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->tot_len = htons(skb->len); 263 iph->frag_off = df; 264 iph->protocol = proto; 265 iph->tos = tos; 266 iph->daddr = daddr; 267 iph->saddr = saddr; 268 iph->ttl = ttl; 269 __ip_select_ident(dev_net(dst->dev), iph, 270 skb_shinfo(skb)->gso_segs ?: 1); 271 } 272 273 static void prepare_ipv6_hdr(struct dst_entry *dst, struct sk_buff *skb, 274 struct in6_addr *saddr, struct in6_addr *daddr, 275 __u8 proto, __u8 prio, __u8 ttl) 276 { 277 struct ipv6hdr *ip6h; 278 279 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt)); 280 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED 281 | IPSKB_REROUTED); 282 skb_dst_set(skb, dst_clone(dst)); 283 284 __skb_push(skb, sizeof(*ip6h)); 285 skb_reset_network_header(skb); 286 ip6h = ipv6_hdr(skb); 287 ip6_flow_hdr(ip6h, prio, htonl(0)); 288 ip6h->payload_len = htons(skb->len); 289 ip6h->nexthdr = proto; 290 ip6h->hop_limit = ttl; 291 ip6h->daddr = *daddr; 292 ip6h->saddr = *saddr; 293 ip6h->payload_len = htons(skb->len - sizeof(*ip6h)); 294 } 295 296 static int prepare4(struct rxe_pkt_info *pkt, struct sk_buff *skb) 297 { 298 struct rxe_qp *qp = pkt->qp; 299 struct dst_entry *dst; 300 bool xnet = false; 301 __be16 df = htons(IP_DF); 302 struct rxe_av *av = rxe_get_av(pkt); 303 struct in_addr *saddr = &av->sgid_addr._sockaddr_in.sin_addr; 304 struct in_addr *daddr = &av->dgid_addr._sockaddr_in.sin_addr; 305 306 dst = rxe_find_route(skb->dev, qp, av); 307 if (!dst) { 308 pr_err("Host not reachable\n"); 309 return -EHOSTUNREACH; 310 } 311 312 prepare_udp_hdr(skb, cpu_to_be16(qp->src_port), 313 cpu_to_be16(ROCE_V2_UDP_DPORT)); 314 315 prepare_ipv4_hdr(dst, skb, saddr->s_addr, daddr->s_addr, IPPROTO_UDP, 316 av->grh.traffic_class, av->grh.hop_limit, df, xnet); 317 318 dst_release(dst); 319 return 0; 320 } 321 322 static int prepare6(struct rxe_pkt_info *pkt, struct sk_buff *skb) 323 { 324 struct rxe_qp *qp = pkt->qp; 325 struct dst_entry *dst; 326 struct rxe_av *av = rxe_get_av(pkt); 327 struct in6_addr *saddr = &av->sgid_addr._sockaddr_in6.sin6_addr; 328 struct in6_addr *daddr = &av->dgid_addr._sockaddr_in6.sin6_addr; 329 330 dst = rxe_find_route(skb->dev, qp, av); 331 if (!dst) { 332 pr_err("Host not reachable\n"); 333 return -EHOSTUNREACH; 334 } 335 336 prepare_udp_hdr(skb, cpu_to_be16(qp->src_port), 337 cpu_to_be16(ROCE_V2_UDP_DPORT)); 338 339 prepare_ipv6_hdr(dst, skb, saddr, daddr, IPPROTO_UDP, 340 av->grh.traffic_class, 341 av->grh.hop_limit); 342 343 dst_release(dst); 344 return 0; 345 } 346 347 int rxe_prepare(struct rxe_pkt_info *pkt, struct sk_buff *skb, u32 *crc) 348 { 349 int err = 0; 350 351 if (skb->protocol == htons(ETH_P_IP)) 352 err = prepare4(pkt, skb); 353 else if (skb->protocol == htons(ETH_P_IPV6)) 354 err = prepare6(pkt, skb); 355 356 *crc = rxe_icrc_hdr(pkt, skb); 357 358 if (ether_addr_equal(skb->dev->dev_addr, rxe_get_av(pkt)->dmac)) 359 pkt->mask |= RXE_LOOPBACK_MASK; 360 361 return err; 362 } 363 364 static void rxe_skb_tx_dtor(struct sk_buff *skb) 365 { 366 struct sock *sk = skb->sk; 367 struct rxe_qp *qp = sk->sk_user_data; 368 int skb_out = atomic_dec_return(&qp->skb_out); 369 370 if (unlikely(qp->need_req_skb && 371 skb_out < RXE_INFLIGHT_SKBS_PER_QP_LOW)) 372 rxe_run_task(&qp->req.task, 1); 373 374 rxe_drop_ref(qp); 375 } 376 377 int rxe_send(struct rxe_pkt_info *pkt, struct sk_buff *skb) 378 { 379 int err; 380 381 