
1663                 ni_count += rc;
1664         }
1665
1666         return ni_count;
1667 failed:
1668         lnet_shutdown_lndnets();
1669
1670         return rc;
1671 }
1672
1673 /**
1674  * Initialize LNet library.
1675  *
1676  * Automatically called at module loading time. Caller has to call
1677  * lnet_lib_exit() after a call to lnet_lib_init(), if and only if the
1678  * latter returned 0. It must be called exactly once.
1679  *
1680  * \retval 0 on success
1681  * \retval -ve on failures.
1682  */
1683 int lnet_lib_init(void)
1684 {
1685         int rc;
1686
1687         lnet_assert_wire_constants();
1688
1689         memset(&the_lnet, 0, sizeof(the_lnet));
1690
1691         /* refer to global cfs_cpt_table for now */
1692         the_lnet.ln_cpt_table   = cfs_cpt_table;
1693         the_lnet.ln_cpt_number  = cfs_cpt_number(cfs_cpt_table);
1694
1695         LASSERT(the_lnet.ln_cpt_number > 0);
1696         if (the_lnet.ln_cpt_number > LNET_CPT_MAX) {
1697                 /* we are under risk of consuming all lh_cookie */
1698                 CERROR("Can't have %d CPTs for LNet (max allowed is %d), "
1699                        "please change setting of CPT-table and retry\n",
1700                        the_lnet.ln_cpt_number, LNET_CPT_MAX);
1701                 return -E2BIG;
1702         }
1703
1704         while ((1 << the_lnet.ln_cpt_bits) < the_lnet.ln_cpt_number)
1705                 the_lnet.ln_cpt_bits++;
1706
1707         rc = lnet_create_locks();
1708         if (rc != 0) {
1709                 CERROR("Can't create LNet global locks: %d\n", rc);
1710                 return rc;
1711         }
1712
1713         the_lnet.ln_refcount = 0;
1714         LNetInvalidateHandle(&the_lnet.ln_rc_eqh);
1715         INIT_LIST_HEAD(&the_lnet.ln_lnds);
1716         INIT_LIST_HEAD(&the_lnet.ln_net_zombie);
1717         INIT_LIST_HEAD(&the_lnet.ln_rcd_zombie);
1718         INIT_LIST_HEAD(&the_lnet.ln_rcd_deathrow);
1719
1720         /* The hash table size is the number of bits it takes to express the set
1721          * ln_num_routes, minus 1 (better to under estimate than over so we
1722          * don't waste memory). */
1723         if (rnet_htable_size <= 0)
1724                 rnet_htable_size = LNET_REMOTE_NETS_HASH_DEFAULT;
1725         else if (rnet_htable_size > LNET_REMOTE_NETS_HASH_MAX)
1726                 rnet_htable_size = LNET_REMOTE_NETS_HASH_MAX;
1727         the_lnet.ln_remote_nets_hbits = max_t(int, 1,
1728                                            order_base_2(rnet_htable_size) - 1);
1729
1730         /* All LNDs apart from the LOLND are in separate modules.  They
1731          * register themselves when their module loads, and unregister
1732          * themselves when their module is unloaded. */
1733         lnet_register_lnd(&the_lolnd);
1734         return 0;
1735 }
1736
1737 /**
1738  * Finalize LNet library.
1739  *
1740  * \pre lnet_lib_init() called with success.
1741  * \pre All LNet users called LNetNIFini() for matching LNetNIInit() calls.
1742  */
1743 void lnet_lib_exit(void)
1744 {
1745         LASSERT(the_lnet.ln_refcount == 0);
1746
1747         while (!list_empty(&the_lnet.ln_lnds))
1748                 lnet_unregister_lnd(list_entry(the_lnet.ln_lnds.next,
1749                                                lnd_t, lnd_list));
1750         lnet_destroy_locks();
1751 }
1752
1753 /**
1754  * Set LNet PID and start LNet interfaces, routing, and forwarding.
1755  *
1756  * Users must call this function at least once before any other functions.
1757  * For each successful call there must be a corresponding call to
1758  * LNetNIFini(). For subsequent calls to LNetNIInit(), \a requested_pid is
1759  * ignored.
1760  *
1761  * The PID used by LNet may be different from the one requested.
1762  * See LNetGetId().
1763  *
1764  * \param requested_pid PID requested by the caller.
1765  *
1766  * \return >= 0 on success, and < 0 error code on failures.
1767  */
1768 int
1769 LNetNIInit(lnet_pid_t requested_pid)
1770 {
1771         int                     im_a_router = 0;
1772         int                     rc;
1773         int                     ni_count;
1774         struct lnet_ping_info   *pinfo;
1775         lnet_handle_md_t        md_handle;
1776         struct list_head        net_head;
1777         struct lnet_net         *net;
1778
1779         INIT_LIST_HEAD(&net_head);
1780
1781         mutex_lock(&the_lnet.ln_api_mutex);
1782
1783         CDEBUG(D_OTHER, "refs %d\n", the_lnet.ln_refcount);
1784
1785         if (the_lnet.ln_refcount > 0) {
1786                 rc = the_lnet.ln_refcount++;
1787                 mutex_unlock(&the_lnet.ln_api_mutex);
1788                 return rc;
1789         }
1790
1791         rc = lnet_prepare(requested_pid);
1792         if (rc != 0) {
1793                 mutex_unlock(&the_lnet.ln_api_mutex);
1794                 return rc;
1795         }
1796
1797         /* create a network for Loopback network */
1798         net = lnet_net_alloc(LNET_MKNET(LOLND, 0), &net_head);
1799         if (net == NULL) {
1800                 rc = -ENOMEM;
1801                 goto err_empty_list;
1802         }
1803
1804         /* Add in the loopback NI */
1805         if (lnet_ni_alloc(net, NULL, NULL) == NULL) {
1806                 rc = -ENOMEM;
1807                 goto err_empty_list;
1808         }
1809
1810         /* If LNet is being initialized via DLC it is possible
1811          * that the user requests not to load module parameters (ones which
1812          * are supported by DLC) on initialization.  Therefore, make sure not
1813          * to load networks, routes and forwarding from module parameters
1814          * in this case.  On cleanup in case of failure only clean up
1815          * routes if it has been loaded */
1816         if (!the_lnet.ln_nis_from_mod_params) {
1817                 rc = lnet_parse_networks(&net_head, lnet_get_networks(),
1818                                          use_tcp_bonding);
1819                 if (rc < 0)
1820                         goto err_empty_list;
1821         }
1822
1823         ni_count = lnet_startup_lndnets(&net_head);
1824         if (ni_count < 0) {
1825                 rc = ni_count;
1826                 goto err_empty_list;
1827         }
1828
1829         if (!the_lnet.ln_nis_from_mod_params) {
1830                 rc = lnet_parse_routes(lnet_get_routes(), &im_a_router);
1831                 if (rc != 0)
1832                         goto err_shutdown_lndnis;
1833
1834                 rc = lnet_check_routes();
1835                 if (rc != 0)
1836                         goto err_destroy_routes;
1837
1838                 rc = lnet_rtrpools_alloc(im_a_router);
1839                 if (rc != 0)
1840                         goto err_destroy_routes;
1841         }
1842
1843         rc = lnet_acceptor_start();
1844         if (rc != 0)
1845                 goto err_destroy_routes;
1846
1847         the_lnet.ln_refcount = 1;
1848         /* Now I may use my own API functions... */
1849
1850         rc = lnet_ping_info_setup(&pinfo, &md_handle, ni_count, true);
1851         if (rc != 0)
1852                 goto err_acceptor_stop;
1853
1854         lnet_ping_target_update(pinfo, md_handle);
1855
1856         rc = lnet_router_checker_start();
1857         if (rc != 0)
1858                 goto err_stop_ping;
1859
1860         lnet_fault_init();
1861         lnet_proc_init();
1862
1863         mutex_unlock(&the_lnet.ln_api_mutex);
1864
1865         return 0;
1866
1867 err_stop_ping:
1868         lnet_ping_target_fini();
1869 err_acceptor_stop:
1870         the_lnet.ln_refcount = 0;
1871         lnet_acceptor_stop();
1872 err_destroy_routes:
1873         if (!the_lnet.ln_nis_from_mod_params)
1874                 lnet_destroy_routes();
1875 err_shutdown_lndnis:
1876         lnet_shutdown_lndnets();
1877 err_empty_list:
1878         lnet_unprepare();
1879         LASSERT(rc < 0);
1880         mutex_unlock(&the_lnet.ln_api_mutex);
1881         while (!list_empty(&net_head)) {
1882                 struct lnet_net *net;
1883
1884                 net = list_entry(net_head.next, struct lnet_net, net_list);
1885                 list_del_init(&net->net_list);
1886                 lnet_net_free(net);
1887         }
1888         return rc;
1889 }
1890 EXPORT_SYMBOL(LNetNIInit);
1891
1892 /**
1893  * Stop LNet interfaces, routing, and forwarding.
