2 * Copyright (c) 2016 Mellanox Technologies Ltd. All rights reserved.
3 * Copyright (c) 2015 System Fabric Works, Inc. All rights reserved.
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:
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
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.
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
34 #include <linux/skbuff.h>
35 #include <linux/if_arp.h>
36 #include <linux/netdevice.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>
48 static LIST_HEAD(rxe_dev_list);
49 static DEFINE_SPINLOCK(dev_list_lock); /* spinlock for device list */
51 struct rxe_dev *net_to_rxe(struct net_device *ndev)
54 struct rxe_dev *found = NULL;
56 spin_lock_bh(&dev_list_lock);
57 list_for_each_entry(rxe, &rxe_dev_list, list) {
58 if (rxe->ndev == ndev) {
63 spin_unlock_bh(&dev_list_lock);
68 struct rxe_dev *get_rxe_by_name(const char *name)
71 struct rxe_dev *found = NULL;
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)) {
80 spin_unlock_bh(&dev_list_lock);
85 static struct rxe_recv_sockets recv_sockets;
87 struct device *rxe_dma_device(struct rxe_dev *rxe)
89 struct net_device *ndev;
93 if (is_vlan_dev(ndev))
94 ndev = vlan_dev_real_dev(ndev);
96 return ndev->dev.parent;
99 int rxe_mcast_add(struct rxe_dev *rxe, union ib_gid *mgid)
102 unsigned char ll_addr[ETH_ALEN];
104 ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr);
105 err = dev_mc_add(rxe->ndev, ll_addr);
110 int rxe_mcast_delete(struct rxe_dev *rxe, union ib_gid *mgid)
113 unsigned char ll_addr[ETH_ALEN];
115 ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr);
116 err = dev_mc_del(rxe->ndev, ll_addr);
121 static struct dst_entry *rxe_find_route4(struct net_device *ndev,
122 struct in_addr *saddr,
123 struct in_addr *daddr)
126 struct flowi4 fl = { { 0 } };
128 memset(&fl, 0, sizeof(fl));
129 fl.flowi4_oif = ndev->ifindex;
130 memcpy(&fl.saddr, saddr, sizeof(*saddr));
131 memcpy(&fl.daddr, daddr, sizeof(*daddr));
132 fl.flowi4_proto = IPPROTO_UDP;
134 rt = ip_route_output_key(&init_net, &fl);
136 pr_err_ratelimited("no route to %pI4\n", &daddr->s_addr);
143 #if IS_ENABLED(CONFIG_IPV6)
144 static struct dst_entry *rxe_find_route6(struct net_device *ndev,
145 struct in6_addr *saddr,
146 struct in6_addr *daddr)
148 struct dst_entry *ndst;
149 struct flowi6 fl6 = { { 0 } };
151 memset(&fl6, 0, sizeof(fl6));
152 fl6.flowi6_oif = ndev->ifindex;
153 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
154 memcpy(&fl6.daddr, daddr, sizeof(*daddr));
155 fl6.flowi6_proto = IPPROTO_UDP;
157 ndst = ipv6_stub->ipv6_dst_lookup_flow(sock_net(recv_sockets.sk6->sk),
158 recv_sockets.sk6->sk, &fl6,
160 if (unlikely(IS_ERR(ndst))) {
161 pr_err_ratelimited("no route to %pI6\n", daddr);
165 if (unlikely(ndst->error)) {
166 pr_err("no route to %pI6\n", daddr);
178 static struct dst_entry *rxe_find_route6(struct net_device *ndev,
179 struct in6_addr *saddr,
180 struct in6_addr *daddr)
187 static struct dst_entry *rxe_find_route(struct rxe_dev *rxe,
191 const struct ib_gid_attr *attr;
192 struct dst_entry *dst = NULL;
193 struct net_device *ndev;
195 attr = rdma_get_gid_attr(&rxe->ib_dev, qp->attr.port_num,
201 if (qp_type(qp) == IB_QPT_RC)
202 dst = sk_dst_get(qp->sk->sk);
204 if (!dst || !dst_check(dst, qp->dst_cookie)) {
208 if (av->network_type == RDMA_NETWORK_IPV4) {
209 struct in_addr *saddr;
210 struct in_addr *daddr;
212 saddr = &av->sgid_addr._sockaddr_in.sin_addr;
213 daddr = &av->dgid_addr._sockaddr_in.sin_addr;
214 dst = rxe_find_route4(ndev, saddr, daddr);
