1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB 2 /* 3 * Copyright (c) 2016 Mellanox Technologies Ltd. All rights reserved. 4 * Copyright (c) 2015 System Fabric Works, Inc. All rights reserved. 5 */ 6 7 #include <linux/skbuff.h> 8 #include <linux/if_arp.h> 9 #include <linux/netdevice.h> 10 #include <linux/if.h> 11 #include <linux/if_vlan.h> 12 #include <net/udp_tunnel.h> 13 #include <net/sch_generic.h> 14 #include <linux/netfilter.h> 15 #include <rdma/ib_addr.h> 16 17 #include "rxe.h" 18 #include "rxe_net.h" 19 #include "rxe_loc.h" 20 21 static struct rxe_recv_sockets recv_sockets; 22 23 int rxe_mcast_add(struct rxe_dev *rxe, union ib_gid *mgid) 24 { 25 int err; 26 unsigned char ll_addr[ETH_ALEN]; 27 28 ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr); 29 err = dev_mc_add(rxe->ndev, ll_addr); 30 31 return err; 32 } 33 34 int rxe_mcast_delete(struct rxe_dev *rxe, union ib_gid *mgid) 35 { 36 int err; 37 unsigned char ll_addr[ETH_ALEN]; 38 39 ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr); 40 err = dev_mc_del(rxe->ndev, ll_addr); 41 42 return err; 43 } 44 45 static struct dst_entry *rxe_find_route4(struct net_device *ndev, 46 struct in_addr *saddr, 47 struct in_addr *daddr) 48 { 49 struct rtable *rt; 50 struct flowi4 fl = { { 0 } }; 51 52 memset(&fl, 0, sizeof(fl)); 53 fl.flowi4_oif = ndev->ifindex; 54 memcpy(&fl.saddr, saddr, sizeof(*saddr)); 55 memcpy(&fl.daddr, daddr, sizeof(*daddr)); 56 fl.flowi4_proto = IPPROTO_UDP; 57 58 rt = ip_route_output_key(&init_net, &fl); 59 if (IS_ERR(rt)) { 60 pr_err_ratelimited("no route to %pI4\n", &daddr->s_addr); 61 return NULL; 62 } 63 64 return &rt->dst; 65 } 66 67 #if IS_ENABLED(CONFIG_IPV6) 68 static struct dst_entry *rxe_find_route6(struct net_device *ndev, 69 struct in6_addr *saddr, 70 struct in6_addr *daddr) 71 { 72 struct dst_entry *ndst; 73 struct flowi6 fl6 = { { 0 } }; 74 75 memset(&fl6, 0, sizeof(fl6)); 76 fl6.flowi6_oif = ndev->ifindex; 77 memcpy(&fl6.saddr, saddr, sizeof(*saddr)); 78 memcpy(&fl6.daddr, daddr, sizeof(*daddr)); 79 fl6.flowi6_proto = IPPROTO_UDP; 80 81 ndst = ipv6_stub->ipv6_dst_lookup_flow(sock_net(recv_sockets.sk6->sk), 82 recv_sockets.sk6->sk, &fl6, 83 NULL); 84 if (IS_ERR(ndst)) { 85 pr_err_ratelimited("no route to %pI6\n", daddr); 86 return NULL; 87 } 88 89 if (unlikely(ndst->error)) { 90 pr_err("no route to %pI6\n", daddr); 91 goto put; 92 } 93 94 return ndst; 95 put: 96 dst_release(ndst); 97 return NULL; 98 } 99 100 #else 101 102 static struct dst_entry *rxe_find_route6(struct net_device *ndev, 103 struct in6_addr *saddr, 104 struct in6_addr *daddr) 105 { 106 return NULL; 107 } 108 109 #endif 110 111 static struct dst_entry *rxe_find_route(struct net_device *ndev, 112 struct rxe_qp *qp, 113 struct