drivers/infiniband/sw/rxe/rxe_net.c

   1 /*
   2  * Copyright (c) 2016 Mellanox Technologies Ltd. All rights reserved.
   3  * Copyright (c) 2015 System Fabric Works, Inc. All rights reserved.
   4  *
   5  * This software is available to you under a choice of one of two
   6  * licenses.  You may choose to be licensed under the terms of the GNU
   7  * General Public License (GPL) Version 2, available from the file
   8  * COPYING in the main directory of this source tree, or the
   9  * OpenIB.org BSD license below:
  10  *
  11  *     Redistribution and use in source and binary forms, with or
  12  *     without modification, are permitted provided that the following
  13  *     conditions are met:
  14  *
  15  *      - Redistributions of source code must retain the above
  16  *        copyright notice, this list of conditions and the following
  17  *        disclaimer.
  18  *
  19  *      - Redistributions in binary form must reproduce the above
  20  *        copyright notice, this list of conditions and the following
  21  *        disclaimer in the documentation and/or other materials
  22  *        provided with the distribution.
  23  *
  24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  31  * SOFTWARE.
  32  */
  33
  34 #include <linux/skbuff.h>
  35 #include <linux/if_arp.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/if.h>
  38 #include <linux/if_vlan.h>
  39 #include <net/udp_tunnel.h>
  40 #include <net/sch_generic.h>
  41 #include <linux/netfilter.h>
  42 #include <rdma/ib_addr.h>
  43
  44 #include "rxe.h"
  45 #include "rxe_net.h"
  46 #include "rxe_loc.h"
  47
  48 static LIST_HEAD(rxe_dev_list);
  49 static DEFINE_SPINLOCK(dev_list_lock); /* spinlock for device list */
  50
  51 struct rxe_dev *net_to_rxe(struct net_device *ndev)
  52 {
  53         struct rxe_dev *rxe;
  54         struct rxe_dev *found = NULL;
  55
  56         spin_lock_bh(&dev_list_lock);
  57         list_for_each_entry(rxe, &rxe_dev_list, list) {
  58                 if (rxe->ndev == ndev) {
  59                         found = rxe;
  60                         break;
  61                 }
  62         }
  63         spin_unlock_bh(&dev_list_lock);
  64
  65         return found;
  66 }
  67
  68 struct rxe_dev *get_rxe_by_name(const char *name)
  69 {
  70         struct rxe_dev *rxe;
  71         struct rxe_dev *found = NULL;
  72
  73         spin_lock_bh(&dev_list_lock);
  74         list_for_each_entry(rxe, &rxe_dev_list, list) {
  75                 if (!strcmp(name, rxe->ib_dev.name)) {
  76                         found = rxe;
  77                         break;
  78                 }
  79         }
  80         spin_unlock_bh(&dev_list_lock);
  81         return found;
  82 }
  83
  84
  85 struct rxe_recv_sockets recv_sockets;
  86
  87 static __be64 rxe_mac_to_eui64(struct net_device *ndev)
  88 {
  89         unsigned char *mac_addr = ndev->dev_addr;
  90         __be64 eui64;
  91         unsigned char *dst = (unsigned char *)&eui64;
  92
  93         dst[0] = mac_addr[0] ^ 2;
  94         dst[1] = mac_addr[1];
  95         dst[2] = mac_addr[2];
  96         dst[3] = 0xff;
  97         dst[4] = 0xfe;
  98         dst[5] = mac_addr[3];
  99         dst[6] = mac_addr[4];
 100         dst[7] = mac_addr[5];
 101
 102         return eui64;
 103 }
 104
 105 __be64 rxe_node_guid(struct rxe_dev *rxe)
 106 {
 107         return rxe_mac_to_eui64(rxe->ndev);
 108 }
 109
 110 __be64 rxe_port_guid(struct rxe_dev *rxe)
 111 {
 112         return rxe_mac_to_eui64(rxe->ndev);
 113 }
 114
 115 struct device *rxe_dma_device(struct rxe_dev *rxe)
 116 {
 117         struct net_device *ndev;
 118
 119         ndev = rxe->ndev;
 120
 121         if (is_vlan_dev(ndev))
 122                 ndev = vlan_dev_real_dev(ndev);
 123
 124         return ndev->dev.parent;
 125 }
 126
