1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 /*
3 * Copyright (c) 2016 Mellanox Technologies Ltd. All rights reserved.
4 * Copyright (c) 2015 System Fabric Works, Inc. All rights reserved.
5 */
6
7 #include <linux/skbuff.h>
8 #include <linux/if_arp.h>
9 #include <linux/netdevice.h>
10 #include <linux/if.h>
11 #include <linux/if_vlan.h>
12 #include <net/udp_tunnel.h>
13 #include <net/sch_generic.h>
14 #include <linux/netfilter.h>
15 #include <rdma/ib_addr.h>
16
17 #include "rxe.h"
18 #include "rxe_net.h"
19 #include "rxe_loc.h"
20
21 static struct rxe_recv_sockets recv_sockets;
22
23 #ifdef CONFIG_DEBUG_LOCK_ALLOC
24 /*
25 * lockdep can detect false positive circular dependencies
26 * when there are user-space socket API users or in kernel
27 * users switching between a tcp and rdma transport.
28 * Maybe also switching between siw and rxe may cause
29 * problems as per default sockets are only classified
30 * by family and not by ip protocol. And there might
31 * be different locks used between the application
32 * and the low level sockets.
33 *
34 * Problems were seen with ksmbd.ko and cifs.ko,
35 * switching transports, use git blame to find
36 * more details.
37 */
38 static struct lock_class_key rxe_recv_sk_key[2];
39 static struct lock_class_key rxe_recv_slock_key[2];
40 #endif /* CONFIG_DEBUG_LOCK_ALLOC */
41
rxe_reclassify_recv_socket(struct socket * sock)42 static inline void rxe_reclassify_recv_socket(struct socket *sock)
43 {
44 #ifdef CONFIG_DEBUG_LOCK_ALLOC
45 struct sock *sk = sock->sk;
46
47 if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
48 return;
49
50 switch (sk->sk_family) {
51 case AF_INET:
52 sock_lock_init_class_and_name(sk,
53 "slock-AF_INET-RDMA-RXE-RECV",
54 &rxe_recv_slock_key[0],
55 "sk_lock-AF_INET-RDMA-RXE-RECV",
56 &rxe_recv_sk_key[0]);
57 break;
58 case AF_INET6:
59 sock_lock_init_class_and_name(sk,
60 "slock-AF_INET6-RDMA-RXE-RECV",
61 &rxe_recv_slock_key[1],
62 "sk_lock-AF_INET6-RDMA-RXE-RECV",
63 &rxe_recv_sk_key[1]);
64 break;
65 default:
66 WARN_ON_ONCE(1);
67 }
68 #endif /* CONFIG_DEBUG_LOCK_ALLOC */
69 }
70
rxe_find_route4(struct rxe_qp * qp,struct net_device * ndev,struct in_addr * saddr,struct in_addr * daddr)71 static struct dst_entry *rxe_find_route4(struct rxe_qp *qp,
72 struct net_device *ndev,
73 struct in_addr *saddr,
74 struct in_addr *daddr)
75 {
76 struct rtable *rt;
77 struct flowi4 fl = { { 0 } };
78
79 memset(&fl, 0, sizeof(fl));
80 fl.flowi4_oif = ndev->ifindex;
81 memcpy(&fl.saddr, saddr, sizeof(*saddr));
82 memcpy(&fl.daddr, daddr, sizeof(*daddr));
83 fl.flowi4_proto = IPPROTO_UDP;
84
85 rt = ip_route_output_key(&init_net, &fl);
86 if (IS_ERR(rt)) {
87 rxe_dbg_qp(qp, "no route to %pI4\n", &daddr->s_addr);
88 return NULL;
89 }
90
91 return &rt->dst;
92 }
93
94 #if IS_ENABLED(CONFIG_IPV6)
rxe_find_route6(struct rxe_qp * qp,struct net_device * ndev,struct in6_addr * saddr,struct in6_addr * daddr)95 static struct dst_entry *rxe_find_route6(struct rxe_qp *qp,