skb->destructor = rxe_skb_tx_dtor; 382 skb->sk = pkt->qp->sk->sk; 383 384 rxe_add_ref(pkt->qp); 385 atomic_inc(&pkt->qp->skb_out); 386 387 if (skb->protocol == htons(ETH_P_IP)) { 388 err = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 389 } else if (skb->protocol == htons(ETH_P_IPV6)) { 390 err = ip6_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb); 391 } else { 392 pr_err("Unknown layer 3 protocol: %d\n", skb->protocol); 393 atomic_dec(&pkt->qp->skb_out); 394 rxe_drop_ref(pkt->qp); 395 kfree_skb(skb); 396 return -EINVAL; 397 } 398 399 if (unlikely(net_xmit_eval(err))) { 400 pr_debug("error sending packet: %d\n", err); 401 return -EAGAIN; 402 } 403 404 return 0; 405 } 406 407 /* fix up a send packet to match the packets 408 * received from UDP before looping them back 409 */ 410 void rxe_loopback(struct sk_buff *skb) 411 { 412 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb); 413 414 if (skb->protocol == htons(ETH_P_IP)) 415 skb_pull(skb, sizeof(struct iphdr)); 416 else 417 skb_pull(skb, sizeof(struct ipv6hdr)); 418 419 if (WARN_ON(!ib_device_try_get(&pkt->rxe->ib_dev))) 420 kfree_skb(skb); 421 else 422 rxe_rcv(skb); 423 } 424 425 struct sk_buff *rxe_init_packet(struct rxe_dev *rxe, struct rxe_av *av, 426 int paylen, struct rxe_pkt_info *pkt) 427 { 428 unsigned int hdr_len; 429 struct sk_buff *skb = NULL; 430 struct net_device *ndev; 431 const struct ib_gid_attr *attr; 432 const int port_num = 1; 433 434 attr = rdma_get_gid_attr(&rxe->ib_dev, port_num, av->grh.sgid_index); 435 if (IS_ERR(attr)) 436 return NULL; 437 438 if (av->network_type == RXE_NETWORK_TYPE_IPV4) 439 hdr_len = ETH_HLEN + sizeof(struct udphdr) + 440 sizeof(struct iphdr); 441 else 442 hdr_len = ETH_HLEN + sizeof(struct udphdr) + 443 sizeof(struct ipv6hdr); 444 445 rcu_read_lock(); 446 ndev = rdma_read_gid_attr_ndev_rcu(attr); 447 if (IS_ERR(ndev)) { 448 rcu_read_unlock(); 449 goto out; 450 } 451 skb = alloc_skb(paylen + hdr_len + LL_RESERVED_SPACE(ndev), 452 GFP_ATOMIC); 453 454 if (unlikely(!skb)) { 455 rcu_read_unlock(); 456 goto out; 457 } 458 459 skb_reserve(skb, hdr_len + LL_RESERVED_SPACE(ndev)); 460 461 /* FIXME: hold reference to this netdev until life of this skb. */ 462 skb->dev = ndev; 463 rcu_read_unlock(); 464 465 if (av->network_type == RXE_NETWORK_TYPE_IPV4) 466 skb->protocol = htons(ETH_P_IP); 467 else 468 skb->protocol = htons(ETH_P_IPV6); 469 470 pkt->rxe = rxe; 471 pkt->port_num = port_num; 472 pkt->hdr = skb_put(skb, paylen); 473 pkt->mask |= RXE_GRH_MASK; 474 475 out: 476 rdma_put_gid_attr(attr); 477 return skb; 478 } 479 480 /* 481 * this is required by rxe_cfg to match rxe devices in 482 * /sys/class/infiniband up with their underlying ethernet devices 483 */ 484 const char *rxe_parent_name(struct rxe_dev *rxe, unsigned int port_num) 485 { 486 return rxe->ndev->name; 487 } 488 489 int rxe_net_add(const char *ibdev_name, struct net_device *ndev) 490 { 491 int err; 492 struct rxe_dev *rxe = NULL; 493 494 rxe = ib_alloc_device(rxe_dev, ib_dev); 495 if (!rxe) 496 return -ENOMEM; 497 498 rxe->ndev = ndev; 499 500 err = rxe_add(rxe, ndev->mtu, ibdev_name); 501 if (err) { 502 ib_dealloc_device(&rxe->ib_dev); 503 return err; 504 } 505 506 return 0; 507 } 508 509 static void rxe_port_event(struct rxe_dev *rxe, 510 enum ib_event_type event) 511 { 512 struct ib_event ev; 513 514 ev.device = &rxe->ib_dev; 515 ev.element.port_num = 1; 516 