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 #include "rxe_ns.h"
21
22 #ifndef SK_REF_FOR_TUNNEL
23 #define SK_REF_FOR_TUNNEL 2
24 #endif
25
26 #ifdef CONFIG_DEBUG_LOCK_ALLOC
27 /*
28 * lockdep can detect false positive circular dependencies
29 * when there are user-space socket API users or in kernel
30 * users switching between a tcp and rdma transport.
31 * Maybe also switching between siw and rxe may cause
32 * problems as per default sockets are only classified
33 * by family and not by ip protocol. And there might
34 * be different locks used between the application
35 * and the low level sockets.
36 *
37 * Problems were seen with ksmbd.ko and cifs.ko,
38 * switching transports, use git blame to find
39 * more details.
40 */
41 static struct lock_class_key rxe_recv_sk_key[2];
42 static struct lock_class_key rxe_recv_slock_key[2];
43 #endif /* CONFIG_DEBUG_LOCK_ALLOC */
44
rxe_reclassify_recv_socket(struct socket * sock)45 static inline void rxe_reclassify_recv_socket(struct socket *sock)
46 {
47 #ifdef CONFIG_DEBUG_LOCK_ALLOC
48 struct sock *sk = sock->sk;
49
50 if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
51 return;
52
53 switch (sk->sk_family) {
54 case AF_INET:
55 sock_lock_init_class_and_name(sk,
56 "slock-AF_INET-RDMA-RXE-RECV",
57 &rxe_recv_slock_key[0],
58 "sk_lock-AF_INET-RDMA-RXE-RECV",
59 &rxe_recv_sk_key[0]);
60 break;
61 case AF_INET6:
62 sock_lock_init_class_and_name(sk,
63 "slock-AF_INET6-RDMA-RXE-RECV",
64 &rxe_recv_slock_key[1],
65 "sk_lock-AF_INET6-RDMA-RXE-RECV",
66 &rxe_recv_sk_key[1]);
67 break;
68 default:
69 WARN_ON_ONCE(1);
70 return;
71 }
72 /*
73 * sock_lock_init_class_and_name() calls
74 * sk_owner_set(sk, THIS_MODULE); in order
75 * to make sure the referenced global
76 * variables rxe_recv_slock_key and
77 * rxe_recv_sk_key are not removed
78 * before the socket is closed.
79 *
80 * However this prevents rxe_net_exit()
81 * from being called and 'rmmod rdma_rxe'
82 * is refused because of the references.
83 *
84 * For the global sockets in recv_sockets,
85 * we are sure that rxe_net_exit() will call
86 * rxe_release_udp_tunnel -> udp_tunnel_sock_release.
87 *
88 * So we don't need the additional reference to
89 * our own (THIS_MODULE).
90 */
91 sk_owner_put(sk);
92 /*
93 * We also call sk_owner_clear() otherwise
94 * sk_owner_put(sk) in sk_prot_free will
95 * fail, which is called via
96 * sk_free -> __sk_free -> sk_destruct
97 * and sk_destruct calls __sk_destruct
98 * directly or via call_rcu()
99 * so sk_prot_free() might be called
100 * after rxe_net_exit().