1894  *
1895  * Users must call this function once for each successful call to LNetNIInit().
1896  * Once the LNetNIFini() operation has been started, the results of pending
1897  * API operations are undefined.
1898  *
1899  * \return always 0 for current implementation.
1900  */
1901 int
1902 LNetNIFini()
1903 {
1904         mutex_lock(&the_lnet.ln_api_mutex);
1905
1906         LASSERT(the_lnet.ln_refcount > 0);
1907
1908         if (the_lnet.ln_refcount != 1) {
1909                 the_lnet.ln_refcount--;
1910         } else {
1911                 LASSERT(!the_lnet.ln_niinit_self);
1912
1913                 lnet_fault_fini();
1914
1915                 lnet_proc_fini();
1916                 lnet_router_checker_stop();
1917                 lnet_ping_target_fini();
1918
1919                 /* Teardown fns that use my own API functions BEFORE here */
1920                 the_lnet.ln_refcount = 0;
1921
1922                 lnet_acceptor_stop();
1923                 lnet_destroy_routes();
1924                 lnet_shutdown_lndnets();
1925                 lnet_unprepare();
1926         }
1927
1928         mutex_unlock(&the_lnet.ln_api_mutex);
1929         return 0;
1930 }
1931 EXPORT_SYMBOL(LNetNIFini);
1932
1933
1934 static int lnet_handle_dbg_task(struct lnet_ioctl_dbg *dbg,
1935                                 struct lnet_dbg_task_info *dbg_info)
1936 {
1937         switch (dbg->dbg_task) {
1938         case LNET_DBG_INCR_DLC_SEQ:
1939                 lnet_incr_dlc_seq();
1940         }
1941
1942         return 0;
1943 }
1944 /**
1945  * Grabs the ni data from the ni structure and fills the out
1946  * parameters
1947  *
1948  * \param[in] ni network        interface structure
1949  * \param[out] cfg_ni           NI config information
1950  * \param[out] tun              network and LND tunables
1951  */
1952 static void
1953 lnet_fill_ni_info(struct lnet_ni *ni, struct lnet_ioctl_config_ni *cfg_ni,
1954                    struct lnet_ioctl_config_lnd_tunables *tun,
1955                    __u32 tun_size)
1956 {
1957         size_t min_size = 0;
1958         int i;
1959
1960         if (!ni || !cfg_ni || !tun)
1961                 return;
1962
1963         if (ni->ni_interfaces[0] != NULL) {
1964                 for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
1965                         if (ni->ni_interfaces[i] != NULL) {
1966                                 strncpy(cfg_ni->lic_ni_intf[i],
1967                                         ni->ni_interfaces[i],
1968                                         sizeof(cfg_ni->lic_ni_intf[i]));
1969                         }
1970                 }
1971         }
1972
1973         cfg_ni->lic_nid = ni->ni_nid;
1974         cfg_ni->lic_status = ni->ni_status->ns_status;
1975         cfg_ni->lic_tcp_bonding = use_tcp_bonding;
1976         cfg_ni->lic_dev_cpt = ni->dev_cpt;
1977
1978         memcpy(&tun->lt_cmn, &ni->ni_net->net_tunables, sizeof(tun->lt_cmn));
1979
1980         /*
1981          * tun->lt_tun will always be present, but in order to be
1982          * backwards compatible, we need to deal with the cases when
1983          * tun->lt_tun is smaller than what the kernel has, because it
1984          * comes from an older version of a userspace program, then we'll
1985          * need to copy as much information as we have available space.
1986          */
1987         min_size = tun_size - sizeof(tun->lt_cmn);
1988         memcpy(&tun->lt_tun, &ni->ni_lnd_tunables, min_size);
1989
1990         /* copy over the cpts */
1991         if (ni->ni_ncpts == LNET_CPT_NUMBER &&
1992             ni->ni_cpts == NULL)  {
1993                 for (i = 0; i < ni->ni_ncpts; i++)
1994                         cfg_ni->lic_cpts[i] = i;
1995         } else {
1996                 for (i = 0;
1997                      ni->ni_cpts != NULL && i < ni->ni_ncpts &&
1998                      i < LNET_MAX_SHOW_NUM_CPT;
1999                      i++)
2000                         cfg_ni->lic_cpts[i] = ni->ni_cpts[i];
2001         }
2002         cfg_ni->lic_ncpts = ni->ni_ncpts;
2003 }
2004
2005 /**
2006  * NOTE: This is a legacy function left in the code to be backwards
2007  * compatible with older userspace programs. It should eventually be
2008  * removed.
