sys/ofed/drivers/infiniband/ulp/ipoib/ipoib_main.c

   1 /*
   2  * Copyright (c) 2004 Topspin Communications.  All rights reserved.
   3  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
   4  * Copyright (c) 2004 Voltaire, Inc. All rights reserved.
   5  *
   6  * This software is available to you under a choice of one of two
   7  * licenses.  You may choose to be licensed under the terms of the GNU
   8  * General Public License (GPL) Version 2, available from the file
   9  * COPYING in the main directory of this source tree, or the
  10  * OpenIB.org BSD license below:
  11  *
  12  *     Redistribution and use in source and binary forms, with or
  13  *     without modification, are permitted provided that the following
  14  *     conditions are met:
  15  *
  16  *      - Redistributions of source code must retain the above
  17  *        copyright notice, this list of conditions and the following
  18  *        disclaimer.
  19  *
  20  *      - Redistributions in binary form must reproduce the above
  21  *        copyright notice, this list of conditions and the following
  22  *        disclaimer in the documentation and/or other materials
  23  *        provided with the distribution.
  24  *
  25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  32  * SOFTWARE.
  33  */
  34
  35 #include "ipoib.h"
  36
  37 static  int ipoib_resolvemulti(struct ifnet *, struct sockaddr **,
  38                 struct sockaddr *);
  39
  40
  41 #include <linux/module.h>
  42
  43 #include <linux/slab.h>
  44 #include <linux/kernel.h>
  45 #include <linux/vmalloc.h>
  46
  47 #include <linux/if_arp.h>       /* For ARPHRD_xxx */
  48 #include <linux/if_vlan.h>
  49 #include <net/ip.h>
  50 #include <net/ipv6.h>
  51
  52 MODULE_AUTHOR("Roland Dreier");
  53 MODULE_DESCRIPTION("IP-over-InfiniBand net driver");
  54 MODULE_LICENSE("Dual BSD/GPL");
  55
  56 int ipoib_sendq_size = IPOIB_TX_RING_SIZE;
  57 int ipoib_recvq_size = IPOIB_RX_RING_SIZE;
  58
  59 module_param_named(send_queue_size, ipoib_sendq_size, int, 0444);
  60 MODULE_PARM_DESC(send_queue_size, "Number of descriptors in send queue");
  61 module_param_named(recv_queue_size, ipoib_recvq_size, int, 0444);
  62 MODULE_PARM_DESC(recv_queue_size, "Number of descriptors in receive queue");
  63
  64 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
  65 int ipoib_debug_level = 1;
  66
  67 module_param_named(debug_level, ipoib_debug_level, int, 0644);
  68 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
  69 #endif
  70
  71 struct ipoib_path_iter {
  72         struct ipoib_dev_priv *priv;
  73         struct ipoib_path  path;
  74 };
  75
  76 static const u8 ipv4_bcast_addr[] = {
  77         0x00, 0xff, 0xff, 0xff,
  78         0xff, 0x12, 0x40, 0x1b, 0x00, 0x00, 0x00, 0x00,
  79         0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff
  80 };
  81
  82 struct workqueue_struct *ipoib_workqueue;
  83
  84 struct ib_sa_client ipoib_sa_client;
  85
  86 static void ipoib_add_one(struct ib_device *device);
  87 static void ipoib_remove_one(struct ib_device *device);
  88 static void ipoib_start(struct ifnet *dev);
  89 static int ipoib_output(struct ifnet *ifp, struct mbuf *m,
  90             struct sockaddr *dst, struct route *ro);
  91 static int ipoib_ioctl(struct ifnet *ifp, u_long command, caddr_t data);
  92 static void ipoib_input(struct ifnet *ifp, struct mbuf *m);
  93
  94 #define IPOIB_MTAP(_ifp, _m)                                    \
  95 do {                                                            \
  96         if (bpf_peers_present((_ifp)->if_bpf)) {                \
  97                 M_ASSERTVALID(_m);                              \
  98                 ipoib_mtap_mb((_ifp), (_m));                    \
  99         }                                                       \
 100 } while (0)
 101
 102 /*
 103  * This is for clients that have an ipoib_header in the mbuf.
 104  */
 105 static void
 106 ipoib_mtap_mb(struct ifnet *ifp, struct mbuf *mb)
 107 {
 108         struct ipoib_header *ih;
 109         struct ether_header eh;
 110
 111         ih = mtod(mb, struct ipoib_header *);
 112         eh.ether_type = ih->proto;
 113         bcopy(ih->hwaddr, &eh.ether_dhost, ETHER_ADDR_LEN);
 114         bzero(&eh.ether_shost, ETHER_ADDR_LEN);
 115         mb->m_data += sizeof(struct ipoib_header);
 116         mb->m_len -= sizeof(struct ipoib_header);
 117         bpf_mtap2(ifp->if_bpf, &eh, sizeof(eh), mb);
 118         mb->m_data -= sizeof(struct ipoib_header);
 119         mb->m_len += sizeof(struct ipoib_header);
 120 }
 121
 122 void
 123 ipoib_mtap_proto(struct ifnet *ifp, struct mbuf *mb, uint16_t proto)
 124 {
 125         struct ether_header eh;
 126
 127         eh.ether_type = proto;
 128         bzero(&eh.ether_shost, ETHER_ADDR_LEN);
 129         bzero(&eh.ether_dhost, ETHER_ADDR_LEN);
 130         bpf_mtap2(ifp->if_bpf, &eh, sizeof(eh), mb);
 131 }
 132
 133 static struct ib_client ipoib_client = {
 134         .name   = "ipoib",
 135         .add    = ipoib_add_one,
 136         .remove = ipoib_remove_one
 137 };
 138
 139 int
 140 ipoib_open(struct ipoib_dev_priv *priv)
 141 {
 142         struct ifnet *dev = priv->dev;
 143
 144         ipoib_dbg(priv, "bringing up interface\n");
 145
 146         set_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
 147
 148         if (ipoib_pkey_dev_delay_open(priv))
 149                 return 0;
 150
 151         if (ipoib_ib_dev_open(priv))
 152                 goto err_disable;
 153
 154         if (ipoib_ib_dev_up(priv))
 155                 goto err_stop;
 156
 157         if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
 158                 struct ipoib_dev_priv *cpriv;
 159
 160                 /* Bring up any child interfaces too */
 161                 mutex_lock(&priv->vlan_mutex);
 162                 list_for_each_entry(cpriv, &priv->child_intfs, list)
 163                         if ((cpriv->dev->if_drv_flags & IFF_DRV_RUNNING) == 0)
 164                                 ipoib_open(cpriv);
 165                 mutex_unlock(&priv->vlan_mutex);
 166         }
 167         dev->if_drv_flags |= IFF_DRV_RUNNING;
 168         dev->if_drv_flags &= ~IFF_DRV_OACTIVE;
 169
 170         return 0;
 171
 172 err_stop:
 173         ipoib_ib_dev_stop(priv, 1);
 174
 175 err_disable:
 176         clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
 177
 178         return -EINVAL;
 179 }
 180
 181 static void
 182 ipoib_init(void *arg)
 183 {
 184         struct ifnet *dev;
 185         struct ipoib_dev_priv *priv;
 186
 187         priv = arg;
 188         dev = priv->dev;
 189         if ((dev->if_drv_flags & IFF_DRV_RUNNING) == 0)
 190                 ipoib_open(priv);
 191         queue_work(ipoib_workqueue, &priv->flush_light);
 192 }
 193
 194
 195 static int
