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 return;
68 }
69 /*
70 * sock_lock_init_class_and_name() calls
71 * sk_owner_set(sk, THIS_MODULE); in order
72 * to make sure the referenced global
73 * variables rxe_recv_slock_key and
74 * rxe_recv_sk_key are not removed
75 * before the socket is closed.
76 *
77 * However this prevents rxe_net_exit()
78 * from being called and 'rmmod rdma_rxe'
79 * is refused because of the references.
80 *
81 * For the global sockets in recv_sockets,
82 * we are sure that rxe_net_exit() will call
83 * rxe_release_udp_tunnel -> udp_tunnel_sock_release.
84 *
85 * So we don't need the additional reference to
86 * our own (THIS_MODULE).
87 */
88 sk_owner_put(sk);
89 /*
90 * We also call sk_owner_clear() otherwise
91 * sk_owner_put(sk) in sk_prot_free will
92 * fail, which is called via
93 * sk_free -> __sk_free -> sk_destruct
94 * and sk_destruct calls __sk_destruct
95 * directly or via call_rcu()
96 * so sk_prot_free() might be called
97 * after rxe_net_exit().
98 */
99 sk_owner_clear(sk);
100 #endif /* CONFIG_DEBUG_LOCK_ALLOC */
101 }
102
rxe_find_route4(struct rxe_qp * qp,struct net_device * ndev,struct in_addr * saddr,struct in_addr * daddr)103 static struct dst_entry *rxe_find_route4(struct rxe_qp *qp,
104 struct net_device *ndev,
105 struct in_addr *saddr,
106 struct in_addr *daddr)
107 {
108 struct rtable *rt;
109 struct flowi4 fl = { { 0 } };
110
111 memset(&fl, 0, sizeof(fl));
112 fl.flowi4_oif = ndev->ifindex;
113 memcpy(&fl.saddr, saddr, sizeof(*saddr));
114 memcpy(&fl.daddr, daddr, sizeof(*daddr));
115 fl.flowi4_proto = IPPROTO_UDP;
116
117 rt = ip_route_output_key(&init_net, &fl);
118 if (IS_ERR(rt)) {
119 rxe_dbg_qp(qp, "no route to %pI4\n", &daddr->s_addr);
120 return NULL;
121 }
122
123 return &rt->dst;
124 }
125
126 #if IS_ENABLED(CONFIG_IPV6)
rxe_find_route6(struct rxe_qp * qp,struct net_device * ndev,struct in6_addr * saddr,struct in6_addr * daddr)127 static struct dst_entry *rxe_find_route6(struct rxe_qp *qp,
128 struct net_device *ndev,
129 struct in6_addr *saddr,
130 struct in6_addr *daddr)
131 {
132 struct dst_entry *ndst;
133 struct flowi6 fl6 = { { 0 } };
134
135 memset(&fl6, 0, sizeof(fl6));
136 fl6.flowi6_oif = ndev->ifindex;
137 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
138 memcpy(&fl6.daddr, daddr, sizeof(*daddr));
139 fl6.flowi6_proto = IPPROTO_UDP;
140
141 ndst = ipv6_stub->ipv6_dst_lookup_flow(sock_net(recv_sockets.sk6->sk),
142 recv_sockets.sk6->sk, &fl6,
143 NULL);
144 if (IS_ERR(ndst)) {
145 rxe_dbg_qp(qp, "no route to %pI6\n", daddr);
146 return NULL;
147 }
148
149 if (unlikely(ndst->error)) {
150 rxe_dbg_qp(qp, "no route to %pI6\n", daddr);
151 goto put;
152 }
153
154 return ndst;
155 put:
156 dst_release(ndst);
157 return NULL;
158 }
159
160 #else
161
