1
2
3
4
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
21static struct rxe_recv_sockets recv_sockets;
22
23int rxe_mcast_add(struct rxe_dev *rxe, union ib_gid *mgid)
24{
25 int err;
26 unsigned char ll_addr[ETH_ALEN];
27
28 ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr);
29 err = dev_mc_add(rxe->ndev, ll_addr);
30
31 return err;
32}
33
34int rxe_mcast_delete(struct rxe_dev *rxe, union ib_gid *mgid)
35{
36 int err;
37 unsigned char ll_addr[ETH_ALEN];
38
39 ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr);
40 err = dev_mc_del(rxe->ndev, ll_addr);
41
42 return err;
43}
44
45static struct dst_entry *rxe_find_route4(struct net_device *ndev,
46 struct in_addr *saddr,
47 struct in_addr *daddr)
48{
49 struct rtable *rt;
50 struct flowi4 fl = { { 0 } };
51
52 memset(&fl, 0, sizeof(fl));
53 fl.flowi4_oif = ndev->ifindex;
54 memcpy(&fl.saddr, saddr, sizeof(*saddr));
55 memcpy(&fl.daddr, daddr, sizeof(*daddr));
56 fl.flowi4_proto = IPPROTO_UDP;
57
58 rt = ip_route_output_key(&init_net, &fl);
59 if (IS_ERR(rt)) {
60 pr_err_ratelimited("no route to %pI4\n", &daddr->s_addr);
61 return NULL;
62 }
63
64 return &rt->dst;
65}
66
67#if IS_ENABLED(CONFIG_IPV6)
68static struct dst_entry *rxe_find_route6(struct net_device *ndev,
69 struct in6_addr *saddr,
70 struct in6_addr *daddr)
71{
72 struct dst_entry *ndst;
73 struct flowi6 fl6 = { { 0 } };
74
75 memset(&fl6, 0, sizeof(fl6));
76 fl6.flowi6_oif = ndev->ifindex;
77 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
78 memcpy(&fl6.daddr, daddr, sizeof(*daddr));
79 fl6.flowi6_proto = IPPROTO_UDP;
80
81 ndst = ipv6_stub->ipv6_dst_lookup_flow(sock_net(recv_sockets.sk6->sk),
82 recv_sockets.sk6->sk, &fl6,
83 NULL);
84 if (IS_ERR(ndst)) {
85 pr_err_ratelimited("no route to %pI6\n", daddr);
86 return NULL;
87 }
88
89 if (unlikely(ndst->error)) {
90 pr_err("no route to %pI6\n", daddr);
91 goto put;
92 }
93
94 return ndst;
95put:
96 dst_release(ndst);
97 return NULL;
98}
99
100#else
101
102static struct dst_entry *rxe_find_route6(struct net_device *ndev,
103 struct in6_addr *saddr,
104 struct in6_addr *daddr)
105{
106 return NULL;
107}
108
109#endif
110
111static struct dst_entry *rxe_find_route(struct net_device *ndev,
112 struct rxe_qp *qp,
113 struct rxe_av *av)
114{
115 struct dst_entry *dst = NULL;
116
117 if (qp_type(qp) == IB_QPT_RC)
118 dst = sk_dst_get(qp->sk->sk);
119
120 if (!dst || !dst_check(dst, qp->dst_cookie)) {
121 if (dst)
122 dst_release(dst);
123
124 if (av->network_type == RXE_NETWORK_TYPE_IPV4) {
125 struct in_addr *saddr;
126 struct in_addr *daddr;
127
128 saddr = &av->sgid_addr._sockaddr_in.sin_addr;
129 daddr = &av->dgid_addr._sockaddr_in.sin_addr;
130 dst = rxe_find_route4(ndev, saddr, daddr);
131 } else if (av->network_type == RXE_NETWORK_TYPE_IPV6) {
132 struct in6_addr *saddr6;
133 struct in6_addr *daddr6;
134
135 saddr6 = &av->sgid_addr._sockaddr_in6.sin6_addr;
136 daddr6 = &av->dgid_addr._sockaddr_in6.sin6_addr;