96 struct net_device *ndev,
97 struct in6_addr *saddr,
98 struct in6_addr *daddr)
99 {
100 struct dst_entry *ndst;
101 struct flowi6 fl6 = { { 0 } };
102
103 memset(&fl6, 0, sizeof(fl6));
104 fl6.flowi6_oif = ndev->ifindex;
105 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
106 memcpy(&fl6.daddr, daddr, sizeof(*daddr));
107 fl6.flowi6_proto = IPPROTO_UDP;
108
109 ndst = ipv6_stub->ipv6_dst_lookup_flow(sock_net(recv_sockets.sk6->sk),
110 recv_sockets.sk6->sk, &fl6,
111 NULL);
112 if (IS_ERR(ndst)) {
113 rxe_dbg_qp(qp, "no route to %pI6\n", daddr);
114 return NULL;
115 }
116
117 if (unlikely(ndst->error)) {
118 rxe_dbg_qp(qp, "no route to %pI6\n", daddr);
119 goto put;
120 }
121
122 return ndst;
123 put:
124 dst_release(ndst);
125 return NULL;
126 }
127
128 #else
129
rxe_find_route6(struct rxe_qp * qp,struct net_device * ndev,struct in6_addr * saddr,struct in6_addr * daddr)130 static struct dst_entry *rxe_find_route6(struct rxe_qp *qp,
131 struct net_device *ndev,
132 struct in6_addr *saddr,
133 struct in6_addr *daddr)
134 {
135 return NULL;
136 }
137
138 #endif
139
rxe_find_route(struct net_device * ndev,struct rxe_qp * qp,struct rxe_av * av)140 static struct dst_entry *rxe_find_route(struct net_device *ndev,
141 struct rxe_qp *qp,
142 struct rxe_av *av)
143 {
144 struct dst_entry *dst = NULL;
145
146 if (qp_type(qp) == IB_QPT_RC)
147 dst = sk_dst_get(qp->sk->sk);
148
149 if (!dst || !dst_check(dst, qp->dst_cookie)) {
150 if (dst)
151 dst_release(dst);
152
153 if (av->network_type == RXE_NETWORK_TYPE_IPV4) {
154 struct in_addr *saddr;
155 struct in_addr *daddr;
156
157 saddr = &av->sgid_addr._sockaddr_in.sin_addr;
158 daddr = &av->dgid_addr._sockaddr_in.sin_addr;
159 dst = rxe_find_route4(qp, ndev, saddr, daddr);
160 } else if (av->network_type == RXE_NETWORK_TYPE_IPV6) {
161 struct in6_addr *saddr6;
162 struct in6_addr *daddr6;
163
164 saddr6 = &av->sgid_addr._sockaddr_in6.sin6_addr;
165 daddr6 = &av->dgid_addr._sockaddr_in6.sin6_addr;
166 dst = rxe_find_route6(qp, ndev, saddr6, daddr6);
167 #if IS_ENABLED(CONFIG_IPV6)
168 if (dst)
169 qp->dst_cookie =
170 rt6_get_cookie((struct rt6_info *)dst);
171 #endif
172 }
173
174 if (dst && (qp_type(qp) == IB_QPT_RC)) {
175 dst_hold(dst);
176 sk_dst_set(qp->sk->sk, dst);
177 }
178 }
179 return dst;
180 }
181
rxe_udp_encap_recv(struct sock * sk,struct sk_buff * skb)182 static int rxe_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
183 {
184 struct udphdr *udph;
185 struct rxe_dev *rxe;
186 struct net_device *ndev = skb->dev;
187 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
188
189 /* takes a reference on rxe->ib_dev
190 * drop when skb is freed
191 */
192 rxe = rxe_get_dev_from_net(ndev);
193 if (!rxe && is_vlan_dev(ndev))
194 rxe = rxe_get_dev_from_net(vlan_dev_real_dev(ndev));
195 if (!rxe)
196 goto drop;
197
198 if (skb_linearize(skb)) {
199 ib_device_put(&rxe->ib_dev);
200 goto drop;
201 }
202
203 udph = udp_hdr(skb);
204 pkt->rxe = rxe;