rxe_av *av) 114 { 115 struct dst_entry *dst = NULL; 116 117 if (qp_type(qp) == IB_QPT_RC) 118 dst = sk_dst_get(qp->sk->sk); 119 120 if (!dst || !dst_check(dst, qp->dst_cookie)) { 121 if (dst) 122 dst_release(dst); 123 124 if (av->network_type == RXE_NETWORK_TYPE_IPV4) { 125 struct in_addr *saddr; 126 struct in_addr *daddr; 127 128 saddr = &av->sgid_addr._sockaddr_in.sin_addr; 129 daddr = &av->dgid_addr._sockaddr_in.sin_addr; 130 dst = rxe_find_route4(ndev, saddr, daddr); 131 } else if (av->network_type == RXE_NETWORK_TYPE_IPV6) { 132 struct in6_addr *saddr6; 133 struct in6_addr *daddr6; 134 135 saddr6 = &av->sgid_addr._sockaddr_in6.sin6_addr; 136 daddr6 = &av->dgid_addr._sockaddr_in6.sin6_addr; 137 dst = rxe_find_route6(ndev, saddr6, daddr6); 138 #if IS_ENABLED(CONFIG_IPV6) 139 if (dst) 140 qp->dst_cookie = 141 rt6_get_cookie((struct rt6_info *)dst); 142 #endif 143 } 144 145 if (dst && (qp_type(qp) == IB_QPT_RC)) { 146 dst_hold(dst); 147 sk_dst_set(qp->sk->sk, dst); 148 } 149 } 150 return dst; 151 } 152 153 static int rxe_udp_encap_recv(struct sock *sk, struct sk_buff *skb) 154 { 155 struct udphdr *udph; 156 struct rxe_dev *rxe; 157 struct net_device *ndev = skb->dev; 158 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb); 159 160 /* takes a reference on rxe->ib_dev 161 * drop when skb is freed 162 */ 163 rxe = rxe_get_dev_from_net(ndev); 164 if (!rxe && is_vlan_dev(ndev)) 165 rxe = rxe_get_dev_from_net(vlan_dev_real_dev(ndev)); 166 if (!rxe) 167 goto drop; 168 169 if (skb_linearize(skb)) { 170 pr_err("skb_linearize failed\n"); 171 ib_device_put(&rxe->ib_dev); 172 goto drop; 173 } 174 175 udph = udp_hdr(skb); 176 pkt->rxe = rxe; 177 pkt->port_num = 1; 178 pkt->hdr = (u8 *)(udph + 1); 179 pkt->mask = RXE_GRH_MASK; 180 pkt->paylen = be16_to_cpu(udph->len) - sizeof(*udph); 181 182 rxe_rcv(skb); 183 184 return 0; 185 drop: 186 kfree_skb(skb); 187 188 return 0; 189 } 190 191 static struct socket *rxe_setup_udp_tunnel(struct net *net, __be16 port, 192 bool ipv6) 193 { 194 int err; 195 struct socket *sock; 196 struct udp_port_cfg udp_cfg = { }; 197 struct udp_tunnel_sock_cfg tnl_cfg = { }; 198 199 if (ipv6) { 200 udp_cfg.family = AF_INET6; 201 udp_cfg.ipv6_v6only = 1; 202 } else { 203 udp_cfg.family = AF_INET; 204 } 205 206 udp_cfg.local_udp_port = port; 207 208 /* Create UDP socket */ 209 err = udp_sock_create(net, &udp_cfg, &sock); 210 if (err < 0) 211 return ERR_PTR(err); 212 213 tnl_cfg.encap_type = 1; 214 tnl_cfg.encap_rcv = rxe_udp_encap_recv; 215 216 /* Setup UDP tunnel */ 217 setup_udp_tunnel_sock(net, sock, &tnl_cfg); 218 219 return sock; 220 } 221 222 static void rxe_release_udp_tunnel(struct socket *sk) 223 { 224 if (sk) 225 udp_tunnel_sock_release(sk); 226 } 227 228 static void