101 */
102 sk_owner_clear(sk);
103 #endif /* CONFIG_DEBUG_LOCK_ALLOC */
104 }
105
rxe_find_route4(struct rxe_qp * qp,struct net * net,struct net_device * ndev,struct in_addr * saddr,struct in_addr * daddr)106 static struct dst_entry *rxe_find_route4(struct rxe_qp *qp,
107 struct net *net,
108 struct net_device *ndev,
109 struct in_addr *saddr,
110 struct in_addr *daddr)
111 {
112 struct rtable *rt;
113 struct flowi4 fl = {};
114
115 fl.flowi4_oif = ndev->ifindex;
116 memcpy(&fl.saddr, saddr, sizeof(*saddr));
117 memcpy(&fl.daddr, daddr, sizeof(*daddr));
118 fl.flowi4_proto = IPPROTO_UDP;
119
120 rt = ip_route_output_key(net, &fl);
121 if (IS_ERR(rt)) {
122 rxe_dbg_qp(qp, "no route to %pI4\n", &daddr->s_addr);
123 return NULL;
124 }
125
126 return &rt->dst;
127 }
128
129 #if IS_ENABLED(CONFIG_IPV6)
rxe_find_route6(struct rxe_qp * qp,struct net * net,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 *net,
132 struct net_device *ndev,
133 struct in6_addr *saddr,
134 struct in6_addr *daddr)
135 {
136 struct dst_entry *ndst;
137 struct flowi6 fl6 = {};
138
139 fl6.flowi6_oif = ndev->ifindex;
140 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
141 memcpy(&fl6.daddr, daddr, sizeof(*daddr));
142 fl6.flowi6_proto = IPPROTO_UDP;
143
144 ndst = ip6_dst_lookup_flow(net, rxe_ns_pernet_sk6(net), &fl6, NULL);
145 if (IS_ERR(ndst)) {
146 rxe_dbg_qp(qp, "no route to %pI6\n", daddr);
147 return NULL;
148 }
149
150 if (unlikely(ndst->error)) {
151 rxe_dbg_qp(qp, "no route to %pI6\n", daddr);
152 goto put;
153 }
154
155 return ndst;
156 put:
157 dst_release(ndst);
158 return NULL;
159 }
160
161 #else
162
rxe_find_route6(struct rxe_qp * qp,struct net * net,struct net_device * ndev,struct in6_addr * saddr,struct in6_addr * daddr)163 static struct dst_entry *rxe_find_route6(struct rxe_qp *qp,
164 struct net *net,
165 struct net_device *ndev,
166 struct in6_addr *saddr,
167 struct in6_addr *daddr)
168 {
169 return NULL;
170 }
171
172 #endif
173
rxe_find_route(struct net_device * ndev,struct rxe_qp * qp,struct rxe_av * av)174 static struct dst_entry *rxe_find_route(struct net_device *ndev,
175 struct rxe_qp *qp,
176 struct rxe_av *av)
177 {
178 struct dst_entry *dst = NULL;
179 struct net *net;
180
181 if (qp_type(qp) == IB_QPT_RC)
182 dst = sk_dst_get(qp->sk->sk);
183
184 if (!dst || !dst_check(dst, qp->dst_cookie)) {
185 if (dst)
186 dst_release(dst);
187
188 net = dev_net(ndev);
189
190 if (av->network_type == RXE_NETWORK_TYPE_IPV4) {
191 struct in_addr *saddr;
192 struct in_addr *daddr;
193
194 saddr = &av->sgid_addr._sockaddr_in.sin_addr;
195 daddr = &av->dgid_addr._sockaddr_in.sin_addr;
196 dst = rxe_find_route4(qp, net, ndev, saddr, daddr);
197 } else if (av->network_type == RXE_NETWORK_TYPE_IPV6) {
198 struct in6_addr *saddr6;
199 struct in6_addr *daddr6;
200
201 saddr6 = &av->sgid_addr._sockaddr_in6.sin6_addr;
