1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
3 * Copyright (c) 2016 Mellanox Technologies Ltd. All rights reserved.
4 * Copyright (c) 2015 System Fabric Works, Inc. All rights reserved.
7 #include <linux/skbuff.h>
8 #include <linux/if_arp.h>
9 #include <linux/netdevice.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>
21 static struct rxe_recv_sockets recv_sockets;
23 int rxe_mcast_add(struct rxe_dev *rxe, union ib_gid *mgid)
26 unsigned char ll_addr[ETH_ALEN];
28 ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr);
29 err = dev_mc_add(rxe->ndev, ll_addr);
34 int rxe_mcast_delete(struct rxe_dev *rxe, union ib_gid *mgid)
37 unsigned char ll_addr[ETH_ALEN];
39 ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr);
40 err = dev_mc_del(rxe->ndev, ll_addr);
45 static struct dst_entry *rxe_find_route4(struct net_device *ndev,
46 struct in_addr *saddr,
47 struct in_addr *daddr)
50 struct flowi4 fl = { { 0 } };
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;
58 rt = ip_route_output_key(&init_net, &fl);
60 pr_err_ratelimited("no route to %pI4\n", &daddr->s_addr);
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)
72 struct dst_entry *ndst;
73 struct flowi6 fl6 = { { 0 } };
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;
81 ndst = ipv6_stub->ipv6_dst_lookup_flow(sock_net(recv_sockets.sk6->sk),
82 recv_sockets.sk6->sk, &fl6,
85 pr_err_ratelimited("no route to %pI6\n", daddr);
89 if (unlikely(ndst->error)) {
90 pr_err("no route to %pI6\n", daddr);
102 static struct dst_entry *rxe_find_route6(struct net_device *ndev,
103 struct in6_addr *saddr,
104 struct in6_addr *daddr)
111 static struct dst_entry *rxe_find_route(struct net_device *ndev,
115 struct dst_entry *dst = NULL;
117 if (qp_type(qp) == IB_QPT_RC)
118 dst = sk_dst_get(qp->sk->sk);
120 if (!dst || !dst_check(dst, qp->dst_cookie)) {
124 if (av->network_type == RXE_NETWORK_TYPE_IPV4) {
125 struct in_addr *saddr;
126 struct in_addr *daddr;
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;
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)
141 rt6_get_cookie((struct rt6_info *)dst);
145 if (dst && (qp_type(qp) == IB_QPT_RC)) {
147 sk_dst_set(qp->sk->sk, dst);
153 static int rxe_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
156 struct net_device *ndev = skb->dev;
157 struct net_device *rdev = ndev;
158 struct rxe_dev *rxe = rxe_get_dev_from_net(ndev);
159 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
161 if (!rxe && is_vlan_dev(rdev)) {
162 rdev = vlan_dev_real_dev(ndev);
163 rxe = rxe_get_dev_from_net(rdev);
168 if (skb_linearize(skb)) {
169 pr_err("skb_linearize failed\n");
170 ib_device_put(&rxe->ib_dev);
177 pkt->hdr = (u8 *)(udph + 1);
178 pkt->mask = RXE_GRH_MASK;
179 pkt->paylen = be16_to_cpu(udph->len) - sizeof(*udph);
184 * FIXME: this is in the wrong place, it needs to be done when pkt is
187 ib_device_put(&rxe->ib_dev);
196 static struct socket *rxe_setup_udp_tunnel(struct net *net, __be16 port,