 196 ipoib_stop(struct ipoib_dev_priv *priv)
 197 {
 198         struct ifnet *dev = priv->dev;
 199
 200         ipoib_dbg(priv, "stopping interface\n");
 201
 202         clear_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags);
 203
 204         dev->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
 205
 206         ipoib_ib_dev_down(priv, 0);
 207         ipoib_ib_dev_stop(priv, 0);
 208
 209         if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
 210                 struct ipoib_dev_priv *cpriv;
 211
 212                 /* Bring down any child interfaces too */
 213                 mutex_lock(&priv->vlan_mutex);
 214                 list_for_each_entry(cpriv, &priv->child_intfs, list)
 215                         if ((cpriv->dev->if_drv_flags & IFF_DRV_RUNNING) != 0)
 216                                 ipoib_stop(cpriv);
 217                 mutex_unlock(&priv->vlan_mutex);
 218         }
 219
 220         return 0;
 221 }
 222
 223 int
 224 ipoib_change_mtu(struct ipoib_dev_priv *priv, int new_mtu)
 225 {
 226         struct ifnet *dev = priv->dev;
 227
 228         /* dev->if_mtu > 2K ==> connected mode */
 229         if (ipoib_cm_admin_enabled(priv)) {
 230                 if (new_mtu > IPOIB_CM_MTU(ipoib_cm_max_mtu(priv)))
 231                         return -EINVAL;
 232
 233                 if (new_mtu > priv->mcast_mtu)
 234                         ipoib_warn(priv, "mtu > %d will cause multicast packet drops.\n",
 235                                    priv->mcast_mtu);
 236
 237                 dev->if_mtu = new_mtu;
 238                 return 0;
 239         }
 240
 241         if (new_mtu > IPOIB_UD_MTU(priv->max_ib_mtu))
 242                 return -EINVAL;
 243
 244         priv->admin_mtu = new_mtu;
 245
 246         dev->if_mtu = min(priv->mcast_mtu, priv->admin_mtu);
 247
 248         queue_work(ipoib_workqueue, &priv->flush_light);
 249
 250         return 0;
 251 }
 252
 253 static int
 254 ipoib_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
 255 {
 256         struct ipoib_dev_priv *priv = ifp->if_softc;
 257         struct ifaddr *ifa = (struct ifaddr *) data;
 258         struct ifreq *ifr = (struct ifreq *) data;
 259         int error = 0;
 260
 261         /* check if detaching */
 262         if (priv == NULL || priv->gone != 0)
 263                 return (ENXIO);
 264
 265         switch (command) {
 266         case SIOCSIFFLAGS:
 267                 if (ifp->if_flags & IFF_UP) {
 268                         if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0)
 269                                 error = -ipoib_open(priv);
 270                 } else
 271                         if (ifp->if_drv_flags & IFF_DRV_RUNNING)
 272                                 ipoib_stop(priv);
 273                 break;
 274         case SIOCADDMULTI:
 275         case SIOCDELMULTI:
 276                 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
 277                         queue_work(ipoib_workqueue, &priv->restart_task);
 278                 break;
 279         case SIOCSIFADDR:
 280                 ifp->if_flags |= IFF_UP;
 281
 282                 switch (ifa->ifa_addr->sa_family) {
 283 #ifdef INET
 284                 case AF_INET:
 285                         ifp->if_init(ifp->if_softc);    /* before arpwhohas */
 286                         arp_ifinit(ifp, ifa);
 287                         break;
 288 #endif
 289                 default:
 290                         ifp->if_init(ifp->if_softc);
 291                         break;
 292                 }
 293                 break;
 294
 295         case SIOCGIFADDR:
 296                 {
 297                         struct sockaddr *sa;
 298
 299                         sa = (struct sockaddr *) & ifr->ifr_data;
 300                         bcopy(IF_LLADDR(ifp),
 301                               (caddr_t) sa->sa_data, INFINIBAND_ALEN);
 302                 }
 303                 break;
 304
 305         case SIOCSIFMTU:
 306                 /*
 307                  * Set the interface MTU.
 308                  */
 309                 error = -ipoib_change_mtu(priv, ifr->ifr_mtu);
 310                 break;
 311         default:
 312                 error = EINVAL;
 313                 break;
 314         }
 315         return (error);
 316 }
 317
 318
 319 static struct ipoib_path *
 320 __path_find(struct ipoib_dev_priv *priv, void *gid)
 321 {
 322         struct rb_node *n = priv->path_tree.rb_node;
 323         struct ipoib_path *path;
 324         int ret;
 325
 326         while (n) {
 327                 path = rb_entry(n, struct ipoib_path, rb_node);
 328
 329                 ret = memcmp(gid, path->pathrec.dgid.raw,
 330                              sizeof (union ib_gid));
 331
 332                 if (ret < 0)
 333                         n = n->rb_left;
 334                 else if (ret > 0)
 335                         n = n->rb_right;
 336                 else
 337                         return path;
 338         }
 339
 340         return NULL;
 341 }
 342
 343 static int
 344 __path_add(struct ipoib_dev_priv *priv, struct ipoib_path *path)
 345 {
 346         struct rb_node **n = &priv->path_tree.rb_node;
 347         struct rb_node *pn = NULL;
 348         struct ipoib_path *tpath;
 349         int ret;
 350
 351         while (*n) {
 352                 pn = *n;
 353                 tpath = rb_entry(pn, struct ipoib_path, rb_node);
 354
 355                 ret = memcmp(path->pathrec.dgid.raw, tpath->pathrec.dgid.raw,
 356                              sizeof (union ib_gid));
 357                 if (ret < 0)
 358                         n = &pn->rb_left;
 359                 else if (ret > 0)
 360                         n = &pn->rb_right;
 361                 else
 362                         return -EEXIST;
 363         }
 364
 365         rb_link_node(&path->rb_node, pn, n);
 366         rb_insert_color(&path->rb_node, &priv->path_tree);
 367
 368         list_add_tail(&path->list, &priv->path_list);
 369
 370         return 0;
 371 }
 372
 373 void
 374 ipoib_path_free(struct ipoib_dev_priv *priv, struct ipoib_path *path)
 375 {
 376
 377         _IF_DRAIN(&path->queue);
 378
 379         if (path->ah)
 380                 ipoib_put_ah(path->ah);
 381         if (ipoib_cm_get(path))
 382                 ipoib_cm_destroy_tx(ipoib_cm_get(path));
 383
 384         kfree(path);
 385 }
 386
 387 #ifdef CONFIG_INFINIBAND_IPOIB_DEBUG
 388
 389 struct ipoib_path_iter *
 390 ipoib_path_iter_init(struct ipoib_dev_priv *priv)
 391 {
 392         struct ipoib_path_iter *iter;
 393
 394         iter = kmalloc(sizeof *iter, GFP_KERNEL);
 395         if (!iter)
 396                 return NULL;
 397
 398         iter->priv = priv;
 399         memset(iter->path.pathrec.dgid.raw, 0, 16);
 400
 401         if (ipoib_path_iter_next(iter)) {