ev.event = event; 517 518 ib_dispatch_event(&ev); 519 } 520 521 /* Caller must hold net_info_lock */ 522 void rxe_port_up(struct rxe_dev *rxe) 523 { 524 struct rxe_port *port; 525 526 port = &rxe->port; 527 port->attr.state = IB_PORT_ACTIVE; 528 529 rxe_port_event(rxe, IB_EVENT_PORT_ACTIVE); 530 dev_info(&rxe->ib_dev.dev, "set active\n"); 531 } 532 533 /* Caller must hold net_info_lock */ 534 void rxe_port_down(struct rxe_dev *rxe) 535 { 536 struct rxe_port *port; 537 538 port = &rxe->port; 539 port->attr.state = IB_PORT_DOWN; 540 541 rxe_port_event(rxe, IB_EVENT_PORT_ERR); 542 rxe_counter_inc(rxe, RXE_CNT_LINK_DOWNED); 543 dev_info(&rxe->ib_dev.dev, "set down\n"); 544 } 545 546 void rxe_set_port_state(struct rxe_dev *rxe) 547 { 548 if (netif_running(rxe->ndev) && netif_carrier_ok(rxe->ndev)) 549 rxe_port_up(rxe); 550 else 551 rxe_port_down(rxe); 552 } 553 554 static int rxe_notify(struct notifier_block *not_blk, 555 unsigned long event, 556 void *arg) 557 { 558 struct net_device *ndev = netdev_notifier_info_to_dev(arg); 559 struct rxe_dev *rxe = rxe_get_dev_from_net(ndev); 560 561 if (!rxe) 562 return NOTIFY_OK; 563 564 switch (event) { 565 case NETDEV_UNREGISTER: 566 ib_unregister_device_queued(&rxe->ib_dev); 567 break; 568 case NETDEV_UP: 569 rxe_port_up(rxe); 570 break; 571 case NETDEV_DOWN: 572 rxe_port_down(rxe); 573 break; 574 case NETDEV_CHANGEMTU: 575 pr_info("%s changed mtu to %d\n", ndev->name, ndev->mtu); 576 rxe_set_mtu(rxe, ndev->mtu); 577 break; 578 case NETDEV_CHANGE: 579 rxe_set_port_state(rxe); 580 break; 581 case NETDEV_REBOOT: 582 case NETDEV_GOING_DOWN: 583 case NETDEV_CHANGEADDR: 584 case NETDEV_CHANGENAME: 585 case NETDEV_FEAT_CHANGE: 586 default: 587 pr_info("ignoring netdev event = %ld for %s\n", 588 event, ndev->name); 589 break; 590 } 591 592 ib_device_put(&rxe->ib_dev); 593 return NOTIFY_OK; 594 } 595 596 static struct notifier_block rxe_net_notifier = { 597 .notifier_call = rxe_notify, 598 }; 599 600 static int rxe_net_ipv4_init(void) 601 { 602 recv_sockets.sk4 = rxe_setup_udp_tunnel(&init_net, 603 htons(ROCE_V2_UDP_DPORT), false); 604 if (IS_ERR(recv_sockets.sk4)) { 605 recv_sockets.sk4 = NULL; 606 pr_err("Failed to create IPv4 UDP tunnel\n"); 607 return -1; 608 } 609 610 return 0; 611 } 612 613 static int rxe_net_ipv6_init(void) 614 { 615 #if IS_ENABLED(CONFIG_IPV6) 616 617 recv_sockets.sk6 = rxe_setup_udp_tunnel(&init_net, 618 htons(ROCE_V2_UDP_DPORT), true); 619 if (PTR_ERR(recv_sockets.sk6) == -EAFNOSUPPORT) { 620 recv_sockets.sk6 = NULL; 621 pr_warn("IPv6 is not supported, can not create a UDPv6 socket\n"); 622 return 0; 623 } 624 625 if (IS_ERR(recv_sockets.sk6)) { 626 recv_sockets.sk6 = NULL; 627 pr_err("Failed to create IPv6 UDP tunnel\n"); 628 return -1; 629 } 630 #endif 631 return 0; 632 } 633 634 void rxe_net_exit(void) 635 { 636 rxe_release_udp_tunnel(recv_sockets.sk6); 637 rxe_release_udp_tunnel(recv_sockets.sk4); 638 unregister_netdevice_notifier(&rxe_net_notifier); 639 } 640 641 int rxe_net_init(void) 642 { 643 int err; 644 645 recv_sockets.sk6 = NULL; 646 647 err = rxe_net_ipv4_init(); 648 if (err) 649 return err; 650 err = rxe_net_ipv6_init(); 651 if (err) 652 goto err_out; 653 err = register_netdevice_notifier(&rxe_net_notifier); 654 if (err) { 655 pr_err("Failed to register netdev notifier\n"); 656 goto err_out; 657 } 658 return 0; 659 err_out: 660 rxe_net_exit(); 661 return err; 662 } 663