202 daddr6 = &av->dgid_addr._sockaddr_in6.sin6_addr;
203 dst = rxe_find_route6(qp, net, ndev, saddr6, daddr6);
204 #if IS_ENABLED(CONFIG_IPV6)
205 if (dst)
206 qp->dst_cookie =
207 rt6_get_cookie((struct rt6_info *)dst);
208 #endif
209 }
210
211 if (dst && (qp_type(qp) == IB_QPT_RC)) {
212 dst_hold(dst);
213 sk_dst_set(qp->sk->sk, dst);
214 }
215 }
216 return dst;
217 }
218
rxe_udp_encap_recv(struct sock * sk,struct sk_buff * skb)219 static int rxe_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
220 {
221 struct udphdr *udph;
222 struct rxe_dev *rxe;
223 struct net_device *ndev = skb->dev;
224 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
225
226 /* takes a reference on rxe->ib_dev
227 * drop when skb is freed
228 */
229 rxe = rxe_get_dev_from_net(ndev);
230 if (!rxe && is_vlan_dev(ndev))
231 rxe = rxe_get_dev_from_net(vlan_dev_real_dev(ndev));
232 if (!rxe)
233 goto drop;
234
235 if (skb_linearize(skb)) {
236 ib_device_put(&rxe->ib_dev);
237 goto drop;
238 }
239
240 udph = udp_hdr(skb);
241 pkt->rxe = rxe;
242 pkt->port_num = 1;
243 pkt->hdr = (u8 *)(udph + 1);
244 pkt->mask = RXE_GRH_MASK;
245 pkt->paylen = udp_get_len_short(udph) - sizeof(*udph);
246
247 /* remove udp header */
248 skb_pull(skb, sizeof(struct udphdr));
249
250 rxe_rcv(skb);
251
252 return 0;
253 drop:
254 kfree_skb(skb);
255
256 return 0;
257 }
258
rxe_setup_udp_tunnel(struct net * net,__be16 port,bool ipv6)259 static struct socket *rxe_setup_udp_tunnel(struct net *net, __be16 port,
260 bool ipv6)
261 {
262 int err;
263 struct socket *sock;
264 struct udp_port_cfg udp_cfg = { };
265 struct udp_tunnel_sock_cfg tnl_cfg = { };
266
267 if (ipv6) {
268 udp_cfg.family = AF_INET6;
269 udp_cfg.ipv6_v6only = 1;
270 } else {
271 udp_cfg.family = AF_INET;
272 }
273
274 udp_cfg.local_udp_port = port;
275
276 /* Create UDP socket */
277 err = udp_sock_create(net, &udp_cfg, &sock);
278 if (err < 0)
279 return ERR_PTR(err);
280 rxe_reclassify_recv_socket(sock);
281
282 tnl_cfg.encap_type = 1;
283 tnl_cfg.encap_rcv = rxe_udp_encap_recv;
284
285 /* Setup UDP tunnel */
286 setup_udp_tunnel_sock(net, sock->sk, &tnl_cfg);
287
288 return sock;
289 }
290
rxe_release_udp_tunnel(struct sock * sk)291 static void rxe_release_udp_tunnel(struct sock *sk)
292 {
293 if (sk)
294 udp_tunnel_sock_release(sk);
295 }
296
prepare_udp_hdr(struct sk_buff * skb,__be16 src_port,__be16 dst_port)297 static void prepare_udp_hdr(struct sk_buff *skb, __be16 src_port,
298 __be16 dst_port)
299 {
300 struct udphdr *udph;
301
302 __skb_push(skb, sizeof(*udph));
303 skb_reset_transport_header(skb);
304 udph = udp_hdr(skb);
305
306 udph->dest = dst_port;
307 udph->source = src_port;
308 udp_set_len_short(udph, skb->len);
309 udph->check = 0;
310 }
311
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)312 static void prepare_ipv4_hdr(struct dst_entry *dst, struct sk_buff *skb,