205 pkt->port_num = 1;
206 pkt->hdr = (u8 *)(udph + 1);
207 pkt->mask = RXE_GRH_MASK;
208 pkt->paylen = be16_to_cpu(udph->len) - sizeof(*udph);
209
210 /* remove udp header */
211 skb_pull(skb, sizeof(struct udphdr));
212
213 rxe_rcv(skb);
214
215 return 0;
216 drop:
217 kfree_skb(skb);
218
219 return 0;
220 }
221
rxe_setup_udp_tunnel(struct net * net,__be16 port,bool ipv6)222 static struct socket *rxe_setup_udp_tunnel(struct net *net, __be16 port,
223 bool ipv6)
224 {
225 int err;
226 struct socket *sock;
227 struct udp_port_cfg udp_cfg = { };
228 struct udp_tunnel_sock_cfg tnl_cfg = { };
229
230 if (ipv6) {
231 udp_cfg.family = AF_INET6;
232 udp_cfg.ipv6_v6only = 1;
233 } else {
234 udp_cfg.family = AF_INET;
235 }
236
237 udp_cfg.local_udp_port = port;
238
239 /* Create UDP socket */
240 err = udp_sock_create(net, &udp_cfg, &sock);
241 if (err < 0)
242 return ERR_PTR(err);
243 rxe_reclassify_recv_socket(sock);
244
245 tnl_cfg.encap_type = 1;
246 tnl_cfg.encap_rcv = rxe_udp_encap_recv;
247
248 /* Setup UDP tunnel */
249 setup_udp_tunnel_sock(net, sock, &tnl_cfg);
250
251 return sock;
252 }
253
rxe_release_udp_tunnel(struct socket * sk)254 static void rxe_release_udp_tunnel(struct socket *sk)
255 {
256 if (sk)
257 udp_tunnel_sock_release(sk);
258 }
259
prepare_udp_hdr(struct sk_buff * skb,__be16 src_port,__be16 dst_port)260 static void prepare_udp_hdr(struct sk_buff *skb, __be16 src_port,
261 __be16 dst_port)
262 {
263 struct udphdr *udph;
264
265 __skb_push(skb, sizeof(*udph));
266 skb_reset_transport_header(skb);
267 udph = udp_hdr(skb);
268
269 udph->dest = dst_port;
270 udph->source = src_port;
271 udph->len = htons(skb->len);
272 udph->check = 0;
273 }
274
prepare_ipv4_hdr(struct dst_entry * dst,struct sk_buff * skb,__be32 saddr,__be32 daddr,__u8 proto,__u8 tos,__u8 ttl,__be16 df,bool xnet)275 static void prepare_ipv4_hdr(struct dst_entry *dst, struct sk_buff *skb,
276 __be32 saddr, __be32 daddr, __u8 proto,
277 __u8 tos, __u8 ttl, __be16 df, bool xnet)
278 {
279 struct iphdr *iph;
280
281 skb_scrub_packet(skb, xnet);
282
283 skb_clear_hash(skb);
284 skb_dst_set(skb, dst_clone(dst));
285 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
286
287 skb_push(skb, sizeof(struct iphdr));
288 skb_reset_network_header(skb);
289
290 iph = ip_hdr(skb);
291
292 iph->version = IPVERSION;
293 iph->ihl = sizeof(struct iphdr) >> 2;
294 iph->tot_len = htons(skb->len);
295 iph->frag_off = df;
296 iph->protocol = proto;
297 iph->tos = tos;
298 iph->daddr = daddr;
299 iph->saddr = saddr;
300 iph->ttl = ttl;
301 __ip_select_ident(dev_net(dst->dev), iph,
302 skb_shinfo(skb)->gso_segs ?: 1);
303 }
304
prepare_ipv6_hdr(struct dst_entry * dst,struct sk_buff * skb,struct in6_addr * saddr,struct in6_addr * daddr,__u8 proto,__u8 prio,__u8 ttl)305 static void prepare_ipv6_hdr(struct dst_entry *dst, struct sk_buff *skb,
306 struct in6_addr *saddr, struct in6_addr *daddr,
307 __u8 proto, __u8 prio, __u8 ttl)
308 {