137 dst = rxe_find_route6(ndev, saddr6, daddr6);
138#if IS_ENABLED(CONFIG_IPV6)
139 if (dst)
140 qp->dst_cookie =
141 rt6_get_cookie((struct rt6_info *)dst);
142#endif
143 }
144
145 if (dst && (qp_type(qp) == IB_QPT_RC)) {
146 dst_hold(dst);
147 sk_dst_set(qp->sk->sk, dst);
148 }
149 }
150 return dst;
151}
152
153static int rxe_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
154{
155 struct udphdr *udph;
156 struct rxe_dev *rxe;
157 struct net_device *ndev = skb->dev;
158 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
159
160
161
162
163 rxe = rxe_get_dev_from_net(ndev);
164 if (!rxe && is_vlan_dev(ndev))
165 rxe = rxe_get_dev_from_net(vlan_dev_real_dev(ndev));
166 if (!rxe)
167 goto drop;
168
169 if (skb_linearize(skb)) {
170 pr_err("skb_linearize failed\n");
171 ib_device_put(&rxe->ib_dev);
172 goto drop;
173 }
174
175 udph = udp_hdr(skb);
176 pkt->rxe = rxe;
177 pkt->port_num = 1;
178 pkt->hdr = (u8 *)(udph + 1);
179 pkt->mask = RXE_GRH_MASK;
180 pkt->paylen = be16_to_cpu(udph->len) - sizeof(*udph);
181
182 rxe_rcv(skb);
183
184 return 0;
185drop:
186 kfree_skb(skb);
187
188 return 0;
189}
190
191static struct socket *rxe_setup_udp_tunnel(struct net *net, __be16 port,
192 bool ipv6)
193{
194 int err;
195 struct socket *sock;
196 struct udp_port_cfg udp_cfg = { };
197 struct udp_tunnel_sock_cfg tnl_cfg = { };
198
199 if (ipv6) {
200 udp_cfg.family = AF_INET6;
201 udp_cfg.ipv6_v6only = 1;
202 } else {
203 udp_cfg.family = AF_INET;
204 }
205
206 udp_cfg.local_udp_port = port;
207
208
209 err = udp_sock_create(net, &udp_cfg, &sock);
210 if (err < 0)
211 return ERR_PTR(err);
212
213 tnl_cfg.encap_type = 1;
214 tnl_cfg.encap_rcv = rxe_udp_encap_recv;
215
216
217 setup_udp_tunnel_sock(net, sock, &tnl_cfg);
218
219 return sock;
220}
221
222static void rxe_release_udp_tunnel(struct socket *sk)
223{
224 if (sk)
225 udp_tunnel_sock_release(sk);
226}
227
228static void prepare_udp_hdr(struct sk_buff *skb, __be16 src_port,
229 __be16 dst_port)
230{
231 struct udphdr *udph;
232
233 __skb_push(skb, sizeof(*udph));
234 skb_reset_transport_header(skb);
235 udph = udp_hdr(skb);
236
237 udph->dest = dst_port;
238 udph->source = src_port;
239 udph->len = htons(skb->len);
240 udph->check = 0;
241}
242
243static void prepare_ipv4_hdr(struct dst_entry *dst, struct sk_buff *skb,
244 __be32 saddr, __be32 daddr, __u8 proto,
245 __u8 tos, __u8 ttl, __be16 df, bool xnet)
246{
247 struct iphdr *iph;
248
249 skb_scrub_packet(skb, xnet);
250
251 skb_clear_hash(skb);
252 skb_dst_set(skb, dst_clone(dst));
253 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
254
255 skb_push(skb, sizeof(struct iphdr));
256 skb_reset_network_header(skb);
257
258 iph = ip_hdr(skb);
259
260 iph->version = IPVERSION;
261 iph->ihl = sizeof(struct iphdr) >> 2;
262 iph->frag_off = df;
263 iph->protocol = proto;
264 iph->tos = tos;
265 iph->daddr = daddr;
266 iph->saddr = saddr;
267 iph->ttl = ttl;
268 __ip_select_ident(dev_net(dst->dev), iph,