prepare_udp_hdr(struct sk_buff *skb, __be16 src_port, 229 __be16 dst_port) 230 { 231 struct udphdr *udph; 232 233 __skb_push(skb, sizeof(*udph)); 234 skb_reset_transport_header(skb); 235 udph = udp_hdr(skb); 236 237 udph->dest = dst_port; 238 udph->source = src_port; 239 udph->len = htons(skb->len); 240 udph->check = 0; 241 } 242 243 static void prepare_ipv4_hdr(struct dst_entry *dst, struct sk_buff *skb, 244 __be32 saddr, __be32 daddr, __u8 proto, 245 __u8 tos, __u8 ttl, __be16 df, bool xnet) 246 { 247 struct iphdr *iph; 248 249 skb_scrub_packet(skb, xnet); 250 251 skb_clear_hash(skb); 252 skb_dst_set(skb, dst_clone(dst)); 253 memset(IPCB(skb), 0, sizeof(*IPCB(skb))); 254 255 skb_push(skb, sizeof(struct iphdr)); 256 skb_reset_network_header(skb); 257 258 iph = ip_hdr(skb); 259 260 iph->version = IPVERSION; 261 iph->ihl = sizeof(struct iphdr) >> 2; 262 iph->frag_off = df; 263 iph->protocol = proto; 264 iph->tos = tos; 265 iph->daddr = daddr; 266 iph->saddr = saddr; 267 iph->ttl = ttl; 268 __ip_select_ident(dev_net(dst->dev), iph, 269 skb_shinfo(skb)->gso_segs ?: 1); 270 } 271 272 static void prepare_ipv6_hdr(struct dst_entry *dst, struct sk_buff *skb, 273 struct in6_addr *saddr, struct in6_addr *daddr, 274 __u8 proto, __u8 prio, __u8 ttl) 275 { 276 struct ipv6hdr *ip6h; 277 278 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); 279 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED 280 | IPSKB_REROUTED); 281 skb_dst_set(skb, dst_clone(dst)); 282 283 __skb_push(skb, sizeof(*ip6h)); 284 skb_reset_network_header(skb); 285 ip6h = ipv6_hdr(skb); 286 ip6_flow_hdr(ip6h, prio, htonl(0)); 287 ip6h->payload_len = htons(skb->len); 288 ip6h->nexthdr = proto; 289 ip6h->hop_limit = ttl; 290 ip6h->daddr = *daddr; 291 ip6h->saddr = *saddr; 292 ip6h->payload_len = htons(skb->len - sizeof(*ip6h)); 293 } 294 295 static int prepare4(struct rxe_pkt_info *pkt, struct sk_buff *skb) 296 { 297 struct rxe_qp *qp = pkt->qp; 298 struct dst_entry *dst; 299 bool xnet = false; 300 __be16 df = htons(IP_DF); 301 struct rxe_av *av = rxe_get_av(pkt); 302 struct in_addr *saddr = &av->sgid_addr._sockaddr_in.sin_addr; 303 struct in_addr *daddr = &av->dgid_addr._sockaddr_in.sin_addr; 304 305 dst = rxe_find_route(skb->dev, qp, av); 306 if (!dst) { 307 pr_err("Host not reachable\n"); 308 return -EHOSTUNREACH; 309 } 310 311 prepare_udp_hdr(skb, cpu_to_be16(qp->src_port), 312 cpu_to_be16(ROCE_V2_UDP_DPORT)); 313 314 prepare_ipv4_hdr(dst, skb, saddr->s_addr, daddr->s_addr, IPPROTO_UDP, 315 av->grh.traffic_class, av->grh.hop_limit, df, xnet); 316 317 dst_release(dst); 318 return 0; 319 } 320 321 static int prepare6(struct rxe_pkt_info *pkt, struct sk_buff *skb) 322 { 323 struct rxe_qp *qp = pkt->qp; 324 struct dst_entry *dst; 325 