309 struct ipv6hdr *ip6h;
310
311 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
312 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED
313 | IPSKB_REROUTED);
314 skb_dst_set(skb, dst_clone(dst));
315
316 __skb_push(skb, sizeof(*ip6h));
317 skb_reset_network_header(skb);
318 ip6h = ipv6_hdr(skb);
319 ip6_flow_hdr(ip6h, prio, htonl(0));
320 ip6h->payload_len = htons(skb->len);
321 ip6h->nexthdr = proto;
322 ip6h->hop_limit = ttl;
323 ip6h->daddr = *daddr;
324 ip6h->saddr = *saddr;
325 ip6h->payload_len = htons(skb->len - sizeof(*ip6h));
326 }
327
prepare4(struct rxe_av * av,struct rxe_pkt_info * pkt,struct sk_buff * skb)328 static int prepare4(struct rxe_av *av, struct rxe_pkt_info *pkt,
329 struct sk_buff *skb)
330 {
331 struct rxe_qp *qp = pkt->qp;
332 struct dst_entry *dst;
333 bool xnet = false;
334 __be16 df = htons(IP_DF);
335 struct in_addr *saddr = &av->sgid_addr._sockaddr_in.sin_addr;
336 struct in_addr *daddr = &av->dgid_addr._sockaddr_in.sin_addr;
337
338 dst = rxe_find_route(skb->dev, qp, av);
339 if (!dst) {
340 rxe_dbg_qp(qp, "Host not reachable\n");
341 return -EHOSTUNREACH;
342 }
343
344 prepare_udp_hdr(skb, cpu_to_be16(qp->src_port),
345 cpu_to_be16(ROCE_V2_UDP_DPORT));
346
347 prepare_ipv4_hdr(dst, skb, saddr->s_addr, daddr->s_addr, IPPROTO_UDP,
348 av->grh.traffic_class, av->grh.hop_limit, df, xnet);
349
350 dst_release(dst);
351 return 0;
352 }
353
prepare6(struct rxe_av * av,struct rxe_pkt_info * pkt,struct sk_buff * skb)354 static int prepare6(struct rxe_av *av, struct rxe_pkt_info *pkt,
355 struct sk_buff *skb)
356 {
357 struct rxe_qp *qp = pkt->qp;
358 struct dst_entry *dst;
359 struct in6_addr *saddr = &av->sgid_addr._sockaddr_in6.sin6_addr;
360 struct in6_addr *daddr = &av->dgid_addr._sockaddr_in6.sin6_addr;
361
362 dst = rxe_find_route(skb->dev, qp, av);
363 if (!dst) {
364 rxe_dbg_qp(qp, "Host not reachable\n");
365 return -EHOSTUNREACH;
366 }
367
368 prepare_udp_hdr(skb, cpu_to_be16(qp->src_port),
369 cpu_to_be16(ROCE_V2_UDP_DPORT));
370
371 prepare_ipv6_hdr(dst, skb, saddr, daddr, IPPROTO_UDP,
372 av->grh.traffic_class,
373 av->grh.hop_limit);
374
375 dst_release(dst);
376 return 0;
377 }
378
rxe_prepare(struct rxe_av * av,struct rxe_pkt_info * pkt,struct sk_buff * skb)379 int rxe_prepare(struct rxe_av *av, struct rxe_pkt_info *pkt,
380 struct sk_buff *skb)
381 {
382 int err = 0;
383
384 if (skb->protocol == htons(ETH_P_IP))
385 err = prepare4(av, pkt, skb);
386 else if (skb->protocol == htons(ETH_P_IPV6))
387 err = prepare6(av, pkt, skb);
388
389 if (ether_addr_equal(skb->dev->dev_addr, av->dmac))
390 pkt->mask |= RXE_LOOPBACK_MASK;
391
392 return err;
393 }
394
rxe_skb_tx_dtor(struct sk_buff * skb)395 static void rxe_skb_tx_dtor(struct sk_buff *skb)
396 {
397 struct rxe_qp *qp = skb->sk->sk_user_data;
398 int skb_out;
399
400 skb_out = atomic_dec_return(&qp->skb_out);
401 if (unlikely(qp->need_req_skb &&
402 skb_out < RXE_INFLIGHT_SKBS_PER_QP_LOW))
403 rxe_sched_task(&qp->send_task);