313 __be32 saddr, __be32 daddr, __u8 proto,
314 __u8 tos, __u8 ttl, __be16 df, bool xnet)
315 {
316 struct iphdr *iph;
317
318 skb_scrub_packet(skb, xnet);
319
320 skb_clear_hash(skb);
321 skb_dst_set(skb, dst_clone(dst));
322 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
323
324 skb_push(skb, sizeof(struct iphdr));
325 skb_reset_network_header(skb);
326
327 iph = ip_hdr(skb);
328
329 iph->version = IPVERSION;
330 iph->ihl = sizeof(struct iphdr) >> 2;
331 iph->tot_len = htons(skb->len);
332 iph->frag_off = df;
333 iph->protocol = proto;
334 iph->tos = tos;
335 iph->daddr = daddr;
336 iph->saddr = saddr;
337 iph->ttl = ttl;
338 __ip_select_ident(dev_net(dst->dev), iph,
339 skb_shinfo(skb)->gso_segs ?: 1);
340 }
341
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)342 static void prepare_ipv6_hdr(struct dst_entry *dst, struct sk_buff *skb,
343 struct in6_addr *saddr, struct in6_addr *daddr,
344 __u8 proto, __u8 prio, __u8 ttl)
345 {
346 struct ipv6hdr *ip6h;
347
348 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
349 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED
350 | IPSKB_REROUTED);
351 skb_dst_set(skb, dst_clone(dst));
352
353 __skb_push(skb, sizeof(*ip6h));
354 skb_reset_network_header(skb);
355 ip6h = ipv6_hdr(skb);
356 ip6_flow_hdr(ip6h, prio, htonl(0));
357 ip6h->payload_len = htons(skb->len);
358 ip6h->nexthdr = proto;
359 ip6h->hop_limit = ttl;
360 ip6h->daddr = *daddr;
361 ip6h->saddr = *saddr;
362 ip6h->payload_len = htons(skb->len - sizeof(*ip6h));
363 }
364
prepare4(struct rxe_av * av,struct rxe_pkt_info * pkt,struct sk_buff * skb)365 static int prepare4(struct rxe_av *av, struct rxe_pkt_info *pkt,
366 struct sk_buff *skb)
367 {
368 struct rxe_qp *qp = pkt->qp;
369 struct dst_entry *dst;
370 bool xnet = false;
371 __be16 df = htons(IP_DF);
372 struct in_addr *saddr = &av->sgid_addr._sockaddr_in.sin_addr;
373 struct in_addr *daddr = &av->dgid_addr._sockaddr_in.sin_addr;
374
375 dst = rxe_find_route(skb->dev, qp, av);
376 if (!dst) {
377 rxe_dbg_qp(qp, "Host not reachable\n");
378 return -EHOSTUNREACH;
379 }
380
381 prepare_udp_hdr(skb, cpu_to_be16(qp->src_port),
382 cpu_to_be16(ROCE_V2_UDP_DPORT));
383
384 prepare_ipv4_hdr(dst, skb, saddr->s_addr, daddr->s_addr, IPPROTO_UDP,
385 av->grh.traffic_class, av->grh.hop_limit, df, xnet);
386
387 dst_release(dst);
388 return 0;
389 }
390
prepare6(struct rxe_av * av,struct rxe_pkt_info * pkt,struct sk_buff * skb)391 static int prepare6(struct rxe_av *av, struct rxe_pkt_info *pkt,
392 struct sk_buff *skb)
393 {
394 struct rxe_qp *qp = pkt->qp;
395 struct dst_entry *dst;
396 struct in6_addr *saddr = &av->sgid_addr._sockaddr_in6.sin6_addr;
397 struct in6_addr *daddr = &av->dgid_addr._sockaddr_in6.sin6_addr;
398
399 dst = rxe_find_route(skb->dev, qp, av);
400 if (!dst) {
401 rxe_dbg_qp(qp, "Host not reachable\n");
402 return -EHOSTUNREACH;