269 skb_shinfo(skb)->gso_segs ?: 1);
270 iph->tot_len = htons(skb->len);
271 ip_send_check(iph);
272}
273
274static void prepare_ipv6_hdr(struct dst_entry *dst, struct sk_buff *skb,
275 struct in6_addr *saddr, struct in6_addr *daddr,
276 __u8 proto, __u8 prio, __u8 ttl)
277{
278 struct ipv6hdr *ip6h;
279
280 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
281 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED
282 | IPSKB_REROUTED);
283 skb_dst_set(skb, dst_clone(dst));
284
285 __skb_push(skb, sizeof(*ip6h));
286 skb_reset_network_header(skb);
287 ip6h = ipv6_hdr(skb);
288 ip6_flow_hdr(ip6h, prio, htonl(0));
289 ip6h->payload_len = htons(skb->len);
290 ip6h->nexthdr = proto;
291 ip6h->hop_limit = ttl;
292 ip6h->daddr = *daddr;
293 ip6h->saddr = *saddr;
294 ip6h->payload_len = htons(skb->len - sizeof(*ip6h));
295}
296
297static int prepare4(struct rxe_pkt_info *pkt, struct sk_buff *skb)
298{
299 struct rxe_qp *qp = pkt->qp;
300 struct dst_entry *dst;
301 bool xnet = false;
302 __be16 df = htons(IP_DF);
303 struct rxe_av *av = rxe_get_av(pkt);
304 struct in_addr *saddr = &av->sgid_addr._sockaddr_in.sin_addr;
305 struct in_addr *daddr = &av->dgid_addr._sockaddr_in.sin_addr;
306
307 dst = rxe_find_route(skb->dev, qp, av);
308 if (!dst) {
309 pr_err("Host not reachable\n");
310 return -EHOSTUNREACH;
311 }
312
313 prepare_udp_hdr(skb, cpu_to_be16(qp->src_port),
314 cpu_to_be16(ROCE_V2_UDP_DPORT));
315
316 prepare_ipv4_hdr(dst, skb, saddr->s_addr, daddr->s_addr, IPPROTO_UDP,
317 av->grh.traffic_class, av->grh.hop_limit, df, xnet);
318
319 dst_release(dst);
320 return 0;
321}
322
323static int prepare6(struct rxe_pkt_info *pkt, struct sk_buff *skb)
324{
325 struct rxe_qp *qp = pkt->qp;
326 struct dst_entry *dst;
327 struct rxe_av *av = rxe_get_av(pkt);
328 struct in6_addr *saddr = &av->sgid_addr._sockaddr_in6.sin6_addr;
329 struct in6_addr *daddr = &av->dgid_addr._sockaddr_in6.sin6_addr;
330
331 dst = rxe_find_route(skb->dev, qp, av);
332 if (!dst) {
333 pr_err("Host not reachable\n");
334 return -EHOSTUNREACH;
335 }
336
337 prepare_udp_hdr(skb, cpu_to_be16(qp->src_port),
338 cpu_to_be16(ROCE_V2_UDP_DPORT));
339
340 prepare_ipv6_hdr(dst, skb, saddr, daddr, IPPROTO_UDP,
341 av->grh.traffic_class,
342 av->grh.hop_limit);
343
344 dst_release(dst);
345 return 0;
346}
347
348int rxe_prepare(struct rxe_pkt_info *pkt, struct sk_buff *skb, u32 *crc)
349{
350 int err = 0;
351
352 if (skb->protocol == htons(ETH_P_IP))
353 err = prepare4(pkt, skb);
354 else if (skb->protocol == htons(ETH_P_IPV6))
355 err = prepare6(pkt, skb);
356
357 *crc = rxe_icrc_hdr(pkt, skb);
358
359 if (ether_addr_equal(skb->dev->dev_addr, rxe_get_av(pkt)->dmac))
360 pkt->mask |= RXE_LOOPBACK_MASK;
361
362 return err;
363}
364
365static void rxe_skb_tx_dtor(struct sk_buff *skb)
366{
367 struct sock *sk = skb->sk;
368 struct rxe_qp *qp = sk->sk_user_data;
369 int skb_out = atomic_dec_return(&qp->skb_out);
370