2009  *
2010  * Grabs the ni data from the ni structure and fills the out
2011  * parameters
2012  *
2013  * \param[in] ni network        interface structure
2014  * \param[out] config           config information
2015  */
2016 static void
2017 lnet_fill_ni_info_legacy(struct lnet_ni *ni,
2018                          struct lnet_ioctl_config_data *config)
2019 {
2020         struct lnet_ioctl_net_config *net_config;
2021         struct lnet_ioctl_config_lnd_tunables *lnd_cfg = NULL;
2022         size_t min_size, tunable_size = 0;
2023         int i;
2024
2025         if (!ni || !config)
2026                 return;
2027
2028         net_config = (struct lnet_ioctl_net_config *) config->cfg_bulk;
2029         if (!net_config)
2030                 return;
2031
2032         BUILD_BUG_ON(ARRAY_SIZE(ni->ni_interfaces) !=
2033                      ARRAY_SIZE(net_config->ni_interfaces));
2034
2035         for (i = 0; i < ARRAY_SIZE(ni->ni_interfaces); i++) {
2036                 if (!ni->ni_interfaces[i])
2037                         break;
2038
2039                 strncpy(net_config->ni_interfaces[i],
2040                         ni->ni_interfaces[i],
2041                         sizeof(net_config->ni_interfaces[i]));
2042         }
2043
2044         config->cfg_nid = ni->ni_nid;
2045         config->cfg_config_u.cfg_net.net_peer_timeout =
2046                 ni->ni_net->net_tunables.lct_peer_timeout;
2047         config->cfg_config_u.cfg_net.net_max_tx_credits =
2048                 ni->ni_net->net_tunables.lct_max_tx_credits;
2049         config->cfg_config_u.cfg_net.net_peer_tx_credits =
2050                 ni->ni_net->net_tunables.lct_peer_tx_credits;
2051         config->cfg_config_u.cfg_net.net_peer_rtr_credits =
2052                 ni->ni_net->net_tunables.lct_peer_rtr_credits;
2053
2054         net_config->ni_status = ni->ni_status->ns_status;
2055
2056         if (ni->ni_cpts) {
2057                 int num_cpts = min(ni->ni_ncpts, LNET_MAX_SHOW_NUM_CPT);
2058
2059                 for (i = 0; i < num_cpts; i++)
2060                         net_config->ni_cpts[i] = ni->ni_cpts[i];
2061
2062                 config->cfg_ncpts = num_cpts;
2063         }
2064
2065         /*
2066          * See if user land tools sent in a newer and larger version
2067          * of struct lnet_tunables than what the kernel uses.
2068          */
2069         min_size = sizeof(*config) + sizeof(*net_config);
2070
2071         if (config->cfg_hdr.ioc_len > min_size)
2072                 tunable_size = config->cfg_hdr.ioc_len - min_size;
2073
2074         /* Don't copy too much data to user space */
2075         min_size = min(tunable_size, sizeof(ni->ni_lnd_tunables));
2076         lnd_cfg = (struct lnet_ioctl_config_lnd_tunables *)net_config->cfg_bulk;
2077
2078         if (lnd_cfg && min_size) {
2079                 memcpy(&lnd_cfg->lt_tun, &ni->ni_lnd_tunables, min_size);
2080                 config->cfg_config_u.cfg_net.net_interface_count = 1;
2081
2082                 /* Tell user land that kernel side has less data */
2083                 if (tunable_size > sizeof(ni->ni_lnd_tunables)) {
2084                         min_size = tunable_size - sizeof(ni->ni_lnd_tunables);
2085                         config->cfg_hdr.ioc_len -= min_size;
2086                 }
2087         }
2088 }
2089
2090 struct lnet_ni *
2091 lnet_get_ni_idx_locked(int idx)
2092 {
2093         struct lnet_ni          *ni;
2094         struct lnet_net         *net;
2095
2096         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2097                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2098                         if (idx-- == 0)
2099                                 return ni;
2100                 }
2101         }
2102
2103         return NULL;
2104 }
2105
2106 struct lnet_ni *
2107 lnet_get_next_ni_locked(struct lnet_net *mynet, struct lnet_ni *prev)
2108 {
2109         struct lnet_ni          *ni;
2110         struct lnet_net         *net = mynet;
2111
2112         if (prev == NULL) {
2113                 if (net == NULL)
2114                         net = list_entry(the_lnet.ln_nets.next, struct lnet_net,
2115                                         net_list);
2116                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2117                                 ni_netlist);
2118
2119                 return ni;
2120         }
2121
2122         if (prev->ni_netlist.next == &prev->ni_net->net_ni_list) {
2123                 /* if you reached the end of the ni list and the net is
2124                  * specified, then there are no more nis in that net */
2125                 if (net != NULL)
2126                         return NULL;
2127
2128                 /* we reached the end of this net ni list. move to the
2129                  * next net */
2130                 if (prev->ni_net->net_list.next == &the_lnet.ln_nets)
2131                         /* no more nets and no more NIs. */
2132                         return NULL;
2133
2134                 /* get the next net */
2135                 net = list_entry(prev->ni_net->net_list.next, struct lnet_net,
2136                                  net_list);
2137                 /* get the ni on it */
2138                 ni = list_entry(net->net_ni_list.next, struct lnet_ni,
2139                                 ni_netlist);
2140
2141                 return ni;
2142         }
2143
2144         /* there are more nis left */
2145         ni = list_entry(prev->ni_netlist.next, struct lnet_ni, ni_netlist);
2146
2147         return ni;
2148 }
2149
2150 int
2151 lnet_get_net_config(struct lnet_ioctl_config_data *config)
2152 {
2153         struct lnet_ni *ni;
2154         int cpt;
2155         int rc = -ENOENT;
2156         int idx = config->cfg_count;
2157
2158         cpt = lnet_net_lock_current();
2159
2160         ni = lnet_get_ni_idx_locked(idx);
2161
2162         if (ni != NULL) {
2163                 rc = 0;
2164                 lnet_ni_lock(ni);
2165                 lnet_fill_ni_info_legacy(ni, config);
2166                 lnet_ni_unlock(ni);
2167         }
2168
2169         lnet_net_unlock(cpt);
2170         return rc;
2171 }
2172
2173 int
2174 lnet_get_ni_config(struct lnet_ioctl_config_ni *cfg_ni,
2175                    struct lnet_ioctl_config_lnd_tunables *tun,
2176                    __u32 tun_size)
2177 {
2178         struct lnet_ni          *ni;
2179         int                     cpt;
2180         int                     rc = -ENOENT;
2181
2182         if (!cfg_ni || !tun)
2183                 return -EINVAL;
2184
2185         cpt = lnet_net_lock_current();
2186
2187         ni = lnet_get_ni_idx_locked(cfg_ni->lic_idx);
2188
2189         if (ni) {
2190                 rc = 0;
2191                 lnet_ni_lock(ni);
2192                 lnet_fill_ni_info(ni, cfg_ni, tun, tun_size);
2193                 lnet_ni_unlock(ni);
2194         }
2195
2196         lnet_net_unlock(cpt);
2197         return rc;
2198 }
2199
2200 static int lnet_add_net_common(struct lnet_net *net,
2201                                struct lnet_ioctl_config_lnd_tunables *tun)
2202 {
2203         struct lnet_net         *netl = NULL;
2204         __u32                   net_id;
2205         lnet_ping_info_t        *pinfo;
2206         lnet_handle_md_t        md_handle;
2207         int                     rc;
2208         lnet_remotenet_t        *rnet;
2209         int                     net_ni_count;
2210         int                     num_acceptor_nets;
2211
2212         lnet_net_lock(LNET_LOCK_EX);
2213         rnet = lnet_find_rnet_locked(net->net_id);
2214         lnet_net_unlock(LNET_LOCK_EX);
2215         /*
2216          * make sure that the net added doesn't invalidate the current
2217          * configuration LNet is keeping
2218          */
2219         if (rnet) {
2220                 CERROR("Adding net %s will invalidate routing configuration\n",
2221                        libcfs_net2str(net->net_id));
2222                 rc = -EUSERS;
2223                 goto failed1;
2224         }
2225
2226         /*
2227          * make sure you calculate the correct number of slots in the ping
2228          * info. Since the ping info is a flattened list of all the NIs,
2229          * we should allocate enough slots to accomodate the number of NIs
2230          * which will be added.