 127 int rxe_mcast_add(struct rxe_dev *rxe, union ib_gid *mgid)
 128 {
 129         int err;
 130         unsigned char ll_addr[ETH_ALEN];
 131
 132         ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr);
 133         err = dev_mc_add(rxe->ndev, ll_addr);
 134
 135         return err;
 136 }
 137
 138 int rxe_mcast_delete(struct rxe_dev *rxe, union ib_gid *mgid)
 139 {
 140         int err;
 141         unsigned char ll_addr[ETH_ALEN];
 142
 143         ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr);
 144         err = dev_mc_del(rxe->ndev, ll_addr);
 145
 146         return err;
 147 }
 148
 149 static struct dst_entry *rxe_find_route4(struct net_device *ndev,
 150                                   struct in_addr *saddr,
 151                                   struct in_addr *daddr)
 152 {
 153         struct rtable *rt;
 154         struct flowi4 fl = { { 0 } };
 155
 156         memset(&fl, 0, sizeof(fl));
 157         fl.flowi4_oif = ndev->ifindex;
 158         memcpy(&fl.saddr, saddr, sizeof(*saddr));
 159         memcpy(&fl.daddr, daddr, sizeof(*daddr));
 160         fl.flowi4_proto = IPPROTO_UDP;
 161
 162         rt = ip_route_output_key(&init_net, &fl);
 163         if (IS_ERR(rt)) {
 164                 pr_err_ratelimited("no route to %pI4\n", &daddr->s_addr);
 165                 return NULL;
 166         }
 167
 168         return &rt->dst;
 169 }
 170
 171 #if IS_ENABLED(CONFIG_IPV6)
 172 static struct dst_entry *rxe_find_route6(struct net_device *ndev,
 173                                          struct in6_addr *saddr,
 174                                          struct in6_addr *daddr)
 175 {
 176         struct dst_entry *ndst;
 177         struct flowi6 fl6 = { { 0 } };
 178
 179         memset(&fl6, 0, sizeof(fl6));
 180         fl6.flowi6_oif = ndev->ifindex;
 181         memcpy(&fl6.saddr, saddr, sizeof(*saddr));
 182         memcpy(&fl6.daddr, daddr, sizeof(*daddr));
 183         fl6.flowi6_proto = IPPROTO_UDP;
 184
 185         if (unlikely(ipv6_stub->ipv6_dst_lookup(sock_net(recv_sockets.sk6->sk),
 186                                                 recv_sockets.sk6->sk, &ndst, &fl6))) {
 187                 pr_err_ratelimited("no route to %pI6\n", daddr);
 188                 goto put;
 189         }
 190
 191         if (unlikely(ndst->error)) {
 192                 pr_err("no route to %pI6\n", daddr);
 193                 goto put;
 194         }
 195
 196         return ndst;
 197 put:
 198         dst_release(ndst);
 199         return NULL;
 200 }
 201
 202 #else
 203
 204 static struct dst_entry *rxe_find_route6(struct net_device *ndev,
 205                                          struct in6_addr *saddr,
 206                                          struct in6_addr *daddr)
 207 {
 208         return NULL;
 209 }
 210
 211 #endif
 212
 213 static int rxe_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 214 {
 215         struct udphdr *udph;
 216         struct net_device *ndev = skb->dev;
 217         struct rxe_dev *rxe = net_to_rxe(ndev);
 218         struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
 219
 220         if (!rxe)
 221                 goto drop;
 222
 223         if (skb_linearize(skb)) {
 224                 pr_err("skb_linearize failed\n");
 225                 goto drop;
 226         }
 227
 228         udph = udp_hdr(skb);
 229         pkt->rxe = rxe;
 230         pkt->port_num = 1;
 231         pkt->hdr = (u8 *)(udph + 1);
 232         pkt->mask = RXE_GRH_MASK;
 233         pkt->paylen = be16_to_cpu(udph->len) - sizeof(*udph);
 234
 235         return rxe_rcv(skb);
 236 drop:
 237         kfree_skb(skb);
 238         return 0;
 239 }
 240
 241 static struct socket *rxe_setup_udp_tunnel(struct net *net, __be16 port,
 242                                            bool ipv6)
 243 {
 244         int err;