215 } else if (av->network_type == RDMA_NETWORK_IPV6) {
216 struct in6_addr *saddr6;
217 struct in6_addr *daddr6;
219 saddr6 = &av->sgid_addr._sockaddr_in6.sin6_addr;
220 daddr6 = &av->dgid_addr._sockaddr_in6.sin6_addr;
221 dst = rxe_find_route6(ndev, saddr6, daddr6);
222 #if IS_ENABLED(CONFIG_IPV6)
225 rt6_get_cookie((struct rt6_info *)dst);
229 if (dst && (qp_type(qp) == IB_QPT_RC)) {
231 sk_dst_set(qp->sk->sk, dst);
234 rdma_put_gid_attr(attr);
238 static int rxe_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
241 struct net_device *ndev = skb->dev;
242 struct net_device *rdev = ndev;
243 struct rxe_dev *rxe = net_to_rxe(ndev);
244 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
246 if (!rxe && is_vlan_dev(rdev)) {
247 rdev = vlan_dev_real_dev(ndev);
248 rxe = net_to_rxe(rdev);
253 if (skb_linearize(skb)) {
254 pr_err("skb_linearize failed\n");
261 pkt->hdr = (u8 *)(udph + 1);
262 pkt->mask = RXE_GRH_MASK;
263 pkt->paylen = be16_to_cpu(udph->len) - sizeof(*udph);
274 static struct socket *rxe_setup_udp_tunnel(struct net *net, __be16 port,
279 struct udp_port_cfg udp_cfg = { };
280 struct udp_tunnel_sock_cfg tnl_cfg = { };
283 udp_cfg.family = AF_INET6;
284 udp_cfg.ipv6_v6only = 1;
286 udp_cfg.family = AF_INET;
289 udp_cfg.local_udp_port = port;
291 /* Create UDP socket */
292 err = udp_sock_create(net, &udp_cfg, &sock);
296 tnl_cfg.encap_type = 1;
297 tnl_cfg.encap_rcv = rxe_udp_encap_recv;
299 /* Setup UDP tunnel */
300 setup_udp_tunnel_sock(net, sock, &tnl_cfg);
305 static void rxe_release_udp_tunnel(struct socket *sk)
308 udp_tunnel_sock_release(sk);
311 static void prepare_udp_hdr(struct sk_buff *skb, __be16 src_port,
316 __skb_push(skb, sizeof(*udph));
317 skb_reset_transport_header(skb);
320 udph->dest = dst_port;
321 udph->source = src_port;
322 udph->len = htons(skb->len);
326 static void prepare_ipv4_hdr(struct dst_entry *dst, struct sk_buff *skb,
327 __be32 saddr, __be32 daddr, __u8 proto,
328 __u8 tos, __u8 ttl, __be16 df, bool xnet)
332 skb_scrub_packet(skb, xnet);
335 skb_dst_set(skb, dst_clone(dst));
336 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
338 skb_push(skb, sizeof(struct iphdr));
339 skb_reset_network_header(skb);
343 iph->version = IPVERSION;
344 iph->ihl = sizeof(struct iphdr) >> 2;
346 iph->protocol = proto;
351 __ip_select_ident(dev_net(dst->dev), iph,
352 skb_shinfo(skb)->gso_segs ?: 1);
353 iph->tot_len = htons(skb->len);
357 static void prepare_ipv6_hdr(struct dst_entry *dst, struct sk_buff *skb,
358 struct in6_addr *saddr, struct in6_addr *daddr,
359 __u8 proto, __u8 prio, __u8 ttl)
361 struct ipv6hdr *ip6h;
363 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
364 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED
366 skb_dst_set(skb, dst_clone(dst));
368 __skb_push(skb, sizeof(*ip6h));
369 skb_reset_network_header(skb);
370 ip6h = ipv6_hdr(skb);
371 ip6_flow_hdr(ip6h, prio, htonl(0));
372 ip6h->payload_len = htons(skb->len);
373 ip6h->nexthdr = proto;
374 ip6h->hop_limit = ttl;
375 ip6h->daddr = *daddr;
376 ip6h->saddr = *saddr;
377 ip6h->payload_len = htons(skb->len - sizeof(*ip6h));
380 static int prepare4(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
381 struct sk_buff *skb, struct rxe_av *av)
383 struct rxe_qp *qp = pkt->qp;
384 struct dst_entry *dst;
386 __be16 df = htons(IP_DF);
387 struct in_addr *saddr = &av->sgid_addr._sockaddr_in.sin_addr;
388 struct in_addr *daddr = &av->dgid_addr._sockaddr_in.sin_addr;
390 dst = rxe_find_route(rxe, qp, av);
392 pr_err("Host not reachable\n");
393 return -EHOSTUNREACH;
