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 spinlock_t 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 static __be64 rxe_mac_to_eui64(struct net_device *ndev)
89 unsigned char *mac_addr = ndev->dev_addr;
91 unsigned char *dst = (unsigned char *)&eui64;
93 dst[0] = mac_addr[0] ^ 2;
100 dst[7] = mac_addr[5];
105 static __be64 node_guid(struct rxe_dev *rxe)
107 return rxe_mac_to_eui64(rxe->ndev);
110 static __be64 port_guid(struct rxe_dev *rxe)
112 return rxe_mac_to_eui64(rxe->ndev);
115 static struct device *dma_device(struct rxe_dev *rxe)
117 struct net_device *ndev;
121 if (ndev->priv_flags & IFF_802_1Q_VLAN)
122 ndev = vlan_dev_real_dev(ndev);
124 return ndev->dev.parent;
127 static int mcast_add(struct rxe_dev *rxe, union ib_gid *mgid)
130 unsigned char ll_addr[ETH_ALEN];
132 ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr);
133 err = dev_mc_add(rxe->ndev, ll_addr);
138 static int mcast_delete(struct rxe_dev *rxe, union ib_gid *mgid)
141 unsigned char ll_addr[ETH_ALEN];
143 ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr);
144 err = dev_mc_del(rxe->ndev, ll_addr);
149 static struct dst_entry *rxe_find_route4(struct net_device *ndev,
150 struct in_addr *saddr,
151 struct in_addr *daddr)
154 struct flowi4 fl = { { 0 } };
156 memset(&fl, 0, sizeof(fl));
157 fl.flowi4_oif = ndev->ifindex;
158 memcpy(&fl.saddr, saddr, sizeof(*saddr));
159 memcpy(&fl.daddr, daddr, sizeof(*daddr));
160 fl.flowi4_proto = IPPROTO_UDP;
162 rt = ip_route_output_key(&init_net, &fl);
164 pr_err_ratelimited("no route to %pI4\n", &daddr->s_addr);
171 #if IS_ENABLED(CONFIG_IPV6)
172 static struct dst_entry *rxe_find_route6(struct net_device *ndev,
173 struct in6_addr *saddr,
174 struct in6_addr *daddr)
176 struct dst_entry *ndst;
177 struct flowi6 fl6 = { { 0 } };
179 memset(&fl6, 0, sizeof(fl6));
180 fl6.flowi6_oif = ndev->ifindex;
181 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
182 memcpy(&fl6.daddr, daddr, sizeof(*daddr));
183 fl6.flowi6_proto = IPPROTO_UDP;
185 if (unlikely(ipv6_stub->ipv6_dst_lookup(sock_net(recv_sockets.sk6->sk),
186 recv_sockets.sk6->sk, &ndst, &fl6))) {
187 pr_err_ratelimited("no route to %pI6\n", daddr);
191 if (unlikely(ndst->error)) {
192 pr_err("no route to %pI6\n", daddr);
204 static struct dst_entry *rxe_find_route6(struct net_device *ndev,
205 struct in6_addr *saddr,
206 struct in6_addr *daddr)
213 static int rxe_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
216 struct net_device *ndev = skb->dev;
217 struct rxe_dev *rxe = net_to_rxe(ndev);
218 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
223 if (skb_linearize(skb)) {
224 pr_err("skb_linearize failed\n");
231 pkt->hdr = (u8 *)(udph + 1);
232 pkt->mask = RXE_GRH_MASK;
233 pkt->paylen = be16_to_cpu(udph->len) - sizeof(*udph);
241 static struct socket *rxe_setup_udp_tunnel(struct net *net, __be16 port,
246 struct udp_port_cfg udp_cfg;