201 struct udp_port_cfg udp_cfg = { };
202 struct udp_tunnel_sock_cfg tnl_cfg = { };
205 udp_cfg.family = AF_INET6;
206 udp_cfg.ipv6_v6only = 1;
208 udp_cfg.family = AF_INET;
211 udp_cfg.local_udp_port = port;
213 /* Create UDP socket */
214 err = udp_sock_create(net, &udp_cfg, &sock);
218 tnl_cfg.encap_type = 1;
219 tnl_cfg.encap_rcv = rxe_udp_encap_recv;
221 /* Setup UDP tunnel */
222 setup_udp_tunnel_sock(net, sock, &tnl_cfg);
227 static void rxe_release_udp_tunnel(struct socket *sk)
230 udp_tunnel_sock_release(sk);
233 static void prepare_udp_hdr(struct sk_buff *skb, __be16 src_port,
238 __skb_push(skb, sizeof(*udph));
239 skb_reset_transport_header(skb);
242 udph->dest = dst_port;
243 udph->source = src_port;
244 udph->len = htons(skb->len);
248 static void prepare_ipv4_hdr(struct dst_entry *dst, struct sk_buff *skb,
249 __be32 saddr, __be32 daddr, __u8 proto,
250 __u8 tos, __u8 ttl, __be16 df, bool xnet)
254 skb_scrub_packet(skb, xnet);
257 skb_dst_set(skb, dst_clone(dst));
258 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
260 skb_push(skb, sizeof(struct iphdr));
261 skb_reset_network_header(skb);
265 iph->version = IPVERSION;
266 iph->ihl = sizeof(struct iphdr) >> 2;
268 iph->protocol = proto;
273 __ip_select_ident(dev_net(dst->dev), iph,
274 skb_shinfo(skb)->gso_segs ?: 1);
275 iph->tot_len = htons(skb->len);
279 static void prepare_ipv6_hdr(struct dst_entry *dst, struct sk_buff *skb,
280 struct in6_addr *saddr, struct in6_addr *daddr,
281 __u8 proto, __u8 prio, __u8 ttl)
283 struct ipv6hdr *ip6h;
285 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
286 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED
288 skb_dst_set(skb, dst_clone(dst));
290 __skb_push(skb, sizeof(*ip6h));
291 skb_reset_network_header(skb);
292 ip6h = ipv6_hdr(skb);
293 ip6_flow_hdr(ip6h, prio, htonl(0));
294 ip6h->payload_len = htons(skb->len);
295 ip6h->nexthdr = proto;
296 ip6h->hop_limit = ttl;
297 ip6h->daddr = *daddr;
298 ip6h->saddr = *saddr;
299 ip6h->payload_len = htons(skb->len - sizeof(*ip6h));
302 static int prepare4(struct rxe_pkt_info *pkt, struct sk_buff *skb)
304 struct rxe_qp *qp = pkt->qp;
305 struct dst_entry *dst;
307 __be16 df = htons(IP_DF);
308 struct rxe_av *av = rxe_get_av(pkt);
309 struct in_addr *saddr = &av->sgid_addr._sockaddr_in.sin_addr;
310 struct in_addr *daddr = &av->dgid_addr._sockaddr_in.sin_addr;
312 dst = rxe_find_route(skb->dev, qp, av);
314 pr_err("Host not reachable\n");
315 return -EHOSTUNREACH;
318 prepare_udp_hdr(skb, cpu_to_be16(qp->src_port),
319 cpu_to_be16(ROCE_V2_UDP_DPORT));
321 prepare_ipv4_hdr(dst, skb, saddr->s_addr, daddr->s_addr, IPPROTO_UDP,
322 av->grh.traffic_class, av->grh.hop_limit, df, xnet);
328 static int prepare6(struct rxe_pkt_info *pkt, struct sk_buff *skb)
330 struct rxe_qp *qp = pkt->qp;
331 struct dst_entry *dst;
332 struct rxe_av *av = rxe_get_av(pkt);
333 struct in6_addr *saddr = &av->sgid_addr._sockaddr_in6.sin6_addr;