 402                 kfree(iter);
 403                 return NULL;
 404         }
 405
 406         return iter;
 407 }
 408
 409 int
 410 ipoib_path_iter_next(struct ipoib_path_iter *iter)
 411 {
 412         struct ipoib_dev_priv *priv = iter->priv;
 413         struct rb_node *n;
 414         struct ipoib_path *path;
 415         int ret = 1;
 416
 417         spin_lock_irq(&priv->lock);
 418
 419         n = rb_first(&priv->path_tree);
 420
 421         while (n) {
 422                 path = rb_entry(n, struct ipoib_path, rb_node);
 423
 424                 if (memcmp(iter->path.pathrec.dgid.raw, path->pathrec.dgid.raw,
 425                            sizeof (union ib_gid)) < 0) {
 426                         iter->path = *path;
 427                         ret = 0;
 428                         break;
 429                 }
 430
 431                 n = rb_next(n);
 432         }
 433
 434         spin_unlock_irq(&priv->lock);
 435
 436         return ret;
 437 }
 438
 439 void
 440 ipoib_path_iter_read(struct ipoib_path_iter *iter, struct ipoib_path *path)
 441 {
 442         *path = iter->path;
 443 }
 444
 445 #endif /* CONFIG_INFINIBAND_IPOIB_DEBUG */
 446
 447 void
 448 ipoib_mark_paths_invalid(struct ipoib_dev_priv *priv)
 449 {
 450         struct ipoib_path *path, *tp;
 451
 452         spin_lock_irq(&priv->lock);
 453
 454         list_for_each_entry_safe(path, tp, &priv->path_list, list) {
 455                 ipoib_dbg(priv, "mark path LID 0x%04x GID %16D invalid\n",
 456                         be16_to_cpu(path->pathrec.dlid),
 457                         path->pathrec.dgid.raw, ":");
 458                 path->valid =  0;
 459         }
 460
 461         spin_unlock_irq(&priv->lock);
 462 }
 463
 464 void
 465 ipoib_flush_paths(struct ipoib_dev_priv *priv)
 466 {
 467         struct ipoib_path *path, *tp;
 468         LIST_HEAD(remove_list);
 469         unsigned long flags;
 470
 471         spin_lock_irqsave(&priv->lock, flags);
 472
 473         list_splice_init(&priv->path_list, &remove_list);
 474
 475         list_for_each_entry(path, &remove_list, list)
 476                 rb_erase(&path->rb_node, &priv->path_tree);
 477
 478         list_for_each_entry_safe(path, tp, &remove_list, list) {
 479                 if (path->query)
 480                         ib_sa_cancel_query(path->query_id, path->query);
 481                 spin_unlock_irqrestore(&priv->lock, flags);
 482                 wait_for_completion(&path->done);
 483                 ipoib_path_free(priv, path);
 484                 spin_lock_irqsave(&priv->lock, flags);
 485         }
 486
 487         spin_unlock_irqrestore(&priv->lock, flags);
 488 }
 489
 490 static void
 491 path_rec_completion(int status, struct ib_sa_path_rec *pathrec, void *path_ptr)
 492 {
 493         struct ipoib_path *path = path_ptr;
 494         struct ipoib_dev_priv *priv = path->priv;
 495         struct ifnet *dev = priv->dev;
 496         struct ipoib_ah *ah = NULL;
 497         struct ipoib_ah *old_ah = NULL;
 498         struct ifqueue mbqueue;
 499         struct mbuf *mb;
 500         unsigned long flags;
 501
 502         if (!status)
 503                 ipoib_dbg(priv, "PathRec LID 0x%04x for GID %16D\n",
 504                           be16_to_cpu(pathrec->dlid), pathrec->dgid.raw, ":");
 505         else
 506                 ipoib_dbg(priv, "PathRec status %d for GID %16D\n",
 507                           status, path->pathrec.dgid.raw, ":");
 508
 509         bzero(&mbqueue, sizeof(mbqueue));
 510
 511         if (!status) {
 512                 struct ib_ah_attr av;
 513
 514                 if (!ib_init_ah_from_path(priv->ca, priv->port, pathrec, &av))
 515                         ah = ipoib_create_ah(priv, priv->pd, &av);
 516         }
 517
 518         spin_lock_irqsave(&priv->lock, flags);
 519
 520         if (ah) {
 521                 path->pathrec = *pathrec;
 522
 523                 old_ah   = path->ah;
 524                 path->ah = ah;
 525
 526                 ipoib_dbg(priv, "created address handle %p for LID 0x%04x, SL %d\n",
 527                           ah, be16_to_cpu(pathrec->dlid), pathrec->sl);
 528
 529                 for (;;) {
 530                         _IF_DEQUEUE(&path->queue, mb);
 531                         if (mb == NULL)
 532                                 break;
 533                         _IF_ENQUEUE(&mbqueue, mb);
 534                 }
 535
 536 #ifdef CONFIG_INFINIBAND_IPOIB_CM
 537                 if (ipoib_cm_enabled(priv, path->hwaddr) && !ipoib_cm_get(path))
 538                         ipoib_cm_set(path, ipoib_cm_create_tx(priv, path));
 539 #endif
 540
 541                 path->valid = 1;
 542         }
 543
 544         path->query = NULL;
 545         complete(&path->done);
 546
 547         spin_unlock_irqrestore(&priv->lock, flags);
 548
 549         if (old_ah)
 550                 ipoib_put_ah(old_ah);
 551
 552         for (;;) {
 553                 _IF_DEQUEUE(&mbqueue, mb);
 554                 if (mb == NULL)
 555                         break;
 556                 mb->m_pkthdr.rcvif = dev;
 557                 if (dev->if_transmit(dev, mb))
 558                         ipoib_warn(priv, "dev_queue_xmit failed "
 559                                    "to requeue packet\n");
 560         }
 561 }
 562
 563 static struct ipoib_path *
 564 path_rec_create(struct ipoib_dev_priv *priv, uint8_t *hwaddr)
 565 {
 566         struct ipoib_path *path;
 567
 568         if (!priv->broadcast)
 569                 return NULL;
 570
 571         path = kzalloc(sizeof *path, GFP_ATOMIC);
 572         if (!path)
 573                 return NULL;
 574
 575         path->priv = priv;
 576
 577         bzero(&path->queue, sizeof(path->queue));
 578
 579 #ifdef CONFIG_INFINIBAND_IPOIB_CM
 580         memcpy(&path->hwaddr, hwaddr, INFINIBAND_ALEN);
 581 #endif
 582         memcpy(path->pathrec.dgid.raw, &hwaddr[4], sizeof (union ib_gid));
 583         path->pathrec.sgid          = priv->local_gid;
 584         path->pathrec.pkey          = cpu_to_be16(priv->pkey);
 585         path->pathrec.numb_path     = 1;
 586         path->pathrec.traffic_class = priv->broadcast->mcmember.traffic_class;
 587
 588         return path;
 589 }
 590
 591 static int
 592 path_rec_start(struct ipoib_dev_priv *priv, struct ipoib_path *path)
 593 {
 594         struct ifnet *dev = priv->dev;
 595
 596         ib_sa_comp_mask comp_mask = IB_SA_PATH_REC_MTU_SELECTOR | IB_SA_PATH_REC_MTU;
 597         struct ib_sa_path_rec p_rec;
 598
 599         p_rec = path->pathrec;
 600         p_rec.mtu_selector = IB_SA_GT;
 601
 602         switch (roundup_pow_of_two(dev->if_mtu + IPOIB_ENCAP_LEN)) {
 603         case 512:
 604                 p_rec.mtu = IB_MTU_256;
 605                 break;
 606         case 1024:
 607                 p_rec.mtu = IB_MTU_512;
