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 spinlock_t 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 static __be64 node_guid(struct rxe_dev *rxe)
 106 {
 107         return rxe_mac_to_eui64(rxe->ndev);
 108 }
 109
 110 static __be64 port_guid(struct rxe_dev *rxe)
 111 {
 112         return rxe_mac_to_eui64(rxe->ndev);
 113 }
 114
 115 static struct device *dma_device(struct rxe_dev *rxe)
 116 {
 117         struct net_device *ndev;
 118
 119         ndev = rxe->ndev;
 120
 121         if (ndev->priv_flags & IFF_802_1Q_VLAN)
 122                 ndev = vlan_dev_real_dev(ndev);
 123
 124         return ndev->dev.parent;
 125 }
 126
 127 static int 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 static int 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         memset(&udp_cfg, 0, sizeof(udp_cfg));
 250
 251         if (ipv6) {
 252                 udp_cfg.family = AF_INET6;
 253                 udp_cfg.ipv6_v6only = 1;
 254         } else {
 255                 udp_cfg.family = AF_INET;
 256         }
 257
 258         udp_cfg.local_udp_port = port;
 259
 260         /* Create UDP socket */
 261         err = udp_sock_create(net, &udp_cfg, &sock);
 262         if (err < 0) {
 263                 pr_err("failed to create udp socket. err = %d\n", err);
 264                 return ERR_PTR(err);
 265         }
 266
 267         tnl_cfg.sk_user_data = NULL;
 268         tnl_cfg.encap_type = 1;
 269         tnl_cfg.encap_rcv = rxe_udp_encap_recv;
 270         tnl_cfg.encap_destroy = NULL;
 271
 272         /* Setup UDP tunnel */
 273         setup_udp_tunnel_sock(net, sock, &tnl_cfg);
 274
 275         return sock;
 276 }
 277
 278 void rxe_release_udp_tunnel(struct socket *sk)
 279 {
 280         if (sk)
 281                 udp_tunnel_sock_release(sk);
 282 }
 283
 284 static void prepare_udp_hdr(struct sk_buff *skb, __be16 src_port,
 285                             __be16 dst_port)
 286 {
 287         struct udphdr *udph;
 288
 289         __skb_push(skb, sizeof(*udph));
 290         skb_reset_transport_header(skb);
 291         udph = udp_hdr(skb);
 292
 293         udph->dest = dst_port;
 294         udph->source = src_port;
 295         udph->len = htons(skb->len);
 296         udph->check = 0;
 297 }
 298
 299 static void prepare_ipv4_hdr(struct dst_entry *dst, struct sk_buff *skb,
 300                              __be32 saddr, __be32 daddr, __u8 proto,
 301                              __u8 tos, __u8 ttl, __be16 df, bool xnet)
 302 {
 303         struct iphdr *iph;
 304
 305         skb_scrub_packet(skb, xnet);
 306
 307         skb_clear_hash(skb);
 308         skb_dst_set(skb, dst);
 309         memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
 310
 311         skb_push(skb, sizeof(struct iphdr));
 312         skb_reset_network_header(skb);
 313
 314         iph = ip_hdr(skb);
 315
 316         iph->version    =       IPVERSION;
 317         iph->ihl        =       sizeof(struct iphdr) >> 2;
 318         iph->frag_off   =       df;
 319         iph->protocol   =       proto;
 320         iph->tos        =       tos;
 321         iph->daddr      =       daddr;
 322         iph->saddr      =       saddr;
 323         iph->ttl        =       ttl;
 324         __ip_select_ident(dev_net(dst->dev), iph,
 325                           skb_shinfo(skb)->gso_segs ?: 1);
 326         iph->tot_len = htons(skb->len);
 327         ip_send_check(iph);
 328 }
 329
 330 static void prepare_ipv6_hdr(struct dst_entry *dst, struct sk_buff *skb,
 331                              struct in6_addr *saddr, struct in6_addr *daddr,
 332                              __u8 proto, __u8 prio, __u8 ttl)
 333 {
 334         struct ipv6hdr *ip6h;
 335
 336         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
 337         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED
 338                             | IPSKB_REROUTED);
 339         skb_dst_set(skb, dst);
 340
 341         __skb_push(skb, sizeof(*ip6h));
 342         skb_reset_network_header(skb);
 343         ip6h              = ipv6_hdr(skb);
 344         ip6_flow_hdr(ip6h, prio, htonl(0));
 345         ip6h->payload_len = htons(skb->len);
 346         ip6h->nexthdr     = proto;