334 struct in6_addr *daddr = &av->dgid_addr._sockaddr_in6.sin6_addr;
336 dst = rxe_find_route(skb->dev, qp, av);
338 pr_err("Host not reachable\n");
339 return -EHOSTUNREACH;
342 prepare_udp_hdr(skb, cpu_to_be16(qp->src_port),
343 cpu_to_be16(ROCE_V2_UDP_DPORT));
345 prepare_ipv6_hdr(dst, skb, saddr, daddr, IPPROTO_UDP,
346 av->grh.traffic_class,
353 int rxe_prepare(struct rxe_pkt_info *pkt, struct sk_buff *skb, u32 *crc)
357 if (skb->protocol == htons(ETH_P_IP))
358 err = prepare4(pkt, skb);
359 else if (skb->protocol == htons(ETH_P_IPV6))
360 err = prepare6(pkt, skb);
362 *crc = rxe_icrc_hdr(pkt, skb);
364 if (ether_addr_equal(skb->dev->dev_addr, rxe_get_av(pkt)->dmac))
365 pkt->mask |= RXE_LOOPBACK_MASK;
370 static void rxe_skb_tx_dtor(struct sk_buff *skb)
372 struct sock *sk = skb->sk;
373 struct rxe_qp *qp = sk->sk_user_data;
374 int skb_out = atomic_dec_return(&qp->skb_out);
376 if (unlikely(qp->need_req_skb &&
377 skb_out < RXE_INFLIGHT_SKBS_PER_QP_LOW))
378 rxe_run_task(&qp->req.task, 1);
383 int rxe_send(struct rxe_pkt_info *pkt, struct sk_buff *skb)
387 skb->destructor = rxe_skb_tx_dtor;
388 skb->sk = pkt->qp->sk->sk;
390 rxe_add_ref(pkt->qp);
391 atomic_inc(&pkt->qp->skb_out);
393 if (skb->protocol == htons(ETH_P_IP)) {
394 err = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
395 } else if (skb->protocol == htons(ETH_P_IPV6)) {
396 err = ip6_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
398 pr_err("Unknown layer 3 protocol: %d\n", skb->protocol);
399 atomic_dec(&pkt->qp->skb_out);
400 rxe_drop_ref(pkt->qp);
405 if (unlikely(net_xmit_eval(err))) {
406 pr_debug("error sending packet: %d\n", err);
413 void rxe_loopback(struct sk_buff *skb)
415 if (skb->protocol == htons(ETH_P_IP))
416 skb_pull(skb, sizeof(struct iphdr));
418 skb_pull(skb, sizeof(struct ipv6hdr));
423 struct sk_buff *rxe_init_packet(struct rxe_dev *rxe, struct rxe_av *av,
424 int paylen, struct rxe_pkt_info *pkt)
426 unsigned int hdr_len;
427 struct sk_buff *skb = NULL;
428 struct net_device *ndev;
429 const struct ib_gid_attr *attr;
430 const int port_num = 1;
432 attr = rdma_get_gid_attr(&rxe->ib_dev, port_num, av->grh.sgid_index);
436 if (av->network_type == RXE_NETWORK_TYPE_IPV4)
437 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
438 sizeof(struct iphdr);
440 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
441 sizeof(struct ipv6hdr);
444 ndev = rdma_read_gid_attr_ndev_rcu(attr);
449 skb = alloc_skb(paylen + hdr_len + LL_RESERVED_SPACE(ndev),
452 if (unlikely(!skb)) {
457 skb_reserve(skb, hdr_len + LL_RESERVED_SPACE(ndev));
459 /* FIXME: hold reference to this netdev until life of this skb. */
463 if (av->network_type == RXE_NETWORK_TYPE_IPV4)
464 skb->protocol = htons(ETH_P_IP);
466 skb->protocol = htons(ETH_P_IPV6);
469 pkt->port_num = port_num;
470 pkt->hdr = skb_put_zero(skb, paylen);
471 pkt->mask |= RXE_GRH_MASK;
474 rdma_put_gid_attr(attr);
479 * this is required by rxe_cfg to match rxe devices in