247 struct udp_tunnel_sock_cfg tnl_cfg;
249 memset(&udp_cfg, 0, sizeof(udp_cfg));
252 udp_cfg.family = AF_INET6;
253 udp_cfg.ipv6_v6only = 1;
255 udp_cfg.family = AF_INET;
258 udp_cfg.local_udp_port = port;
260 /* Create UDP socket */
261 err = udp_sock_create(net, &udp_cfg, &sock);
263 pr_err("failed to create udp socket. err = %d\n", err);
267 tnl_cfg.sk_user_data = NULL;
268 tnl_cfg.encap_type = 1;
269 tnl_cfg.encap_rcv = rxe_udp_encap_recv;
270 tnl_cfg.encap_destroy = NULL;
272 /* Setup UDP tunnel */
273 setup_udp_tunnel_sock(net, sock, &tnl_cfg);
278 static void rxe_release_udp_tunnel(struct socket *sk)
280 udp_tunnel_sock_release(sk);
283 static void prepare_udp_hdr(struct sk_buff *skb, __be16 src_port,
288 __skb_push(skb, sizeof(*udph));
289 skb_reset_transport_header(skb);
292 udph->dest = dst_port;
293 udph->source = src_port;
294 udph->len = htons(skb->len);
298 static void prepare_ipv4_hdr(struct dst_entry *dst, struct sk_buff *skb,
299 __be32 saddr, __be32 daddr, __u8 proto,
300 __u8 tos, __u8 ttl, __be16 df, bool xnet)
304 skb_scrub_packet(skb, xnet);
307 skb_dst_set(skb, dst);
308 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
310 skb_push(skb, sizeof(struct iphdr));
311 skb_reset_network_header(skb);
315 iph->version = IPVERSION;
316 iph->ihl = sizeof(struct iphdr) >> 2;
318 iph->protocol = proto;
323 __ip_select_ident(dev_net(dst->dev), iph,
324 skb_shinfo(skb)->gso_segs ?: 1);
325 iph->tot_len = htons(skb->len);
329 static void prepare_ipv6_hdr(struct dst_entry *dst, struct sk_buff *skb,
330 struct in6_addr *saddr, struct in6_addr *daddr,
331 __u8 proto, __u8 prio, __u8 ttl)
333 struct ipv6hdr *ip6h;
335 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
336 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED
338 skb_dst_set(skb, dst);
340 __skb_push(skb, sizeof(*ip6h));
341 skb_reset_network_header(skb);
342 ip6h = ipv6_hdr(skb);
343 ip6_flow_hdr(ip6h, prio, htonl(0));
344 ip6h->payload_len = htons(skb->len);
345 ip6h->nexthdr = proto;
346 ip6h->hop_limit = ttl;
347 ip6h->daddr = *daddr;
348 ip6h->saddr = *saddr;
349 ip6h->payload_len = htons(skb->len - sizeof(*ip6h));
352 static int prepare4(struct rxe_dev *rxe, struct sk_buff *skb, struct rxe_av *av)
354 struct dst_entry *dst;
356 __be16 df = htons(IP_DF);
357 struct in_addr *saddr = &av->sgid_addr._sockaddr_in.sin_addr;
358 struct in_addr *daddr = &av->dgid_addr._sockaddr_in.sin_addr;
359 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
361 dst = rxe_find_route4(rxe->ndev, saddr, daddr);
363 pr_err("Host not reachable\n");
364 return -EHOSTUNREACH;
367 if (!memcmp(saddr, daddr, sizeof(*daddr)))
368 pkt->mask |= RXE_LOOPBACK_MASK;
370 prepare_udp_hdr(skb, htons(RXE_ROCE_V2_SPORT),
371 htons(ROCE_V2_UDP_DPORT));
373 prepare_ipv4_hdr(dst, skb, saddr->s_addr, daddr->s_addr, IPPROTO_UDP,
374 av->grh.traffic_class, av->grh.hop_limit, df, xnet);