 347         ip6h->hop_limit   = ttl;
 348         ip6h->daddr       = *daddr;
 349         ip6h->saddr       = *saddr;
 350         ip6h->payload_len = htons(skb->len - sizeof(*ip6h));
 351 }
 352
 353 static int prepare4(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
 354                     struct sk_buff *skb, struct rxe_av *av)
 355 {
 356         struct dst_entry *dst;
 357         bool xnet = false;
 358         __be16 df = htons(IP_DF);
 359         struct in_addr *saddr = &av->sgid_addr._sockaddr_in.sin_addr;
 360         struct in_addr *daddr = &av->dgid_addr._sockaddr_in.sin_addr;
 361
 362         dst = rxe_find_route4(rxe->ndev, saddr, daddr);
 363         if (!dst) {
 364                 pr_err("Host not reachable\n");
 365                 return -EHOSTUNREACH;
 366         }
 367
 368         if (!memcmp(saddr, daddr, sizeof(*daddr)))
 369                 pkt->mask |= RXE_LOOPBACK_MASK;
 370
 371         prepare_udp_hdr(skb, htons(RXE_ROCE_V2_SPORT),
 372                         htons(ROCE_V2_UDP_DPORT));
 373
 374         prepare_ipv4_hdr(dst, skb, saddr->s_addr, daddr->s_addr, IPPROTO_UDP,
 375                          av->grh.traffic_class, av->grh.hop_limit, df, xnet);
 376         return 0;
 377 }
 378
 379 static int prepare6(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
 380                     struct sk_buff *skb, struct rxe_av *av)
 381 {
 382         struct dst_entry *dst;
 383         struct in6_addr *saddr = &av->sgid_addr._sockaddr_in6.sin6_addr;
 384         struct in6_addr *daddr = &av->dgid_addr._sockaddr_in6.sin6_addr;
 385
 386         dst = rxe_find_route6(rxe->ndev, saddr, daddr);
 387         if (!dst) {
 388                 pr_err("Host not reachable\n");
 389                 return -EHOSTUNREACH;
 390         }
 391
 392         if (!memcmp(saddr, daddr, sizeof(*daddr)))
 393                 pkt->mask |= RXE_LOOPBACK_MASK;
 394
 395         prepare_udp_hdr(skb, htons(RXE_ROCE_V2_SPORT),
 396                         htons(ROCE_V2_UDP_DPORT));
 397
 398         prepare_ipv6_hdr(dst, skb, saddr, daddr, IPPROTO_UDP,
 399                          av->grh.traffic_class,
 400                          av->grh.hop_limit);
 401         return 0;
 402 }
 403
 404 static int prepare(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
 405                    struct sk_buff *skb, u32 *crc)
 406 {
 407         int err = 0;
 408         struct rxe_av *av = rxe_get_av(pkt);
 409
 410         if (av->network_type == RDMA_NETWORK_IPV4)
 411                 err = prepare4(rxe, pkt, skb, av);
 412         else if (av->network_type == RDMA_NETWORK_IPV6)
 413                 err = prepare6(rxe, pkt, skb, av);
 414
 415         *crc = rxe_icrc_hdr(pkt, skb);
 416
 417         return err;
 418 }
 419
 420 static void rxe_skb_tx_dtor(struct sk_buff *skb)
 421 {
 422         struct sock *sk = skb->sk;
 423         struct rxe_qp *qp = sk->sk_user_data;
 424         int skb_out = atomic_dec_return(&qp->skb_out);
 425
 426         if (unlikely(qp->need_req_skb &&
 427                      skb_out < RXE_INFLIGHT_SKBS_PER_QP_LOW))
 428                 rxe_run_task(&qp->req.task, 1);
 429 }
 430
 431 static int send(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
 432                 struct sk_buff *skb)
 433 {
 434         struct sk_buff *nskb;
 435         struct rxe_av *av;
 436         int err;
 437
 438         av = rxe_get_av(pkt);
 439
 440         nskb = skb_clone(skb, GFP_ATOMIC);
 441         if (!nskb)
 442                 return -ENOMEM;
 443
 444         nskb->destructor = rxe_skb_tx_dtor;
 445         nskb->sk = pkt->qp->sk->sk;
 446
 447         if (av->network_type == RDMA_NETWORK_IPV4) {
 448                 err = ip_local_out(dev_net(skb_dst(skb)->dev), nskb->sk, nskb);
 449         } else if (av->network_type == RDMA_NETWORK_IPV6) {
 450                 err = ip6_local_out(dev_net(skb_dst(skb)->dev), nskb->sk, nskb);
 451         } else {
 452                 pr_err("Unknown layer 3 protocol: %d\n", av->network_type);
 453                 kfree_skb(nskb);
 454                 return -EINVAL;
 455         }
 456
 457         if (unlikely(net_xmit_eval(err))) {