rxe_find_route6(struct rxe_qp * qp,struct net_device * ndev,struct in6_addr * saddr,struct in6_addr * daddr)162 static struct dst_entry *rxe_find_route6(struct rxe_qp *qp,
163 struct net_device *ndev,
164 struct in6_addr *saddr,
165 struct in6_addr *daddr)
166 {
167 return NULL;
168 }
169
170 #endif
171
rxe_find_route(struct net_device * ndev,struct rxe_qp * qp,struct rxe_av * av)172 static struct dst_entry *rxe_find_route(struct net_device *ndev,
173 struct rxe_qp *qp,
174 struct rxe_av *av)
175 {
176 struct dst_entry *dst = NULL;
177
178 if (qp_type(qp) == IB_QPT_RC)
179 dst = sk_dst_get(qp->sk->sk);
180
181 if (!dst || !dst_check(dst, qp->dst_cookie)) {
182 if (dst)
183 dst_release(dst);
184
185 if (av->network_type == RXE_NETWORK_TYPE_IPV4) {
186 struct in_addr *saddr;
187 struct in_addr *daddr;
188
189 saddr = &av->sgid_addr._sockaddr_in.sin_addr;
190 daddr = &av->dgid_addr._sockaddr_in.sin_addr;
191 dst = rxe_find_route4(qp, ndev, saddr, daddr);
192 } else if (av->network_type == RXE_NETWORK_TYPE_IPV6) {
193 struct in6_addr *saddr6;
194 struct in6_addr *daddr6;
195
196 saddr6 = &av->sgid_addr._sockaddr_in6.sin6_addr;
197 daddr6 = &av->dgid_addr._sockaddr_in6.sin6_addr;
198 dst = rxe_find_route6(qp, ndev, saddr6, daddr6);
199 #if IS_ENABLED(CONFIG_IPV6)
200 if (dst)
201 qp->dst_cookie =
202 rt6_get_cookie((struct rt6_info *)dst);
203 #endif
204 }
205
206 if (dst && (qp_type(qp) == IB_QPT_RC)) {
207 dst_hold(dst);
208 sk_dst_set(qp->sk->sk, dst);
209 }
210 }
211 return dst;
212 }
213
rxe_udp_encap_recv(struct sock * sk,struct sk_buff * skb)214 static int rxe_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
215 {
216 struct udphdr *udph;
217 struct rxe_dev *rxe;
218 struct net_device *ndev = skb->dev;
219 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
220
221 /* takes a reference on rxe->ib_dev
222 * drop when skb is freed
223 */
224 rxe = rxe_get_dev_from_net(ndev);
225 if (!rxe && is_vlan_dev(ndev))
226 rxe = rxe_get_dev_from_net(vlan_dev_real_dev(ndev));
227 if (!rxe)
228 goto drop;
229
230 if (skb_linearize(skb)) {
231 ib_device_put(&rxe->ib_dev);
232 goto drop;
233 }
234
235 udph = udp_hdr(skb);
236 pkt->rxe = rxe;
237 pkt->port_num = 1;
238 pkt->hdr = (u8 *)(udph + 1);
239 pkt->mask = RXE_GRH_MASK;
240 pkt->paylen = be16_to_cpu(udph->len) - sizeof(*udph);
241
242 /* remove udp header */
243 skb_pull(skb, sizeof(struct udphdr));
244
245 rxe_rcv(skb);
246
247 return 0;
248 drop:
249 kfree_skb(skb);
250
251 return 0;
252 }
253
rxe_setup_udp_tunnel(struct net * net,__be16 port,bool ipv6)254 static struct socket *rxe_setup_udp_tunnel(struct net *net, __be16 port,
255 bool ipv6)
256 {
257 int err;
258 struct socket *sock;
259 struct udp_port_cfg udp_cfg = { };
260 struct udp_tunnel_sock_cfg tnl_cfg = { };
261
262 if (ipv6) {
263 udp_cfg.family = AF_INET6;