2231          *
2232          * since ni hasn't been configured yet, use
2233          * lnet_get_net_ni_count_pre() which checks the net_ni_added list
2234          */
2235         net_ni_count = lnet_get_net_ni_count_pre(net);
2236
2237         rc = lnet_ping_info_setup(&pinfo, &md_handle,
2238                                   net_ni_count + lnet_get_ni_count(),
2239                                   false);
2240         if (rc < 0)
2241                 goto failed1;
2242
2243         if (tun)
2244                 memcpy(&net->net_tunables,
2245                        &tun->lt_cmn, sizeof(net->net_tunables));
2246         else
2247                 memset(&net->net_tunables, -1, sizeof(net->net_tunables));
2248
2249         /*
2250          * before starting this network get a count of the current TCP
2251          * networks which require the acceptor thread running. If that
2252          * count is == 0 before we start up this network, then we'd want to
2253          * start up the acceptor thread after starting up this network
2254          */
2255         num_acceptor_nets = lnet_count_acceptor_nets();
2256
2257         net_id = net->net_id;
2258
2259         rc = lnet_startup_lndnet(net,
2260                                  (tun) ? &tun->lt_tun : NULL);
2261         if (rc < 0)
2262                 goto failed;
2263
2264         lnet_net_lock(LNET_LOCK_EX);
2265         netl = lnet_get_net_locked(net_id);
2266         lnet_net_unlock(LNET_LOCK_EX);
2267
2268         LASSERT(netl);
2269
2270         /*
2271          * Start the acceptor thread if this is the first network
2272          * being added that requires the thread.
2273          */
2274         if (netl->net_lnd->lnd_accept &&
2275             num_acceptor_nets == 0)
2276         {
2277                 rc = lnet_acceptor_start();
2278                 if (rc < 0) {
2279                         /* shutdown the net that we just started */
2280                         CERROR("Failed to start up acceptor thread\n");
2281                         lnet_shutdown_lndnet(net);
2282                         goto failed;
2283                 }
2284         }
2285
2286         lnet_net_lock(LNET_LOCK_EX);
2287         lnet_peer_net_added(netl);
2288         lnet_net_unlock(LNET_LOCK_EX);
2289
2290         lnet_ping_target_update(pinfo, md_handle);
2291
2292         return 0;
2293
2294 failed:
2295         lnet_ping_md_unlink(pinfo, &md_handle);
2296         lnet_ping_info_free(pinfo);
2297 failed1:
2298         lnet_net_free(net);
2299         return rc;
2300 }
2301
2302 static int lnet_handle_legacy_ip2nets(char *ip2nets,
2303                                       struct lnet_ioctl_config_lnd_tunables *tun)
2304 {
2305         struct lnet_net *net;
2306         char *nets;
2307         int rc;
2308         struct list_head net_head;
2309
2310         INIT_LIST_HEAD(&net_head);
2311
2312         rc = lnet_parse_ip2nets(&nets, ip2nets);
2313         if (rc < 0)
2314                 return rc;
2315
2316         rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
2317         if (rc < 0)
2318                 return rc;
2319
2320         mutex_lock(&the_lnet.ln_api_mutex);
2321         while (!list_empty(&net_head)) {
2322                 net = list_entry(net_head.next, struct lnet_net, net_list);
2323                 list_del_init(&net->net_list);
2324                 rc = lnet_add_net_common(net, tun);
2325                 if (rc < 0)
2326                         goto out;
2327         }
2328
2329 out:
2330         mutex_unlock(&the_lnet.ln_api_mutex);
2331
2332         while (!list_empty(&net_head)) {
2333                 net = list_entry(net_head.next, struct lnet_net, net_list);
2334                 list_del_init(&net->net_list);
2335                 lnet_net_free(net);
2336         }
2337         return rc;
2338 }
2339
2340 int lnet_dyn_add_ni(struct lnet_ioctl_config_ni *conf)
2341 {
2342         struct lnet_net *net;
2343         struct lnet_ni *ni;
2344         struct lnet_ioctl_config_lnd_tunables *tun = NULL;
2345         int rc;
2346         __u32 net_id;
2347
2348         /* get the tunables if they are available */
2349         if (conf->lic_cfg_hdr.ioc_len >=
2350             sizeof(*conf) + sizeof(*tun))
2351                 tun = (struct lnet_ioctl_config_lnd_tunables *)
2352                         conf->lic_bulk;
2353
2354         /* handle legacy ip2nets from DLC */
2355         if (conf->lic_legacy_ip2nets[0] != '\0')
2356                 return lnet_handle_legacy_ip2nets(conf->lic_legacy_ip2nets,
2357                                                   tun);
2358
2359         net_id = LNET_NIDNET(conf->lic_nid);
2360
2361         net = lnet_net_alloc(net_id, NULL);
2362         if (!net)
2363                 return -ENOMEM;
2364
2365         ni = lnet_ni_alloc_w_cpt_array(net, conf->lic_cpts, conf->lic_ncpts,
2366                                        conf->lic_ni_intf[0]);
2367         if (!ni)
2368                 return -ENOMEM;
2369
2370         mutex_lock(&the_lnet.ln_api_mutex);
2371
2372         rc = lnet_add_net_common(net, tun);
2373
2374         mutex_unlock(&the_lnet.ln_api_mutex);
2375
2376         return rc;
2377 }
2378
2379 int lnet_dyn_del_ni(struct lnet_ioctl_config_ni *conf)
2380 {
2381         struct lnet_net  *net;
2382         struct lnet_ni *ni;
2383         __u32 net_id = LNET_NIDNET(conf->lic_nid);
2384         lnet_ping_info_t *pinfo;
2385         lnet_handle_md_t  md_handle;
2386         int               rc;
2387         int               net_count;
2388         __u32             addr;
2389
2390         /* don't allow userspace to shutdown the LOLND */
2391         if (LNET_NETTYP(net_id) == LOLND)
2392                 return -EINVAL;
2393
2394         mutex_lock(&the_lnet.ln_api_mutex);
2395
2396         lnet_net_lock(0);
2397
2398         net = lnet_get_net_locked(net_id);
2399         if (!net) {
2400                 CERROR("net %s not found\n",
2401                        libcfs_net2str(net_id));
2402                 rc = -ENOENT;
2403                 goto net_unlock;
2404         }
2405
2406         addr = LNET_NIDADDR(conf->lic_nid);
2407         if (addr == 0) {
2408                 /* remove the entire net */
2409                 net_count = lnet_get_net_ni_count_locked(net);
2410
2411                 lnet_net_unlock(0);
2412
2413                 /* create and link a new ping info, before removing the old one */
2414                 rc = lnet_ping_info_setup(&pinfo, &md_handle,
2415                                         lnet_get_ni_count() - net_count,
2416                                         false);
2417                 if (rc != 0)
2418                         goto out;
2419
2420                 lnet_shutdown_lndnet(net);
2421
2422                 if (lnet_count_acceptor_nets() == 0)
2423                         lnet_acceptor_stop();
2424
2425                 lnet_ping_target_update(pinfo, md_handle);
2426
2427                 goto out;
2428         }
2429
2430         ni = lnet_nid2ni_locked(conf->lic_nid, 0);
2431         if (!ni) {
2432                 CERROR("nid %s not found \n",
2433                        libcfs_nid2str(conf->lic_nid));
2434                 rc = -ENOENT;
2435                 goto net_unlock;
2436         }
2437
2438         net_count = lnet_get_net_ni_count_locked(net);
2439
2440         lnet_net_unlock(0);
2441
2442         /* create and link a new ping info, before removing the old one */
2443         rc = lnet_ping_info_setup(&pinfo, &md_handle,
2444                                   lnet_get_ni_count() - 1, false);
2445         if (rc != 0)
2446                 goto out;
2447
2448         lnet_shutdown_lndni(ni);
2449
2450         if (lnet_count_acceptor_nets() == 0)
2451                 lnet_acceptor_stop();
2452
2453         lnet_ping_target_update(pinfo, md_handle);
2454
2455         /* check if the net is empty and remove it if it is */
2456         if (net_count == 1)
2457                 lnet_shutdown_lndnet(net);
2458
2459         goto out;
2460
2461 net_unlock:
2462         lnet_net_unlock(0);
2463 out:
2464         mutex_unlock(&the_lnet.ln_api_mutex);
2465
2466         return rc;
2467 }
2468
2469 /*
2470  * lnet_dyn_add_net and lnet_dyn_del_net are now deprecated.
