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 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 if (unlikely(ipv6_stub->ipv6_dst_lookup(sock_net(recv_sockets.sk6->sk),
158 recv_sockets.sk6->sk, &ndst, &fl6))) {
159 pr_err_ratelimited("no route to %pI6\n", daddr);
163 if (unlikely(ndst->error)) {
164 pr_err("no route to %pI6\n", daddr);
176 static struct dst_entry *rxe_find_route6(struct net_device *ndev,
177 struct in6_addr *saddr,
178 struct in6_addr *daddr)
185 static struct dst_entry *rxe_find_route(struct rxe_dev *rxe,
189 struct dst_entry *dst = NULL;
191 if (qp_type(qp) == IB_QPT_RC)
192 dst = sk_dst_get(qp->sk->sk);
194 if (!dst || !(dst->obsolete && dst->ops->check(dst, 0))) {
198 if (av->network_type == RDMA_NETWORK_IPV4) {
199 struct in_addr *saddr;
200 struct in_addr *daddr;
202 saddr = &av->sgid_addr._sockaddr_in.sin_addr;
203 daddr = &av->dgid_addr._sockaddr_in.sin_addr;
204 dst = rxe_find_route4(rxe->ndev, saddr, daddr);
205 } else if (av->network_type == RDMA_NETWORK_IPV6) {
206 struct in6_addr *saddr6;
207 struct in6_addr *daddr6;
209 saddr6 = &av->sgid_addr._sockaddr_in6.sin6_addr;
210 daddr6 = &av->dgid_addr._sockaddr_in6.sin6_addr;
211 dst = rxe_find_route6(rxe->ndev, saddr6, daddr6);
218 static int rxe_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
221 struct net_device *ndev = skb->dev;
222 struct rxe_dev *rxe = net_to_rxe(ndev);
223 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
228 if (skb_linearize(skb)) {
229 pr_err("skb_linearize failed\n");
236 pkt->hdr = (u8 *)(udph + 1);
237 pkt->mask = RXE_GRH_MASK;
238 pkt->paylen = be16_to_cpu(udph->len) - sizeof(*udph);
246 static struct socket *rxe_setup_udp_tunnel(struct net *net, __be16 port,
251 struct udp_port_cfg udp_cfg = { };
252 struct udp_tunnel_sock_cfg tnl_cfg = { };
255 udp_cfg.family = AF_INET6;
256 udp_cfg.ipv6_v6only = 1;
258 udp_cfg.family = AF_INET;
261 udp_cfg.local_udp_port = port;
263 /* Create UDP socket */
264 err = udp_sock_create(net, &udp_cfg, &sock);
266 pr_err("failed to create udp socket. err = %d\n", err);
270 tnl_cfg.encap_type = 1;
271 tnl_cfg.encap_rcv = rxe_udp_encap_recv;
273 /* Setup UDP tunnel */
274 setup_udp_tunnel_sock(net, sock, &tnl_cfg);
279 void rxe_release_udp_tunnel(struct socket *sk)
282 udp_tunnel_sock_release(sk);
285 static void prepare_udp_hdr(struct sk_buff *skb, __be16 src_port,
290 __skb_push(skb, sizeof(*udph));
291 skb_reset_transport_header(skb);
294 udph->dest = dst_port;
295 udph->source = src_port;
296 udph->len = htons(skb->len);
300 static void prepare_ipv4_hdr(struct dst_entry *dst, struct sk_buff *skb,
301 __be32 saddr, __be32 daddr, __u8 proto,
302 __u8 tos, __u8 ttl, __be16 df, bool xnet)
306 skb_scrub_packet(skb, xnet);
309 skb_dst_set(skb, dst_clone(dst));
310 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
312 skb_push(skb, sizeof(struct iphdr));
313 skb_reset_network_header(skb);
317 iph->version = IPVERSION;
318 iph->ihl = sizeof(struct iphdr) >> 2;
320 iph->protocol = proto;
325 __ip_select_ident(dev_net(dst->dev), iph,
326 skb_shinfo(skb)->gso_segs ?: 1);
327 iph->tot_len = htons(skb->len);
331 static void prepare_ipv6_hdr(struct dst_entry *dst, struct sk_buff *skb,
332 struct in6_addr *saddr, struct in6_addr *daddr,
333 __u8 proto, __u8 prio, __u8 ttl)
335 struct ipv6hdr *ip6h;
337 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
338 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED
340 skb_dst_set(skb, dst);
342 __skb_push(skb, sizeof(*ip6h));
343 skb_reset_network_header(skb);
344 ip6h = ipv6_hdr(skb);
345 ip6_flow_hdr(ip6h, prio, htonl(0));
346 ip6h->payload_len = htons(skb->len);