264 udp_cfg.ipv6_v6only = 1;
265 } else {
266 udp_cfg.family = AF_INET;
267 }
268
269 udp_cfg.local_udp_port = port;
270
271 /* Create UDP socket */
272 err = udp_sock_create(net, &udp_cfg, &sock);
273 if (err < 0)
274 return ERR_PTR(err);
275 rxe_reclassify_recv_socket(sock);
276
277 tnl_cfg.encap_type = 1;
278 tnl_cfg.encap_rcv = rxe_udp_encap_recv;
279
280 /* Setup UDP tunnel */
281 setup_udp_tunnel_sock(net, sock, &tnl_cfg);
282
283 return sock;
284 }
285
rxe_release_udp_tunnel(struct socket * sk)286 static void rxe_release_udp_tunnel(struct socket *sk)
287 {
288 if (sk)
289 udp_tunnel_sock_release(sk);
290 }
291
prepare_udp_hdr(struct sk_buff * skb,__be16 src_port,__be16 dst_port)292 static void prepare_udp_hdr(struct sk_buff *skb, __be16 src_port,
293 __be16 dst_port)
294 {
295 struct udphdr *udph;
296
297 __skb_push(skb, sizeof(*udph));
298 skb_reset_transport_header(skb);
299 udph = udp_hdr(skb);
300
301 udph->dest = dst_port;
302 udph->source = src_port;
303 udph->len = htons(skb->len);
304 udph->check = 0;
305 }
306
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)307 static void prepare_ipv4_hdr(struct dst_entry *dst, struct sk_buff *skb,
308 __be32 saddr, __be32 daddr, __u8 proto,
309 __u8 tos, __u8 ttl, __be16 df, bool xnet)
310 {
311 struct iphdr *iph;
312
313 skb_scrub_packet(skb, xnet);
314
315 skb_clear_hash(skb);
316 skb_dst_set(skb, dst_clone(dst));
317 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
318
319 skb_push(skb, sizeof(struct iphdr));
320 skb_reset_network_header(skb);
321
322 iph = ip_hdr(skb);
323
324 iph->version = IPVERSION;
325 iph->ihl = sizeof(struct iphdr) >> 2;
326 iph->tot_len = htons(skb->len);
327 iph->frag_off = df;
328 iph->protocol = proto;
329 iph->tos = tos;
330 iph->daddr = daddr;
331 iph->saddr = saddr;
332 iph->ttl = ttl;
333 __ip_select_ident(dev_net(dst->dev), iph,
334 skb_shinfo(skb)->gso_segs ?: 1);
335 }
336
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)337 static void prepare_ipv6_hdr(struct dst_entry *dst, struct sk_buff *skb,
338 struct in6_addr *saddr, struct in6_addr *daddr,
339 __u8 proto, __u8 prio, __u8 ttl)
340 {
341 struct ipv6hdr *ip6h;
342
343 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
344 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED
345 | IPSKB_REROUTED);
346 skb_dst_set(skb, dst_clone(dst));
347
348 __skb_push(skb, sizeof(*ip6h));
349 skb_reset_network_header(skb);
350 ip6h = ipv6_hdr(skb);
351 ip6_flow_hdr(ip6h, prio, htonl(0));
352 ip6h->payload_len = htons(skb->len);
353 ip6h->nexthdr = proto;
354 ip6h->hop_limit = ttl;
355 ip6h->daddr = *daddr;
356 ip6h->saddr = *saddr;
357 ip6h->payload_len = htons(skb->len - sizeof(*ip6h));
358 }
359
prepare4(struct rxe_av * av,struct rxe_pkt_info * pkt,struct sk_buff * skb)360 static int prepare4(struct rxe_av *av, struct rxe_pkt_info *pkt,
