1 /*
2  * Copyright (c) 2016 Mellanox Technologies Ltd. All rights reserved.
3  * Copyright (c) 2015 System Fabric Works, Inc. All rights reserved.
4  *
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:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *	- Redistributions of source code must retain the above
16  *	  copyright notice, this list of conditions and the following
17  *	  disclaimer.
18  *
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.
23  *
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
31  * SOFTWARE.
32  */
33 
34 #include <linux/skbuff.h>
35 #include <linux/if_arp.h>
36 #include <linux/netdevice.h>
37 #include <linux/if.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>
43 
44 #include "rxe.h"
45 #include "rxe_net.h"
46 #include "rxe_loc.h"
47 
48 static LIST_HEAD(rxe_dev_list);
49 static spinlock_t dev_list_lock; /* spinlock for device list */
50 
51 struct rxe_dev *net_to_rxe(struct net_device *ndev)
52 {
53 	struct rxe_dev *rxe;
54 	struct rxe_dev *found = NULL;
55 
56 	spin_lock_bh(&dev_list_lock);
57 	list_for_each_entry(rxe, &rxe_dev_list, list) {
58 		if (rxe->ndev == ndev) {
59 			found = rxe;
60 			break;
61 		}
62 	}
63 	spin_unlock_bh(&dev_list_lock);
64 
65 	return found;
66 }
67 
68 struct rxe_dev *get_rxe_by_name(const char* name)
69 {
70 	struct rxe_dev *rxe;
71 	struct rxe_dev *found = NULL;
72 
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)) {
76 			found = rxe;
77 			break;
78 		}
79 	}
80 	spin_unlock_bh(&dev_list_lock);
81 	return found;
82 }
83 
84 
85 struct rxe_recv_sockets recv_sockets;
86 
87 static __be64 rxe_mac_to_eui64(struct net_device *ndev)
88 {
89 	unsigned char *mac_addr = ndev->dev_addr;
90 	__be64 eui64;
91 	unsigned char *dst = (unsigned char *)&eui64;
92 
93 	dst[0] = mac_addr[0] ^ 2;
94 	dst[1] = mac_addr[1];
95 	dst[2] = mac_addr[2];
96 	dst[3] = 0xff;
97 	dst[4] = 0xfe;
98 	dst[5] = mac_addr[3];
99 	dst[6] = mac_addr[4];
100 	dst[7] = mac_addr[5];
101 
102 	return eui64;
103 }
104 
105 static __be64 node_guid(struct rxe_dev *rxe)
106 {
107 	return rxe_mac_to_eui64(rxe->ndev);
108 }
109 
110 static __be64 port_guid(struct rxe_dev *rxe)
111 {
112 	return rxe_mac_to_eui64(rxe->ndev);
113 }
114 
115 static struct device *dma_device(struct rxe_dev *rxe)
116 {
117 	struct net_device *ndev;
118 
119 	ndev = rxe->ndev;
120 
121 	if (ndev->priv_flags & IFF_802_1Q_VLAN)
122 		ndev = vlan_dev_real_dev(ndev);
123 
124 	return ndev->dev.parent;
125 }
126 
127 static int mcast_add(struct rxe_dev *rxe, union ib_gid *mgid)
128 {
129 	int err;
130 	unsigned char ll_addr[ETH_ALEN];
131 
132 	ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr);
133 	err = dev_mc_add(rxe->ndev, ll_addr);
134 
135 	return err;
136 }
137 
138 static int mcast_delete(struct rxe_dev *rxe, union ib_gid *mgid)
139 {
140 	int err;
141 	unsigned char ll_addr[ETH_ALEN];
142 
143 	ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr);
144 	err = dev_mc_del(rxe->ndev, ll_addr);
145 
146 	return err;
147 }
148 