396 if (!memcmp(saddr, daddr, sizeof(*daddr)))
397 pkt->mask |= RXE_LOOPBACK_MASK;
399 prepare_udp_hdr(skb, htons(RXE_ROCE_V2_SPORT),
400 htons(ROCE_V2_UDP_DPORT));
402 prepare_ipv4_hdr(dst, skb, saddr->s_addr, daddr->s_addr, IPPROTO_UDP,
403 av->grh.traffic_class, av->grh.hop_limit, df, xnet);
409 static int prepare6(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
410 struct sk_buff *skb, struct rxe_av *av)
412 struct rxe_qp *qp = pkt->qp;
413 struct dst_entry *dst;
414 struct in6_addr *saddr = &av->sgid_addr._sockaddr_in6.sin6_addr;
415 struct in6_addr *daddr = &av->dgid_addr._sockaddr_in6.sin6_addr;
417 dst = rxe_find_route(rxe, qp, av);
419 pr_err("Host not reachable\n");
420 return -EHOSTUNREACH;
423 if (!memcmp(saddr, daddr, sizeof(*daddr)))
424 pkt->mask |= RXE_LOOPBACK_MASK;
426 prepare_udp_hdr(skb, htons(RXE_ROCE_V2_SPORT),
427 htons(ROCE_V2_UDP_DPORT));
429 prepare_ipv6_hdr(dst, skb, saddr, daddr, IPPROTO_UDP,
430 av->grh.traffic_class,
437 int rxe_prepare(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
438 struct sk_buff *skb, u32 *crc)
441 struct rxe_av *av = rxe_get_av(pkt);
443 if (av->network_type == RDMA_NETWORK_IPV4)
444 err = prepare4(rxe, pkt, skb, av);
445 else if (av->network_type == RDMA_NETWORK_IPV6)
446 err = prepare6(rxe, pkt, skb, av);
448 *crc = rxe_icrc_hdr(pkt, skb);
453 static void rxe_skb_tx_dtor(struct sk_buff *skb)
455 struct sock *sk = skb->sk;
456 struct rxe_qp *qp = sk->sk_user_data;
457 int skb_out = atomic_dec_return(&qp->skb_out);
459 if (unlikely(qp->need_req_skb &&
460 skb_out < RXE_INFLIGHT_SKBS_PER_QP_LOW))
461 rxe_run_task(&qp->req.task, 1);
466 int rxe_send(struct rxe_pkt_info *pkt, struct sk_buff *skb)
471 av = rxe_get_av(pkt);
473 skb->destructor = rxe_skb_tx_dtor;
474 skb->sk = pkt->qp->sk->sk;
476 rxe_add_ref(pkt->qp);
477 atomic_inc(&pkt->qp->skb_out);
479 if (av->network_type == RDMA_NETWORK_IPV4) {
480 err = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
481 } else if (av->network_type == RDMA_NETWORK_IPV6) {
482 err = ip6_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
484 pr_err("Unknown layer 3 protocol: %d\n", av->network_type);
485 atomic_dec(&pkt->qp->skb_out);
486 rxe_drop_ref(pkt->qp);
491 if (unlikely(net_xmit_eval(err))) {
492 pr_debug("error sending packet: %d\n", err);
499 void rxe_loopback(struct sk_buff *skb)
501 if (skb->protocol == htons(ETH_P_IP))
502 skb_pull(skb, sizeof(struct iphdr));
504 skb_pull(skb, sizeof(struct ipv6hdr));
509 static inline int addr_same(struct rxe_dev *rxe, struct rxe_av *av)
511 return rxe->port.port_guid == av->grh.dgid.global.interface_id;
514 struct sk_buff *rxe_init_packet(struct rxe_dev *rxe, struct rxe_av *av,
515 int paylen, struct rxe_pkt_info *pkt)
517 unsigned int hdr_len;
519 struct net_device *ndev;
520 const struct ib_gid_attr *attr;
521 const int port_num = 1;
523 attr = rdma_get_gid_attr(&rxe->ib_dev, port_num, av->grh.sgid_index);
528 if (av->network_type == RDMA_NETWORK_IPV4)
529 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
530 sizeof(struct iphdr);
532 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
533 sizeof(struct ipv6hdr);
535 skb = alloc_skb(paylen + hdr_len + LL_RESERVED_SPACE(ndev),
541 skb_reserve(skb, hdr_len + LL_RESERVED_SPACE(ndev));
543 /* FIXME: hold reference to this netdev until life of this skb. */
545 if (av->network_type == RDMA_NETWORK_IPV4)
546 skb->protocol = htons(ETH_P_IP);
548 skb->protocol = htons(ETH_P_IPV6);
551 pkt->port_num = port_num;
552 pkt->hdr = skb_put_zero(skb, paylen);
553 pkt->mask |= RXE_GRH_MASK;