347 ip6h->nexthdr = proto;
348 ip6h->hop_limit = ttl;
349 ip6h->daddr = *daddr;
350 ip6h->saddr = *saddr;
351 ip6h->payload_len = htons(skb->len - sizeof(*ip6h));
354 static int prepare4(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
355 struct sk_buff *skb, struct rxe_av *av)
357 struct rxe_qp *qp = pkt->qp;
358 struct dst_entry *dst;
360 __be16 df = htons(IP_DF);
361 struct in_addr *saddr = &av->sgid_addr._sockaddr_in.sin_addr;
362 struct in_addr *daddr = &av->dgid_addr._sockaddr_in.sin_addr;
364 dst = rxe_find_route(rxe, qp, av);
366 pr_err("Host not reachable\n");
367 return -EHOSTUNREACH;
370 if (!memcmp(saddr, daddr, sizeof(*daddr)))
371 pkt->mask |= RXE_LOOPBACK_MASK;
373 prepare_udp_hdr(skb, htons(RXE_ROCE_V2_SPORT),
374 htons(ROCE_V2_UDP_DPORT));
376 prepare_ipv4_hdr(dst, skb, saddr->s_addr, daddr->s_addr, IPPROTO_UDP,
377 av->grh.traffic_class, av->grh.hop_limit, df, xnet);
379 if (qp_type(qp) == IB_QPT_RC)
380 sk_dst_set(qp->sk->sk, dst);
387 static int prepare6(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
388 struct sk_buff *skb, struct rxe_av *av)
390 struct rxe_qp *qp = pkt->qp;
391 struct dst_entry *dst = NULL;
392 struct in6_addr *saddr = &av->sgid_addr._sockaddr_in6.sin6_addr;
393 struct in6_addr *daddr = &av->dgid_addr._sockaddr_in6.sin6_addr;
395 dst = rxe_find_route(rxe, qp, av);
397 pr_err("Host not reachable\n");
398 return -EHOSTUNREACH;
401 if (!memcmp(saddr, daddr, sizeof(*daddr)))
402 pkt->mask |= RXE_LOOPBACK_MASK;
404 prepare_udp_hdr(skb, htons(RXE_ROCE_V2_SPORT),
405 htons(ROCE_V2_UDP_DPORT));
407 prepare_ipv6_hdr(dst, skb, saddr, daddr, IPPROTO_UDP,
408 av->grh.traffic_class,
411 if (qp_type(qp) == IB_QPT_RC)
412 sk_dst_set(qp->sk->sk, dst);
419 int rxe_prepare(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
420 struct sk_buff *skb, u32 *crc)
423 struct rxe_av *av = rxe_get_av(pkt);
425 if (av->network_type == RDMA_NETWORK_IPV4)
426 err = prepare4(rxe, pkt, skb, av);
427 else if (av->network_type == RDMA_NETWORK_IPV6)
428 err = prepare6(rxe, pkt, skb, av);
430 *crc = rxe_icrc_hdr(pkt, skb);
435 static void rxe_skb_tx_dtor(struct sk_buff *skb)
437 struct sock *sk = skb->sk;
438 struct rxe_qp *qp = sk->sk_user_data;
439 int skb_out = atomic_dec_return(&qp->skb_out);
441 if (unlikely(qp->need_req_skb &&
442 skb_out < RXE_INFLIGHT_SKBS_PER_QP_LOW))
443 rxe_run_task(&qp->req.task, 1);
446 int rxe_send(struct rxe_dev *rxe, struct rxe_pkt_info *pkt, struct sk_buff *skb)
448 struct sk_buff *nskb;
452 av = rxe_get_av(pkt);
454 nskb = skb_clone(skb, GFP_ATOMIC);
458 nskb->destructor = rxe_skb_tx_dtor;
459 nskb->sk = pkt->qp->sk->sk;
461 if (av->network_type == RDMA_NETWORK_IPV4) {
462 err = ip_local_out(dev_net(skb_dst(skb)->dev), nskb->sk, nskb);
463 } else if (av->network_type == RDMA_NETWORK_IPV6) {
464 err = ip6_local_out(dev_net(skb_dst(skb)->dev), nskb->sk, nskb);
466 pr_err("Unknown layer 3 protocol: %d\n", av->network_type);
471 if (unlikely(net_xmit_eval(err))) {
472 pr_debug("error sending packet: %d\n", err);
476 atomic_inc(&pkt->qp->skb_out);
482 int rxe_loopback(struct sk_buff *skb)
487 static inline int addr_same(struct rxe_dev *rxe, struct rxe_av *av)
489 return rxe->port.port_guid == av->grh.dgid.global.interface_id;
492 struct sk_buff *rxe_init_packet(struct rxe_dev *rxe, struct rxe_av *av,
493 int paylen, struct rxe_pkt_info *pkt)
495 unsigned int hdr_len;
498 if (av->network_type == RDMA_NETWORK_IPV4)
499 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
500 sizeof(struct iphdr);
502 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
503 sizeof(struct ipv6hdr);
505 skb = alloc_skb(paylen + hdr_len + LL_RESERVED_SPACE(rxe->ndev),
