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 DEFINE_SPINLOCK(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 = {0};
247 	struct udp_tunnel_sock_cfg tnl_cfg = {0};
248 
249 	if (ipv6) {
250 		udp_cfg.family = AF_INET6;
251 		udp_cfg.ipv6_v6only = 1;
252 	} else {
253 		udp_cfg.family = AF_INET;
254 	}
255 
256 	udp_cfg.local_udp_port = port;
257 
258 	/* Create UDP socket */
259 	err = udp_sock_create(net, &udp_cfg, &sock);
260 	if (err < 0) {
261 		pr_err("failed to create udp socket. err = %d\n", err);
262 		return ERR_PTR(err);
263 	}
264 
265 	tnl_cfg.encap_type = 1;
266 	tnl_cfg.encap_rcv = rxe_udp_encap_recv;
267 
268 	/* Setup UDP tunnel */
269 	setup_udp_tunnel_sock(net, sock, &tnl_cfg);
270 
271 	return sock;
272 }
273 
274 void rxe_release_udp_tunnel(struct socket *sk)
275 {
276 	if (sk)
277 		udp_tunnel_sock_release(sk);
278 }
279 
280 static void prepare_udp_hdr(struct sk_buff *skb, __be16 src_port,
281 			    __be16 dst_port)
282 {
283 	struct udphdr *udph;
284 
285 	__skb_push(skb, sizeof(*udph));
286 	skb_reset_transport_header(skb);
287 	udph = udp_hdr(skb);
288 
289 	udph->dest = dst_port;
290 	udph->source = src_port;
291 	udph->len = htons(skb->len);
292 	udph->check = 0;
293 }
294 
295 static void prepare_ipv4_hdr(struct dst_entry *dst, struct sk_buff *skb,
296 			     __be32 saddr, __be32 daddr, __u8 proto,
297 			     __u8 tos, __u8 ttl, __be16 df, bool xnet)
298 {
299 	struct iphdr *iph;
300 
301 	skb_scrub_packet(skb, xnet);
302 
303 	skb_clear_hash(skb);
304 	skb_dst_set(skb, dst);
305 	memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
306 
307 	skb_push(skb, sizeof(struct iphdr));
308 	skb_reset_network_header(skb);
309 
310 	iph = ip_hdr(skb);
311 
312 	iph->version	=	IPVERSION;
313 	iph->ihl	=	sizeof(struct iphdr) >> 2;
314 	iph->frag_off	=	df;
315 	iph->protocol	=	proto;
316 	iph->tos	=	tos;
317 	iph->daddr	=	daddr;
318 	iph->saddr	=	saddr;
319 	iph->ttl	=	ttl;
320 	__ip_select_ident(dev_net(dst->dev), iph,
321 			  skb_shinfo(skb)->gso_segs ?: 1);
322 	iph->tot_len = htons(skb->len);
323 	ip_send_check(iph);
324 }
325 
326 static void prepare_ipv6_hdr(struct dst_entry *dst, struct sk_buff *skb,
327 			     struct in6_addr *saddr, struct in6_addr *daddr,
328 			     __u8 proto, __u8 prio, __u8 ttl)
329 {
330 	struct ipv6hdr *ip6h;
331 
332 	memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
333 	IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED
334 			    | IPSKB_REROUTED);
335 	skb_dst_set(skb, dst);
336 
337 	__skb_push(skb, sizeof(*ip6h));
338 	skb_reset_network_header(skb);
339 	ip6h		  = ipv6_hdr(skb);
340 	ip6_flow_hdr(ip6h, prio, htonl(0));
341 	ip6h->payload_len = htons(skb->len);
342 	ip6h->nexthdr     = proto;
343 	ip6h->hop_limit   = ttl;
344 	ip6h->daddr	  = *daddr;
345 	ip6h->saddr	  = *saddr;
346 	ip6h->payload_len = htons(skb->len - sizeof(*ip6h));
347 }
348 
349 static int prepare4(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
350 		    struct sk_buff *skb, struct rxe_av *av)
351 {
352 	struct dst_entry *dst;
353 	bool xnet = false;
354 	__be16 df = htons(IP_DF);
355 	struct in_addr *saddr = &av->sgid_addr._sockaddr_in.sin_addr;