403 }
404
405 prepare_udp_hdr(skb, cpu_to_be16(qp->src_port),
406 cpu_to_be16(ROCE_V2_UDP_DPORT));
407
408 prepare_ipv6_hdr(dst, skb, saddr, daddr, IPPROTO_UDP,
409 av->grh.traffic_class,
410 av->grh.hop_limit);
411
412 dst_release(dst);
413 return 0;
414 }
415
rxe_prepare(struct rxe_av * av,struct rxe_pkt_info * pkt,struct sk_buff * skb)416 int rxe_prepare(struct rxe_av *av, struct rxe_pkt_info *pkt,
417 struct sk_buff *skb)
418 {
419 int err = 0;
420
421 if (skb->protocol == htons(ETH_P_IP))
422 err = prepare4(av, pkt, skb);
423 else if (skb->protocol == htons(ETH_P_IPV6))
424 err = prepare6(av, pkt, skb);
425
426 if (ether_addr_equal(skb->dev->dev_addr, av->dmac))
427 pkt->mask |= RXE_LOOPBACK_MASK;
428
429 return err;
430 }
431
rxe_skb_tx_dtor(struct sk_buff * skb)432 static void rxe_skb_tx_dtor(struct sk_buff *skb)
433 {
434 struct rxe_qp *qp = skb->sk->sk_user_data;
435 int skb_out;
436
437 skb_out = atomic_dec_return(&qp->skb_out);
438 if (unlikely(qp->need_req_skb &&
439 skb_out < RXE_INFLIGHT_SKBS_PER_QP_LOW))
440 rxe_sched_task(&qp->send_task);
441
442 rxe_put(qp);
443 sock_put(skb->sk);
444 }
445
rxe_send(struct sk_buff * skb,struct rxe_pkt_info * pkt)446 static int rxe_send(struct sk_buff *skb, struct rxe_pkt_info *pkt)
447 {
448 int err;
449 struct sock *sk = pkt->qp->sk->sk;
450
451 sock_hold(sk);
452 skb->sk = sk;
453 skb->destructor = rxe_skb_tx_dtor;
454 rxe_get(pkt->qp);
455 atomic_inc(&pkt->qp->skb_out);
456
457 if (skb->protocol == htons(ETH_P_IP))
458 err = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
459 else
460 err = ip6_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
461
462 return err;
463 }
464
465 /* fix up a send packet to match the packets
466 * received from UDP before looping them back
467 */
rxe_loopback(struct sk_buff * skb,struct rxe_pkt_info * pkt)468 static int rxe_loopback(struct sk_buff *skb, struct rxe_pkt_info *pkt)
469 {
470 struct sock *sk = pkt->qp->sk->sk;
471
472 memcpy(SKB_TO_PKT(skb), pkt, sizeof(*pkt));
473
474 sock_hold(sk);
475 skb->sk = sk;
476 skb->destructor = rxe_skb_tx_dtor;
477 rxe_get(pkt->qp);
478 atomic_inc(&pkt->qp->skb_out);
479
480 if (skb->protocol == htons(ETH_P_IP))
481 skb_pull(skb, sizeof(struct iphdr));
482 else
483 skb_pull(skb, sizeof(struct ipv6hdr));
484
485 if (WARN_ON(!ib_device_try_get(&pkt->rxe->ib_dev))) {
486 kfree_skb(skb);
487 return -EIO;
488 }
489
490 /* remove udp header */
491 skb_pull(skb, sizeof(struct udphdr));
492
493 rxe_rcv(skb);
494
495 return 0;
496 }
497
rxe_xmit_packet(struct rxe_qp * qp,struct rxe_pkt_info * pkt,struct sk_buff * skb)498 int rxe_xmit_packet(struct rxe_qp *qp, struct rxe_pkt_info *pkt,
499 struct sk_buff *skb)
500 {
501 int err;
502 int is_request = pkt->mask & RXE_REQ_MASK;
503 struct rxe_dev *rxe = to_rdev(qp->ibqp.device);
504 unsigned int skblen = skb->len;
505 unsigned long flags;