556 rdma_put_gid_attr(attr);
561 * this is required by rxe_cfg to match rxe devices in
562 * /sys/class/infiniband up with their underlying ethernet devices
564 const char *rxe_parent_name(struct rxe_dev *rxe, unsigned int port_num)
566 return rxe->ndev->name;
569 enum rdma_link_layer rxe_link_layer(struct rxe_dev *rxe, unsigned int port_num)
571 return IB_LINK_LAYER_ETHERNET;
574 struct rxe_dev *rxe_net_add(struct net_device *ndev)
577 struct rxe_dev *rxe = NULL;
579 rxe = (struct rxe_dev *)ib_alloc_device(sizeof(*rxe));
585 err = rxe_add(rxe, ndev->mtu);
587 ib_dealloc_device(&rxe->ib_dev);
591 spin_lock_bh(&dev_list_lock);
592 list_add_tail(&rxe->list, &rxe_dev_list);
593 spin_unlock_bh(&dev_list_lock);
597 void rxe_remove_all(void)
599 spin_lock_bh(&dev_list_lock);
600 while (!list_empty(&rxe_dev_list)) {
601 struct rxe_dev *rxe =
602 list_first_entry(&rxe_dev_list, struct rxe_dev, list);
604 list_del(&rxe->list);
605 spin_unlock_bh(&dev_list_lock);
607 spin_lock_bh(&dev_list_lock);
609 spin_unlock_bh(&dev_list_lock);
612 static void rxe_port_event(struct rxe_dev *rxe,
613 enum ib_event_type event)
617 ev.device = &rxe->ib_dev;
618 ev.element.port_num = 1;
621 ib_dispatch_event(&ev);
624 /* Caller must hold net_info_lock */
625 void rxe_port_up(struct rxe_dev *rxe)
627 struct rxe_port *port;
630 port->attr.state = IB_PORT_ACTIVE;
631 port->attr.phys_state = IB_PHYS_STATE_LINK_UP;
633 rxe_port_event(rxe, IB_EVENT_PORT_ACTIVE);
634 pr_info("set %s active\n", rxe->ib_dev.name);
637 /* Caller must hold net_info_lock */
638 void rxe_port_down(struct rxe_dev *rxe)
640 struct rxe_port *port;
643 port->attr.state = IB_PORT_DOWN;
644 port->attr.phys_state = IB_PHYS_STATE_LINK_DOWN;
646 rxe_port_event(rxe, IB_EVENT_PORT_ERR);
647 pr_info("set %s down\n", rxe->ib_dev.name);
650 static int rxe_notify(struct notifier_block *not_blk,
654 struct net_device *ndev = netdev_notifier_info_to_dev(arg);
655 struct rxe_dev *rxe = net_to_rxe(ndev);
661 case NETDEV_UNREGISTER:
662 list_del(&rxe->list);
671 case NETDEV_CHANGEMTU:
672 pr_info("%s changed mtu to %d\n", ndev->name, ndev->mtu);
673 rxe_set_mtu(rxe, ndev->mtu);
676 if (netif_running(ndev) && netif_carrier_ok(ndev))
682 case NETDEV_GOING_DOWN:
683 case NETDEV_CHANGEADDR:
684 case NETDEV_CHANGENAME:
685 case NETDEV_FEAT_CHANGE:
687 pr_info("ignoring netdev event = %ld for %s\n",
695 static struct notifier_block rxe_net_notifier = {
696 .notifier_call = rxe_notify,
699 static int rxe_net_ipv4_init(void)
701 recv_sockets.sk4 = rxe_setup_udp_tunnel(&init_net,
702 htons(ROCE_V2_UDP_DPORT), false);
703 if (IS_ERR(recv_sockets.sk4)) {
704 recv_sockets.sk4 = NULL;
705 pr_err("Failed to create IPv4 UDP tunnel\n");
712 static int rxe_net_ipv6_init(void)
714 #if IS_ENABLED(CONFIG_IPV6)
716 recv_sockets.sk6 = rxe_setup_udp_tunnel(&init_net,
717 htons(ROCE_V2_UDP_DPORT), true);
718 if (PTR_ERR(recv_sockets.sk6) == -EAFNOSUPPORT) {
719 recv_sockets.sk6 = NULL;
720 pr_warn("IPv6 is not supported, can not create a UDPv6 socket\n");
724 if (IS_ERR(recv_sockets.sk6)) {
725 recv_sockets.sk6 = NULL;
726 pr_err("Failed to create IPv6 UDP tunnel\n");
733 void rxe_net_exit(void)
735 rxe_release_udp_tunnel(recv_sockets.sk6);
736 rxe_release_udp_tunnel(recv_sockets.sk4);
737 unregister_netdevice_notifier(&rxe_net_notifier);
740 int rxe_net_init(void)
744 recv_sockets.sk6 = NULL;
746 err = rxe_net_ipv4_init();
749 err = rxe_net_ipv6_init();
752 err = register_netdevice_notifier(&rxe_net_notifier);
754 pr_err("Failed to register netdev notifier\n");