404
405 rxe_put(qp);
406 sock_put(skb->sk);
407 }
408
rxe_send(struct sk_buff * skb,struct rxe_pkt_info * pkt)409 static int rxe_send(struct sk_buff *skb, struct rxe_pkt_info *pkt)
410 {
411 int err;
412 struct sock *sk = pkt->qp->sk->sk;
413
414 sock_hold(sk);
415 skb->sk = sk;
416 skb->destructor = rxe_skb_tx_dtor;
417 rxe_get(pkt->qp);
418 atomic_inc(&pkt->qp->skb_out);
419
420 if (skb->protocol == htons(ETH_P_IP))
421 err = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
422 else
423 err = ip6_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
424
425 return err;
426 }
427
428 /* fix up a send packet to match the packets
429 * received from UDP before looping them back
430 */
rxe_loopback(struct sk_buff * skb,struct rxe_pkt_info * pkt)431 static int rxe_loopback(struct sk_buff *skb, struct rxe_pkt_info *pkt)
432 {
433 struct sock *sk = pkt->qp->sk->sk;
434
435 memcpy(SKB_TO_PKT(skb), pkt, sizeof(*pkt));
436
437 sock_hold(sk);
438 skb->sk = sk;
439 skb->destructor = rxe_skb_tx_dtor;
440 rxe_get(pkt->qp);
441 atomic_inc(&pkt->qp->skb_out);
442
443 if (skb->protocol == htons(ETH_P_IP))
444 skb_pull(skb, sizeof(struct iphdr));
445 else
446 skb_pull(skb, sizeof(struct ipv6hdr));
447
448 if (WARN_ON(!ib_device_try_get(&pkt->rxe->ib_dev))) {
449 kfree_skb(skb);
450 return -EIO;
451 }
452
453 /* remove udp header */
454 skb_pull(skb, sizeof(struct udphdr));
455
456 rxe_rcv(skb);
457
458 return 0;
459 }
460
rxe_xmit_packet(struct rxe_qp * qp,struct rxe_pkt_info * pkt,struct sk_buff * skb)461 int rxe_xmit_packet(struct rxe_qp *qp, struct rxe_pkt_info *pkt,
462 struct sk_buff *skb)
463 {
464 int err;
465 int is_request = pkt->mask & RXE_REQ_MASK;
466 struct rxe_dev *rxe = to_rdev(qp->ibqp.device);
467 unsigned long flags;
468
469 spin_lock_irqsave(&qp->state_lock, flags);
470 if ((is_request && (qp_state(qp) < IB_QPS_RTS)) ||
471 (!is_request && (qp_state(qp) < IB_QPS_RTR))) {
472 spin_unlock_irqrestore(&qp->state_lock, flags);
473 rxe_dbg_qp(qp, "Packet dropped. QP is not in ready state\n");
474 goto drop;
475 }
476 spin_unlock_irqrestore(&qp->state_lock, flags);
477
478 rxe_icrc_generate(skb, pkt);
479
480 if (pkt->mask & RXE_LOOPBACK_MASK)
481 err = rxe_loopback(skb, pkt);
482 else
483 err = rxe_send(skb, pkt);
484 if (err) {
485 rxe_counter_inc(rxe, RXE_CNT_SEND_ERR);
486 return err;
487 }
488
489 rxe_counter_inc(rxe, RXE_CNT_SENT_PKTS);
490 goto done;
491
492 drop:
493 kfree_skb(skb);
494 err = 0;
495 done:
496 return err;
497 }
498
rxe_init_packet(struct rxe_dev * rxe,struct rxe_av * av,int paylen,struct rxe_pkt_info * pkt)499 struct sk_buff *rxe_init_packet(struct rxe_dev *rxe, struct rxe_av *av,
500 int paylen, struct rxe_pkt_info *pkt)
501 {
502 unsigned int hdr_len;
503 struct sk_buff *skb = NULL;
504 struct net_device *ndev;
505 const struct ib_gid_attr *attr;
506 const int port_num = 1;
507
508 attr = rdma_get_gid_attr(&rxe->ib_dev, port_num, av->grh.sgid_index);
509 if (IS_ERR(attr))
510 return NULL;