361 struct sk_buff *skb)
362 {
363 struct rxe_qp *qp = pkt->qp;
364 struct dst_entry *dst;
365 bool xnet = false;
366 __be16 df = htons(IP_DF);
367 struct in_addr *saddr = &av->sgid_addr._sockaddr_in.sin_addr;
368 struct in_addr *daddr = &av->dgid_addr._sockaddr_in.sin_addr;
369
370 dst = rxe_find_route(skb->dev, qp, av);
371 if (!dst) {
372 rxe_dbg_qp(qp, "Host not reachable\n");
373 return -EHOSTUNREACH;
374 }
375
376 prepare_udp_hdr(skb, cpu_to_be16(qp->src_port),
377 cpu_to_be16(ROCE_V2_UDP_DPORT));
378
379 prepare_ipv4_hdr(dst, skb, saddr->s_addr, daddr->s_addr, IPPROTO_UDP,
380 av->grh.traffic_class, av->grh.hop_limit, df, xnet);
381
382 dst_release(dst);
383 return 0;
384 }
385
prepare6(struct rxe_av * av,struct rxe_pkt_info * pkt,struct sk_buff * skb)386 static int prepare6(struct rxe_av *av, struct rxe_pkt_info *pkt,
387 struct sk_buff *skb)
388 {
389 struct rxe_qp *qp = pkt->qp;
390 struct dst_entry *dst;
391 struct in6_addr *saddr = &av->sgid_addr._sockaddr_in6.sin6_addr;
392 struct in6_addr *daddr = &av->dgid_addr._sockaddr_in6.sin6_addr;
393
394 dst = rxe_find_route(skb->dev, qp, av);
395 if (!dst) {
396 rxe_dbg_qp(qp, "Host not reachable\n");
397 return -EHOSTUNREACH;
398 }
399
400 prepare_udp_hdr(skb, cpu_to_be16(qp->src_port),
401 cpu_to_be16(ROCE_V2_UDP_DPORT));
402
403 prepare_ipv6_hdr(dst, skb, saddr, daddr, IPPROTO_UDP,
404 av->grh.traffic_class,
405 av->grh.hop_limit);
406
407 dst_release(dst);
408 return 0;
409 }
410
rxe_prepare(struct rxe_av * av,struct rxe_pkt_info * pkt,struct sk_buff * skb)411 int rxe_prepare(struct rxe_av *av, struct rxe_pkt_info *pkt,
412 struct sk_buff *skb)
413 {
414 int err = 0;
415
416 if (skb->protocol == htons(ETH_P_IP))
417 err = prepare4(av, pkt, skb);
418 else if (skb->protocol == htons(ETH_P_IPV6))
419 err = prepare6(av, pkt, skb);
420
421 if (ether_addr_equal(skb->dev->dev_addr, av->dmac))
422 pkt->mask |= RXE_LOOPBACK_MASK;
423
424 return err;
425 }
426
rxe_skb_tx_dtor(struct sk_buff * skb)427 static void rxe_skb_tx_dtor(struct sk_buff *skb)
428 {
429 struct rxe_qp *qp = skb->sk->sk_user_data;
430 int skb_out;
431
432 skb_out = atomic_dec_return(&qp->skb_out);
433 if (unlikely(qp->need_req_skb &&
434 skb_out < RXE_INFLIGHT_SKBS_PER_QP_LOW))
435 rxe_sched_task(&qp->send_task);
436
437 rxe_put(qp);
438 sock_put(skb->sk);
439 }
440
rxe_send(struct sk_buff * skb,struct rxe_pkt_info * pkt)441 static int rxe_send(struct sk_buff *skb, struct rxe_pkt_info *pkt)
442 {
443 int err;
444 struct sock *sk = pkt->qp->sk->sk;
445
446 sock_hold(sk);
447 skb->sk = sk;
448 skb->destructor = rxe_skb_tx_dtor;
449 rxe_get(pkt->qp);
450 atomic_inc(&pkt->qp->skb_out);
451
452 if (skb->protocol == htons(ETH_P_IP))
453 err = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