2471  * They are only expected to be called for unique networks.
2472  * That can be as a result of older DLC library
2473  * calls. Multi-Rail DLC and beyond no longer uses these APIs.
2474  */
2475 int
2476 lnet_dyn_add_net(struct lnet_ioctl_config_data *conf)
2477 {
2478         struct lnet_net         *net;
2479         struct list_head        net_head;
2480         int                     rc;
2481         struct lnet_ioctl_config_lnd_tunables tun;
2482         char *nets = conf->cfg_config_u.cfg_net.net_intf;
2483
2484         INIT_LIST_HEAD(&net_head);
2485
2486         /* Create a net/ni structures for the network string */
2487         rc = lnet_parse_networks(&net_head, nets, use_tcp_bonding);
2488         if (rc <= 0)
2489                 return rc == 0 ? -EINVAL : rc;
2490
2491         mutex_lock(&the_lnet.ln_api_mutex);
2492
2493         if (rc > 1) {
2494                 rc = -EINVAL; /* only add one network per call */
2495                 goto failed;
2496         }
2497
2498         net = list_entry(net_head.next, struct lnet_net, net_list);
2499         list_del_init(&net->net_list);
2500
2501         LASSERT(lnet_net_unique(net->net_id, &the_lnet.ln_nets, NULL));
2502
2503         memset(&tun, sizeof(tun), 0);
2504
2505         tun.lt_cmn.lct_peer_timeout =
2506           conf->cfg_config_u.cfg_net.net_peer_timeout;
2507         tun.lt_cmn.lct_peer_tx_credits =
2508           conf->cfg_config_u.cfg_net.net_peer_tx_credits;
2509         tun.lt_cmn.lct_peer_rtr_credits =
2510           conf->cfg_config_u.cfg_net.net_peer_rtr_credits;
2511         tun.lt_cmn.lct_max_tx_credits =
2512           conf->cfg_config_u.cfg_net.net_max_tx_credits;
2513
2514         rc = lnet_add_net_common(net, &tun);
2515         if (rc != 0)
2516                 goto failed;
2517
2518         return 0;
2519
2520 failed:
2521         mutex_unlock(&the_lnet.ln_api_mutex);
2522         while (!list_empty(&net_head)) {
2523                 net = list_entry(net_head.next, struct lnet_net, net_list);
2524                 list_del_init(&net->net_list);
2525                 lnet_net_free(net);
2526         }
2527         return rc;
2528 }
2529
2530 int
2531 lnet_dyn_del_net(__u32 net_id)
2532 {
2533         struct lnet_net  *net;
2534         struct lnet_ping_info *pinfo;
2535         lnet_handle_md_t  md_handle;
2536         int               rc;
2537         int               net_ni_count;
2538
2539         /* don't allow userspace to shutdown the LOLND */
2540         if (LNET_NETTYP(net_id) == LOLND)
2541                 return -EINVAL;
2542
2543         mutex_lock(&the_lnet.ln_api_mutex);
2544
2545         lnet_net_lock(0);
2546
2547         net = lnet_get_net_locked(net_id);
2548         if (net == NULL) {
2549                 rc = -EINVAL;
2550                 goto out;
2551         }
2552
2553         net_ni_count = lnet_get_net_ni_count_locked(net);
2554
2555         lnet_net_unlock(0);
2556
2557         /* create and link a new ping info, before removing the old one */
2558         rc = lnet_ping_info_setup(&pinfo, &md_handle,
2559                                   lnet_get_ni_count() - net_ni_count, false);
2560         if (rc != 0)
2561                 goto out;
2562
2563         lnet_shutdown_lndnet(net);
2564
2565         if (lnet_count_acceptor_nets() == 0)
2566                 lnet_acceptor_stop();
2567
2568         lnet_ping_target_update(pinfo, md_handle);
2569
2570 out:
2571         mutex_unlock(&the_lnet.ln_api_mutex);
2572
2573         return rc;
2574 }
2575
2576 void lnet_incr_dlc_seq(void)
2577 {
2578         atomic_inc(&lnet_dlc_seq_no);
2579 }
2580
2581 __u32 lnet_get_dlc_seq_locked(void)
2582 {
2583         return atomic_read(&lnet_dlc_seq_no);
2584 }
2585
2586 inline __u32 lnet_get_numa_range(void)
2587 {
2588         return lnet_numa_range;
2589 }
2590
2591 /**
2592  * LNet ioctl handler.
2593  *
2594  */
2595 int
2596 LNetCtl(unsigned int cmd, void *arg)
2597 {
2598         struct libcfs_ioctl_data *data = arg;
2599         struct lnet_ioctl_config_data *config;
2600         lnet_process_id_t         id = {0};
2601         lnet_ni_t                *ni;
2602         int                       rc;
2603
2604         BUILD_BUG_ON(sizeof(struct lnet_ioctl_net_config) +
2605                      sizeof(struct lnet_ioctl_config_data) > LIBCFS_IOC_DATA_MAX);
2606
2607         switch (cmd) {
2608         case IOC_LIBCFS_GET_NI:
2609                 rc = LNetGetId(data->ioc_count, &id);
2610                 data->ioc_nid = id.nid;
2611                 return rc;
2612
2613         case IOC_LIBCFS_FAIL_NID:
2614                 return lnet_fail_nid(data->ioc_nid, data->ioc_count);
2615
2616         case IOC_LIBCFS_ADD_ROUTE:
2617                 config = arg;
2618
2619                 if (config->cfg_hdr.ioc_len < sizeof(*config))
2620                         return -EINVAL;
2621
2622                 mutex_lock(&the_lnet.ln_api_mutex);
2623                 rc = lnet_add_route(config->cfg_net,
2624                                     config->cfg_config_u.cfg_route.rtr_hop,
2625                                     config->cfg_nid,
2626                                     config->cfg_config_u.cfg_route.