149 static struct dst_entry *rxe_find_route4(struct net_device *ndev,
150 				  struct in_addr *saddr,
151 				  struct in_addr *daddr)
152 {
153 	struct rtable *rt;
154 	struct flowi4 fl = { { 0 } };
155 
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;
161 
162 	rt = ip_route_output_key(&init_net, &fl);
163 	if (IS_ERR(rt)) {
164 		pr_err_ratelimited("no route to %pI4\n", &daddr->s_addr);
165 		return NULL;
166 	}
167 
168 	return &rt->dst;
169 }
170 
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)
175 {
176 	struct dst_entry *ndst;
177 	struct flowi6 fl6 = { { 0 } };
178 
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;
184 
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);
188 		goto put;
189 	}
190 
191 	if (unlikely(ndst->error)) {
192 		pr_err("no route to %pI6\n", daddr);
193 		goto put;
194 	}
195 
196 	return ndst;
197 put:
198 	dst_release(ndst);
199 	return NULL;
200 }
201 
202 #else
203 
204 static struct dst_entry *rxe_find_route6(struct net_device *ndev,
205 					 struct in6_addr *saddr,
206 					 struct in6_addr *daddr)
207 {
208 	return NULL;
209 }
210 
211 #endif
212 
213 static int rxe_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
214 {
215 	struct udphdr *udph;
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);
219 
220 	if (!rxe)
221 		goto drop;
222 
223 	if (skb_linearize(skb)) {
224 		pr_err("skb_linearize failed\n");
225 		goto drop;
226 	}
227 
228 	udph = udp_hdr(skb);
229 	pkt->rxe = rxe;
230 	pkt->port_num = 1;
231 	pkt->hdr = (u8 *)(udph + 1);
232 	pkt->mask = RXE_GRH_MASK;
233 	pkt->paylen = be16_to_cpu(udph->len) - sizeof(*udph);
234 
235 	return rxe_rcv(skb);
236 drop:
237 	kfree_skb(skb);
238 	return 0;
239 }
240 
241 static struct socket *rxe_setup_udp_tunnel(struct net *net, __be16 port,
242 					   bool ipv6)
243 {
244 	int err;
245 	struct socket *sock;
246 	struct udp_port_cfg udp_cfg;
247 	struct udp_tunnel_sock_cfg tnl_cfg;
248 
249 	memset(&udp_cfg, 0, sizeof(udp_cfg));
250 
251 	if (ipv6) {
252 		udp_cfg.family = AF_INET6;
253 		udp_cfg.ipv6_v6only = 1;
254 	} else {
255 		udp_cfg.family = AF_INET;
256 	}
257 
258 	udp_cfg.local_udp_port = port;
259 
260 	/* Create UDP socket */
261 	err = udp_sock_create(net, &udp_cfg, &sock);
262 	if (err < 0) {
263 		pr_err("failed to create udp socket. err = %d\n", err);
264 		return ERR_PTR(err);
265 	}
266 
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;
271 
272 	/* Setup UDP tunnel */
273 	setup_udp_tunnel_sock(net, sock, &tnl_cfg);
274 
275 	return sock;
276 }
277 
278 static void rxe_release_udp_tunnel(struct socket *sk)
279 {
280 	udp_tunnel_sock_release(sk);
281 }
282 
283 static void prepare_udp_hdr(struct sk_buff *skb, __be16 src_port,
284 			    __be16 dst_port)
285 {
286 	struct udphdr *udph;
287 
288 	__skb_push(skb, sizeof(*udph));
289 	skb_reset_transport_header(skb);
290 	udph = udp_hdr(skb);
291 
292 	udph->dest = dst_port;
293 	udph->source = src_port;
294 	udph->len = htons(skb->len);
295 	udph->check = 0;
296 }
297 