 608                 break;
 609         case 2048:
 610                 p_rec.mtu = IB_MTU_1024;
 611                 break;
 612         case 4096:
 613                 p_rec.mtu = IB_MTU_2048;
 614                 break;
 615         default:
 616                 /* Wildcard everything */
 617                 comp_mask = 0;
 618                 p_rec.mtu = 0;
 619                 p_rec.mtu_selector = 0;
 620         }
 621
 622         ipoib_dbg(priv, "Start path record lookup for %16D MTU > %d\n",
 623                   p_rec.dgid.raw, ":",
 624                   comp_mask ? ib_mtu_enum_to_int(p_rec.mtu) : 0);
 625
 626         init_completion(&path->done);
 627
 628         path->query_id =
 629                 ib_sa_path_rec_get(&ipoib_sa_client, priv->ca, priv->port,
 630                                    &p_rec, comp_mask            |
 631                                    IB_SA_PATH_REC_DGID          |
 632                                    IB_SA_PATH_REC_SGID          |
 633                                    IB_SA_PATH_REC_NUMB_PATH     |
 634                                    IB_SA_PATH_REC_TRAFFIC_CLASS |
 635                                    IB_SA_PATH_REC_PKEY,
 636                                    1000, GFP_ATOMIC,
 637                                    path_rec_completion,
 638                                    path, &path->query);
 639         if (path->query_id < 0) {
 640                 ipoib_warn(priv, "ib_sa_path_rec_get failed: %d\n", path->query_id);
 641                 path->query = NULL;
 642                 complete(&path->done);
 643                 return path->query_id;
 644         }
 645
 646         return 0;
 647 }
 648
 649 static void
 650 ipoib_unicast_send(struct mbuf *mb, struct ipoib_dev_priv *priv, struct ipoib_header *eh)
 651 {
 652         struct ipoib_path *path;
 653
 654         path = __path_find(priv, eh->hwaddr + 4);
 655         if (!path || !path->valid) {
 656                 int new_path = 0;
 657
 658                 if (!path) {
 659                         path = path_rec_create(priv, eh->hwaddr);
 660                         new_path = 1;
 661                 }
 662                 if (path) {
 663                         _IF_ENQUEUE(&path->queue, mb);
 664                         if (!path->query && path_rec_start(priv, path)) {
 665                                 spin_unlock_irqrestore(&priv->lock, flags);
 666                                 if (new_path)
 667                                         ipoib_path_free(priv, path);
 668                                 return;
 669                         } else
 670                                 __path_add(priv, path);
 671                 } else {
 672                         ++priv->dev->if_oerrors;
 673                         m_freem(mb);
 674                 }
 675
 676                 return;
 677         }
 678
 679         if (ipoib_cm_get(path) && ipoib_cm_up(path)) {
 680                 ipoib_cm_send(priv, mb, ipoib_cm_get(path));
 681         } else if (path->ah) {
 682                 ipoib_send(priv, mb, path->ah, IPOIB_QPN(eh->hwaddr));
 683         } else if ((path->query || !path_rec_start(priv, path)) &&
 684                     path->queue.ifq_len < IPOIB_MAX_PATH_REC_QUEUE) {
 685                 _IF_ENQUEUE(&path->queue, mb);
 686         } else {
 687                 ++priv->dev->if_oerrors;
 688                 m_freem(mb);
 689         }
 690 }
 691
 692 static int
 693 ipoib_send_one(struct ipoib_dev_priv *priv, struct mbuf *mb)
 694 {
 695         struct ipoib_header *eh;
 696
 697         eh = mtod(mb, struct ipoib_header *);
 698         if (IPOIB_IS_MULTICAST(eh->hwaddr)) {
 699                 /* Add in the P_Key for multicast*/
 700                 eh->hwaddr[8] = (priv->pkey >> 8) & 0xff;
 701                 eh->hwaddr[9] = priv->pkey & 0xff;
 702
 703                 ipoib_mcast_send(priv, eh->hwaddr + 4, mb);
 704         } else
 705                 ipoib_unicast_send(mb, priv, eh);
 706
 707         return 0;
 708 }
 709
 710
 711 static void
 712 _ipoib_start(struct ifnet *dev, struct ipoib_dev_priv *priv)
 713 {
 714         struct mbuf *mb;
 715
 716         if ((dev->if_drv_flags & (IFF_DRV_RUNNING|IFF_DRV_OACTIVE)) !=
 717             IFF_DRV_RUNNING)
 718                 return;
 719
 720         spin_lock(&priv->lock);
 721         while (!IFQ_DRV_IS_EMPTY(&dev->if_snd) &&
 722             (dev->if_drv_flags & IFF_DRV_OACTIVE) == 0) {
 723                 IFQ_DRV_DEQUEUE(&dev->if_snd, mb);
 724                 if (mb == NULL)
 725                         break;
 726                 IPOIB_MTAP(dev, mb);
 727                 ipoib_send_one(priv, mb);
 728         }
 729         spin_unlock(&priv->lock);
 730 }
 731
 732 static void
 733 ipoib_start(struct ifnet *dev)
 734 {
 735         _ipoib_start(dev, dev->if_softc);
 736 }
 737
 738 static void
 739 ipoib_vlan_start(struct ifnet *dev)
 740 {
 741         struct ipoib_dev_priv *priv;
 742         struct mbuf *mb;
 743
 744         priv = VLAN_COOKIE(dev);
 745         if (priv != NULL)
 746                 return _ipoib_start(dev, priv);
 747         while (!IFQ_DRV_IS_EMPTY(&dev->if_snd)) {
 748                 IFQ_DRV_DEQUEUE(&dev->if_snd, mb);
 749                 if (mb == NULL)
 750                         break;
 751                 m_freem(mb);
 752                 dev->if_oerrors++;
 753         }
 754 }
 755
 756 int
 757 ipoib_dev_init(struct ipoib_dev_priv *priv, struct ib_device *ca, int port)
 758 {
 759
 760         /* Allocate RX/TX "rings" to hold queued mbs */
 761         priv->rx_ring = kzalloc(ipoib_recvq_size * sizeof *priv->rx_ring,
 762                                 GFP_KERNEL);
 763         if (!priv->rx_ring) {
 764                 printk(KERN_WARNING "%s: failed to allocate RX ring (%d entries)\n",
 765                        ca->name, ipoib_recvq_size);
 766                 goto out;
 767         }
 768
 769         priv->tx_ring = kzalloc(ipoib_sendq_size * sizeof *priv->tx_ring, GFP_KERNEL);
 770         if (!priv->tx_ring) {
 771                 printk(KERN_WARNING "%s: failed to allocate TX ring (%d entries)\n",
 772                        ca->name, ipoib_sendq_size);
 773                 goto out_rx_ring_cleanup;
 774         }
 775         memset(priv->tx_ring, 0, ipoib_sendq_size * sizeof *priv->tx_ring);
 776
 777         /* priv->tx_head, tx_tail & tx_outstanding are already 0 */
 778
 779         if (ipoib_ib_dev_init(priv, ca, port))
 780                 goto out_tx_ring_cleanup;
 781
 782         return 0;
 783
 784 out_tx_ring_cleanup:
 785         kfree(priv->tx_ring);
 786
 787 out_rx_ring_cleanup:
 788         kfree(priv->rx_ring);
 789
 790 out:
 791         return -ENOMEM;
 792 }
 793
 794 static void
 795 ipoib_detach(struct ipoib_dev_priv *priv)
 796 {
 797         struct ifnet *dev;
 798