2627                                         rtr_priority);
2628                 if (rc == 0) {
2629                         rc = lnet_check_routes();
2630                         if (rc != 0)
2631                                 lnet_del_route(config->cfg_net,
2632                                                config->cfg_nid);
2633                 }
2634                 mutex_unlock(&the_lnet.ln_api_mutex);
2635                 return rc;
2636
2637         case IOC_LIBCFS_DEL_ROUTE:
2638                 config = arg;
2639
2640                 if (config->cfg_hdr.ioc_len < sizeof(*config))
2641                         return -EINVAL;
2642
2643                 mutex_lock(&the_lnet.ln_api_mutex);
2644                 rc = lnet_del_route(config->cfg_net, config->cfg_nid);
2645                 mutex_unlock(&the_lnet.ln_api_mutex);
2646                 return rc;
2647
2648         case IOC_LIBCFS_GET_ROUTE:
2649                 config = arg;
2650
2651                 if (config->cfg_hdr.ioc_len < sizeof(*config))
2652                         return -EINVAL;
2653
2654                 return lnet_get_route(config->cfg_count,
2655                                       &config->cfg_net,
2656                                       &config->cfg_config_u.cfg_route.rtr_hop,
2657                                       &config->cfg_nid,
2658                                       &config->cfg_config_u.cfg_route.rtr_flags,
2659                                       &config->cfg_config_u.cfg_route.
2660                                         rtr_priority);
2661
2662         case IOC_LIBCFS_GET_LOCAL_NI: {
2663                 struct lnet_ioctl_config_ni *cfg_ni;
2664                 struct lnet_ioctl_config_lnd_tunables *tun = NULL;
2665                 __u32 tun_size;
2666
2667                 cfg_ni = arg;
2668                 /* get the tunables if they are available */
2669                 if (cfg_ni->lic_cfg_hdr.ioc_len <
2670                     sizeof(*cfg_ni) + sizeof(*tun))
2671                         return -EINVAL;
2672
2673                 tun = (struct lnet_ioctl_config_lnd_tunables *)
2674                                 cfg_ni->lic_bulk;
2675
2676                 tun_size = cfg_ni->lic_cfg_hdr.ioc_len - sizeof(*cfg_ni);
2677
2678                 return lnet_get_ni_config(cfg_ni, tun, tun_size);
2679         }
2680
2681         case IOC_LIBCFS_GET_NET: {
2682                 size_t total = sizeof(*config) +
2683                                sizeof(struct lnet_ioctl_net_config);
2684                 config = arg;
2685
2686                 if (config->cfg_hdr.ioc_len < total)
2687                         return -EINVAL;
2688
2689                 return lnet_get_net_config(config);
2690         }
2691
2692         case IOC_LIBCFS_GET_LNET_STATS:
2693         {
2694                 struct lnet_ioctl_lnet_stats *lnet_stats = arg;
2695
2696                 if (lnet_stats->st_hdr.ioc_len < sizeof(*lnet_stats))
2697                         return -EINVAL;
2698
2699                 lnet_counters_get(&lnet_stats->st_cntrs);
2700                 return 0;
2701         }
2702
2703         case IOC_LIBCFS_CONFIG_RTR:
2704                 config = arg;
2705
2706                 if (config->cfg_hdr.ioc_len < sizeof(*config))
2707                         return -EINVAL;
2708
2709                 mutex_lock(&the_lnet.ln_api_mutex);
2710                 if (config->cfg_config_u.cfg_buffers.buf_enable) {
2711                         rc = lnet_rtrpools_enable();
2712                         mutex_unlock(&the_lnet.ln_api_mutex);
2713                         return rc;
2714                 }
2715                 lnet_rtrpools_disable();
2716                 mutex_unlock(&the_lnet.ln_api_mutex);
2717                 return 0;
2718
2719         case IOC_LIBCFS_ADD_BUF:
2720                 config = arg;
2721
2722                 if (config->cfg_hdr.ioc_len < sizeof(*config))
2723                         return -EINVAL;
2724
2725                 mutex_lock(&the_lnet.ln_api_mutex);
2726                 rc = lnet_rtrpools_adjust(config->cfg_config_u.cfg_buffers.
2727                                                 buf_tiny,
2728                                           config->cfg_config_u.cfg_buffers.
2729                                                 buf_small,
2730                                           config->cfg_config_u.cfg_buffers.
2731                                                 buf_large);
2732                 mutex_unlock(&the_lnet.ln_api_mutex);
2733                 return rc;
2734
2735         case IOC_LIBCFS_SET_NUMA_RANGE: {
2736                 struct lnet_ioctl_numa_range *numa;
2737                 numa = arg;
2738                 if (numa->nr_hdr.ioc_len != sizeof(*numa))
2739                         return -EINVAL;
2740                 lnet_numa_range = numa->nr_range;
2741                 return 0;
2742         }
2743
2744         case IOC_LIBCFS_GET_NUMA_RANGE: {
2745                 struct lnet_ioctl_numa_range *numa;
2746                 numa = arg;
2747                 if (numa->nr_hdr.ioc_len != sizeof(*numa))
2748                         return -EINVAL;
2749                 numa->nr_range = lnet_numa_range;
2750                 return 0;
2751         }
2752
2753         case IOC_LIBCFS_GET_BUF: {
2754                 struct lnet_ioctl_pool_cfg *pool_cfg;
2755                 size_t total = sizeof(*config) + sizeof(*pool_cfg);
2756
2757                 config = arg;
2758
2759                 if (config->cfg_hdr.ioc_len < total)
2760                         return -EINVAL;
2761
2762                 pool_cfg = (struct lnet_ioctl_pool_cfg *)config->cfg_bulk;
2763                 return lnet_get_rtr_pool_cfg(config->cfg_count, pool_cfg);
2764         }
2765
2766         case IOC_LIBCFS_ADD_PEER_NI: {
2767                 struct lnet_ioctl_peer_cfg *cfg = arg;
2768
2769                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
2770                         return -EINVAL;
2771
2772                 return lnet_add_peer_ni_to_peer(cfg->prcfg_key_nid,
2773                                                 cfg->prcfg_cfg_nid,
2774                                                 cfg->prcfg_mr);
2775         }
2776
2777         case IOC_LIBCFS_DEL_PEER_NI: {
2778                 struct lnet_ioctl_peer_cfg *cfg = arg;
2779
2780                 if (cfg->prcfg_hdr.ioc_len < sizeof(*cfg))
2781                         return -EINVAL;
2782
2783                 return lnet_del_peer_ni_from_peer(cfg->prcfg_key_nid,