356 	struct in_addr *daddr = &av->dgid_addr._sockaddr_in.sin_addr;
357 
358 	dst = rxe_find_route4(rxe->ndev, saddr, daddr);
359 	if (!dst) {
360 		pr_err("Host not reachable\n");
361 		return -EHOSTUNREACH;
362 	}
363 
364 	if (!memcmp(saddr, daddr, sizeof(*daddr)))
365 		pkt->mask |= RXE_LOOPBACK_MASK;
366 
367 	prepare_udp_hdr(skb, htons(RXE_ROCE_V2_SPORT),
368 			htons(ROCE_V2_UDP_DPORT));
369 
370 	prepare_ipv4_hdr(dst, skb, saddr->s_addr, daddr->s_addr, IPPROTO_UDP,
371 			 av->grh.traffic_class, av->grh.hop_limit, df, xnet);
372 	return 0;
373 }
374 
375 static int prepare6(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
376 		    struct sk_buff *skb, struct rxe_av *av)
377 {
378 	struct dst_entry *dst;
379 	struct in6_addr *saddr = &av->sgid_addr._sockaddr_in6.sin6_addr;
380 	struct in6_addr *daddr = &av->dgid_addr._sockaddr_in6.sin6_addr;
381 
382 	dst = rxe_find_route6(rxe->ndev, saddr, daddr);
383 	if (!dst) {
384 		pr_err("Host not reachable\n");
385 		return -EHOSTUNREACH;
386 	}
387 
388 	if (!memcmp(saddr, daddr, sizeof(*daddr)))
389 		pkt->mask |= RXE_LOOPBACK_MASK;
390 
391 	prepare_udp_hdr(skb, htons(RXE_ROCE_V2_SPORT),
392 			htons(ROCE_V2_UDP_DPORT));
393 
394 	prepare_ipv6_hdr(dst, skb, saddr, daddr, IPPROTO_UDP,
395 			 av->grh.traffic_class,
396 			 av->grh.hop_limit);
397 	return 0;
398 }
399 
400 static int prepare(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
401 		   struct sk_buff *skb, u32 *crc)
402 {
403 	int err = 0;
404 	struct rxe_av *av = rxe_get_av(pkt);
405 
406 	if (av->network_type == RDMA_NETWORK_IPV4)
407 		err = prepare4(rxe, pkt, skb, av);
408 	else if (av->network_type == RDMA_NETWORK_IPV6)
409 		err = prepare6(rxe, pkt, skb, av);
410 
411 	*crc = rxe_icrc_hdr(pkt, skb);
412 
413 	return err;
414 }
415 
416 static void rxe_skb_tx_dtor(struct sk_buff *skb)
417 {
418 	struct sock *sk = skb->sk;
419 	struct rxe_qp *qp = sk->sk_user_data;
420 	int skb_out = atomic_dec_return(&qp->skb_out);
421 
422 	if (unlikely(qp->need_req_skb &&
423 		     skb_out < RXE_INFLIGHT_SKBS_PER_QP_LOW))
424 		rxe_run_task(&qp->req.task, 1);
425 }
426 
427 static int send(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
428 		struct sk_buff *skb)
429 {
430 	struct sk_buff *nskb;
431 	struct rxe_av *av;
432 	int err;
433 
434 	av = rxe_get_av(pkt);
435 
436 	nskb = skb_clone(skb, GFP_ATOMIC);
437 	if (!nskb)
438 		return -ENOMEM;
439 
440 	nskb->destructor = rxe_skb_tx_dtor;
441 	nskb->sk = pkt->qp->sk->sk;
442 
443 	if (av->network_type == RDMA_NETWORK_IPV4) {
444 		err = ip_local_out(dev_net(skb_dst(skb)->dev), nskb->sk, nskb);
445 	} else if (av->network_type == RDMA_NETWORK_IPV6) {
446 		err = ip6_local_out(dev_net(skb_dst(skb)->dev), nskb->sk, nskb);
447 	} else {
448 		pr_err("Unknown layer 3 protocol: %d\n", av->network_type);
449 		kfree_skb(nskb);
450 		return -EINVAL;
451 	}
452 
453 	if (unlikely(net_xmit_eval(err))) {
454 		pr_debug("error sending packet: %d\n", err);
455 		return -EAGAIN;
456 	}
457 
458 	atomic_inc(&pkt->qp->skb_out);
459 	kfree_skb(skb);
460 
461 	return 0;
462 }
463 
464 static int loopback(struct sk_buff *skb)