454 else
455 err = ip6_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
456
457 return err;
458 }
459
460 /* fix up a send packet to match the packets
461 * received from UDP before looping them back
462 */
rxe_loopback(struct sk_buff * skb,struct rxe_pkt_info * pkt)463 static int rxe_loopback(struct sk_buff *skb, struct rxe_pkt_info *pkt)
464 {
465 struct sock *sk = pkt->qp->sk->sk;
466
467 memcpy(SKB_TO_PKT(skb), pkt, sizeof(*pkt));
468
469 sock_hold(sk);
470 skb->sk = sk;
471 skb->destructor = rxe_skb_tx_dtor;
472 rxe_get(pkt->qp);
473 atomic_inc(&pkt->qp->skb_out);
474
475 if (skb->protocol == htons(ETH_P_IP))
476 skb_pull(skb, sizeof(struct iphdr));
477 else
478 skb_pull(skb, sizeof(struct ipv6hdr));
479
480 if (WARN_ON(!ib_device_try_get(&pkt->rxe->ib_dev))) {
481 kfree_skb(skb);
482 return -EIO;
483 }
484
485 /* remove udp header */
486 skb_pull(skb, sizeof(struct udphdr));
487
488 rxe_rcv(skb);
489
490 return 0;
491 }
492
rxe_xmit_packet(struct rxe_qp * qp,struct rxe_pkt_info * pkt,struct sk_buff * skb)493 int rxe_xmit_packet(struct rxe_qp *qp, struct rxe_pkt_info *pkt,
494 struct sk_buff *skb)
495 {
496 int err;
497 int is_request = pkt->mask & RXE_REQ_MASK;
498 struct rxe_dev *rxe = to_rdev(qp->ibqp.device);
499 unsigned long flags;
500
501 spin_lock_irqsave(&qp->state_lock, flags);
502 if ((is_request && (qp_state(qp) < IB_QPS_RTS)) ||
503 (!is_request && (qp_state(qp) < IB_QPS_RTR))) {
504 spin_unlock_irqrestore(&qp->state_lock, flags);
505 rxe_dbg_qp(qp, "Packet dropped. QP is not in ready state\n");
506 goto drop;
507 }
508 spin_unlock_irqrestore(&qp->state_lock, flags);
509
510 rxe_icrc_generate(skb, pkt);
511
512 if (pkt->mask & RXE_LOOPBACK_MASK)
513 err = rxe_loopback(skb, pkt);
514 else
515 err = rxe_send(skb, pkt);
516 if (err) {
517 rxe_counter_inc(rxe, RXE_CNT_SEND_ERR);
518 return err;
519 }
520
521 rxe_counter_inc(rxe, RXE_CNT_SENT_PKTS);
522 goto done;
523
524 drop:
525 kfree_skb(skb);
526 err = 0;
527 done:
528 return err;
529 }
530
rxe_init_packet(struct rxe_dev * rxe,struct rxe_av * av,int paylen,struct rxe_pkt_info * pkt)531 struct sk_buff *rxe_init_packet(struct rxe_dev *rxe, struct rxe_av *av,
532 int paylen, struct rxe_pkt_info *pkt)
533 {
534 unsigned int hdr_len;
535 struct sk_buff *skb = NULL;
536 struct net_device *ndev;
537 const struct ib_gid_attr *attr;
538 const int port_num = 1;
539
540 attr = rdma_get_gid_attr(&rxe->ib_dev, port_num, av->grh.sgid_index);
541 if (IS_ERR(attr))
542 return NULL;
543
544 if (av->network_type == RXE_NETWORK_TYPE_IPV4)
545 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
546 sizeof(struct iphdr);
547 else
548 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
549 sizeof(struct ipv6hdr);
550
551 rcu_read_lock();
552 ndev = rdma_read_gid_attr_ndev_rcu(attr);
553 if (IS_ERR(ndev)) {