2784                                                   cfg->prcfg_cfg_nid);
2785         }
2786
2787         case IOC_LIBCFS_GET_PEER_INFO: {
2788                 struct lnet_ioctl_peer *peer_info = arg;
2789
2790                 if (peer_info->pr_hdr.ioc_len < sizeof(*peer_info))
2791                         return -EINVAL;
2792
2793                 return lnet_get_peer_ni_info(
2794                    peer_info->pr_count,
2795                    &peer_info->pr_nid,
2796                    peer_info->pr_lnd_u.pr_peer_credits.cr_aliveness,
2797                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ncpt,
2798                    &peer_info->pr_lnd_u.pr_peer_credits.cr_refcount,
2799                    &peer_info->pr_lnd_u.pr_peer_credits.cr_ni_peer_tx_credits,
2800                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_credits,
2801                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_rtr_credits,
2802                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_min_rtr_credits,
2803                    &peer_info->pr_lnd_u.pr_peer_credits.cr_peer_tx_qnob);
2804         }
2805
2806         case IOC_LIBCFS_GET_PEER_NI: {
2807                 struct lnet_ioctl_peer_cfg *cfg = arg;
2808                 struct lnet_peer_ni_credit_info *lpni_cri;
2809                 size_t total = sizeof(*cfg) + sizeof(*lpni_cri);
2810
2811                 if (cfg->prcfg_hdr.ioc_len < total)
2812                         return -EINVAL;
2813
2814                 lpni_cri = (struct lnet_peer_ni_credit_info*) cfg->prcfg_bulk;
2815
2816                 return lnet_get_peer_info(cfg->prcfg_idx, &cfg->prcfg_key_nid,
2817                                           &cfg->prcfg_cfg_nid, lpni_cri);
2818         }
2819
2820         case IOC_LIBCFS_NOTIFY_ROUTER: {
2821                 unsigned long jiffies_passed;
2822
2823                 jiffies_passed = ktime_get_real_seconds() - data->ioc_u64[0];
2824                 jiffies_passed = cfs_time_seconds(jiffies_passed);
2825
2826                 return lnet_notify(NULL, data->ioc_nid, data->ioc_flags,
2827                                    jiffies - jiffies_passed);
2828         }
2829
2830         case IOC_LIBCFS_LNET_DIST:
2831                 rc = LNetDist(data->ioc_nid, &data->ioc_nid, &data->ioc_u32[1]);
2832                 if (rc < 0 && rc != -EHOSTUNREACH)
2833                         return rc;
2834
2835                 data->ioc_u32[0] = rc;
2836                 return 0;
2837
2838         case IOC_LIBCFS_TESTPROTOCOMPAT:
2839                 lnet_net_lock(LNET_LOCK_EX);
2840                 the_lnet.ln_testprotocompat = data->ioc_flags;
2841                 lnet_net_unlock(LNET_LOCK_EX);
2842                 return 0;
2843
2844         case IOC_LIBCFS_LNET_FAULT:
2845                 return lnet_fault_ctl(data->ioc_flags, data);
2846
2847         case IOC_LIBCFS_PING: {
2848                 signed long timeout;
2849
2850                 id.nid = data->ioc_nid;
2851                 id.pid = data->ioc_u32[0];
2852
2853                 /* Don't block longer than 2 minutes */
2854                 if (data->ioc_u32[1] > 120 * MSEC_PER_SEC)
2855                         return -EINVAL;
2856
2857                 /* If timestamp is negative then disable timeout */
2858                 if ((s32)data->ioc_u32[1] < 0)
2859                         timeout = MAX_SCHEDULE_TIMEOUT;
2860                 else
2861                         timeout = msecs_to_jiffies(data->ioc_u32[1]);
2862
2863                 rc = lnet_ping(id, timeout, data->ioc_pbuf1,
2864                                data->ioc_plen1 / sizeof(lnet_process_id_t));
2865                 if (rc < 0)
2866                         return rc;
2867                 data->ioc_count = rc;
2868                 return 0;
2869         }
2870
2871         case IOC_LIBCFS_DBG: {
2872                 struct lnet_ioctl_dbg *dbg = arg;
2873                 struct lnet_dbg_task_info *dbg_info;
2874                 size_t total = sizeof(*dbg) + sizeof(*dbg_info);
2875
2876                 if (dbg->dbg_hdr.ioc_len < total)
2877                         return -EINVAL;
2878
2879                 dbg_info = (struct lnet_dbg_task_info*) dbg->dbg_bulk;
2880
2881                 return lnet_handle_dbg_task(dbg, dbg_info);
2882         }
2883
2884         default:
2885                 ni = lnet_net2ni_addref(data->ioc_net);
2886                 if (ni == NULL)
2887                         return -EINVAL;
2888
2889                 if (ni->ni_net->net_lnd->lnd_ctl == NULL)
2890                         rc = -EINVAL;
2891                 else
2892                         rc = ni->ni_net->net_lnd->lnd_ctl(ni, cmd, arg);
2893
2894                 lnet_ni_decref(ni);
2895                 return rc;
2896         }
2897         /* not reached */
2898 }
2899 EXPORT_SYMBOL(LNetCtl);
2900
2901 void LNetDebugPeer(lnet_process_id_t id)
2902 {
2903         lnet_debug_peer(id.nid);
2904 }
2905 EXPORT_SYMBOL(LNetDebugPeer);
2906
2907 /**
2908  * Retrieve the lnet_process_id_t ID of LNet interface at \a index. Note that
2909  * all interfaces share a same PID, as requested by LNetNIInit().
2910  *
2911  * \param index Index of the interface to look up.
2912  * \param id On successful return, this location will hold the
2913  * lnet_process_id_t ID of the interface.
2914  *
2915  * \retval 0 If an interface exists at \a index.
2916  * \retval -ENOENT If no interface has been found.
2917  */
2918 int
2919 LNetGetId(unsigned int index, lnet_process_id_t *id)
2920 {
2921         struct lnet_ni   *ni;
2922         struct lnet_net  *net;
2923         int               cpt;
2924         int               rc = -ENOENT;
2925
2926         LASSERT(the_lnet.ln_refcount > 0);
2927
2928         cpt = lnet_net_lock_current();
2929
2930         list_for_each_entry(net, &the_lnet.ln_nets, net_list) {
2931                 list_for_each_entry(ni, &net->net_ni_list, ni_netlist) {
2932                         if (index-- != 0)
2933                                 continue;
2934
2935                         id->nid = ni->ni_nid;
2936                         id->pid = the_lnet.ln_pid;
2937                         rc = 0;
2938                         break;
2939                 }
2940         }
2941
2942         lnet_net_unlock(cpt);
2943         return rc;
2944 }
2945 EXPORT_SYMBOL(LNetGetId);
2946
2947 /**
2948  * Print a string representation of handle \a h into buffer \a str of
2949  * \a len bytes.