 245         struct socket *sock;
 246         struct udp_port_cfg udp_cfg = { };
 247         struct udp_tunnel_sock_cfg tnl_cfg = { };
 248
 249         if (ipv6) {
 250                 udp_cfg.family = AF_INET6;
 251                 udp_cfg.ipv6_v6only = 1;
 252         } else {
 253                 udp_cfg.family = AF_INET;
 254         }
 255
 256         udp_cfg.local_udp_port = port;
 257
 258         /* Create UDP socket */
 259         err = udp_sock_create(net, &udp_cfg, &sock);
 260         if (err < 0) {
 261                 pr_err("failed to create udp socket. err = %d\n", err);
 262                 return ERR_PTR(err);
 263         }
 264
 265         tnl_cfg.encap_type = 1;
 266         tnl_cfg.encap_rcv = rxe_udp_encap_recv;
 267
 268         /* Setup UDP tunnel */
 269         setup_udp_tunnel_sock(net, sock, &tnl_cfg);
 270
 271         return sock;
 272 }
 273
 274 void rxe_release_udp_tunnel(struct socket *sk)
 275 {
 276         if (sk)
 277                 udp_tunnel_sock_release(sk);
 278 }
 279
 280 static void prepare_udp_hdr(struct sk_buff *skb, __be16 src_port,
 281                             __be16 dst_port)
 282 {
 283         struct udphdr *udph;
 284
 285         __skb_push(skb, sizeof(*udph));
 286         skb_reset_transport_header(skb);
 287         udph = udp_hdr(skb);
 288
 289         udph->dest = dst_port;
 290         udph->source = src_port;
 291         udph->len = htons(skb->len);
 292         udph->check = 0;
 293 }
 294
 295 static void prepare_ipv4_hdr(struct dst_entry *dst, struct sk_buff *skb,
 296                              __be32 saddr, __be32 daddr, __u8 proto,
 297                              __u8 tos, __u8 ttl, __be16 df, bool xnet)
 298 {
 299         struct iphdr *iph;
 300
 301         skb_scrub_packet(skb, xnet);
 302
 303         skb_clear_hash(skb);
 304         skb_dst_set(skb, dst);
 305         memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
 306
 307         skb_push(skb, sizeof(struct iphdr));
 308         skb_reset_network_header(skb);
 309
 310         iph = ip_hdr(skb);
 311
 312         iph->version    =       IPVERSION;
 313         iph->ihl        =       sizeof(struct iphdr) >> 2;
 314         iph->frag_off   =       df;
 315         iph->protocol   =       proto;
 316         iph->tos        =       tos;
 317         iph->daddr      =       daddr;
 318         iph->saddr      =       saddr;
 319         iph->ttl        =       ttl;
 320         __ip_select_ident(dev_net(dst->dev), iph,
 321                           skb_shinfo(skb)->gso_segs ?: 1);
 322         iph->tot_len = htons(skb->len);
 323         ip_send_check(iph);
 324 }
 325
 326 static void prepare_ipv6_hdr(struct dst_entry *dst, struct sk_buff *skb,
 327                              struct in6_addr *saddr, struct in6_addr *daddr,
 328                              __u8 proto, __u8 prio, __u8 ttl)
 329 {
 330         struct ipv6hdr *ip6h;
 331
 332         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 333         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED
 334                             | IPSKB_REROUTED);
 335         skb_dst_set(skb, dst);
 336
 337         __skb_push(skb, sizeof(*ip6h));
 338         skb_reset_network_header(skb);
 339         ip6h              = ipv6_hdr(skb);
 340         ip6_flow_hdr(ip6h, prio, htonl(0));
 341         ip6h->payload_len = htons(skb->len);
 342         ip6h->nexthdr     = proto;
 343         ip6h->hop_limit   = ttl;
 344         ip6h->daddr       = *daddr;
 345         ip6h->saddr       = *saddr;
 346         ip6h->payload_len = htons(skb->len - sizeof(*ip6h));
 347 }
 348
 349 static int prepare4(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
 350                     struct sk_buff *skb, struct rxe_av *av)
 351 {
 352         struct dst_entry *dst;