 799         dev = priv->dev;
 800         if (!test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
 801                 priv->gone = 1;
 802                 bpfdetach(dev);
 803                 if_detach(dev);
 804                 if_free(dev);
 805         } else
 806                 VLAN_SETCOOKIE(priv->dev, NULL);
 807
 808         free(priv, M_TEMP);
 809 }
 810
 811 void
 812 ipoib_dev_cleanup(struct ipoib_dev_priv *priv)
 813 {
 814         struct ipoib_dev_priv *cpriv, *tcpriv;
 815
 816         /* Delete any child interfaces first */
 817         list_for_each_entry_safe(cpriv, tcpriv, &priv->child_intfs, list) {
 818                 ipoib_dev_cleanup(cpriv);
 819                 ipoib_detach(cpriv);
 820         }
 821
 822         ipoib_ib_dev_cleanup(priv);
 823
 824         kfree(priv->rx_ring);
 825         kfree(priv->tx_ring);
 826
 827         priv->rx_ring = NULL;
 828         priv->tx_ring = NULL;
 829 }
 830
 831 static volatile int ipoib_unit;
 832
 833 static struct ipoib_dev_priv *
 834 ipoib_priv_alloc(void)
 835 {
 836         struct ipoib_dev_priv *priv;
 837
 838         priv = malloc(sizeof(struct ipoib_dev_priv), M_TEMP, M_ZERO|M_WAITOK);
 839         spin_lock_init(&priv->lock);
 840         mutex_init(&priv->vlan_mutex);
 841         INIT_LIST_HEAD(&priv->path_list);
 842         INIT_LIST_HEAD(&priv->child_intfs);
 843         INIT_LIST_HEAD(&priv->dead_ahs);
 844         INIT_LIST_HEAD(&priv->multicast_list);
 845         INIT_DELAYED_WORK(&priv->pkey_poll_task, ipoib_pkey_poll);
 846         INIT_DELAYED_WORK(&priv->mcast_task,   ipoib_mcast_join_task);
 847         INIT_WORK(&priv->carrier_on_task, ipoib_mcast_carrier_on_task);
 848         INIT_WORK(&priv->flush_light,   ipoib_ib_dev_flush_light);
 849         INIT_WORK(&priv->flush_normal,   ipoib_ib_dev_flush_normal);
 850         INIT_WORK(&priv->flush_heavy,   ipoib_ib_dev_flush_heavy);
 851         INIT_WORK(&priv->restart_task, ipoib_mcast_restart_task);
 852         INIT_DELAYED_WORK(&priv->ah_reap_task, ipoib_reap_ah);
 853         memcpy(priv->broadcastaddr, ipv4_bcast_addr, INFINIBAND_ALEN);
 854
 855         return (priv);
 856 }
 857
 858 struct ipoib_dev_priv *
 859 ipoib_intf_alloc(const char *name)
 860 {
 861         struct ipoib_dev_priv *priv;
 862         struct sockaddr_dl *sdl;
 863         struct ifnet *dev;
 864
 865         priv = ipoib_priv_alloc();
 866         dev = priv->dev = if_alloc(IFT_INFINIBAND);
 867         if (!dev) {
 868                 free(priv, M_TEMP);
 869                 return NULL;
 870         }
 871         dev->if_softc = priv;
 872         if_initname(dev, name, atomic_fetchadd_int(&ipoib_unit, 1));
 873         dev->if_flags = IFF_BROADCAST | IFF_MULTICAST;
 874         dev->if_addrlen = INFINIBAND_ALEN;
 875         dev->if_hdrlen = IPOIB_HEADER_LEN;
 876         if_attach(dev);
 877         dev->if_init = ipoib_init;
 878         dev->if_ioctl = ipoib_ioctl;
 879         dev->if_start = ipoib_start;
 880         dev->if_output = ipoib_output;
 881         dev->if_input = ipoib_input;
 882         dev->if_resolvemulti = ipoib_resolvemulti;
 883         dev->if_baudrate = IF_Gbps(10UL);
 884         dev->if_broadcastaddr = priv->broadcastaddr;
 885         dev->if_snd.ifq_maxlen = ipoib_sendq_size * 2;
 886         sdl = (struct sockaddr_dl *)dev->if_addr->ifa_addr;
 887         sdl->sdl_type = IFT_INFINIBAND;
 888         sdl->sdl_alen = dev->if_addrlen;
 889         priv->dev = dev;
 890         if_link_state_change(dev, LINK_STATE_DOWN);
 891         bpfattach(dev, DLT_EN10MB, ETHER_HDR_LEN);
 892
 893         return dev->if_softc;
 894 }
 895
 896 int
 897 ipoib_set_dev_features(struct ipoib_dev_priv *priv, struct ib_device *hca)
 898 {
 899         struct ib_device_attr *device_attr;
 900         int result = -ENOMEM;
 901
 902         device_attr = kmalloc(sizeof *device_attr, GFP_KERNEL);
 903         if (!device_attr) {
 904                 printk(KERN_WARNING "%s: allocation of %zu bytes failed\n",
 905                        hca->name, sizeof *device_attr);
 906                 return result;
 907         }
 908
 909         result = ib_query_device(hca, device_attr);
 910         if (result) {
 911                 printk(KERN_WARNING "%s: ib_query_device failed (ret = %d)\n",
 912                        hca->name, result);
 913                 kfree(device_attr);
 914                 return result;
 915         }
 916         priv->hca_caps = device_attr->device_cap_flags;
 917
 918         kfree(device_attr);
 919
 920         priv->dev->if_hwassist = 0;
 921         priv->dev->if_capabilities = 0;
 922
 923 #ifndef CONFIG_INFINIBAND_IPOIB_CM
 924         if (priv->hca_caps & IB_DEVICE_UD_IP_CSUM) {
 925                 set_bit(IPOIB_FLAG_CSUM, &priv->flags);
 926                 priv->dev->if_hwassist = CSUM_IP | CSUM_TCP | CSUM_UDP;
 927                 priv->dev->if_capabilities = IFCAP_HWCSUM | IFCAP_VLAN_HWCSUM;
 928         }
 929
 930 #if 0
 931         if (priv->dev->features & NETIF_F_SG && priv->hca_caps & IB_DEVICE_UD_TSO) {
 932                 priv->dev->if_capabilities |= IFCAP_TSO4;
 933                 priv->dev->if_hwassist |= CSUM_TSO;
 934         }
 935 #endif
 936 #endif
 937         priv->dev->if_capabilities |=
 938             IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_LINKSTATE;
 939         priv->dev->if_capenable = priv->dev->if_capabilities;
 940
 941         return 0;
 942 }
 943
 944
 945 static struct ifnet *
 946 ipoib_add_port(const char *format, struct ib_device *hca, u8 port)
 947 {
 948         struct ipoib_dev_priv *priv;
 949         struct ib_port_attr attr;
 950         int result = -ENOMEM;
 951
 952         priv = ipoib_intf_alloc(format);
 953         if (!priv)
 954                 goto alloc_mem_failed;
 955
 956         if (!ib_query_port(hca, port, &attr))
 957                 priv->max_ib_mtu = ib_mtu_enum_to_int(attr.max_mtu);
 958         else {
 959                 printk(KERN_WARNING "%s: ib_query_port %d failed\n",
 960                        hca->name, port);
 961                 goto device_init_failed;
 962         }
 963
 964         /* MTU will be reset when mcast join happens */
 965         priv->dev->if_mtu = IPOIB_UD_MTU(priv->max_ib_mtu);
 966         priv->mcast_mtu = priv->admin_mtu = priv->dev->if_mtu;
 967
 968         result = ib_query_pkey(hca, port, 0, &priv->pkey);
 969         if (result) {
 970                 printk(KERN_WARNING "%s: ib_query_pkey port %d failed (ret = %d)\n",
 971                        hca->name, port, result);
 972                 goto device_init_failed;
 973         }
 974
 975         if (ipoib_set_dev_features(priv, hca))
 976                 goto device_init_failed;
 977
 978         /*
 979          * Set the full membership bit, so that we join the right
 980          * broadcast group, etc.
