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 static struct rxe_recv_sockets recv_sockets;
86 
87 struct device *rxe_dma_device(struct rxe_dev *rxe)
88 {
89 	struct net_device *ndev;
90 
91 	ndev = rxe->ndev;
92 
93 	if (is_vlan_dev(ndev))
94 		ndev = vlan_dev_real_dev(ndev);
95 
96 	return ndev->dev.parent;
97 }
98 
99 int rxe_mcast_add(struct rxe_dev *rxe, union ib_gid *mgid)
100 {
101 	int err;
102 	unsigned char ll_addr[ETH_ALEN];
103 
104 	ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr);
105 	err = dev_mc_add(rxe->ndev, ll_addr);
106 
107 	return err;
108 }
109 
110 int rxe_mcast_delete(struct rxe_dev *rxe, union ib_gid *mgid)
111 {
112 	int err;
113 	unsigned char ll_addr[ETH_ALEN];
114 
115 	ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr);
116 	err = dev_mc_del(rxe->ndev, ll_addr);
117 
118 	return err;
119 }
120 
121 static struct dst_entry *rxe_find_route4(struct net_device *ndev,
122 				  struct in_addr *saddr,
123 				  struct in_addr *daddr)
124 {
125 	struct rtable *rt;
126 	struct flowi4 fl = { { 0 } };
127 
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;
133 
134 	rt = ip_route_output_key(&init_net, &fl);
135 	if (IS_ERR(rt)) {
136 		pr_err_ratelimited("no route to %pI4\n", &daddr->s_addr);
137 		return NULL;
138 	}
139 
140 	return &rt->dst;
141 }
142 
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)
147 {
148 	struct dst_entry *ndst;
149 	struct flowi6 fl6 = { { 0 } };
150 
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;
156 
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);
160 		goto put;
161 	}
162 
163 	if (unlikely(ndst->error)) {
164 		pr_err("no route to %pI6\n", daddr);
165 		goto put;
166 	}
167 
168 	return ndst;
169 put:
170 	dst_release(ndst);
171 	return NULL;
172 }
173 
174 #else
175 
176 static struct dst_entry *rxe_find_route6(struct net_device *ndev,
177 					 struct in6_addr *saddr,
178 					 struct in6_addr *daddr)
179 {
180 	return NULL;
181 }
182 
183 #endif
184 
185 static struct dst_entry *rxe_find_route(struct rxe_dev *rxe,
186 					struct rxe_qp *qp,
187 					struct rxe_av *av)
188 {
189 	const struct ib_gid_attr *attr;
190 	struct dst_entry *dst = NULL;
191 	struct net_device *ndev;
192 
193 	attr = rdma_get_gid_attr(&rxe->ib_dev, qp->attr.port_num,
194 				 av->grh.sgid_index);
195 	if (IS_ERR(attr))
196 		return NULL;
197 	ndev = attr->ndev;
198 
199 	if (qp_type(qp) == IB_QPT_RC)
200 		dst = sk_dst_get(qp->sk->sk);
201 
202 	if (!dst || !dst_check(dst, qp->dst_cookie)) {
203 		if (dst)
204 			dst_release(dst);
205 
206 		if (av->network_type == RDMA_NETWORK_IPV4) {
207 			struct in_addr *saddr;
208 			struct in_addr *daddr;
209 
210 			saddr = &av->sgid_addr._sockaddr_in.sin_addr;
211 			daddr = &av->dgid_addr._sockaddr_in.sin_addr;
212 			dst = rxe_find_route4(ndev, saddr, daddr);
213 		} else if (av->network_type == RDMA_NETWORK_IPV6) {
214 			struct in6_addr *saddr6;
215 			struct in6_addr *daddr6;
216 
217 			saddr6 = &av->sgid_addr._sockaddr_in6.sin6_addr;
218 			daddr6 = &av->dgid_addr._sockaddr_in6.sin6_addr;
219 			dst = rxe_find_route6(ndev, saddr6, daddr6);