510 skb_reserve(skb, hdr_len + LL_RESERVED_SPACE(rxe->ndev));
512 skb->dev = rxe->ndev;
513 if (av->network_type == RDMA_NETWORK_IPV4)
514 skb->protocol = htons(ETH_P_IP);
516 skb->protocol = htons(ETH_P_IPV6);
520 pkt->hdr = skb_put(skb, paylen);
521 pkt->mask |= RXE_GRH_MASK;
523 memset(pkt->hdr, 0, paylen);
529 * this is required by rxe_cfg to match rxe devices in
530 * /sys/class/infiniband up with their underlying ethernet devices
532 const char *rxe_parent_name(struct rxe_dev *rxe, unsigned int port_num)
534 return rxe->ndev->name;
537 enum rdma_link_layer rxe_link_layer(struct rxe_dev *rxe, unsigned int port_num)
539 return IB_LINK_LAYER_ETHERNET;
542 struct rxe_dev *rxe_net_add(struct net_device *ndev)
545 struct rxe_dev *rxe = NULL;
547 rxe = (struct rxe_dev *)ib_alloc_device(sizeof(*rxe));
553 err = rxe_add(rxe, ndev->mtu);
555 ib_dealloc_device(&rxe->ib_dev);
559 spin_lock_bh(&dev_list_lock);
560 list_add_tail(&rxe->list, &rxe_dev_list);
561 spin_unlock_bh(&dev_list_lock);
565 void rxe_remove_all(void)
567 spin_lock_bh(&dev_list_lock);
568 while (!list_empty(&rxe_dev_list)) {
569 struct rxe_dev *rxe =
570 list_first_entry(&rxe_dev_list, struct rxe_dev, list);
572 list_del(&rxe->list);
573 spin_unlock_bh(&dev_list_lock);
575 spin_lock_bh(&dev_list_lock);
577 spin_unlock_bh(&dev_list_lock);
579 EXPORT_SYMBOL(rxe_remove_all);
581 static void rxe_port_event(struct rxe_dev *rxe,
582 enum ib_event_type event)
586 ev.device = &rxe->ib_dev;
587 ev.element.port_num = 1;
590 ib_dispatch_event(&ev);
593 /* Caller must hold net_info_lock */
594 void rxe_port_up(struct rxe_dev *rxe)
596 struct rxe_port *port;
599 port->attr.state = IB_PORT_ACTIVE;
600 port->attr.phys_state = IB_PHYS_STATE_LINK_UP;
602 rxe_port_event(rxe, IB_EVENT_PORT_ACTIVE);
603 pr_info("set %s active\n", rxe->ib_dev.name);
606 /* Caller must hold net_info_lock */
607 void rxe_port_down(struct rxe_dev *rxe)
609 struct rxe_port *port;
612 port->attr.state = IB_PORT_DOWN;
613 port->attr.phys_state = IB_PHYS_STATE_LINK_DOWN;
615 rxe_port_event(rxe, IB_EVENT_PORT_ERR);
616 pr_info("set %s down\n", rxe->ib_dev.name);
619 static int rxe_notify(struct notifier_block *not_blk,
623 struct net_device *ndev = netdev_notifier_info_to_dev(arg);
624 struct rxe_dev *rxe = net_to_rxe(ndev);
630 case NETDEV_UNREGISTER:
631 list_del(&rxe->list);
640 case NETDEV_CHANGEMTU:
641 pr_info("%s changed mtu to %d\n", ndev->name, ndev->mtu);
642 rxe_set_mtu(rxe, ndev->mtu);
646 case NETDEV_GOING_DOWN:
647 case NETDEV_CHANGEADDR:
648 case NETDEV_CHANGENAME:
649 case NETDEV_FEAT_CHANGE:
651 pr_info("ignoring netdev event = %ld for %s\n",
659 struct notifier_block rxe_net_notifier = {
660 .notifier_call = rxe_notify,
663 static int rxe_net_ipv4_init(void)
665 recv_sockets.sk4 = rxe_setup_udp_tunnel(&init_net,
666 htons(ROCE_V2_UDP_DPORT), false);
667 if (IS_ERR(recv_sockets.sk4)) {
668 recv_sockets.sk4 = NULL;
669 pr_err("Failed to create IPv4 UDP tunnel\n");
676 static int rxe_net_ipv6_init(void)
678 #if IS_ENABLED(CONFIG_IPV6)
680 recv_sockets.sk6 = rxe_setup_udp_tunnel(&init_net,
681 htons(ROCE_V2_UDP_DPORT), true);
682 if (IS_ERR(recv_sockets.sk6)) {
683 recv_sockets.sk6 = NULL;
684 pr_err("Failed to create IPv6 UDP tunnel\n");
691 void rxe_net_exit(void)
693 rxe_release_udp_tunnel(recv_sockets.sk6);
694 rxe_release_udp_tunnel(recv_sockets.sk4);
695 unregister_netdevice_notifier(&rxe_net_notifier);
698 int rxe_net_init(void)
702 recv_sockets.sk6 = NULL;
704 err = rxe_net_ipv4_init();
707 err = rxe_net_ipv6_init();
710 err = register_netdevice_notifier(&rxe_net_notifier);
712 pr_err("Failed to register netdev notifier\n");