378 static int prepare6(struct rxe_dev *rxe, struct sk_buff *skb, struct rxe_av *av)
380 struct dst_entry *dst;
381 struct in6_addr *saddr = &av->sgid_addr._sockaddr_in6.sin6_addr;
382 struct in6_addr *daddr = &av->dgid_addr._sockaddr_in6.sin6_addr;
383 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
385 dst = rxe_find_route6(rxe->ndev, saddr, daddr);
387 pr_err("Host not reachable\n");
388 return -EHOSTUNREACH;
391 if (!memcmp(saddr, daddr, sizeof(*daddr)))
392 pkt->mask |= RXE_LOOPBACK_MASK;
394 prepare_udp_hdr(skb, htons(RXE_ROCE_V2_SPORT),
395 htons(ROCE_V2_UDP_DPORT));
397 prepare_ipv6_hdr(dst, skb, saddr, daddr, IPPROTO_UDP,
398 av->grh.traffic_class,
403 static int prepare(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
404 struct sk_buff *skb, u32 *crc)
407 struct rxe_av *av = rxe_get_av(pkt);
409 if (av->network_type == RDMA_NETWORK_IPV4)
410 err = prepare4(rxe, skb, av);
411 else if (av->network_type == RDMA_NETWORK_IPV6)
412 err = prepare6(rxe, skb, av);
414 *crc = rxe_icrc_hdr(pkt, skb);
419 static void rxe_skb_tx_dtor(struct sk_buff *skb)
421 struct sock *sk = skb->sk;
422 struct rxe_qp *qp = sk->sk_user_data;
423 int skb_out = atomic_dec_return(&qp->skb_out);
425 if (unlikely(qp->need_req_skb &&
426 skb_out < RXE_INFLIGHT_SKBS_PER_QP_LOW))
427 rxe_run_task(&qp->req.task, 1);
430 static int send(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
433 struct sk_buff *nskb;
437 av = rxe_get_av(pkt);
439 nskb = skb_clone(skb, GFP_ATOMIC);
443 nskb->destructor = rxe_skb_tx_dtor;
444 nskb->sk = pkt->qp->sk->sk;
446 if (av->network_type == RDMA_NETWORK_IPV4) {
447 err = ip_local_out(dev_net(skb_dst(skb)->dev), nskb->sk, nskb);
448 } else if (av->network_type == RDMA_NETWORK_IPV6) {
449 err = ip6_local_out(dev_net(skb_dst(skb)->dev), nskb->sk, nskb);
451 pr_err("Unknown layer 3 protocol: %d\n", av->network_type);
456 if (unlikely(net_xmit_eval(err))) {
457 pr_debug("error sending packet: %d\n", err);
466 static int loopback(struct sk_buff *skb)
471 static inline int addr_same(struct rxe_dev *rxe, struct rxe_av *av)
473 return rxe->port.port_guid == av->grh.dgid.global.interface_id;
476 static struct sk_buff *init_packet(struct rxe_dev *rxe, struct rxe_av *av,
477 int paylen, struct rxe_pkt_info *pkt)
479 unsigned int hdr_len;
482 if (av->network_type == RDMA_NETWORK_IPV4)
483 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
484 sizeof(struct iphdr);
486 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
487 sizeof(struct ipv6hdr);
489 skb = alloc_skb(paylen + hdr_len + LL_RESERVED_SPACE(rxe->ndev),
494 skb_reserve(skb, hdr_len + LL_RESERVED_SPACE(rxe->ndev));
496 skb->dev = rxe->ndev;
497 if (av->network_type == RDMA_NETWORK_IPV4)
498 skb->protocol = htons(ETH_P_IP);
500 skb->protocol = htons(ETH_P_IPV6);
504 pkt->hdr = skb_put(skb, paylen);
505 pkt->mask |= RXE_GRH_MASK;
507 memset(pkt->hdr, 0, paylen);