220 #if IS_ENABLED(CONFIG_IPV6)
221 			if (dst)
222 				qp->dst_cookie =
223 					rt6_get_cookie((struct rt6_info *)dst);
224 #endif
225 		}
226 
227 		if (dst && (qp_type(qp) == IB_QPT_RC)) {
228 			dst_hold(dst);
229 			sk_dst_set(qp->sk->sk, dst);
230 		}
231 	}
232 	rdma_put_gid_attr(attr);
233 	return dst;
234 }
235 
236 static int rxe_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
237 {
238 	struct udphdr *udph;
239 	struct net_device *ndev = skb->dev;
240 	struct net_device *rdev = ndev;
241 	struct rxe_dev *rxe = net_to_rxe(ndev);
242 	struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
243 
244 	if (!rxe && is_vlan_dev(rdev)) {
245 		rdev = vlan_dev_real_dev(ndev);
246 		rxe = net_to_rxe(rdev);
247 	}
248 	if (!rxe)
249 		goto drop;
250 
251 	if (skb_linearize(skb)) {
252 		pr_err("skb_linearize failed\n");
253 		goto drop;
254 	}
255 
256 	udph = udp_hdr(skb);
257 	pkt->rxe = rxe;
258 	pkt->port_num = 1;
259 	pkt->hdr = (u8 *)(udph + 1);
260 	pkt->mask = RXE_GRH_MASK;
261 	pkt->paylen = be16_to_cpu(udph->len) - sizeof(*udph);
262 
263 	rxe_rcv(skb);
264 
265 	return 0;
266 drop:
267 	kfree_skb(skb);
268 
269 	return 0;
270 }
271 
272 static struct socket *rxe_setup_udp_tunnel(struct net *net, __be16 port,
273 					   bool ipv6)
274 {
275 	int err;
276 	struct socket *sock;
277 	struct udp_port_cfg udp_cfg = { };
278 	struct udp_tunnel_sock_cfg tnl_cfg = { };
279 
280 	if (ipv6) {
281 		udp_cfg.family = AF_INET6;
282 		udp_cfg.ipv6_v6only = 1;
283 	} else {
284 		udp_cfg.family = AF_INET;
285 	}
286 
287 	udp_cfg.local_udp_port = port;
288 
289 	/* Create UDP socket */
290 	err = udp_sock_create(net, &udp_cfg, &sock);
291 	if (err < 0) {
292 		pr_err("failed to create udp socket. err = %d\n", err);
293 		return ERR_PTR(err);
294 	}
295 
296 	tnl_cfg.encap_type = 1;
297 	tnl_cfg.encap_rcv = rxe_udp_encap_recv;
298 
299 	/* Setup UDP tunnel */
300 	setup_udp_tunnel_sock(net, sock, &tnl_cfg);
301 
302 	return sock;
303 }
304 
305 static void rxe_release_udp_tunnel(struct socket *sk)
306 {
307 	if (sk)
308 		udp_tunnel_sock_release(sk);
309 }
310 
311 static void prepare_udp_hdr(struct sk_buff *skb, __be16 src_port,
312 			    __be16 dst_port)
313 {
314 	struct udphdr *udph;
315 
316 	__skb_push(skb, sizeof(*udph));
317 	skb_reset_transport_header(skb);
318 	udph = udp_hdr(skb);
319 
320 	udph->dest = dst_port;
321 	udph->source = src_port;
322 	udph->len = htons(skb->len);
323 	udph->check = 0;
324 }
325 
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)
329 {
330 	struct iphdr *iph;
331 
332 	skb_scrub_packet(skb, xnet);
333 
334 	skb_clear_hash(skb);
335 	skb_dst_set(skb, dst_clone(dst));
336 	memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
337 
338 	skb_push(skb, sizeof(struct iphdr));
339 	skb_reset_network_header(skb);
340 
341 	iph = ip_hdr(skb);
342 
343 	iph->version	=	IPVERSION;
344 	iph->ihl	=	sizeof(struct iphdr) >> 2;
345 	iph->frag_off	=	df;
346 	iph->protocol	=	proto;
347 	iph->tos	=	tos;
348 	iph->daddr	=	daddr;
349 	iph->saddr	=	saddr;
350 	iph->ttl	=	ttl;
351 	__ip_select_ident(dev_net(dst->dev), iph,