506
507 spin_lock_irqsave(&qp->state_lock, flags);
508 if ((is_request && (qp_state(qp) < IB_QPS_RTS)) ||
509 (!is_request && (qp_state(qp) < IB_QPS_RTR))) {
510 spin_unlock_irqrestore(&qp->state_lock, flags);
511 rxe_dbg_qp(qp, "Packet dropped. QP is not in ready state\n");
512 goto drop;
513 }
514 spin_unlock_irqrestore(&qp->state_lock, flags);
515
516 rxe_icrc_generate(skb, pkt);
517
518 if (pkt->mask & RXE_LOOPBACK_MASK)
519 err = rxe_loopback(skb, pkt);
520 else
521 err = rxe_send(skb, pkt);
522 if (err) {
523 rxe_counter_inc(rxe, RXE_CNT_SEND_ERR);
524 return err;
525 }
526
527 rxe_counter_inc(rxe, RXE_CNT_SENT_PKTS);
528 rxe_counter_add(rxe, RXE_CNT_SENT_BYTES, skblen);
529 goto done;
530
531 drop:
532 kfree_skb(skb);
533 err = 0;
534 done:
535 return err;
536 }
537
rxe_init_packet(struct rxe_dev * rxe,struct rxe_av * av,int paylen,struct rxe_pkt_info * pkt)538 struct sk_buff *rxe_init_packet(struct rxe_dev *rxe, struct rxe_av *av,
539 int paylen, struct rxe_pkt_info *pkt)
540 {
541 unsigned int hdr_len;
542 struct sk_buff *skb = NULL;
543 struct net_device *ndev;
544 const struct ib_gid_attr *attr;
545 const int port_num = 1;
546
547 attr = rdma_get_gid_attr(&rxe->ib_dev, port_num, av->grh.sgid_index);
548 if (IS_ERR(attr))
549 return NULL;
550
551 if (av->network_type == RXE_NETWORK_TYPE_IPV4)
552 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
553 sizeof(struct iphdr);
554 else
555 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
556 sizeof(struct ipv6hdr);
557
558 rcu_read_lock();
559 ndev = rdma_read_gid_attr_ndev_rcu(attr);
560 if (IS_ERR(ndev)) {
561 rcu_read_unlock();
562 goto out;
563 }
564 skb = alloc_skb(paylen + hdr_len + LL_RESERVED_SPACE(ndev),
565 GFP_ATOMIC);
566
567 if (unlikely(!skb)) {
568 rcu_read_unlock();
569 goto out;
570 }
571
572 /* Add time stamp to skb. */
573 skb->tstamp = ktime_get();
574
575 skb_reserve(skb, hdr_len + LL_RESERVED_SPACE(ndev));
576
577 /* FIXME: hold reference to this netdev until life of this skb. */
578 skb->dev = ndev;
579 rcu_read_unlock();
580
581 if (av->network_type == RXE_NETWORK_TYPE_IPV4)
582 skb->protocol = htons(ETH_P_IP);
583 else
584 skb->protocol = htons(ETH_P_IPV6);
585
586 pkt->rxe = rxe;
587 pkt->port_num = port_num;
588 pkt->hdr = skb_put(skb, paylen);
589 pkt->mask |= RXE_GRH_MASK;
590
591 out:
592 rdma_put_gid_attr(attr);
593 return skb;
594 }
595
596 /*
597 * this is required by rxe_cfg to match rxe devices in
598 * /sys/class/infiniband up with their underlying ethernet devices
599 */
rxe_parent_name(struct rxe_dev * rxe,unsigned int port_num)600 const char *rxe_parent_name(struct rxe_dev *rxe, unsigned int port_num)
601 {
602 struct net_device *ndev;
603 char *ndev_name;
604
605 ndev = ib_device_get_netdev(&rxe->ib_dev, 1);
606 if (!ndev)
607 return NULL;
608 ndev_name = ndev->name;
609 dev_put(ndev);
610
611 return ndev_name;