 981          */
 982         priv->pkey |= 0x8000;
 983
 984         priv->broadcastaddr[8] = priv->pkey >> 8;
 985         priv->broadcastaddr[9] = priv->pkey & 0xff;
 986
 987         result = ib_query_gid(hca, port, 0, &priv->local_gid);
 988         if (result) {
 989                 printk(KERN_WARNING "%s: ib_query_gid port %d failed (ret = %d)\n",
 990                        hca->name, port, result);
 991                 goto device_init_failed;
 992         }
 993         memcpy(IF_LLADDR(priv->dev) + 4, priv->local_gid.raw, sizeof (union ib_gid));
 994
 995         result = ipoib_dev_init(priv, hca, port);
 996         if (result < 0) {
 997                 printk(KERN_WARNING "%s: failed to initialize port %d (ret = %d)\n",
 998                        hca->name, port, result);
 999                 goto device_init_failed;
1000         }
1001         if (ipoib_cm_admin_enabled(priv))
1002                 priv->dev->if_mtu = IPOIB_CM_MTU(ipoib_cm_max_mtu(priv));
1003
1004         INIT_IB_EVENT_HANDLER(&priv->event_handler,
1005                               priv->ca, ipoib_event);
1006         result = ib_register_event_handler(&priv->event_handler);
1007         if (result < 0) {
1008                 printk(KERN_WARNING "%s: ib_register_event_handler failed for "
1009                        "port %d (ret = %d)\n",
1010                        hca->name, port, result);
1011                 goto event_failed;
1012         }
1013         if_printf(priv->dev, "Attached to %s port %d\n", hca->name, port);
1014
1015         return priv->dev;
1016
1017 event_failed:
1018         ipoib_dev_cleanup(priv);
1019
1020 device_init_failed:
1021         ipoib_detach(priv);
1022
1023 alloc_mem_failed:
1024         return ERR_PTR(result);
1025 }
1026
1027 static void
1028 ipoib_add_one(struct ib_device *device)
1029 {
1030         struct list_head *dev_list;
1031         struct ifnet *dev;
1032         struct ipoib_dev_priv *priv;
1033         int s, e, p;
1034
1035         if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1036                 return;
1037
1038         dev_list = kmalloc(sizeof *dev_list, GFP_KERNEL);
1039         if (!dev_list)
1040                 return;
1041
1042         INIT_LIST_HEAD(dev_list);
1043
1044         if (device->node_type == RDMA_NODE_IB_SWITCH) {
1045                 s = 0;
1046                 e = 0;
1047         } else {
1048                 s = 1;
1049                 e = device->phys_port_cnt;
1050         }
1051
1052         for (p = s; p <= e; ++p) {
1053                 if (rdma_port_get_link_layer(device, p) != IB_LINK_LAYER_INFINIBAND)
1054                         continue;
1055                 dev = ipoib_add_port("ib", device, p);
1056                 if (!IS_ERR(dev)) {
1057                         priv = dev->if_softc;
1058                         list_add_tail(&priv->list, dev_list);
1059                 }
1060         }
1061
1062         ib_set_client_data(device, &ipoib_client, dev_list);
1063 }
1064
1065 static void
1066 ipoib_remove_one(struct ib_device *device)
1067 {
1068         struct ipoib_dev_priv *priv, *tmp;
1069         struct list_head *dev_list;
1070
1071         if (rdma_node_get_transport(device->node_type) != RDMA_TRANSPORT_IB)
1072                 return;
1073
1074         dev_list = ib_get_client_data(device, &ipoib_client);
1075
1076         list_for_each_entry_safe(priv, tmp, dev_list, list) {
1077                 if (rdma_port_get_link_layer(device, priv->port) != IB_LINK_LAYER_INFINIBAND)
1078                         continue;
1079
1080                 ipoib_stop(priv);
1081
1082                 ib_unregister_event_handler(&priv->event_handler);
1083
1084                 /* dev_change_flags(priv->dev, priv->dev->flags & ~IFF_UP); */
1085
1086                 flush_workqueue(ipoib_workqueue);
1087
1088                 ipoib_dev_cleanup(priv);
1089                 ipoib_detach(priv);
1090         }
1091
1092         kfree(dev_list);
1093 }
1094
1095 static void
1096 ipoib_config_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
1097 {
1098         struct ipoib_dev_priv *parent;
1099         struct ipoib_dev_priv *priv;
1100         struct ifnet *dev;
1101         uint16_t pkey;
1102         int error;
1103
1104         if (ifp->if_type != IFT_INFINIBAND)
1105                 return;
1106         dev = VLAN_DEVAT(ifp, vtag);
1107         if (dev == NULL)
1108                 return;
1109         priv = NULL;
1110         error = 0;
1111         parent = ifp->if_softc;
1112         /* We only support 15 bits of pkey. */
1113         if (vtag & 0x8000)
1114                 return;
1115         pkey = vtag | 0x8000;   /* Set full membership bit. */
1116         if (pkey == parent->pkey)
1117                 return;
1118         /* Check for dups */
1119         mutex_lock(&parent->vlan_mutex);
1120         list_for_each_entry(priv, &parent->child_intfs, list) {
1121                 if (priv->pkey == pkey) {
1122                         priv = NULL;
1123                         error = EBUSY;
1124                         goto out;
1125                 }
1126         }
1127         priv = ipoib_priv_alloc();
1128         priv->dev = dev;
1129         priv->max_ib_mtu = parent->max_ib_mtu;
1130         priv->mcast_mtu = priv->admin_mtu = parent->dev->if_mtu;
1131         set_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags);
1132         error = ipoib_set_dev_features(priv, parent->ca);
1133         if (error)
1134                 goto out;
1135         priv->pkey = pkey;
1136         priv->broadcastaddr[8] = pkey >> 8;
1137         priv->broadcastaddr[9] = pkey & 0xff;
1138         dev->if_broadcastaddr = priv->broadcastaddr;
1139         error = ipoib_dev_init(priv, parent->ca, parent->port);
1140         if (error)
1141                 goto out;
1142         priv->parent = parent->dev;
1143         list_add_tail(&priv->list, &parent->child_intfs);
1144         VLAN_SETCOOKIE(dev, priv);
1145         dev->if_start = ipoib_vlan_start;
1146         dev->if_drv_flags &= ~IFF_DRV_RUNNING;
1147         dev->if_hdrlen = IPOIB_HEADER_LEN;
1148         if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1149                 ipoib_open(priv);
1150         mutex_unlock(&parent->vlan_mutex);
1151         return;
1152 out:
1153         mutex_unlock(&parent->vlan_mutex);
1154         if (priv)
1155                 free(priv, M_TEMP);
1156         if (error)
1157                 ipoib_warn(parent,
1158                     "failed to initialize subinterface: device %s, port %d vtag 0x%X",
1159                     parent->ca->name, parent->port, vtag);
1160         return;
1161 }
1162
1163 static void
1164 ipoib_unconfig_vlan(void *arg, struct ifnet *ifp, u_int16_t vtag)
1165 {
1166         struct ipoib_dev_priv *parent;
1167         struct ipoib_dev_priv *priv;
1168         struct ifnet *dev;