352 			  skb_shinfo(skb)->gso_segs ?: 1);
353 	iph->tot_len = htons(skb->len);
354 	ip_send_check(iph);
355 }
356 
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)
360 {
361 	struct ipv6hdr *ip6h;
362 
363 	memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
364 	IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED
365 			    | IPSKB_REROUTED);
366 	skb_dst_set(skb, dst_clone(dst));
367 
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));
378 }
379 
380 static int prepare4(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
381 		    struct sk_buff *skb, struct rxe_av *av)
382 {
383 	struct rxe_qp *qp = pkt->qp;
384 	struct dst_entry *dst;
385 	bool xnet = false;
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;
389 
390 	dst = rxe_find_route(rxe, qp, av);
391 	if (!dst) {
392 		pr_err("Host not reachable\n");
393 		return -EHOSTUNREACH;
394 	}
395 
396 	if (!memcmp(saddr, daddr, sizeof(*daddr)))
397 		pkt->mask |= RXE_LOOPBACK_MASK;
398 
399 	prepare_udp_hdr(skb, htons(RXE_ROCE_V2_SPORT),
400 			htons(ROCE_V2_UDP_DPORT));
401 
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);
404 
405 	dst_release(dst);
406 	return 0;
407 }
408 
409 static int prepare6(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
410 		    struct sk_buff *skb, struct rxe_av *av)
411 {
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;
416 
417 	dst = rxe_find_route(rxe, qp, av);
418 	if (!dst) {
419 		pr_err("Host not reachable\n");
420 		return -EHOSTUNREACH;
421 	}
422 
423 	if (!memcmp(saddr, daddr, sizeof(*daddr)))
424 		pkt->mask |= RXE_LOOPBACK_MASK;
425 
426 	prepare_udp_hdr(skb, htons(RXE_ROCE_V2_SPORT),
427 			htons(ROCE_V2_UDP_DPORT));
428 
429 	prepare_ipv6_hdr(dst, skb, saddr, daddr, IPPROTO_UDP,
430 			 av->grh.traffic_class,
431 			 av->grh.hop_limit);
432 
433 	dst_release(dst);
434 	return 0;
435 }
436 
437 int rxe_prepare(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
438 		struct sk_buff *skb, u32 *crc)
439 {
440 	int err = 0;
441 	struct rxe_av *av = rxe_get_av(pkt);
442 
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);
447 
448 	*crc = rxe_icrc_hdr(pkt, skb);
449 
450 	return err;
451 }
452 
453 static void rxe_skb_tx_dtor(struct sk_buff *skb)
454 {
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);
458 
459 	if (unlikely(qp->need_req_skb &&
460 		     skb_out < RXE_INFLIGHT_SKBS_PER_QP_LOW))
461 		rxe_run_task(&qp->req.task, 1);
462 
463 	rxe_drop_ref(qp);
464 }
465 
466 int rxe_send(struct rxe_pkt_info *pkt, struct sk_buff *skb)
467 {
468 	struct rxe_av *av;
469 	int err;
470 
471 	av = rxe_get_av(pkt);
472 
473 	skb->destructor = rxe_skb_tx_dtor;
474 	skb->sk = pkt->qp->sk->sk;
475 
476 	rxe_add_ref(pkt->qp);
477 	atomic_inc(&pkt->qp->skb_out);
478 
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);
483 	} else {
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);
487 		kfree_skb(skb);
488 		return -EINVAL;
489 	}
490 
491 	if (unlikely(net_xmit_eval(err))) {
492 		pr_debug("error sending packet: %d\n", err);
493 		return -EAGAIN;
494 	}
495 
496 	return 0;
497 }
498 