 353         bool xnet = false;
 354         __be16 df = htons(IP_DF);
 355         struct in_addr *saddr = &av->sgid_addr._sockaddr_in.sin_addr;
 356         struct in_addr *daddr = &av->dgid_addr._sockaddr_in.sin_addr;
 357
 358         dst = rxe_find_route4(rxe->ndev, saddr, daddr);
 359         if (!dst) {
 360                 pr_err("Host not reachable\n");
 361                 return -EHOSTUNREACH;
 362         }
 363
 364         if (!memcmp(saddr, daddr, sizeof(*daddr)))
 365                 pkt->mask |= RXE_LOOPBACK_MASK;
 366
 367         prepare_udp_hdr(skb, htons(RXE_ROCE_V2_SPORT),
 368                         htons(ROCE_V2_UDP_DPORT));
 369
 370         prepare_ipv4_hdr(dst, skb, saddr->s_addr, daddr->s_addr, IPPROTO_UDP,
 371                          av->grh.traffic_class, av->grh.hop_limit, df, xnet);
 372         return 0;
 373 }
 374
 375 static int prepare6(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
 376                     struct sk_buff *skb, struct rxe_av *av)
 377 {
 378         struct dst_entry *dst;
 379         struct in6_addr *saddr = &av->sgid_addr._sockaddr_in6.sin6_addr;
 380         struct in6_addr *daddr = &av->dgid_addr._sockaddr_in6.sin6_addr;
 381
 382         dst = rxe_find_route6(rxe->ndev, saddr, daddr);
 383         if (!dst) {
 384                 pr_err("Host not reachable\n");
 385                 return -EHOSTUNREACH;
 386         }
 387
 388         if (!memcmp(saddr, daddr, sizeof(*daddr)))
 389                 pkt->mask |= RXE_LOOPBACK_MASK;
 390
 391         prepare_udp_hdr(skb, htons(RXE_ROCE_V2_SPORT),
 392                         htons(ROCE_V2_UDP_DPORT));
 393
 394         prepare_ipv6_hdr(dst, skb, saddr, daddr, IPPROTO_UDP,
 395                          av->grh.traffic_class,
 396                          av->grh.hop_limit);
 397         return 0;
 398 }
 399
 400 int rxe_prepare(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
 401                 struct sk_buff *skb, u32 *crc)
 402 {
 403         int err = 0;
 404         struct rxe_av *av = rxe_get_av(pkt);
 405
 406         if (av->network_type == RDMA_NETWORK_IPV4)
 407                 err = prepare4(rxe, pkt, skb, av);
 408         else if (av->network_type == RDMA_NETWORK_IPV6)
 409                 err = prepare6(rxe, pkt, skb, av);
 410
 411         *crc = rxe_icrc_hdr(pkt, skb);
 412
 413         return err;
 414 }
 415
 416 static void rxe_skb_tx_dtor(struct sk_buff *skb)
 417 {
 418         struct sock *sk = skb->sk;
 419         struct rxe_qp *qp = sk->sk_user_data;
 420         int skb_out = atomic_dec_return(&qp->skb_out);
 421
 422         if (unlikely(qp->need_req_skb &&
 423                      skb_out < RXE_INFLIGHT_SKBS_PER_QP_LOW))
 424                 rxe_run_task(&qp->req.task, 1);
 425 }
 426
 427 int rxe_send(struct rxe_dev *rxe, struct rxe_pkt_info *pkt, struct sk_buff *skb)
 428 {
 429         struct sk_buff *nskb;
 430         struct rxe_av *av;
 431         int err;
 432
 433         av = rxe_get_av(pkt);
 434
 435         nskb = skb_clone(skb, GFP_ATOMIC);
 436         if (!nskb)
 437                 return -ENOMEM;
 438
 439         nskb->destructor = rxe_skb_tx_dtor;
 440         nskb->sk = pkt->qp->sk->sk;
 441
 442         if (av->network_type == RDMA_NETWORK_IPV4) {
 443                 err = ip_local_out(dev_net(skb_dst(skb)->dev), nskb->sk, nskb);
 444         } else if (av->network_type == RDMA_NETWORK_IPV6) {
 445                 err = ip6_local_out(dev_net(skb_dst(skb)->dev), nskb->sk, nskb);
 446         } else {
 447                 pr_err("Unknown layer 3 protocol: %d\n", av->network_type);
 448                 kfree_skb(nskb);
 449                 return -EINVAL;
 450         }
 451
 452         if (unlikely(net_xmit_eval(err))) {
 453                 pr_debug("error sending packet: %d\n", err);
 454                 return -EAGAIN;