371 if (unlikely(qp->need_req_skb &&
372 skb_out < RXE_INFLIGHT_SKBS_PER_QP_LOW))
373 rxe_run_task(&qp->req.task, 1);
374
375 rxe_drop_ref(qp);
376}
377
378int rxe_send(struct rxe_pkt_info *pkt, struct sk_buff *skb)
379{
380 int err;
381
382 skb->destructor = rxe_skb_tx_dtor;
383 skb->sk = pkt->qp->sk->sk;
384
385 rxe_add_ref(pkt->qp);
386 atomic_inc(&pkt->qp->skb_out);
387
388 if (skb->protocol == htons(ETH_P_IP)) {
389 err = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
390 } else if (skb->protocol == htons(ETH_P_IPV6)) {
391 err = ip6_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
392 } else {
393 pr_err("Unknown layer 3 protocol: %d\n", skb->protocol);
394 atomic_dec(&pkt->qp->skb_out);
395 rxe_drop_ref(pkt->qp);
396 kfree_skb(skb);
397 return -EINVAL;
398 }
399
400 if (unlikely(net_xmit_eval(err))) {
401 pr_debug("error sending packet: %d\n", err);
402 return -EAGAIN;
403 }
404
405 return 0;
406}
407
408
409
410
411void rxe_loopback(struct sk_buff *skb)
412{
413 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
414
415 if (skb->protocol == htons(ETH_P_IP))
416 skb_pull(skb, sizeof(struct iphdr));
417 else
418 skb_pull(skb, sizeof(struct ipv6hdr));
419
420 if (WARN_ON(!ib_device_try_get(&pkt->rxe->ib_dev)))
421 kfree_skb(skb);
422 else
423 rxe_rcv(skb);
424}
425
426struct sk_buff *rxe_init_packet(struct rxe_dev *rxe, struct rxe_av *av,
427 int paylen, struct rxe_pkt_info *pkt)
428{
429 unsigned int hdr_len;
430 struct sk_buff *skb = NULL;
431 struct net_device *ndev;
432 const struct ib_gid_attr *attr;
433 const int port_num = 1;
434
435 attr = rdma_get_gid_attr(&rxe->ib_dev, port_num, av->grh.sgid_index);
436 if (IS_ERR(attr))
437 return NULL;
438
439 if (av->network_type == RXE_NETWORK_TYPE_IPV4)
440 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
441 sizeof(struct iphdr);
442 else
443 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
444 sizeof(struct ipv6hdr);
445
446 rcu_read_lock();
447 ndev = rdma_read_gid_attr_ndev_rcu(attr);
448 if (IS_ERR(ndev)) {
449 rcu_read_unlock();
450 goto out;
451 }
452 skb = alloc_skb(paylen + hdr_len + LL_RESERVED_SPACE(ndev),
453 GFP_ATOMIC);
454
455 if (unlikely(!skb)) {
456 rcu_read_unlock();
457 goto out;
458 }
459
460 skb_reserve(skb, hdr_len + LL_RESERVED_SPACE(ndev));
461
462
463 skb->dev = ndev;
464 rcu_read_unlock();
465
466 if (av->network_type == RXE_NETWORK_TYPE_IPV4)
467 skb->protocol = htons(ETH_P_IP);
468 else
469 skb->protocol = htons(ETH_P_IPV6);
470
471 pkt->rxe = rxe;
472 pkt->port_num = port_num;
473 pkt->hdr = skb_put_zero(skb, paylen);
474 pkt->mask |= RXE_GRH_MASK;
475
476out:
477 rdma_put_gid_attr(attr);
478 return skb;
479}
480
481
482
483
484
485const char *rxe_parent_name(struct rxe_dev *rxe, unsigned int port_num)
486{
487 return rxe->ndev->name;
488}
489
490int rxe_net_add(const char *ibdev_name, struct net_device *ndev)
491{
492 int err;
493 struct rxe_dev *rxe = NULL;
494
495 rxe = ib_alloc_device(rxe_dev, ib_dev);