554 rcu_read_unlock();
555 goto out;
556 }
557 skb = alloc_skb(paylen + hdr_len + LL_RESERVED_SPACE(ndev),
558 GFP_ATOMIC);
559
560 if (unlikely(!skb)) {
561 rcu_read_unlock();
562 goto out;
563 }
564
565 /* Add time stamp to skb. */
566 skb->tstamp = ktime_get();
567
568 skb_reserve(skb, hdr_len + LL_RESERVED_SPACE(ndev));
569
570 /* FIXME: hold reference to this netdev until life of this skb. */
571 skb->dev = ndev;
572 rcu_read_unlock();
573
574 if (av->network_type == RXE_NETWORK_TYPE_IPV4)
575 skb->protocol = htons(ETH_P_IP);
576 else
577 skb->protocol = htons(ETH_P_IPV6);
578
579 pkt->rxe = rxe;
580 pkt->port_num = port_num;
581 pkt->hdr = skb_put(skb, paylen);
582 pkt->mask |= RXE_GRH_MASK;
583
584 out:
585 rdma_put_gid_attr(attr);
586 return skb;
587 }
588
589 /*
590 * this is required by rxe_cfg to match rxe devices in
591 * /sys/class/infiniband up with their underlying ethernet devices
592 */
rxe_parent_name(struct rxe_dev * rxe,unsigned int port_num)593 const char *rxe_parent_name(struct rxe_dev *rxe, unsigned int port_num)
594 {
595 struct net_device *ndev;
596 char *ndev_name;
597
598 ndev = rxe_ib_device_get_netdev(&rxe->ib_dev);
599 if (!ndev)
600 return NULL;
601 ndev_name = ndev->name;
602 dev_put(ndev);
603
604 return ndev_name;
605 }
606
rxe_net_add(const char * ibdev_name,struct net_device * ndev)607 int rxe_net_add(const char *ibdev_name, struct net_device *ndev)
608 {
609 int err;
610 struct rxe_dev *rxe = NULL;
611
612 rxe = ib_alloc_device(rxe_dev, ib_dev);
613 if (!rxe)
614 return -ENOMEM;
615
616 ib_mark_name_assigned_by_user(&rxe->ib_dev);
617
618 err = rxe_add(rxe, ndev->mtu, ibdev_name, ndev);
619 if (err) {
620 ib_dealloc_device(&rxe->ib_dev);
621 return err;
622 }
623
624 return 0;
625 }
626
rxe_port_event(struct rxe_dev * rxe,enum ib_event_type event)627 static void rxe_port_event(struct rxe_dev *rxe,
628 enum ib_event_type event)
629 {
630 struct ib_event ev;
631
632 ev.device = &rxe->ib_dev;
633 ev.element.port_num = 1;
634 ev.event = event;
635
636 ib_dispatch_event(&ev);
637 }
638
639 /* Caller must hold net_info_lock */
rxe_port_up(struct rxe_dev * rxe)640 void rxe_port_up(struct rxe_dev *rxe)
641 {
642 rxe_port_event(rxe, IB_EVENT_PORT_ACTIVE);
643 dev_info(&rxe->ib_dev.dev, "set active\n");
644 }
645
646 /* Caller must hold net_info_lock */
rxe_port_down(struct rxe_dev * rxe)647 void rxe_port_down(struct rxe_dev *rxe)
648 {
649 rxe_port_event(rxe, IB_EVENT_PORT_ERR);
650 rxe_counter_inc(rxe, RXE_CNT_LINK_DOWNED);
651 dev_info(&rxe->ib_dev.dev, "set down\n");
652 }
653
rxe_set_port_state(struct rxe_dev * rxe)654 void rxe_set_port_state(struct rxe_dev *rxe)
655 {
656 struct net_device *ndev;
657
658 ndev = rxe_ib_device_get_netdev(&rxe->ib_dev);
659 if (!ndev)
660 return;
661
662 if (ib_get_curr_port_state(ndev) == IB_PORT_ACTIVE)