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)
301 {
302 	struct iphdr *iph;
303 
304 	skb_scrub_packet(skb, xnet);
305 
306 	skb_clear_hash(skb);
307 	skb_dst_set(skb, dst);
308 	memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
309 
310 	skb_push(skb, sizeof(struct iphdr));
311 	skb_reset_network_header(skb);
312 
313 	iph = ip_hdr(skb);
314 
315 	iph->version	=	IPVERSION;
316 	iph->ihl	=	sizeof(struct iphdr) >> 2;
317 	iph->frag_off	=	df;
318 	iph->protocol	=	proto;
319 	iph->tos	=	tos;
320 	iph->daddr	=	daddr;
321 	iph->saddr	=	saddr;
322 	iph->ttl	=	ttl;
323 	__ip_select_ident(dev_net(dst->dev), iph,
324 			  skb_shinfo(skb)->gso_segs ?: 1);
325 	iph->tot_len = htons(skb->len);
326 	ip_send_check(iph);
327 }
328 
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)
332 {
333 	struct ipv6hdr *ip6h;
334 
335 	memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
336 	IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED
337 			    | IPSKB_REROUTED);
338 	skb_dst_set(skb, dst);
339 
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));
350 }
351 
352 static int prepare4(struct rxe_dev *rxe, struct sk_buff *skb, struct rxe_av *av)
353 {
354 	struct dst_entry *dst;
355 	bool xnet = false;
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);
360 
361 	dst = rxe_find_route4(rxe->ndev, saddr, daddr);
362 	if (!dst) {
363 		pr_err("Host not reachable\n");
364 		return -EHOSTUNREACH;
365 	}
366 
367 	if (!memcmp(saddr, daddr, sizeof(*daddr)))
368 		pkt->mask |= RXE_LOOPBACK_MASK;
369 
370 	prepare_udp_hdr(skb, htons(RXE_ROCE_V2_SPORT),
371 			htons(ROCE_V2_UDP_DPORT));
372 
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);
375 	return 0;
376 }
377 
378 static int prepare6(struct rxe_dev *rxe, struct sk_buff *skb, struct rxe_av *av)
379 {
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);
384 
385 	dst = rxe_find_route6(rxe->ndev, saddr, daddr);
386 	if (!dst) {
387 		pr_err("Host not reachable\n");
388 		return -EHOSTUNREACH;
389 	}
390 
391 	if (!memcmp(saddr, daddr, sizeof(*daddr)))
392 		pkt->mask |= RXE_LOOPBACK_MASK;
393 
394 	prepare_udp_hdr(skb, htons(RXE_ROCE_V2_SPORT),
395 			htons(ROCE_V2_UDP_DPORT));
396 
397 	prepare_ipv6_hdr(dst, skb, saddr, daddr, IPPROTO_UDP,
398 			 av->grh.traffic_class,
399 			 av->grh.hop_limit);
400 	return 0;
401 }
402 
403 static int prepare(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
404 		   struct sk_buff *skb, u32 *crc)
405 {
406 	int err = 0;
407 	struct rxe_av *av = rxe_get_av(pkt);
408 
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);
413 
414 	*crc = rxe_icrc_hdr(pkt, skb);
415 
416 	return err;
417 }
418 
419 static void rxe_skb_tx_dtor(struct sk_buff *skb)
420 {
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);
424 
425 	if (unlikely(qp->need_req_skb &&
426 		     skb_out < RXE_INFLIGHT_SKBS_PER_QP_LOW))
427 		rxe_run_task(&qp->req.task, 1);
428 }
429 
430 static int send(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
431 		struct sk_buff *skb)
432 {
433 	struct sk_buff *nskb;