1169         uint16_t pkey;
1170
1171         if (ifp->if_type != IFT_INFINIBAND)
1172                 return;
1173
1174         dev = VLAN_DEVAT(ifp, vtag);
1175         if (dev)
1176                 VLAN_SETCOOKIE(dev, NULL);
1177         pkey = vtag | 0x8000;
1178         parent = ifp->if_softc;
1179         mutex_lock(&parent->vlan_mutex);
1180         list_for_each_entry(priv, &parent->child_intfs, list) {
1181                 if (priv->pkey == pkey) {
1182                         ipoib_dev_cleanup(priv);
1183                         list_del(&priv->list);
1184                         break;
1185                 }
1186         }
1187         mutex_unlock(&parent->vlan_mutex);
1188 }
1189
1190 eventhandler_tag ipoib_vlan_attach;
1191 eventhandler_tag ipoib_vlan_detach;
1192
1193 static int __init
1194 ipoib_init_module(void)
1195 {
1196         int ret;
1197
1198         ipoib_recvq_size = roundup_pow_of_two(ipoib_recvq_size);
1199         ipoib_recvq_size = min(ipoib_recvq_size, IPOIB_MAX_QUEUE_SIZE);
1200         ipoib_recvq_size = max(ipoib_recvq_size, IPOIB_MIN_QUEUE_SIZE);
1201
1202         ipoib_sendq_size = roundup_pow_of_two(ipoib_sendq_size);
1203         ipoib_sendq_size = min(ipoib_sendq_size, IPOIB_MAX_QUEUE_SIZE);
1204         ipoib_sendq_size = max(ipoib_sendq_size, max(2 * MAX_SEND_CQE,
1205                                                      IPOIB_MIN_QUEUE_SIZE));
1206 #ifdef CONFIG_INFINIBAND_IPOIB_CM
1207         ipoib_max_conn_qp = min(ipoib_max_conn_qp, IPOIB_CM_MAX_CONN_QP);
1208 #endif
1209
1210         ipoib_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
1211                 ipoib_config_vlan, NULL, EVENTHANDLER_PRI_FIRST);
1212         ipoib_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
1213                 ipoib_unconfig_vlan, NULL, EVENTHANDLER_PRI_FIRST);
1214
1215         /*
1216          * We create our own workqueue mainly because we want to be
1217          * able to flush it when devices are being removed.  We can't
1218          * use schedule_work()/flush_scheduled_work() because both
1219          * unregister_netdev() and linkwatch_event take the rtnl lock,
1220          * so flush_scheduled_work() can deadlock during device
1221          * removal.
1222          */
1223         ipoib_workqueue = create_singlethread_workqueue("ipoib");
1224         if (!ipoib_workqueue) {
1225                 ret = -ENOMEM;
1226                 goto err_fs;
1227         }
1228
1229         ib_sa_register_client(&ipoib_sa_client);
1230
1231         ret = ib_register_client(&ipoib_client);
1232         if (ret)
1233                 goto err_sa;
1234
1235         return 0;
1236
1237 err_sa:
1238         ib_sa_unregister_client(&ipoib_sa_client);
1239         destroy_workqueue(ipoib_workqueue);
1240
1241 err_fs:
1242         return ret;
1243 }
1244
1245 static void __exit
1246 ipoib_cleanup_module(void)
1247 {
1248
1249         EVENTHANDLER_DEREGISTER(vlan_config, ipoib_vlan_attach);
1250         EVENTHANDLER_DEREGISTER(vlan_unconfig, ipoib_vlan_detach);
1251         ib_unregister_client(&ipoib_client);
1252         ib_sa_unregister_client(&ipoib_sa_client);
1253         destroy_workqueue(ipoib_workqueue);
1254 }
1255
1256 /*
1257  * Infiniband output routine.
1258  */
1259 static int
1260 ipoib_output(struct ifnet *ifp, struct mbuf *m,
1261         struct sockaddr *dst, struct route *ro)
1262 {
1263         u_char edst[INFINIBAND_ALEN];
1264         struct llentry *lle = NULL;
1265         struct rtentry *rt0 = NULL;
1266         struct ipoib_header *eh;
1267         int error = 0;
1268         short type;
1269
1270         if (ro != NULL) {
1271                 if (!(m->m_flags & (M_BCAST | M_MCAST)))
1272                         lle = ro->ro_lle;
1273                 rt0 = ro->ro_rt;
1274         }
1275 #ifdef MAC
1276         error = mac_ifnet_check_transmit(ifp, m);
1277         if (error)
1278                 goto bad;
1279 #endif
1280
1281         M_PROFILE(m);
1282         if (ifp->if_flags & IFF_MONITOR) {
1283                 error = ENETDOWN;
1284                 goto bad;
1285         }
1286         if (!((ifp->if_flags & IFF_UP) &&
1287             (ifp->if_drv_flags & IFF_DRV_RUNNING))) {
1288                 error = ENETDOWN;
1289                 goto bad;
1290         }
1291
1292         switch (dst->sa_family) {
1293 #ifdef INET
1294         case AF_INET:
1295                 if (lle != NULL && (lle->la_flags & LLE_VALID))
1296                         memcpy(edst, &lle->ll_addr.mac8, sizeof(edst));
1297                 else if (m->m_flags & M_MCAST)
1298                         ip_ib_mc_map(((struct sockaddr_in *)dst)->sin_addr.s_addr, ifp->if_broadcastaddr, edst);
1299                 else
1300                         error = arpresolve(ifp, rt0, m, dst, edst, &lle);
1301                 if (error)
1302                         return (error == EWOULDBLOCK ? 0 : error);
1303                 type = htons(ETHERTYPE_IP);
1304                 break;
1305         case AF_ARP:
1306         {
1307                 struct arphdr *ah;
1308                 ah = mtod(m, struct arphdr *);
1309                 ah->ar_hrd = htons(ARPHRD_INFINIBAND);
1310
1311                 switch(ntohs(ah->ar_op)) {
1312                 case ARPOP_REVREQUEST:
1313                 case ARPOP_REVREPLY:
1314                         type = htons(ETHERTYPE_REVARP);
1315                         break;
1316                 case ARPOP_REQUEST:
1317                 case ARPOP_REPLY:
1318                 default:
1319                         type = htons(ETHERTYPE_ARP);
1320                         break;
1321                 }
1322
1323                 if (m->m_flags & M_BCAST)
1324                         bcopy(ifp->if_broadcastaddr, edst, INFINIBAND_ALEN);
1325                 else
1326                         bcopy(ar_tha(ah), edst, INFINIBAND_ALEN);
1327
1328         }
1329         break;
1330 #endif
1331 #ifdef INET6
1332         case AF_INET6:
1333                 if (lle != NULL && (lle->la_flags & LLE_VALID))
1334                         memcpy(edst, &lle->ll_addr.mac8, sizeof(edst));
1335                 else if (m->m_flags & M_MCAST)
1336                         ipv6_ib_mc_map(&((struct sockaddr_in6 *)dst)->sin6_addr, ifp->if_broadcastaddr, edst);
1337                 else
1338                         error = nd6_storelladdr(ifp, m, dst, (u_char *)edst, &lle);
1339                 if (error)
1340                         return error;
1341                 type = htons(ETHERTYPE_IPV6);
1342                 break;
1343 #endif
1344
1345         default:
1346                 if_printf(ifp, "can't handle af%d\n", dst->sa_family);
1347                 error = EAFNOSUPPORT;
1348                 goto bad;
1349         }
1350
1351         /*
1352          * Add local net header.  If no space in first mbuf,
1353          * allocate another.