499 void rxe_loopback(struct sk_buff *skb)
500 {
501 	rxe_rcv(skb);
502 }
503 
504 static inline int addr_same(struct rxe_dev *rxe, struct rxe_av *av)
505 {
506 	return rxe->port.port_guid == av->grh.dgid.global.interface_id;
507 }
508 
509 struct sk_buff *rxe_init_packet(struct rxe_dev *rxe, struct rxe_av *av,
510 				int paylen, struct rxe_pkt_info *pkt)
511 {
512 	unsigned int hdr_len;
513 	struct sk_buff *skb;
514 	struct net_device *ndev;
515 	const struct ib_gid_attr *attr;
516 	const int port_num = 1;
517 
518 	attr = rdma_get_gid_attr(&rxe->ib_dev, port_num, av->grh.sgid_index);
519 	if (IS_ERR(attr))
520 		return NULL;
521 	ndev = attr->ndev;
522 
523 	if (av->network_type == RDMA_NETWORK_IPV4)
524 		hdr_len = ETH_HLEN + sizeof(struct udphdr) +
525 			sizeof(struct iphdr);
526 	else
527 		hdr_len = ETH_HLEN + sizeof(struct udphdr) +
528 			sizeof(struct ipv6hdr);
529 
530 	skb = alloc_skb(paylen + hdr_len + LL_RESERVED_SPACE(ndev),
531 			GFP_ATOMIC);
532 
533 	if (unlikely(!skb))
534 		goto out;
535 
536 	skb_reserve(skb, hdr_len + LL_RESERVED_SPACE(rxe->ndev));
537 
538 	skb->dev	= ndev;
539 	if (av->network_type == RDMA_NETWORK_IPV4)
540 		skb->protocol = htons(ETH_P_IP);
541 	else
542 		skb->protocol = htons(ETH_P_IPV6);
543 
544 	pkt->rxe	= rxe;
545 	pkt->port_num	= port_num;
546 	pkt->hdr	= skb_put_zero(skb, paylen);
547 	pkt->mask	|= RXE_GRH_MASK;
548 
549 out:
550 	rdma_put_gid_attr(attr);
551 	return skb;
552 }
553 
554 /*
555  * this is required by rxe_cfg to match rxe devices in
556  * /sys/class/infiniband up with their underlying ethernet devices
557  */
558 const char *rxe_parent_name(struct rxe_dev *rxe, unsigned int port_num)
559 {
560 	return rxe->ndev->name;
561 }
562 
563 enum rdma_link_layer rxe_link_layer(struct rxe_dev *rxe, unsigned int port_num)
564 {
565 	return IB_LINK_LAYER_ETHERNET;
566 }
567 
568 struct rxe_dev *rxe_net_add(struct net_device *ndev)
569 {
570 	int err;
571 	struct rxe_dev *rxe = NULL;
572 
573 	rxe = (struct rxe_dev *)ib_alloc_device(sizeof(*rxe));
574 	if (!rxe)
575 		return NULL;
576 
577 	rxe->ndev = ndev;
578 
579 	err = rxe_add(rxe, ndev->mtu);
580 	if (err) {
581 		ib_dealloc_device(&rxe->ib_dev);
582 		return NULL;
583 	}
584 
585 	spin_lock_bh(&dev_list_lock);
586 	list_add_tail(&rxe->list, &rxe_dev_list);
587 	spin_unlock_bh(&dev_list_lock);
588 	return rxe;
589 }
590 
591 void rxe_remove_all(void)
592 {
593 	spin_lock_bh(&dev_list_lock);
594 	while (!list_empty(&rxe_dev_list)) {
595 		struct rxe_dev *rxe =
596 			list_first_entry(&rxe_dev_list, struct rxe_dev, list);
597 
598 		list_del(&rxe->list);
599 		spin_unlock_bh(&dev_list_lock);
600 		rxe_remove(rxe);
601 		spin_lock_bh(&dev_list_lock);
602 	}
603 	spin_unlock_bh(&dev_list_lock);
604 }
605 
606 static void rxe_port_event(struct rxe_dev *rxe,
607 			   enum ib_event_type event)
608 {
609 	struct ib_event ev;
610 
611 	ev.device = &rxe->ib_dev;
612 	ev.element.port_num = 1;
613 	ev.event = event;
614 
615 	ib_dispatch_event(&ev);
616 }
617 
618 /* Caller must hold net_info_lock */
619 void rxe_port_up(struct rxe_dev *rxe)
620 {
621 	struct rxe_port *port;