434 	struct rxe_av *av;
435 	int err;
436 
437 	av = rxe_get_av(pkt);
438 
439 	nskb = skb_clone(skb, GFP_ATOMIC);
440 	if (!nskb)
441 		return -ENOMEM;
442 
443 	nskb->destructor = rxe_skb_tx_dtor;
444 	nskb->sk = pkt->qp->sk->sk;
445 
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);
450 	} else {
451 		pr_err("Unknown layer 3 protocol: %d\n", av->network_type);
452 		kfree_skb(nskb);
453 		return -EINVAL;
454 	}
455 
456 	if (unlikely(net_xmit_eval(err))) {
457 		pr_debug("error sending packet: %d\n", err);
458 		return -EAGAIN;
459 	}
460 
461 	kfree_skb(skb);
462 
463 	return 0;
464 }
465 
466 static int loopback(struct sk_buff *skb)
467 {
468 	return rxe_rcv(skb);
469 }
470 
471 static inline int addr_same(struct rxe_dev *rxe, struct rxe_av *av)
472 {
473 	return rxe->port.port_guid == av->grh.dgid.global.interface_id;
474 }
475 
476 static struct sk_buff *init_packet(struct rxe_dev *rxe, struct rxe_av *av,
477 				   int paylen, struct rxe_pkt_info *pkt)
478 {
479 	unsigned int hdr_len;
480 	struct sk_buff *skb;
481 
482 	if (av->network_type == RDMA_NETWORK_IPV4)
483 		hdr_len = ETH_HLEN + sizeof(struct udphdr) +
484 			sizeof(struct iphdr);
485 	else
486 		hdr_len = ETH_HLEN + sizeof(struct udphdr) +
487 			sizeof(struct ipv6hdr);
488 
489 	skb = alloc_skb(paylen + hdr_len + LL_RESERVED_SPACE(rxe->ndev),
490 			GFP_ATOMIC);
491 	if (unlikely(!skb))
492 		return NULL;
493 
494 	skb_reserve(skb, hdr_len + LL_RESERVED_SPACE(rxe->ndev));
495 
496 	skb->dev	= rxe->ndev;
497 	if (av->network_type == RDMA_NETWORK_IPV4)
498 		skb->protocol = htons(ETH_P_IP);
499 	else
500 		skb->protocol = htons(ETH_P_IPV6);
501 
502 	pkt->rxe	= rxe;
503 	pkt->port_num	= 1;
504 	pkt->hdr	= skb_put(skb, paylen);
505 	pkt->mask	|= RXE_GRH_MASK;
506 
507 	memset(pkt->hdr, 0, paylen);
508 
509 	return skb;
510 }
511 
512 /*
513  * this is required by rxe_cfg to match rxe devices in
514  * /sys/class/infiniband up with their underlying ethernet devices
515  */
516 static char *parent_name(struct rxe_dev *rxe, unsigned int port_num)
517 {
518 	return rxe->ndev->name;
519 }
520 
521 static enum rdma_link_layer link_layer(struct rxe_dev *rxe,
522 				       unsigned int port_num)
523 {
524 	return IB_LINK_LAYER_ETHERNET;
525 }
526 
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,
533 	.prepare	= prepare,
534 	.send		= send,
535 	.loopback	= loopback,
536 	.init_packet	= init_packet,
537 	.parent_name	= parent_name,
538 	.link_layer	= link_layer,
539 };
540 
541 struct rxe_dev *rxe_net_add(struct net_device *ndev)
542 {
543 	int err;
544 	struct rxe_dev *rxe = NULL;
545 
546 	rxe = (struct rxe_dev *)ib_alloc_device(sizeof(*rxe));
547 	if (!rxe)
548 		return NULL;
549 
550 	rxe->ifc_ops = &ifc_ops;
551 	rxe->ndev = ndev;
552 
553 	err = rxe_add(rxe, ndev->mtu);
554 	if (err) {
555 		ib_dealloc_device(&rxe->ib_dev);
556 		return NULL;
557 	}
558 
559 	spin_lock_bh(&dev_list_lock);
560 	list_add_tail(&rxe_dev_list, &rxe->list);
561 	spin_unlock_bh(&dev_list_lock);
562 	return rxe;