465 {
466 	return rxe_rcv(skb);
467 }
468 
469 static inline int addr_same(struct rxe_dev *rxe, struct rxe_av *av)
470 {
471 	return rxe->port.port_guid == av->grh.dgid.global.interface_id;
472 }
473 
474 static struct sk_buff *init_packet(struct rxe_dev *rxe, struct rxe_av *av,
475 				   int paylen, struct rxe_pkt_info *pkt)
476 {
477 	unsigned int hdr_len;
478 	struct sk_buff *skb;
479 
480 	if (av->network_type == RDMA_NETWORK_IPV4)
481 		hdr_len = ETH_HLEN + sizeof(struct udphdr) +
482 			sizeof(struct iphdr);
483 	else
484 		hdr_len = ETH_HLEN + sizeof(struct udphdr) +
485 			sizeof(struct ipv6hdr);
486 
487 	skb = alloc_skb(paylen + hdr_len + LL_RESERVED_SPACE(rxe->ndev),
488 			GFP_ATOMIC);
489 	if (unlikely(!skb))
490 		return NULL;
491 
492 	skb_reserve(skb, hdr_len + LL_RESERVED_SPACE(rxe->ndev));
493 
494 	skb->dev	= rxe->ndev;
495 	if (av->network_type == RDMA_NETWORK_IPV4)
496 		skb->protocol = htons(ETH_P_IP);
497 	else
498 		skb->protocol = htons(ETH_P_IPV6);
499 
500 	pkt->rxe	= rxe;
501 	pkt->port_num	= 1;
502 	pkt->hdr	= skb_put(skb, paylen);
503 	pkt->mask	|= RXE_GRH_MASK;
504 
505 	memset(pkt->hdr, 0, paylen);
506 
507 	return skb;
508 }
509 
510 /*
511  * this is required by rxe_cfg to match rxe devices in
512  * /sys/class/infiniband up with their underlying ethernet devices
513  */
514 static char *parent_name(struct rxe_dev *rxe, unsigned int port_num)
515 {
516 	return rxe->ndev->name;
517 }
518 
519 static enum rdma_link_layer link_layer(struct rxe_dev *rxe,
520 				       unsigned int port_num)
521 {
522 	return IB_LINK_LAYER_ETHERNET;
523 }
524 
525 static struct rxe_ifc_ops ifc_ops = {
526 	.node_guid	= node_guid,
527 	.port_guid	= port_guid,
528 	.dma_device	= dma_device,
529 	.mcast_add	= mcast_add,
530 	.mcast_delete	= mcast_delete,
531 	.prepare	= prepare,
532 	.send		= send,
533 	.loopback	= loopback,
534 	.init_packet	= init_packet,
535 	.parent_name	= parent_name,
536 	.link_layer	= link_layer,
537 };
538 
539 struct rxe_dev *rxe_net_add(struct net_device *ndev)
540 {
541 	int err;
542 	struct rxe_dev *rxe = NULL;
543 
544 	rxe = (struct rxe_dev *)ib_alloc_device(sizeof(*rxe));
545 	if (!rxe)
546 		return NULL;
547 
548 	rxe->ifc_ops = &ifc_ops;
549 	rxe->ndev = ndev;
550 
551 	err = rxe_add(rxe, ndev->mtu);
552 	if (err) {
553 		ib_dealloc_device(&rxe->ib_dev);
554 		return NULL;
555 	}
556 
557 	spin_lock_bh(&dev_list_lock);
558 	list_add_tail(&rxe_dev_list, &rxe->list);
559 	spin_unlock_bh(&dev_list_lock);
560 	return rxe;
561 }
562 
563 void rxe_remove_all(void)
564 {
565 	spin_lock_bh(&dev_list_lock);
566 	while (!list_empty(&rxe_dev_list)) {
567 		struct rxe_dev *rxe =
568 			list_first_entry(&rxe_dev_list, struct rxe_dev, list);
569 
570 		list_del(&rxe->list);
571 		spin_unlock_bh(&dev_list_lock);
572 		rxe_remove(rxe);
573 		spin_lock_bh(&dev_list_lock);
574 	}
575 	spin_unlock_bh(&dev_list_lock);
576 }
577 EXPORT_SYMBOL(rxe_remove_all);
578 
579 static void rxe_port_event(struct rxe_dev *rxe,
580 			   enum ib_event_type event)
581 {
582 	struct ib_event ev;
583 
584 	ev.device = &rxe->ib_dev;
585 	ev.element.port_num = 1;