2950  */
2951 void
2952 LNetSnprintHandle(char *str, int len, lnet_handle_any_t h)
2953 {
2954         snprintf(str, len, "%#llx", h.cookie);
2955 }
2956 EXPORT_SYMBOL(LNetSnprintHandle);
2957
2958 static int lnet_ping(lnet_process_id_t id, signed long timeout,
2959                      lnet_process_id_t __user *ids, int n_ids)
2960 {
2961         lnet_handle_eq_t     eqh;
2962         lnet_handle_md_t     mdh;
2963         lnet_event_t         event;
2964         lnet_md_t            md = { NULL };
2965         int                  which;
2966         int                  unlinked = 0;
2967         int                  replied = 0;
2968         const signed long a_long_time = msecs_to_jiffies(60 * MSEC_PER_SEC);
2969         int                  infosz;
2970         struct lnet_ping_info    *info;
2971         lnet_process_id_t    tmpid;
2972         int                  i;
2973         int                  nob;
2974         int                  rc;
2975         int                  rc2;
2976         sigset_t         blocked;
2977
2978         infosz = offsetof(struct lnet_ping_info, pi_ni[n_ids]);
2979
2980         /* n_ids limit is arbitrary */
2981         if (n_ids <= 0 || n_ids > 20 || id.nid == LNET_NID_ANY)
2982                 return -EINVAL;
2983
2984         if (id.pid == LNET_PID_ANY)
2985                 id.pid = LNET_PID_LUSTRE;
2986
2987         LIBCFS_ALLOC(info, infosz);
2988         if (info == NULL)
2989                 return -ENOMEM;
2990
2991         /* NB 2 events max (including any unlink event) */
2992         rc = LNetEQAlloc(2, LNET_EQ_HANDLER_NONE, &eqh);
2993         if (rc != 0) {
2994                 CERROR("Can't allocate EQ: %d\n", rc);
2995                 goto out_0;
2996         }
2997
2998         /* initialize md content */
2999         md.start     = info;
3000         md.length    = infosz;
3001         md.threshold = 2; /*GET/REPLY*/
3002         md.max_size  = 0;
3003         md.options   = LNET_MD_TRUNCATE;
3004         md.user_ptr  = NULL;
3005         md.eq_handle = eqh;
3006
3007         rc = LNetMDBind(md, LNET_UNLINK, &mdh);
3008         if (rc != 0) {
3009                 CERROR("Can't bind MD: %d\n", rc);
3010                 goto out_1;
3011         }
3012
3013         rc = LNetGet(LNET_NID_ANY, mdh, id,
3014                      LNET_RESERVED_PORTAL,
3015                      LNET_PROTO_PING_MATCHBITS, 0);
3016
3017         if (rc != 0) {
3018                 /* Don't CERROR; this could be deliberate! */
3019
3020                 rc2 = LNetMDUnlink(mdh);
3021                 LASSERT(rc2 == 0);
3022
3023                 /* NB must wait for the UNLINK event below... */
3024                 unlinked = 1;
3025                 timeout = a_long_time;
3026         }
3027
3028         do {
3029                 /* MUST block for unlink to complete */
3030                 if (unlinked)
3031                         blocked = cfs_block_allsigs();
3032
3033                 rc2 = LNetEQPoll(&eqh, 1, timeout, &event, &which);
3034
3035                 if (unlinked)
3036                         cfs_restore_sigs(blocked);
3037
3038                 CDEBUG(D_NET, "poll %d(%d %d)%s\n", rc2,
3039                        (rc2 <= 0) ? -1 : event.type,
3040                        (rc2 <= 0) ? -1 : event.status,
3041                        (rc2 > 0 && event.unlinked) ? " unlinked" : "");
3042
3043                 LASSERT(rc2 != -EOVERFLOW);     /* can't miss anything */
3044
3045                 if (rc2 <= 0 || event.status != 0) {
3046                         /* timeout or error */
3047                         if (!replied && rc == 0)
3048                                 rc = (rc2 < 0) ? rc2 :
3049                                      (rc2 == 0) ? -ETIMEDOUT :
3050                                      event.status;
3051
3052                         if (!unlinked) {
3053                                 /* Ensure completion in finite time... */
3054                                 LNetMDUnlink(mdh);
3055                                 /* No assertion (racing with network) */
3056                                 unlinked = 1;
3057                                 timeout = a_long_time;
3058                         } else if (rc2 == 0) {
3059                                 /* timed out waiting for unlink */
3060                                 CWARN("ping %s: late network completion\n",
3061                                       libcfs_id2str(id));
3062                         }
3063                 } else if (event.type == LNET_EVENT_REPLY) {
3064                         replied = 1;
3065                         rc = event.mlength;
3066                 }
3067
3068         } while (rc2 <= 0 || !event.unlinked);
3069
3070         if (!replied) {
3071                 if (rc >= 0)
3072                         CWARN("%s: Unexpected rc >= 0 but no reply!\n",
3073                               libcfs_id2str(id));
3074                 rc = -EIO;
3075                 goto out_1;
3076         }
3077
3078         nob = rc;
3079         LASSERT(nob >= 0 && nob <= infosz);
3080
3081         rc = -EPROTO;                           /* if I can't parse... */
3082
3083         if (nob < 8) {
3084                 /* can't check magic/version */
3085                 CERROR("%s: ping info too short %d\n",
3086                        libcfs_id2str(id), nob);
3087                 goto out_1;
3088         }
3089
3090         if (info->pi_magic == __swab32(LNET_PROTO_PING_MAGIC)) {
3091                 lnet_swap_pinginfo(info);
3092         } else if (info->pi_magic != LNET_PROTO_PING_MAGIC) {
3093                 CERROR("%s: Unexpected magic %08x\n",
3094                        libcfs_id2str(id), info->pi_magic);
3095                 goto out_1;
3096         }
3097
3098         if ((info->pi_features & LNET_PING_FEAT_NI_STATUS) == 0) {
3099                 CERROR("%s: ping w/o NI status: 0x%x\n",
3100                        libcfs_id2str(id), info->pi_features);
3101                 goto out_1;
3102         }
3103
3104         if (nob < offsetof(struct lnet_ping_info, pi_ni[0])) {
3105                 CERROR("%s: Short reply %d(%d min)\n", libcfs_id2str(id),
3106                        nob, (int)offsetof(struct lnet_ping_info, pi_ni[0]));
3107                 goto out_1;
3108         }
3109
3110         if (info->pi_nnis < n_ids)
3111                 n_ids = info->pi_nnis;
3112
3113         if (nob < offsetof(struct lnet_ping_info, pi_ni[n_ids])) {
3114                 CERROR("%s: Short reply %d(%d expected)\n", libcfs_id2str(id),
3115                        nob, (int)offsetof(struct lnet_ping_info, pi_ni[n_ids]));
3116                 goto out_1;
3117         }
3118
3119         rc = -EFAULT;                           /* If I SEGV... */
3120
3121         memset(&tmpid, 0, sizeof(tmpid));
3122         for (i = 0; i < n_ids; i++) {
3123                 tmpid.pid = info->pi_pid;
3124                 tmpid.nid = info->pi_ni[i].ns_nid;
3125                 if (copy_to_user(&ids[i], &tmpid, sizeof(tmpid)))
3126                         goto out_1;
3127         }
3128         rc = info->pi_nnis;
3129
3130  out_1:
3131         rc2 = LNetEQFree(eqh);
3132         if (rc2 != 0)
3133                 CERROR("rc2 %d\n", rc2);
3134         LASSERT(rc2 == 0);
3135
3136  out_0:
3137         LIBCFS_FREE(info, infosz);
3138         return rc;
3139 }