 458                 pr_debug("error sending packet: %d\n", err);
 459                 return -EAGAIN;
 460         }
 461
 462         kfree_skb(skb);
 463
 464         return 0;
 465 }
 466
 467 static int loopback(struct sk_buff *skb)
 468 {
 469         return rxe_rcv(skb);
 470 }
 471
 472 static inline int addr_same(struct rxe_dev *rxe, struct rxe_av *av)
 473 {
 474         return rxe->port.port_guid == av->grh.dgid.global.interface_id;
 475 }
 476
 477 static struct sk_buff *init_packet(struct rxe_dev *rxe, struct rxe_av *av,
 478                                    int paylen, struct rxe_pkt_info *pkt)
 479 {
 480         unsigned int hdr_len;
 481         struct sk_buff *skb;
 482
 483         if (av->network_type == RDMA_NETWORK_IPV4)
 484                 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
 485                         sizeof(struct iphdr);
 486         else
 487                 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
 488                         sizeof(struct ipv6hdr);
 489
 490         skb = alloc_skb(paylen + hdr_len + LL_RESERVED_SPACE(rxe->ndev),
 491                         GFP_ATOMIC);
 492         if (unlikely(!skb))
 493                 return NULL;
 494
 495         skb_reserve(skb, hdr_len + LL_RESERVED_SPACE(rxe->ndev));
 496
 497         skb->dev        = rxe->ndev;
 498         if (av->network_type == RDMA_NETWORK_IPV4)
 499                 skb->protocol = htons(ETH_P_IP);
 500         else
 501                 skb->protocol = htons(ETH_P_IPV6);
 502
 503         pkt->rxe        = rxe;
 504         pkt->port_num   = 1;
 505         pkt->hdr        = skb_put(skb, paylen);
 506         pkt->mask       |= RXE_GRH_MASK;
 507
 508         memset(pkt->hdr, 0, paylen);
 509
 510         return skb;
 511 }
 512
 513 /*
 514  * this is required by rxe_cfg to match rxe devices in
 515  * /sys/class/infiniband up with their underlying ethernet devices
 516  */
 517 static char *parent_name(struct rxe_dev *rxe, unsigned int port_num)
 518 {
 519         return rxe->ndev->name;
 520 }
 521
 522 static enum rdma_link_layer link_layer(struct rxe_dev *rxe,
 523                                        unsigned int port_num)
 524 {
 525         return IB_LINK_LAYER_ETHERNET;
 526 }
 527
 528 static struct rxe_ifc_ops ifc_ops = {
 529         .node_guid      = node_guid,
 530         .port_guid      = port_guid,
 531         .dma_device     = dma_device,
 532         .mcast_add      = mcast_add,
 533         .mcast_delete   = mcast_delete,
 534         .prepare        = prepare,
 535         .send           = send,
 536         .loopback       = loopback,
 537         .init_packet    = init_packet,
 538         .parent_name    = parent_name,
 539         .link_layer     = link_layer,
 540 };
 541
 542 struct rxe_dev *rxe_net_add(struct net_device *ndev)
 543 {
 544         int err;
 545         struct rxe_dev *rxe = NULL;
 546
 547         rxe = (struct rxe_dev *)ib_alloc_device(sizeof(*rxe));
 548         if (!rxe)
 549                 return NULL;
 550
 551         rxe->ifc_ops = &ifc_ops;
 552         rxe->ndev = ndev;
 553
 554         err = rxe_add(rxe, ndev->mtu);
 555         if (err) {
 556                 ib_dealloc_device(&rxe->ib_dev);
 557                 return NULL;
 558         }
 559
 560         spin_lock_bh(&dev_list_lock);
 561         list_add_tail(&rxe_dev_list, &rxe->list);
 562         spin_unlock_bh(&dev_list_lock);
 563         return rxe;
 564 }
 565
 566 void rxe_remove_all(void)
 567 {
 568         spin_lock_bh(&dev_list_lock);
 569         while (!list_empty(&rxe_dev_list)) {
 570                 struct rxe_dev *rxe =
 571                         list_first_entry(&rxe_dev_list, struct rxe_dev, list);
 572
 573                 list_del(&rxe->list);
 574                 spin_unlock_bh(&dev_list_lock);
 575                 rxe_remove(rxe);
 576                 spin_lock_bh(&dev_list_lock);
 577         }
 578         spin_unlock_bh(&dev_list_lock);
 579 }
 580 EXPORT_SYMBOL(rxe_remove_all);
 581
 582 static void rxe_port_event(struct rxe_dev *rxe,
 583                            enum ib_event_type event)
 584 {
 585         struct ib_event ev;
 586
 587         ev.device = &rxe->ib_dev;