513 * this is required by rxe_cfg to match rxe devices in
514 * /sys/class/infiniband up with their underlying ethernet devices
516 static char *parent_name(struct rxe_dev *rxe, unsigned int port_num)
518 return rxe->ndev->name;
521 static enum rdma_link_layer link_layer(struct rxe_dev *rxe,
522 unsigned int port_num)
524 return IB_LINK_LAYER_ETHERNET;
527 static struct rxe_ifc_ops ifc_ops = {
528 .node_guid = node_guid,
529 .port_guid = port_guid,
530 .dma_device = dma_device,
531 .mcast_add = mcast_add,
532 .mcast_delete = mcast_delete,
535 .loopback = loopback,
536 .init_packet = init_packet,
537 .parent_name = parent_name,
538 .link_layer = link_layer,
541 struct rxe_dev *rxe_net_add(struct net_device *ndev)
544 struct rxe_dev *rxe = NULL;
546 rxe = (struct rxe_dev *)ib_alloc_device(sizeof(*rxe));
550 rxe->ifc_ops = &ifc_ops;
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_dev_list, &rxe->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("rxe: set %s active\n", rxe->ib_dev.name);
607 /* Caller must hold net_info_lock */
608 void rxe_port_down(struct rxe_dev *rxe)
610 struct rxe_port *port;
613 port->attr.state = IB_PORT_DOWN;
614 port->attr.phys_state = IB_PHYS_STATE_LINK_DOWN;
616 rxe_port_event(rxe, IB_EVENT_PORT_ERR);
617 pr_info("rxe: set %s down\n", rxe->ib_dev.name);
621 static int rxe_notify(struct notifier_block *not_blk,
625 struct net_device *ndev = netdev_notifier_info_to_dev(arg);
626 struct rxe_dev *rxe = net_to_rxe(ndev);
632 case NETDEV_UNREGISTER:
633 list_del(&rxe->list);
642 case NETDEV_CHANGEMTU:
643 pr_info("rxe: %s changed mtu to %d\n", ndev->name, ndev->mtu);
644 rxe_set_mtu(rxe, ndev->mtu);
648 case NETDEV_GOING_DOWN:
649 case NETDEV_CHANGEADDR:
650 case NETDEV_CHANGENAME:
651 case NETDEV_FEAT_CHANGE:
653 pr_info("rxe: ignoring netdev event = %ld for %s\n",
661 static struct notifier_block rxe_net_notifier = {
662 .notifier_call = rxe_notify,
665 int rxe_net_init(void)
669 spin_lock_init(&dev_list_lock);
671 recv_sockets.sk6 = rxe_setup_udp_tunnel(&init_net,
672 htons(ROCE_V2_UDP_DPORT), true);
673 if (IS_ERR(recv_sockets.sk6)) {
674 recv_sockets.sk6 = NULL;
675 pr_err("rxe: Failed to create IPv6 UDP tunnel\n");
679 recv_sockets.sk4 = rxe_setup_udp_tunnel(&init_net,
680 htons(ROCE_V2_UDP_DPORT), false);
681 if (IS_ERR(recv_sockets.sk4)) {
682 rxe_release_udp_tunnel(recv_sockets.sk6);
683 recv_sockets.sk4 = NULL;
684 recv_sockets.sk6 = NULL;
685 pr_err("rxe: Failed to create IPv4 UDP tunnel\n");
689 err = register_netdevice_notifier(&rxe_net_notifier);
691 rxe_release_udp_tunnel(recv_sockets.sk6);
692 rxe_release_udp_tunnel(recv_sockets.sk4);
693 pr_err("rxe: Failed to rigister netdev notifier\n");
699 void rxe_net_exit(void)
701 if (recv_sockets.sk6)
702 rxe_release_udp_tunnel(recv_sockets.sk6);
704 if (recv_sockets.sk4)
705 rxe_release_udp_tunnel(recv_sockets.sk4);
707 unregister_netdevice_notifier(&rxe_net_notifier);