586 	ev.event = event;
587 
588 	ib_dispatch_event(&ev);
589 }
590 
591 /* Caller must hold net_info_lock */
592 void rxe_port_up(struct rxe_dev *rxe)
593 {
594 	struct rxe_port *port;
595 
596 	port = &rxe->port;
597 	port->attr.state = IB_PORT_ACTIVE;
598 	port->attr.phys_state = IB_PHYS_STATE_LINK_UP;
599 
600 	rxe_port_event(rxe, IB_EVENT_PORT_ACTIVE);
601 	pr_info("set %s active\n", rxe->ib_dev.name);
602 }
603 
604 /* Caller must hold net_info_lock */
605 void rxe_port_down(struct rxe_dev *rxe)
606 {
607 	struct rxe_port *port;
608 
609 	port = &rxe->port;
610 	port->attr.state = IB_PORT_DOWN;
611 	port->attr.phys_state = IB_PHYS_STATE_LINK_DOWN;
612 
613 	rxe_port_event(rxe, IB_EVENT_PORT_ERR);
614 	pr_info("set %s down\n", rxe->ib_dev.name);
615 }
616 
617 static int rxe_notify(struct notifier_block *not_blk,
618 		      unsigned long event,
619 		      void *arg)
620 {
621 	struct net_device *ndev = netdev_notifier_info_to_dev(arg);
622 	struct rxe_dev *rxe = net_to_rxe(ndev);
623 
624 	if (!rxe)
625 		goto out;
626 
627 	switch (event) {
628 	case NETDEV_UNREGISTER:
629 		list_del(&rxe->list);
630 		rxe_remove(rxe);
631 		break;
632 	case NETDEV_UP:
633 		rxe_port_up(rxe);
634 		break;
635 	case NETDEV_DOWN:
636 		rxe_port_down(rxe);
637 		break;
638 	case NETDEV_CHANGEMTU:
639 		pr_info("%s changed mtu to %d\n", ndev->name, ndev->mtu);
640 		rxe_set_mtu(rxe, ndev->mtu);
641 		break;
642 	case NETDEV_REBOOT:
643 	case NETDEV_CHANGE:
644 	case NETDEV_GOING_DOWN:
645 	case NETDEV_CHANGEADDR:
646 	case NETDEV_CHANGENAME:
647 	case NETDEV_FEAT_CHANGE:
648 	default:
649 		pr_info("ignoring netdev event = %ld for %s\n",
650 			event, ndev->name);
651 		break;
652 	}
653 out:
654 	return NOTIFY_OK;
655 }
656 
657 struct notifier_block rxe_net_notifier = {
658 	.notifier_call = rxe_notify,
659 };
660 
661 int rxe_net_ipv4_init(void)
662 {
663 	recv_sockets.sk4 = rxe_setup_udp_tunnel(&init_net,
664 				htons(ROCE_V2_UDP_DPORT), false);
665 	if (IS_ERR(recv_sockets.sk4)) {
666 		recv_sockets.sk4 = NULL;
667 		pr_err("Failed to create IPv4 UDP tunnel\n");
668 		return -1;
669 	}
670 
671 	return 0;
672 }
673 
674 int rxe_net_ipv6_init(void)
675 {
676 #if IS_ENABLED(CONFIG_IPV6)
677 
678 	recv_sockets.sk6 = rxe_setup_udp_tunnel(&init_net,
679 						htons(ROCE_V2_UDP_DPORT), true);
680 	if (IS_ERR(recv_sockets.sk6)) {
681 		recv_sockets.sk6 = NULL;
682 		pr_err("Failed to create IPv6 UDP tunnel\n");
683 		return -1;
684 	}
685 #endif
686 	return 0;
687 }
688 
689 void rxe_net_exit(void)
690 {
691 	rxe_release_udp_tunnel(recv_sockets.sk6);
692 	rxe_release_udp_tunnel(recv_sockets.sk4);
693 	unregister_netdevice_notifier(&rxe_net_notifier);
694 }
695 
696 int rxe_net_init(void)
697 {
698 	int err;
699 
700 	recv_sockets.sk6 = NULL;
701 
702 	err = rxe_net_ipv4_init();
703 	if (err)
704 		return err;
705 	err = rxe_net_ipv6_init();
706 	if (err)
707 		goto err_out;
708 	err = register_netdevice_notifier(&rxe_net_notifier);
709 	if (err) {
710 		pr_err("Failed to register netdev notifier\n");
711 		goto err_out;
712 	}
713 	return 0;
714 err_out:
715 	rxe_net_exit();
716 	return err;
717 }
718