struct rxe_av *av = rxe_get_av(pkt); 326 struct in6_addr *saddr = &av->sgid_addr._sockaddr_in6.sin6_addr; 327 struct in6_addr *daddr = &av->dgid_addr._sockaddr_in6.sin6_addr; 328 329 dst = rxe_find_route(skb->dev, qp, av); 330 if (!dst) { 331 pr_err("Host not reachable\n"); 332 return -EHOSTUNREACH; 333 } 334 335 prepare_udp_hdr(skb, cpu_to_be16(qp->src_port), 336 cpu_to_be16(ROCE_V2_UDP_DPORT)); 337 338 prepare_ipv6_hdr(dst, skb, saddr, daddr, IPPROTO_UDP, 339 av->grh.traffic_class, 340 av->grh.hop_limit); 341 342 dst_release(dst); 343 return 0; 344 } 345 346 int rxe_prepare(struct rxe_pkt_info *pkt, struct sk_buff *skb, u32 *crc) 347 { 348 int err = 0; 349 350 if (skb->protocol == htons(ETH_P_IP)) 351 err = prepare4(pkt, skb); 352 else if (skb->protocol == htons(ETH_P_IPV6)) 353 err = prepare6(pkt, skb); 354 355 *crc = rxe_icrc_hdr(pkt, skb); 356 357 if (ether_addr_equal(skb->dev->dev_addr, rxe_get_av(pkt)->dmac)) 358 pkt->mask |= RXE_LOOPBACK_MASK; 359 360 return err; 361 } 362 363 static void rxe_skb_tx_dtor(struct sk_buff *skb) 364 { 365 struct sock *sk = skb->sk; 366 struct rxe_qp *qp = sk->sk_user_data; 367 int skb_out = atomic_dec_return(&qp->skb_out); 368 369 if (unlikely(qp->need_req_skb && 370 skb_out < RXE_INFLIGHT_SKBS_PER_QP_LOW)) 371 rxe_run_task(&qp->req.task, 1); 372 373 rxe_drop_ref(qp); 374 } 375 376 int rxe_send(struct rxe_pkt_info *pkt, struct sk_buff *skb) 377 { 378 int err; 379 380 skb->destructor = rxe_skb_tx_dtor; 381 skb->sk = pkt->qp->sk->sk; 382 383 rxe_add_ref(pkt->qp); 384 atomic_inc(&pkt->qp->skb_out); 385 386 if (skb->protocol == htons(ETH_P_IP)) { 387 err = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 388 } else if (skb->protocol == htons(ETH_P_IPV6)) { 389 err = ip6_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 390 } else { 391 pr_err("Unknown layer 3 protocol: %d\n", skb->protocol); 392 atomic_dec(&pkt->qp->skb_out); 393 rxe_drop_ref(pkt->qp); 394 kfree_skb(skb); 395 return -EINVAL; 396 } 397 398 if (unlikely(net_xmit_eval(err))) { 399 pr_debug("error sending packet: %d\n", err); 400 return -EAGAIN; 401 } 402 403 return 0; 404 } 405 406 /* fix up a send packet to match the packets 407 * received from UDP before looping them back 408 */ 409 void rxe_loopback(struct sk_buff *skb) 410 { 411 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb); 412 413 if (skb->protocol == htons(ETH_P_IP)) 414 skb_pull(skb, sizeof(struct iphdr)); 415 else 416 skb_pull(skb, sizeof(struct ipv6hdr)); 417 418 if (WARN_ON(!ib_device_try_get(&pkt->rxe->ib_dev))) 419 kfree_skb(skb); 420 else 421 rxe_rcv(skb); 422 } 423 424 struct sk_buff *rxe_init_packet(struct rxe_dev *rxe, struct rxe_av *av, 425 int paylen, struct rxe_pkt_info *pkt) 426 { 427 unsigned int hdr_len; 