622 
623 	port = &rxe->port;
624 	port->attr.state = IB_PORT_ACTIVE;
625 	port->attr.phys_state = IB_PHYS_STATE_LINK_UP;
626 
627 	rxe_port_event(rxe, IB_EVENT_PORT_ACTIVE);
628 	pr_info("set %s active\n", rxe->ib_dev.name);
629 }
630 
631 /* Caller must hold net_info_lock */
632 void rxe_port_down(struct rxe_dev *rxe)
633 {
634 	struct rxe_port *port;
635 
636 	port = &rxe->port;
637 	port->attr.state = IB_PORT_DOWN;
638 	port->attr.phys_state = IB_PHYS_STATE_LINK_DOWN;
639 
640 	rxe_port_event(rxe, IB_EVENT_PORT_ERR);
641 	pr_info("set %s down\n", rxe->ib_dev.name);
642 }
643 
644 static int rxe_notify(struct notifier_block *not_blk,
645 		      unsigned long event,
646 		      void *arg)
647 {
648 	struct net_device *ndev = netdev_notifier_info_to_dev(arg);
649 	struct rxe_dev *rxe = net_to_rxe(ndev);
650 
651 	if (!rxe)
652 		goto out;
653 
654 	switch (event) {
655 	case NETDEV_UNREGISTER:
656 		list_del(&rxe->list);
657 		rxe_remove(rxe);
658 		break;
659 	case NETDEV_UP:
660 		rxe_port_up(rxe);
661 		break;
662 	case NETDEV_DOWN:
663 		rxe_port_down(rxe);
664 		break;
665 	case NETDEV_CHANGEMTU:
666 		pr_info("%s changed mtu to %d\n", ndev->name, ndev->mtu);
667 		rxe_set_mtu(rxe, ndev->mtu);
668 		break;
669 	case NETDEV_CHANGE:
670 		if (netif_running(ndev) && netif_carrier_ok(ndev))
671 			rxe_port_up(rxe);
672 		else
673 			rxe_port_down(rxe);
674 		break;
675 	case NETDEV_REBOOT:
676 	case NETDEV_GOING_DOWN:
677 	case NETDEV_CHANGEADDR:
678 	case NETDEV_CHANGENAME:
679 	case NETDEV_FEAT_CHANGE:
680 	default:
681 		pr_info("ignoring netdev event = %ld for %s\n",
682 			event, ndev->name);
683 		break;
684 	}
685 out:
686 	return NOTIFY_OK;
687 }
688 
689 static struct notifier_block rxe_net_notifier = {
690 	.notifier_call = rxe_notify,
691 };
692 
693 static int rxe_net_ipv4_init(void)
694 {
695 	recv_sockets.sk4 = rxe_setup_udp_tunnel(&init_net,
696 				htons(ROCE_V2_UDP_DPORT), false);
697 	if (IS_ERR(recv_sockets.sk4)) {
698 		recv_sockets.sk4 = NULL;
699 		pr_err("Failed to create IPv4 UDP tunnel\n");
700 		return -1;
701 	}
702 
703 	return 0;
704 }
705 
706 static int rxe_net_ipv6_init(void)
707 {
708 #if IS_ENABLED(CONFIG_IPV6)
709 
710 	recv_sockets.sk6 = rxe_setup_udp_tunnel(&init_net,
711 						htons(ROCE_V2_UDP_DPORT), true);
712 	if (IS_ERR(recv_sockets.sk6)) {
713 		recv_sockets.sk6 = NULL;
714 		pr_err("Failed to create IPv6 UDP tunnel\n");
715 		return -1;
716 	}
717 #endif
718 	return 0;
719 }
720 
721 void rxe_net_exit(void)
722 {
723 	rxe_release_udp_tunnel(recv_sockets.sk6);
724 	rxe_release_udp_tunnel(recv_sockets.sk4);
725 	unregister_netdevice_notifier(&rxe_net_notifier);
726 }
727 
728 int rxe_net_init(void)
729 {
730 	int err;
731 
732 	recv_sockets.sk6 = NULL;
733 
734 	err = rxe_net_ipv4_init();
735 	if (err)
736 		return err;
737 	err = rxe_net_ipv6_init();
738 	if (err)
739 		goto err_out;
740 	err = register_netdevice_notifier(&rxe_net_notifier);
741 	if (err) {
742 		pr_err("Failed to register netdev notifier\n");
743 		goto err_out;
744 	}
745 	return 0;
746 err_out:
747 	rxe_net_exit();
748 	return err;
749 }
750