612 }
613
rxe_net_add(const char * ibdev_name,struct net_device * ndev)614 int rxe_net_add(const char *ibdev_name, struct net_device *ndev)
615 {
616 int err;
617 struct rxe_dev *rxe = NULL;
618
619 rxe = ib_alloc_device(rxe_dev, ib_dev);
620 if (!rxe)
621 return -ENOMEM;
622
623 ib_mark_name_assigned_by_user(&rxe->ib_dev);
624
625 err = rxe_add(rxe, ndev->mtu, ibdev_name, ndev);
626 if (err) {
627 ib_dealloc_device(&rxe->ib_dev);
628 return err;
629 }
630
631 return 0;
632 }
633
rxe_sock_put(struct sock * sk,void (* set_sk)(struct net *,struct sock *),struct net * net)634 static void rxe_sock_put(struct sock *sk,
635 void (*set_sk)(struct net *, struct sock *),
636 struct net *net)
637 {
638 if (refcount_read(&sk->sk_refcnt) > SK_REF_FOR_TUNNEL) {
639 __sock_put(sk);
640 } else {
641 rxe_release_udp_tunnel(sk);
642 sk = NULL;
643 set_sk(net, sk);
644 }
645 }
646
rxe_net_del(struct ib_device * dev)647 void rxe_net_del(struct ib_device *dev)
648 {
649 struct net_device *ndev;
650 struct sock *sk;
651 struct net *net;
652
653 ndev = ib_device_get_netdev(dev, 1);
654 if (!ndev)
655 return;
656
657 net = dev_net(ndev);
658
659 sk = rxe_ns_pernet_sk4(net);
660 if (sk)
661 rxe_sock_put(sk, rxe_ns_pernet_set_sk4, net);
662
663 sk = rxe_ns_pernet_sk6(net);
664 if (sk)
665 rxe_sock_put(sk, rxe_ns_pernet_set_sk6, net);
666
667 dev_put(ndev);
668 }
669
rxe_port_event(struct rxe_dev * rxe,enum ib_event_type event)670 static void rxe_port_event(struct rxe_dev *rxe,
671 enum ib_event_type event)
672 {
673 struct ib_event ev;
674
675 ev.device = &rxe->ib_dev;
676 ev.element.port_num = 1;
677 ev.event = event;
678
679 ib_dispatch_event(&ev);
680 }
681
682 /* Caller must hold net_info_lock */
rxe_port_up(struct rxe_dev * rxe)683 void rxe_port_up(struct rxe_dev *rxe)
684 {
685 rxe_port_event(rxe, IB_EVENT_PORT_ACTIVE);
686 dev_info(&rxe->ib_dev.dev, "set active\n");
687 }
688
689 /* Caller must hold net_info_lock */
rxe_port_down(struct rxe_dev * rxe)690 void rxe_port_down(struct rxe_dev *rxe)
691 {
692 rxe_port_event(rxe, IB_EVENT_PORT_ERR);
693 rxe_counter_inc(rxe, RXE_CNT_LINK_DOWNED);
694 dev_info(&rxe->ib_dev.dev, "set down\n");
695 }
696
rxe_set_port_state(struct rxe_dev * rxe)697 void rxe_set_port_state(struct rxe_dev *rxe)
698 {
699 struct net_device *ndev;
700
701 ndev = ib_device_get_netdev(&rxe->ib_dev, 1);
702 if (!ndev)
703 return;
704
705 if (ib_get_curr_port_state(ndev) == IB_PORT_ACTIVE)
706 rxe_port_up(rxe);
707 else
708 rxe_port_down(rxe);
709
710 dev_put(ndev);
711 }
712
rxe_notify(struct notifier_block * not_blk,unsigned long event,void * arg)713 static int rxe_notify(struct notifier_block *not_blk,
714 unsigned long event,
715 void *arg)
716 {
717 struct net_device *ndev = netdev_notifier_info_to_dev(arg);
718 struct rxe_dev *rxe = rxe_get_dev_from_net(ndev);
719
720 if (!rxe)
721 return NOTIFY_OK;
722