 455         }
 456
 457         atomic_inc(&pkt->qp->skb_out);
 458         kfree_skb(skb);
 459
 460         return 0;
 461 }
 462
 463 int rxe_loopback(struct sk_buff *skb)
 464 {
 465         return rxe_rcv(skb);
 466 }
 467
 468 static inline int addr_same(struct rxe_dev *rxe, struct rxe_av *av)
 469 {
 470         return rxe->port.port_guid == av->grh.dgid.global.interface_id;
 471 }
 472
 473 struct sk_buff *rxe_init_packet(struct rxe_dev *rxe, struct rxe_av *av,
 474                                 int paylen, struct rxe_pkt_info *pkt)
 475 {
 476         unsigned int hdr_len;
 477         struct sk_buff *skb;
 478
 479         if (av->network_type == RDMA_NETWORK_IPV4)
 480                 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
 481                         sizeof(struct iphdr);
 482         else
 483                 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
 484                         sizeof(struct ipv6hdr);
 485
 486         skb = alloc_skb(paylen + hdr_len + LL_RESERVED_SPACE(rxe->ndev),
 487                         GFP_ATOMIC);
 488         if (unlikely(!skb))
 489                 return NULL;
 490
 491         skb_reserve(skb, hdr_len + LL_RESERVED_SPACE(rxe->ndev));
 492
 493         skb->dev        = rxe->ndev;
 494         if (av->network_type == RDMA_NETWORK_IPV4)
 495                 skb->protocol = htons(ETH_P_IP);
 496         else
 497                 skb->protocol = htons(ETH_P_IPV6);
 498
 499         pkt->rxe        = rxe;
 500         pkt->port_num   = 1;
 501         pkt->hdr        = skb_put(skb, paylen);
 502         pkt->mask       |= RXE_GRH_MASK;
 503
 504         memset(pkt->hdr, 0, paylen);
 505
 506         return skb;
 507 }
 508
 509 /*
 510  * this is required by rxe_cfg to match rxe devices in
 511  * /sys/class/infiniband up with their underlying ethernet devices
 512  */
 513 const char *rxe_parent_name(struct rxe_dev *rxe, unsigned int port_num)
 514 {
 515         return rxe->ndev->name;
 516 }
 517
 518 enum rdma_link_layer rxe_link_layer(struct rxe_dev *rxe, unsigned int port_num)
 519 {
 520         return IB_LINK_LAYER_ETHERNET;
 521 }
 522
 523 struct rxe_dev *rxe_net_add(struct net_device *ndev)
 524 {
 525         int err;
 526         struct rxe_dev *rxe = NULL;
 527
 528         rxe = (struct rxe_dev *)ib_alloc_device(sizeof(*rxe));
 529         if (!rxe)
 530                 return NULL;
 531
 532         rxe->ndev = ndev;
 533
 534         err = rxe_add(rxe, ndev->mtu);
 535         if (err) {
 536                 ib_dealloc_device(&rxe->ib_dev);
 537                 return NULL;
 538         }
 539
 540         spin_lock_bh(&dev_list_lock);
 541         list_add_tail(&rxe->list, &rxe_dev_list);
 542         spin_unlock_bh(&dev_list_lock);
 543         return rxe;
 544 }
 545
 546 void rxe_remove_all(void)
 547 {
 548         spin_lock_bh(&dev_list_lock);
 549         while (!list_empty(&rxe_dev_list)) {
 550                 struct rxe_dev *rxe =
 551                         list_first_entry(&rxe_dev_list, struct rxe_dev, list);
 552
 553                 list_del(&rxe->list);
 554                 spin_unlock_bh(&dev_list_lock);
 555                 rxe_remove(rxe);
 556                 spin_lock_bh(&dev_list_lock);
 557         }
 558         spin_unlock_bh(&dev_list_lock);
 559 }
 560 EXPORT_SYMBOL(rxe_remove_all);
 561
 562 static void rxe_port_event(struct rxe_dev *rxe,
 563                            enum ib_event_type event)
 564 {
 565         struct ib_event ev;
 566
 567         ev.device = &rxe->ib_dev;
 568         ev.element.port_num = 1;
 569         ev.event = event;
 570
 571         ib_dispatch_event(&ev);
 572 }
 573
 574 /* Caller must hold net_info_lock */
 575 void rxe_port_up(struct rxe_dev *rxe)
 576 {
 577         struct rxe_port *port;
 578
 579         port = &rxe->port;