 588         ev.element.port_num = 1;
 589         ev.event = event;
 590
 591         ib_dispatch_event(&ev);
 592 }
 593
 594 /* Caller must hold net_info_lock */
 595 void rxe_port_up(struct rxe_dev *rxe)
 596 {
 597         struct rxe_port *port;
 598
 599         port = &rxe->port;
 600         port->attr.state = IB_PORT_ACTIVE;
 601         port->attr.phys_state = IB_PHYS_STATE_LINK_UP;
 602
 603         rxe_port_event(rxe, IB_EVENT_PORT_ACTIVE);
 604         pr_info("set %s active\n", rxe->ib_dev.name);
 605 }
 606
 607 /* Caller must hold net_info_lock */
 608 void rxe_port_down(struct rxe_dev *rxe)
 609 {
 610         struct rxe_port *port;
 611
 612         port = &rxe->port;
 613         port->attr.state = IB_PORT_DOWN;
 614         port->attr.phys_state = IB_PHYS_STATE_LINK_DOWN;
 615
 616         rxe_port_event(rxe, IB_EVENT_PORT_ERR);
 617         pr_info("set %s down\n", rxe->ib_dev.name);
 618 }
 619
 620 static int rxe_notify(struct notifier_block *not_blk,
 621                       unsigned long event,
 622                       void *arg)
 623 {
 624         struct net_device *ndev = netdev_notifier_info_to_dev(arg);
 625         struct rxe_dev *rxe = net_to_rxe(ndev);
 626
 627         if (!rxe)
 628                 goto out;
 629
 630         switch (event) {
 631         case NETDEV_UNREGISTER:
 632                 list_del(&rxe->list);
 633                 rxe_remove(rxe);
 634                 break;
 635         case NETDEV_UP:
 636                 rxe_port_up(rxe);
 637                 break;
 638         case NETDEV_DOWN:
 639                 rxe_port_down(rxe);
 640                 break;
 641         case NETDEV_CHANGEMTU:
 642                 pr_info("%s changed mtu to %d\n", ndev->name, ndev->mtu);
 643                 rxe_set_mtu(rxe, ndev->mtu);
 644                 break;
 645         case NETDEV_REBOOT:
 646         case NETDEV_CHANGE:
 647         case NETDEV_GOING_DOWN:
 648         case NETDEV_CHANGEADDR:
 649         case NETDEV_CHANGENAME:
 650         case NETDEV_FEAT_CHANGE:
 651         default:
 652                 pr_info("ignoring netdev event = %ld for %s\n",
 653                         event, ndev->name);
 654                 break;
 655         }
 656 out:
 657         return NOTIFY_OK;
 658 }
 659
 660 struct notifier_block rxe_net_notifier = {
 661         .notifier_call = rxe_notify,
 662 };
 663
 664 int rxe_net_ipv4_init(void)
 665 {
 666         spin_lock_init(&dev_list_lock);
 667
 668         recv_sockets.sk4 = rxe_setup_udp_tunnel(&init_net,
 669                                 htons(ROCE_V2_UDP_DPORT), false);
 670         if (IS_ERR(recv_sockets.sk4)) {
 671                 recv_sockets.sk4 = NULL;
 672                 pr_err("Failed to create IPv4 UDP tunnel\n");
 673                 return -1;
 674         }
 675
 676         return 0;
 677 }
 678
 679 int rxe_net_ipv6_init(void)
 680 {
 681 #if IS_ENABLED(CONFIG_IPV6)
 682
 683         spin_lock_init(&dev_list_lock);
 684
 685         recv_sockets.sk6 = rxe_setup_udp_tunnel(&init_net,
 686                                                 htons(ROCE_V2_UDP_DPORT), true);
 687         if (IS_ERR(recv_sockets.sk6)) {
 688                 recv_sockets.sk6 = NULL;
 689                 pr_err("Failed to create IPv6 UDP tunnel\n");
 690                 return -1;
 691         }
 692 #endif
 693         return 0;
 694 }
 695
 696 void rxe_net_exit(void)
 697 {
 698         rxe_release_udp_tunnel(recv_sockets.sk6);
 699         rxe_release_udp_tunnel(recv_sockets.sk4);
 700         unregister_netdevice_notifier(&rxe_net_notifier);
 701 }
 702
 703 int rxe_net_init(void)
 704 {
 705         int err;
 706
 707         recv_sockets.sk6 = NULL;
 708
 709         err = rxe_net_ipv4_init();
 710         if (err)
 711                 return err;
 712         err = rxe_net_ipv6_init();
 713         if (err)
 714                 goto err_out;
 715         err = register_netdevice_notifier(&rxe_net_notifier);
 716         if (err) {
 717                 pr_err("Failed to register netdev notifier\n");
 718                 goto err_out;
 719         }
 720         return 0;
 721 err_out:
 722         rxe_net_exit();
 723         return err;
 724 }