663 rxe_port_up(rxe);
664 else
665 rxe_port_down(rxe);
666
667 dev_put(ndev);
668 }
669
rxe_notify(struct notifier_block * not_blk,unsigned long event,void * arg)670 static int rxe_notify(struct notifier_block *not_blk,
671 unsigned long event,
672 void *arg)
673 {
674 struct net_device *ndev = netdev_notifier_info_to_dev(arg);
675 struct rxe_dev *rxe = rxe_get_dev_from_net(ndev);
676
677 if (!rxe)
678 return NOTIFY_OK;
679
680 switch (event) {
681 case NETDEV_UNREGISTER:
682 ib_unregister_device_queued(&rxe->ib_dev);
683 break;
684 case NETDEV_CHANGEMTU:
685 rxe_dbg_dev(rxe, "%s changed mtu to %d\n", ndev->name, ndev->mtu);
686 rxe_set_mtu(rxe, ndev->mtu);
687 break;
688 case NETDEV_DOWN:
689 case NETDEV_CHANGE:
690 if (ib_get_curr_port_state(ndev) == IB_PORT_DOWN)
691 rxe_counter_inc(rxe, RXE_CNT_LINK_DOWNED);
692 break;
693 case NETDEV_REBOOT:
694 case NETDEV_GOING_DOWN:
695 case NETDEV_CHANGEADDR:
696 case NETDEV_CHANGENAME:
697 case NETDEV_FEAT_CHANGE:
698 default:
699 rxe_dbg_dev(rxe, "ignoring netdev event = %ld for %s\n",
700 event, ndev->name);
701 break;
702 }
703
704 ib_device_put(&rxe->ib_dev);
705 return NOTIFY_OK;
706 }
707
708 static struct notifier_block rxe_net_notifier = {
709 .notifier_call = rxe_notify,
710 };
711
rxe_net_ipv4_init(void)712 static int rxe_net_ipv4_init(void)
713 {
714 recv_sockets.sk4 = rxe_setup_udp_tunnel(&init_net,
715 htons(ROCE_V2_UDP_DPORT), false);
716 if (IS_ERR(recv_sockets.sk4)) {
717 recv_sockets.sk4 = NULL;
718 pr_err("Failed to create IPv4 UDP tunnel\n");
719 return -1;
720 }
721
722 return 0;
723 }
724
rxe_net_ipv6_init(void)725 static int rxe_net_ipv6_init(void)
726 {
727 #if IS_ENABLED(CONFIG_IPV6)
728
729 recv_sockets.sk6 = rxe_setup_udp_tunnel(&init_net,
730 htons(ROCE_V2_UDP_DPORT), true);
731 if (PTR_ERR(recv_sockets.sk6) == -EAFNOSUPPORT) {
732 recv_sockets.sk6 = NULL;
733 pr_warn("IPv6 is not supported, can not create a UDPv6 socket\n");
734 return 0;
735 }
736
737 if (IS_ERR(recv_sockets.sk6)) {
738 recv_sockets.sk6 = NULL;
739 pr_err("Failed to create IPv6 UDP tunnel\n");
740 return -1;
741 }
742 #endif
743 return 0;
744 }
745
rxe_net_exit(void)746 void rxe_net_exit(void)
747 {
748 rxe_release_udp_tunnel(recv_sockets.sk6);
749 rxe_release_udp_tunnel(recv_sockets.sk4);
750 unregister_netdevice_notifier(&rxe_net_notifier);
751 }
752
rxe_net_init(void)753 int rxe_net_init(void)
754 {
755 int err;
756
757 recv_sockets.sk6 = NULL;
758
759 err = rxe_net_ipv4_init();
760 if (err)
761 return err;
762 err = rxe_net_ipv6_init();
763 if (err)
764 goto err_out;
765 err = register_netdevice_notifier(&rxe_net_notifier);
766 if (err) {
767 pr_err("Failed to register netdev notifier\n");
768 goto err_out;
769 }
770 return 0;
771 err_out:
772 rxe_net_exit();
773 return err;
774 }
775