511
512 if (av->network_type == RXE_NETWORK_TYPE_IPV4)
513 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
514 sizeof(struct iphdr);
515 else
516 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
517 sizeof(struct ipv6hdr);
518
519 rcu_read_lock();
520 ndev = rdma_read_gid_attr_ndev_rcu(attr);
521 if (IS_ERR(ndev)) {
522 rcu_read_unlock();
523 goto out;
524 }
525 skb = alloc_skb(paylen + hdr_len + LL_RESERVED_SPACE(ndev),
526 GFP_ATOMIC);
527
528 if (unlikely(!skb)) {
529 rcu_read_unlock();
530 goto out;
531 }
532
533 /* Add time stamp to skb. */
534 skb->tstamp = ktime_get();
535
536 skb_reserve(skb, hdr_len + LL_RESERVED_SPACE(ndev));
537
538 /* FIXME: hold reference to this netdev until life of this skb. */
539 skb->dev = ndev;
540 rcu_read_unlock();
541
542 if (av->network_type == RXE_NETWORK_TYPE_IPV4)
543 skb->protocol = htons(ETH_P_IP);
544 else
545 skb->protocol = htons(ETH_P_IPV6);
546
547 pkt->rxe = rxe;
548 pkt->port_num = port_num;
549 pkt->hdr = skb_put(skb, paylen);
550 pkt->mask |= RXE_GRH_MASK;
551
552 out:
553 rdma_put_gid_attr(attr);
554 return skb;
555 }
556
557 /*
558 * this is required by rxe_cfg to match rxe devices in
559 * /sys/class/infiniband up with their underlying ethernet devices
560 */
rxe_parent_name(struct rxe_dev * rxe,unsigned int port_num)561 const char *rxe_parent_name(struct rxe_dev *rxe, unsigned int port_num)
562 {
563 struct net_device *ndev;
564 char *ndev_name;
565
566 ndev = rxe_ib_device_get_netdev(&rxe->ib_dev);
567 if (!ndev)
568 return NULL;
569 ndev_name = ndev->name;
570 dev_put(ndev);
571
572 return ndev_name;
573 }
574
rxe_net_add(const char * ibdev_name,struct net_device * ndev)575 int rxe_net_add(const char *ibdev_name, struct net_device *ndev)
576 {
577 int err;
578 struct rxe_dev *rxe = NULL;
579
580 rxe = ib_alloc_device(rxe_dev, ib_dev);
581 if (!rxe)
582 return -ENOMEM;
583
584 ib_mark_name_assigned_by_user(&rxe->ib_dev);
585
586 err = rxe_add(rxe, ndev->mtu, ibdev_name, ndev);
587 if (err) {
588 ib_dealloc_device(&rxe->ib_dev);
589 return err;
590 }
591
592 return 0;
593 }
594
rxe_port_event(struct rxe_dev * rxe,enum ib_event_type event)595 static void rxe_port_event(struct rxe_dev *rxe,
596 enum ib_event_type event)
597 {
598 struct ib_event ev;
599
600 ev.device = &rxe->ib_dev;
601 ev.element.port_num = 1;
602 ev.event = event;
603
604 ib_dispatch_event(&ev);
605 }
606
607 /* Caller must hold net_info_lock */
rxe_port_up(struct rxe_dev * rxe)608 void rxe_port_up(struct rxe_dev *rxe)
609 {
610 rxe_port_event(rxe, IB_EVENT_PORT_ACTIVE);
611 dev_info(&rxe->ib_dev.dev, "set active\n");
612 }
613
614 /* Caller must hold net_info_lock */
rxe_port_down(struct rxe_dev * rxe)615 void rxe_port_down(struct rxe_dev *rxe)
616 {
617 rxe_port_event(rxe, IB_EVENT_PORT_ERR);
618 rxe_counter_inc(rxe, RXE_CNT_LINK_DOWNED);
619 dev_info(&rxe->ib_dev.dev, "set down\n");
620 }
621
rxe_set_port_state(struct rxe_dev * rxe)622 void rxe_set_port_state(struct rxe_dev *rxe)