723 switch (event) {
724 case NETDEV_UNREGISTER:
725 ib_unregister_device_queued(&rxe->ib_dev);
726 rxe_net_del(&rxe->ib_dev);
727 break;
728 case NETDEV_CHANGEMTU:
729 rxe_dbg_dev(rxe, "%s changed mtu to %d\n", ndev->name, ndev->mtu);
730 rxe_set_mtu(rxe, ndev->mtu);
731 break;
732 case NETDEV_DOWN:
733 case NETDEV_CHANGE:
734 if (ib_get_curr_port_state(ndev) == IB_PORT_DOWN)
735 rxe_counter_inc(rxe, RXE_CNT_LINK_DOWNED);
736 break;
737 case NETDEV_REBOOT:
738 case NETDEV_GOING_DOWN:
739 case NETDEV_CHANGEADDR:
740 case NETDEV_CHANGENAME:
741 case NETDEV_FEAT_CHANGE:
742 default:
743 rxe_dbg_dev(rxe, "ignoring netdev event = %ld for %s\n",
744 event, ndev->name);
745 break;
746 }
747
748 ib_device_put(&rxe->ib_dev);
749 return NOTIFY_OK;
750 }
751
752 static struct notifier_block rxe_net_notifier = {
753 .notifier_call = rxe_notify,
754 };
755
rxe_net_ipv4_init(struct net * net)756 static int rxe_net_ipv4_init(struct net *net)
757 {
758 struct sock *sk;
759 struct socket *sock;
760
761 sk = rxe_ns_pernet_sk4(net);
762 if (sk) {
763 sock_hold(sk);
764 return 0;
765 }
766
767 sock = rxe_setup_udp_tunnel(net, htons(ROCE_V2_UDP_DPORT), false);
768 if (IS_ERR(sock)) {
769 pr_err("Failed to create IPv4 UDP tunnel\n");
770 return -1;
771 }
772 rxe_ns_pernet_set_sk4(net, sock->sk);
773
774 return 0;
775 }
776
rxe_net_ipv6_init(struct net * net)777 static int rxe_net_ipv6_init(struct net *net)
778 {
779 #if IS_ENABLED(CONFIG_IPV6)
780 struct sock *sk;
781 struct socket *sock;
782
783 sk = rxe_ns_pernet_sk6(net);
784 if (sk) {
785 sock_hold(sk);
786 return 0;
787 }
788
789 sock = rxe_setup_udp_tunnel(net, htons(ROCE_V2_UDP_DPORT), true);
790 if (PTR_ERR(sock) == -EAFNOSUPPORT) {
791 pr_warn("IPv6 is not supported, can not create a UDPv6 socket\n");
792 return 0;
793 }
794
795 if (IS_ERR(sock)) {
796 pr_err("Failed to create IPv6 UDP tunnel\n");
797 return -1;
798 }
799
800 rxe_ns_pernet_set_sk6(net, sock->sk);
801
802 #endif
803 return 0;
804 }
805
rxe_register_notifier(void)806 int rxe_register_notifier(void)
807 {
808 int err;
809
810 err = register_netdevice_notifier(&rxe_net_notifier);
811 if (err) {
812 pr_err("Failed to register netdev notifier\n");
813 return -1;
814 }
815
816 return 0;
817 }
818
rxe_net_exit(void)819 void rxe_net_exit(void)
820 {
821 unregister_netdevice_notifier(&rxe_net_notifier);
822 }
823
rxe_net_init(struct net_device * ndev)824 int rxe_net_init(struct net_device *ndev)
825 {
826 struct net *net;
827 struct sock *sk;
828 int err;
829
830 net = dev_net(ndev);
831
832 err = rxe_net_ipv4_init(net);
833 if (err)
834 return err;
835
836 err = rxe_net_ipv6_init(net);
837 if (err)
838 goto err_out;
839
840 return 0;
841
842 err_out:
843 /* If ipv6 error, release ipv4 resource */
844 sk = rxe_ns_pernet_sk4(net);
845 if (sk)
846 rxe_sock_put(sk, rxe_ns_pernet_set_sk4, net);
847
848 return err;
849 }
850