480 * /sys/class/infiniband up with their underlying ethernet devices
482 const char *rxe_parent_name(struct rxe_dev *rxe, unsigned int port_num)
484 return rxe->ndev->name;
487 int rxe_net_add(const char *ibdev_name, struct net_device *ndev)
490 struct rxe_dev *rxe = NULL;
492 rxe = ib_alloc_device(rxe_dev, ib_dev);
498 err = rxe_add(rxe, ndev->mtu, ibdev_name);
500 ib_dealloc_device(&rxe->ib_dev);
507 static void rxe_port_event(struct rxe_dev *rxe,
508 enum ib_event_type event)
512 ev.device = &rxe->ib_dev;
513 ev.element.port_num = 1;
516 ib_dispatch_event(&ev);
519 /* Caller must hold net_info_lock */
520 void rxe_port_up(struct rxe_dev *rxe)
522 struct rxe_port *port;
525 port->attr.state = IB_PORT_ACTIVE;
527 rxe_port_event(rxe, IB_EVENT_PORT_ACTIVE);
528 dev_info(&rxe->ib_dev.dev, "set active\n");
531 /* Caller must hold net_info_lock */
532 void rxe_port_down(struct rxe_dev *rxe)
534 struct rxe_port *port;
537 port->attr.state = IB_PORT_DOWN;
539 rxe_port_event(rxe, IB_EVENT_PORT_ERR);
540 rxe_counter_inc(rxe, RXE_CNT_LINK_DOWNED);
541 dev_info(&rxe->ib_dev.dev, "set down\n");
544 void rxe_set_port_state(struct rxe_dev *rxe)
546 if (netif_running(rxe->ndev) && netif_carrier_ok(rxe->ndev))
552 static int rxe_notify(struct notifier_block *not_blk,
556 struct net_device *ndev = netdev_notifier_info_to_dev(arg);
557 struct rxe_dev *rxe = rxe_get_dev_from_net(ndev);
563 case NETDEV_UNREGISTER:
564 ib_unregister_device_queued(&rxe->ib_dev);
572 case NETDEV_CHANGEMTU:
573 pr_info("%s changed mtu to %d\n", ndev->name, ndev->mtu);
574 rxe_set_mtu(rxe, ndev->mtu);
577 rxe_set_port_state(rxe);
580 case NETDEV_GOING_DOWN:
581 case NETDEV_CHANGEADDR:
582 case NETDEV_CHANGENAME:
583 case NETDEV_FEAT_CHANGE:
585 pr_info("ignoring netdev event = %ld for %s\n",
590 ib_device_put(&rxe->ib_dev);
594 static struct notifier_block rxe_net_notifier = {
595 .notifier_call = rxe_notify,
598 static int rxe_net_ipv4_init(void)
600 recv_sockets.sk4 = rxe_setup_udp_tunnel(&init_net,
601 htons(ROCE_V2_UDP_DPORT), false);
602 if (IS_ERR(recv_sockets.sk4)) {
603 recv_sockets.sk4 = NULL;
604 pr_err("Failed to create IPv4 UDP tunnel\n");
611 static int rxe_net_ipv6_init(void)
613 #if IS_ENABLED(CONFIG_IPV6)
615 recv_sockets.sk6 = rxe_setup_udp_tunnel(&init_net,
616 htons(ROCE_V2_UDP_DPORT), true);
617 if (PTR_ERR(recv_sockets.sk6) == -EAFNOSUPPORT) {
618 recv_sockets.sk6 = NULL;
619 pr_warn("IPv6 is not supported, can not create a UDPv6 socket\n");
623 if (IS_ERR(recv_sockets.sk6)) {
624 recv_sockets.sk6 = NULL;
625 pr_err("Failed to create IPv6 UDP tunnel\n");
632 void rxe_net_exit(void)
634 rxe_release_udp_tunnel(recv_sockets.sk6);
635 rxe_release_udp_tunnel(recv_sockets.sk4);
636 unregister_netdevice_notifier(&rxe_net_notifier);
639 int rxe_net_init(void)
643 recv_sockets.sk6 = NULL;
645 err = rxe_net_ipv4_init();
648 err = rxe_net_ipv6_init();
651 err = register_netdevice_notifier(&rxe_net_notifier);
653 pr_err("Failed to register netdev notifier\n");