1354          */
1355         M_PREPEND(m, IPOIB_HEADER_LEN, M_DONTWAIT);
1356         if (m == NULL) {
1357                 error = ENOBUFS;
1358                 goto bad;
1359         }
1360         eh = mtod(m, struct ipoib_header *);
1361         (void)memcpy(&eh->proto, &type, sizeof(eh->proto));
1362         (void)memcpy(&eh->hwaddr, edst, sizeof (edst));
1363
1364         /*
1365          * Queue message on interface, update output statistics if
1366          * successful, and start output if interface not yet active.
1367          */
1368         return ((ifp->if_transmit)(ifp, m));
1369 bad:
1370         if (m != NULL)
1371                 m_freem(m);
1372         return (error);
1373 }
1374
1375 /*
1376  * Upper layer processing for a received Infiniband packet.
1377  */
1378 void
1379 ipoib_demux(struct ifnet *ifp, struct mbuf *m, u_short proto)
1380 {
1381         int isr;
1382
1383 #ifdef MAC
1384         /*
1385          * Tag the mbuf with an appropriate MAC label before any other
1386          * consumers can get to it.
1387          */
1388         mac_ifnet_create_mbuf(ifp, m);
1389 #endif
1390         /* Allow monitor mode to claim this frame, after stats are updated. */
1391         if (ifp->if_flags & IFF_MONITOR) {
1392                 if_printf(ifp, "discard frame at IFF_MONITOR\n");
1393                 m_freem(m);
1394                 return;
1395         }
1396         /*
1397          * Dispatch frame to upper layer.
1398          */
1399         switch (proto) {
1400 #ifdef INET
1401         case ETHERTYPE_IP:
1402                 isr = NETISR_IP;
1403                 break;
1404
1405         case ETHERTYPE_ARP:
1406                 if (ifp->if_flags & IFF_NOARP) {
1407                         /* Discard packet if ARP is disabled on interface */
1408                         m_freem(m);
1409                         return;
1410                 }
1411                 isr = NETISR_ARP;
1412                 break;
1413 #endif
1414 #ifdef INET6
1415         case ETHERTYPE_IPV6:
1416                 isr = NETISR_IPV6;
1417                 break;
1418 #endif
1419         default:
1420                 goto discard;
1421         }
1422         netisr_dispatch(isr, m);
1423         return;
1424
1425 discard:
1426         m_freem(m);
1427 }
1428
1429 /*
1430  * Process a received Infiniband packet.
1431  */
1432 static void
1433 ipoib_input(struct ifnet *ifp, struct mbuf *m)
1434 {
1435         struct ipoib_header *eh;
1436
1437         if ((ifp->if_flags & IFF_UP) == 0) {
1438                 m_freem(m);
1439                 return;
1440         }
1441         CURVNET_SET_QUIET(ifp->if_vnet);
1442
1443         /* Let BPF have it before we strip the header. */
1444         IPOIB_MTAP(ifp, m);
1445         eh = mtod(m, struct ipoib_header *);
1446         /*
1447          * Reset layer specific mbuf flags to avoid confusing upper layers.
1448          * Strip off Infiniband header.
1449          */
1450         m->m_flags &= ~M_VLANTAG;
1451         m->m_flags &= ~(M_PROTOFLAGS);
1452         m_adj(m, IPOIB_HEADER_LEN);
1453
1454         if (IPOIB_IS_MULTICAST(eh->hwaddr)) {
1455                 if (memcmp(eh->hwaddr, ifp->if_broadcastaddr,
1456                     ifp->if_addrlen) == 0)
1457                         m->m_flags |= M_BCAST;
1458                 else
1459                         m->m_flags |= M_MCAST;
1460                 ifp->if_imcasts++;
1461         }
1462
1463         ipoib_demux(ifp, m, ntohs(eh->proto));
1464         CURVNET_RESTORE();
1465 }
1466
1467 static int
1468 ipoib_resolvemulti(struct ifnet *ifp, struct sockaddr **llsa,
1469         struct sockaddr *sa)
1470 {
1471         struct sockaddr_dl *sdl;
1472 #ifdef INET
1473         struct sockaddr_in *sin;
1474 #endif
1475 #ifdef INET6
1476         struct sockaddr_in6 *sin6;
1477 #endif
1478         u_char *e_addr;
1479
1480         switch(sa->sa_family) {
1481         case AF_LINK:
1482                 /*
1483                  * No mapping needed. Just check that it's a valid MC address.
1484                  */
1485                 sdl = (struct sockaddr_dl *)sa;
1486                 e_addr = LLADDR(sdl);
1487                 if (!IPOIB_IS_MULTICAST(e_addr))
1488                         return EADDRNOTAVAIL;
1489                 *llsa = 0;
1490                 return 0;
1491
1492 #ifdef INET
1493         case AF_INET:
1494                 sin = (struct sockaddr_in *)sa;
1495                 if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
1496                         return EADDRNOTAVAIL;
1497                 sdl = malloc(sizeof *sdl, M_IFMADDR,
1498                        M_NOWAIT|M_ZERO);
1499                 if (sdl == NULL)
1500                         return ENOMEM;
1501                 sdl->sdl_len = sizeof *sdl;
1502                 sdl->sdl_family = AF_LINK;
1503                 sdl->sdl_index = ifp->if_index;
1504                 sdl->sdl_type = IFT_INFINIBAND;
1505                 sdl->sdl_alen = INFINIBAND_ALEN;
1506                 e_addr = LLADDR(sdl);
1507                 ip_ib_mc_map(sin->sin_addr.s_addr, ifp->if_broadcastaddr,
1508                     e_addr);
1509                 *llsa = (struct sockaddr *)sdl;
1510                 return 0;
1511 #endif
1512 #ifdef INET6
1513         case AF_INET6:
1514                 sin6 = (struct sockaddr_in6 *)sa;
1515                 /*
1516                  * An IP6 address of 0 means listen to all
1517                  * of the multicast address used for IP6.
1518                  * This has no meaning in ipoib.
1519                  */
1520                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
1521                         return EADDRNOTAVAIL;
1522                 if (!IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
1523                         return EADDRNOTAVAIL;
1524                 sdl = malloc(sizeof *sdl, M_IFMADDR,
1525                        M_NOWAIT|M_ZERO);
1526                 if (sdl == NULL)
1527                         return (ENOMEM);
1528                 sdl->sdl_len = sizeof *sdl;
1529                 sdl->sdl_family = AF_LINK;
1530                 sdl->sdl_index = ifp->if_index;
1531                 sdl->sdl_type = IFT_INFINIBAND;
1532                 sdl->sdl_alen = INFINIBAND_ALEN;
1533                 e_addr = LLADDR(sdl);
1534                 ipv6_ib_mc_map(&sin6->sin6_addr, ifp->if_broadcastaddr, e_addr);
1535                 *llsa = (struct sockaddr *)sdl;
1536                 return 0;
1537 #endif
1538
1539         default:
1540                 return EAFNOSUPPORT;
1541         }
1542 }
1543
1544 module_init(ipoib_init_module);
1545 module_exit(ipoib_cleanup_module);
1546
1547 #undef MODULE_VERSION
1548 #include <sys/module.h>
1549 static int
1550 ipoib_evhand(module_t mod, int event, void *arg)
1551 {
1552         return (0);
1553 }
1554
1555 static moduledata_t ipoib_mod = {
1556         .name = "ipoib",
1557         .evhand = ipoib_evhand,
1558 };
1559
1560 DECLARE_MODULE(ipoib, ipoib_mod, SI_SUB_SMP, SI_ORDER_ANY);
1561 MODULE_DEPEND(ipoib, ibcore, 1, 1, 1);
1562