428 struct sk_buff *skb = NULL; 429 struct net_device *ndev; 430 const struct ib_gid_attr *attr; 431 const int port_num = 1; 432 433 attr = rdma_get_gid_attr(&rxe->ib_dev, port_num, av->grh.sgid_index); 434 if (IS_ERR(attr)) 435 return NULL; 436 437 if (av->network_type == RXE_NETWORK_TYPE_IPV4) 438 hdr_len = ETH_HLEN + sizeof(struct udphdr) + 439 sizeof(struct iphdr); 440 else 441 hdr_len = ETH_HLEN + sizeof(struct udphdr) + 442 sizeof(struct ipv6hdr); 443 444 rcu_read_lock(); 445 ndev = rdma_read_gid_attr_ndev_rcu(attr); 446 if (IS_ERR(ndev)) { 447 rcu_read_unlock(); 448 goto out; 449 } 450 skb = alloc_skb(paylen + hdr_len + LL_RESERVED_SPACE(ndev), 451 GFP_ATOMIC); 452 453 if (unlikely(!skb)) { 454 rcu_read_unlock(); 455 goto out; 456 } 457 458 skb_reserve(skb, hdr_len + LL_RESERVED_SPACE(ndev)); 459 460 /* FIXME: hold reference to this netdev until life of this skb. */ 461 skb->dev = ndev; 462 rcu_read_unlock(); 463 464 if (av->network_type == RXE_NETWORK_TYPE_IPV4) 465 skb->protocol = htons(ETH_P_IP); 466 else 467 skb->protocol = htons(ETH_P_IPV6); 468 469 pkt->rxe = rxe; 470 pkt->port_num = port_num; 471 pkt->hdr = skb_put(skb, paylen); 472 pkt->mask |= RXE_GRH_MASK; 473 474 out: 475 rdma_put_gid_attr(attr); 476 return skb; 477 } 478 479 /* 480 * this is required by rxe_cfg to match rxe devices in 481 * /sys/class/infiniband up with their underlying ethernet devices 482 */ 483 const char *rxe_parent_name(struct rxe_dev *rxe, unsigned int port_num) 484 { 485 return rxe->ndev->name; 486 } 487 488 int rxe_net_add(const char *ibdev_name, struct net_device *ndev) 489 { 490 int err; 491 struct rxe_dev *rxe = NULL; 492 493 rxe = ib_alloc_device(rxe_dev, ib_dev); 494 if (!rxe) 495 return -ENOMEM; 496 497 rxe->ndev = ndev; 498 499 err = rxe_add(rxe, ndev->mtu, ibdev_name); 500 if (err) { 501 ib_dealloc_device(&rxe->ib_dev); 502 return err; 503 } 504 505 return 0; 506 } 507 508 static void rxe_port_event(struct rxe_dev *rxe, 509 enum ib_event_type event) 510 { 511 struct ib_event ev; 512 513 ev.device = &rxe->ib_dev; 514 ev.element.port_num = 1; 515 ev.event = event; 516 517 ib_dispatch_event(&ev); 518 } 519 520 /* Caller must hold net_info_lock */ 521 void rxe_port_up(struct rxe_dev *rxe) 522 { 523 struct rxe_port *port; 524 525 port = &rxe->port; 526 port->attr.state = IB_PORT_ACTIVE; 527 528 rxe_port_event(rxe, IB_EVENT_PORT_ACTIVE); 529 dev_info(&rxe->ib_dev.dev, "set active\n"); 530 } 531 532 /* Caller must hold net_info_lock */ 533 void rxe_port_down(struct rxe_dev *rxe) 534 { 535 struct rxe_port *port; 536 537 port = &rxe->port; 538 port->attr.state = IB_PORT_DOWN; 539 540 rxe_port_event(rxe, IB_EVENT_PORT_ERR); 541 rxe_counter_inc(rxe, RXE_CNT_LINK_DOWNED); 542 dev_info(&rxe->ib_dev.dev, "set