623 {
624 struct net_device *ndev;
625
626 ndev = rxe_ib_device_get_netdev(&rxe->ib_dev);
627 if (!ndev)
628 return;
629
630 if (ib_get_curr_port_state(ndev) == IB_PORT_ACTIVE)
631 rxe_port_up(rxe);
632 else
633 rxe_port_down(rxe);
634
635 dev_put(ndev);
636 }
637
rxe_notify(struct notifier_block * not_blk,unsigned long event,void * arg)638 static int rxe_notify(struct notifier_block *not_blk,
639 unsigned long event,
640 void *arg)
641 {
642 struct net_device *ndev = netdev_notifier_info_to_dev(arg);
643 struct rxe_dev *rxe = rxe_get_dev_from_net(ndev);
644
645 if (!rxe)
646 return NOTIFY_OK;
647
648 switch (event) {
649 case NETDEV_UNREGISTER:
650 ib_unregister_device_queued(&rxe->ib_dev);
651 break;
652 case NETDEV_CHANGEMTU:
653 rxe_dbg_dev(rxe, "%s changed mtu to %d\n", ndev->name, ndev->mtu);
654 rxe_set_mtu(rxe, ndev->mtu);
655 break;
656 case NETDEV_DOWN:
657 case NETDEV_CHANGE:
658 if (ib_get_curr_port_state(ndev) == IB_PORT_DOWN)
659 rxe_counter_inc(rxe, RXE_CNT_LINK_DOWNED);
660 break;
661 case NETDEV_REBOOT:
662 case NETDEV_GOING_DOWN:
663 case NETDEV_CHANGEADDR:
664 case NETDEV_CHANGENAME:
665 case NETDEV_FEAT_CHANGE:
666 default:
667 rxe_dbg_dev(rxe, "ignoring netdev event = %ld for %s\n",
668 event, ndev->name);
669 break;
670 }
671
672 ib_device_put(&rxe->ib_dev);
673 return NOTIFY_OK;
674 }
675
676 static struct notifier_block rxe_net_notifier = {
677 .notifier_call = rxe_notify,
678 };
679
rxe_net_ipv4_init(void)680 static int rxe_net_ipv4_init(void)
681 {
682 recv_sockets.sk4 = rxe_setup_udp_tunnel(&init_net,
683 htons(ROCE_V2_UDP_DPORT), false);
684 if (IS_ERR(recv_sockets.sk4)) {
685 recv_sockets.sk4 = NULL;
686 pr_err("Failed to create IPv4 UDP tunnel\n");
687 return -1;
688 }
689
690 return 0;
691 }
692
rxe_net_ipv6_init(void)693 static int rxe_net_ipv6_init(void)
694 {
695 #if IS_ENABLED(CONFIG_IPV6)
696
697 recv_sockets.sk6 = rxe_setup_udp_tunnel(&init_net,
698 htons(ROCE_V2_UDP_DPORT), true);
699 if (PTR_ERR(recv_sockets.sk6) == -EAFNOSUPPORT) {
700 recv_sockets.sk6 = NULL;
701 pr_warn("IPv6 is not supported, can not create a UDPv6 socket\n");
702 return 0;
703 }
704
705 if (IS_ERR(recv_sockets.sk6)) {
706 recv_sockets.sk6 = NULL;
707 pr_err("Failed to create IPv6 UDP tunnel\n");
708 return -1;
709 }
710 #endif
711 return 0;
712 }
713
rxe_net_exit(void)714 void rxe_net_exit(void)
715 {
716 rxe_release_udp_tunnel(recv_sockets.sk6);
717 rxe_release_udp_tunnel(recv_sockets.sk4);
718 unregister_netdevice_notifier(&rxe_net_notifier);
719 }
720
rxe_net_init(void)721 int rxe_net_init(void)
722 {
723 int err;
724
725 recv_sockets.sk6 = NULL;
726
727 err = rxe_net_ipv4_init();
728 if (err)
729 return err;
730 err = rxe_net_ipv6_init();
731 if (err)
732 goto err_out;
733 err = register_netdevice_notifier(&rxe_net_notifier);
734 if (err) {
735 pr_err("Failed to register netdev notifier\n");
736 goto err_out;
737 }
738 return 0;
739 err_out:
740 rxe_net_exit();
741 return err;
742 }
743