563 }
564 
565 void rxe_remove_all(void)
566 {
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);
571 
572 		list_del(&rxe->list);
573 		spin_unlock_bh(&dev_list_lock);
574 		rxe_remove(rxe);
575 		spin_lock_bh(&dev_list_lock);
576 	}
577 	spin_unlock_bh(&dev_list_lock);
578 }
579 EXPORT_SYMBOL(rxe_remove_all);
580 
581 static void rxe_port_event(struct rxe_dev *rxe,
582 			   enum ib_event_type event)
583 {
584 	struct ib_event ev;
585 
586 	ev.device = &rxe->ib_dev;
587 	ev.element.port_num = 1;
588 	ev.event = event;
589 
590 	ib_dispatch_event(&ev);
591 }
592 
593 /* Caller must hold net_info_lock */
594 void rxe_port_up(struct rxe_dev *rxe)
595 {
596 	struct rxe_port *port;
597 
598 	port = &rxe->port;
599 	port->attr.state = IB_PORT_ACTIVE;
600 	port->attr.phys_state = IB_PHYS_STATE_LINK_UP;
601 
602 	rxe_port_event(rxe, IB_EVENT_PORT_ACTIVE);
603 	pr_info("rxe: set %s active\n", rxe->ib_dev.name);
604 	return;
605 }
606 
607 /* Caller must hold net_info_lock */
608 void rxe_port_down(struct rxe_dev *rxe)
609 {
610 	struct rxe_port *port;
611 
612 	port = &rxe->port;
613 	port->attr.state = IB_PORT_DOWN;
614 	port->attr.phys_state = IB_PHYS_STATE_LINK_DOWN;
615 
616 	rxe_port_event(rxe, IB_EVENT_PORT_ERR);
617 	pr_info("rxe: set %s down\n", rxe->ib_dev.name);
618 	return;
619 }
620 
621 static int rxe_notify(struct notifier_block *not_blk,
622 		      unsigned long event,
623 		      void *arg)
624 {
625 	struct net_device *ndev = netdev_notifier_info_to_dev(arg);
626 	struct rxe_dev *rxe = net_to_rxe(ndev);
627 
628 	if (!rxe)
629 		goto out;
630 
631 	switch (event) {
632 	case NETDEV_UNREGISTER:
633 		list_del(&rxe->list);
634 		rxe_remove(rxe);
635 		break;
636 	case NETDEV_UP:
637 		rxe_port_up(rxe);
638 		break;
639 	case NETDEV_DOWN:
640 		rxe_port_down(rxe);
641 		break;
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);
645 		break;
646 	case NETDEV_REBOOT:
647 	case NETDEV_CHANGE:
648 	case NETDEV_GOING_DOWN:
649 	case NETDEV_CHANGEADDR:
650 	case NETDEV_CHANGENAME:
651 	case NETDEV_FEAT_CHANGE:
652 	default:
653 		pr_info("rxe: ignoring netdev event = %ld for %s\n",
654 			event, ndev->name);
655 		break;
656 	}
657 out:
658 	return NOTIFY_OK;
659 }
660 
661 static struct notifier_block rxe_net_notifier = {
662 	.notifier_call = rxe_notify,
663 };
664 
665 int rxe_net_init(void)
666 {
667 	int err;
668 
669 	spin_lock_init(&dev_list_lock);
670 
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");
676 		return -1;
677 	}
678 
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");
686 		return -1;
687 	}
688 
689 	err = register_netdevice_notifier(&rxe_net_notifier);
690 	if (err) {
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");
694 	}
695 
696 	return err;
697 }
698 
699 void rxe_net_exit(void)
700 {
701 	if (recv_sockets.sk6)
702 		rxe_release_udp_tunnel(recv_sockets.sk6);
703 
704 	if (recv_sockets.sk4)
705 		rxe_release_udp_tunnel(recv_sockets.sk4);
706 
707 	unregister_netdevice_notifier(&rxe_net_notifier);
708 }
709