down\n"); 543 } 544 545 void rxe_set_port_state(struct rxe_dev *rxe) 546 { 547 if (netif_running(rxe->ndev) && netif_carrier_ok(rxe->ndev)) 548 rxe_port_up(rxe); 549 else 550 rxe_port_down(rxe); 551 } 552 553 static int rxe_notify(struct notifier_block *not_blk, 554 unsigned long event, 555 void *arg) 556 { 557 struct net_device *ndev = netdev_notifier_info_to_dev(arg); 558 struct rxe_dev *rxe = rxe_get_dev_from_net(ndev); 559 560 if (!rxe) 561 return NOTIFY_OK; 562 563 switch (event) { 564 case NETDEV_UNREGISTER: 565 ib_unregister_device_queued(&rxe->ib_dev); 566 break; 567 case NETDEV_UP: 568 rxe_port_up(rxe); 569 break; 570 case NETDEV_DOWN: 571 rxe_port_down(rxe); 572 break; 573 case NETDEV_CHANGEMTU: 574 pr_info("%s changed mtu to %d\n", ndev->name, ndev->mtu); 575 rxe_set_mtu(rxe, ndev->mtu); 576 break; 577 case NETDEV_CHANGE: 578 rxe_set_port_state(rxe); 579 break; 580 case NETDEV_REBOOT: 581 case NETDEV_GOING_DOWN: 582 case NETDEV_CHANGEADDR: 583 case NETDEV_CHANGENAME: 584 case NETDEV_FEAT_CHANGE: 585 default: 586 pr_info("ignoring netdev event = %ld for %s\n", 587 event, ndev->name); 588 break; 589 } 590 591 ib_device_put(&rxe->ib_dev); 592 return NOTIFY_OK; 593 } 594 595 static struct notifier_block rxe_net_notifier = { 596 .notifier_call = rxe_notify, 597 }; 598 599 static int rxe_net_ipv4_init(void) 600 { 601 recv_sockets.sk4 = rxe_setup_udp_tunnel(&init_net, 602 htons(ROCE_V2_UDP_DPORT), false); 603 if (IS_ERR(recv_sockets.sk4)) { 604 recv_sockets.sk4 = NULL; 605 pr_err("Failed to create IPv4 UDP tunnel\n"); 606 return -1; 607 } 608 609 return 0; 610 } 611 612 static int rxe_net_ipv6_init(void) 613 { 614 #if IS_ENABLED(CONFIG_IPV6) 615 616 recv_sockets.sk6 = rxe_setup_udp_tunnel(&init_net, 617 htons(ROCE_V2_UDP_DPORT), true); 618 if (PTR_ERR(recv_sockets.sk6) == -EAFNOSUPPORT) { 619 recv_sockets.sk6 = NULL; 620 pr_warn("IPv6 is not supported, can not create a UDPv6 socket\n"); 621 return 0; 622 } 623 624 if (IS_ERR(recv_sockets.sk6)) { 625 recv_sockets.sk6 = NULL; 626 pr_err("Failed to create IPv6 UDP tunnel\n"); 627 return -1; 628 } 629 #endif 630 return 0; 631 } 632 633 void rxe_net_exit(void) 634 { 635 rxe_release_udp_tunnel(recv_sockets.sk6); 636 rxe_release_udp_tunnel(recv_sockets.sk4); 637 unregister_netdevice_notifier(&rxe_net_notifier); 638 } 639 640 int rxe_net_init(void) 641 { 642 int err; 643 644 recv_sockets.sk6 = NULL; 645 646 err = rxe_net_ipv4_init(); 647 if (err) 648 return err; 649 err = rxe_net_ipv6_init(); 650 if (err) 651 goto err_out; 652 err = register_netdevice_notifier(&rxe_net_notifier); 653 if (err) { 654 pr_err("Failed to register netdev notifier\n"); 655 goto err_out; 656 } 657 return 0; 658 err_out: 659 rxe_net_exit(); 660 return err; 661 } 662