 580         port->attr.state = IB_PORT_ACTIVE;
 581         port->attr.phys_state = IB_PHYS_STATE_LINK_UP;
 582
 583         rxe_port_event(rxe, IB_EVENT_PORT_ACTIVE);
 584         pr_info("set %s active\n", rxe->ib_dev.name);
 585 }
 586
 587 /* Caller must hold net_info_lock */
 588 void rxe_port_down(struct rxe_dev *rxe)
 589 {
 590         struct rxe_port *port;
 591
 592         port = &rxe->port;
 593         port->attr.state = IB_PORT_DOWN;
 594         port->attr.phys_state = IB_PHYS_STATE_LINK_DOWN;
 595
 596         rxe_port_event(rxe, IB_EVENT_PORT_ERR);
 597         pr_info("set %s down\n", rxe->ib_dev.name);
 598 }
 599
 600 static int rxe_notify(struct notifier_block *not_blk,
 601                       unsigned long event,
 602                       void *arg)
 603 {
 604         struct net_device *ndev = netdev_notifier_info_to_dev(arg);
 605         struct rxe_dev *rxe = net_to_rxe(ndev);
 606
 607         if (!rxe)
 608                 goto out;
 609
 610         switch (event) {
 611         case NETDEV_UNREGISTER:
 612                 list_del(&rxe->list);
 613                 rxe_remove(rxe);
 614                 break;
 615         case NETDEV_UP:
 616                 rxe_port_up(rxe);
 617                 break;
 618         case NETDEV_DOWN:
 619                 rxe_port_down(rxe);
 620                 break;
 621         case NETDEV_CHANGEMTU:
 622                 pr_info("%s changed mtu to %d\n", ndev->name, ndev->mtu);
 623                 rxe_set_mtu(rxe, ndev->mtu);
 624                 break;
 625         case NETDEV_REBOOT:
 626         case NETDEV_CHANGE:
 627         case NETDEV_GOING_DOWN:
 628         case NETDEV_CHANGEADDR:
 629         case NETDEV_CHANGENAME:
 630         case NETDEV_FEAT_CHANGE:
 631         default:
 632                 pr_info("ignoring netdev event = %ld for %s\n",
 633                         event, ndev->name);
 634                 break;
 635         }
 636 out:
 637         return NOTIFY_OK;
 638 }
 639
 640 struct notifier_block rxe_net_notifier = {
 641         .notifier_call = rxe_notify,
 642 };
 643
 644 static int rxe_net_ipv4_init(void)
 645 {
 646         recv_sockets.sk4 = rxe_setup_udp_tunnel(&init_net,
 647                                 htons(ROCE_V2_UDP_DPORT), false);
 648         if (IS_ERR(recv_sockets.sk4)) {
 649                 recv_sockets.sk4 = NULL;
 650                 pr_err("Failed to create IPv4 UDP tunnel\n");
 651                 return -1;
 652         }
 653
 654         return 0;
 655 }
 656
 657 static int rxe_net_ipv6_init(void)
 658 {
 659 #if IS_ENABLED(CONFIG_IPV6)
 660
 661         recv_sockets.sk6 = rxe_setup_udp_tunnel(&init_net,
 662                                                 htons(ROCE_V2_UDP_DPORT), true);
 663         if (IS_ERR(recv_sockets.sk6)) {
 664                 recv_sockets.sk6 = NULL;
 665                 pr_err("Failed to create IPv6 UDP tunnel\n");
 666                 return -1;
 667         }
 668 #endif
 669         return 0;
 670 }
 671
 672 void rxe_net_exit(void)
 673 {
 674         rxe_release_udp_tunnel(recv_sockets.sk6);
 675         rxe_release_udp_tunnel(recv_sockets.sk4);
 676         unregister_netdevice_notifier(&rxe_net_notifier);
 677 }
 678
 679 int rxe_net_init(void)
 680 {
 681         int err;
 682
 683         recv_sockets.sk6 = NULL;
 684
 685         err = rxe_net_ipv4_init();
 686         if (err)
 687                 return err;
 688         err = rxe_net_ipv6_init();
 689         if (err)
 690                 goto err_out;
 691         err = register_netdevice_notifier(&rxe_net_notifier);
 692         if (err) {
 693                 pr_err("Failed to register netdev notifier\n");
 694                 goto err_out;
 695         }
 696         return 0;
 697 err_out:
 698         rxe_net_exit();
 699         return err;
 700 }