496 if (!rxe)
497 return -ENOMEM;
498
499 rxe->ndev = ndev;
500
501 err = rxe_add(rxe, ndev->mtu, ibdev_name);
502 if (err) {
503 ib_dealloc_device(&rxe->ib_dev);
504 return err;
505 }
506
507 return 0;
508}
509
510static void rxe_port_event(struct rxe_dev *rxe,
511 enum ib_event_type event)
512{
513 struct ib_event ev;
514
515 ev.device = &rxe->ib_dev;
516 ev.element.port_num = 1;
517 ev.event = event;
518
519 ib_dispatch_event(&ev);
520}
521
522
523void rxe_port_up(struct rxe_dev *rxe)
524{
525 struct rxe_port *port;
526
527 port = &rxe->port;
528 port->attr.state = IB_PORT_ACTIVE;
529
530 rxe_port_event(rxe, IB_EVENT_PORT_ACTIVE);
531 dev_info(&rxe->ib_dev.dev, "set active\n");
532}
533
534
535void rxe_port_down(struct rxe_dev *rxe)
536{
537 struct rxe_port *port;
538
539 port = &rxe->port;
540 port->attr.state = IB_PORT_DOWN;
541
542 rxe_port_event(rxe, IB_EVENT_PORT_ERR);
543 rxe_counter_inc(rxe, RXE_CNT_LINK_DOWNED);
544 dev_info(&rxe->ib_dev.dev, "set down\n");
545}
546
547void rxe_set_port_state(struct rxe_dev *rxe)
548{
549 if (netif_running(rxe->ndev) && netif_carrier_ok(rxe->ndev))
550 rxe_port_up(rxe);
551 else
552 rxe_port_down(rxe);
553}
554
555static int rxe_notify(struct notifier_block *not_blk,
556 unsigned long event,
557 void *arg)
558{
559 struct net_device *ndev = netdev_notifier_info_to_dev(arg);
560 struct rxe_dev *rxe = rxe_get_dev_from_net(ndev);
561
562 if (!rxe)
563 return NOTIFY_OK;
564
565 switch (event) {
566 case NETDEV_UNREGISTER:
567 ib_unregister_device_queued(&rxe->ib_dev);
568 break;
569 case NETDEV_UP:
570 rxe_port_up(rxe);
571 break;
572 case NETDEV_DOWN:
573 rxe_port_down(rxe);
574 break;
575 case NETDEV_CHANGEMTU:
576 pr_info("%s changed mtu to %d\n", ndev->name, ndev->mtu);
577 rxe_set_mtu(rxe, ndev->mtu);
578 break;
579 case NETDEV_CHANGE:
580 rxe_set_port_state(rxe);
581 break;
582 case NETDEV_REBOOT:
583 case NETDEV_GOING_DOWN:
584 case NETDEV_CHANGEADDR:
585 case NETDEV_CHANGENAME:
586 case NETDEV_FEAT_CHANGE:
587 default:
588 pr_info("ignoring netdev event = %ld for %s\n",
589 event, ndev->name);
590 break;
591 }
592
593 ib_device_put(&rxe->ib_dev);
594 return NOTIFY_OK;
595}
596
597static struct notifier_block rxe_net_notifier = {
598 .notifier_call = rxe_notify,
599};
600
601static int rxe_net_ipv4_init(void)
602{
603 recv_sockets.sk4 = rxe_setup_udp_tunnel(&init_net,
604 htons(ROCE_V2_UDP_DPORT), false);
605 if (IS_ERR(recv_sockets.sk4)) {
606 recv_sockets.sk4 = NULL;
607 pr_err("Failed to create IPv4 UDP tunnel\n");
608 return -1;
609 }
610
611 return 0;
612}
613
614static int rxe_net_ipv6_init(void)
615{
616#if IS_ENABLED(CONFIG_IPV6)
617
618 recv_sockets.sk6 = rxe_setup_udp_tunnel(&init_net,
619 htons(ROCE_V2_UDP_DPORT), true);
620 if (PTR_ERR(recv_sockets.sk6) == -EAFNOSUPPORT) {
621 recv_sockets.sk6 = NULL;
622 pr_warn("IPv6 is not supported, can not create a UDPv6 socket\n");
623 return 0;
624 }
625
626 if (IS_ERR(recv_sockets.sk6)) {
627 recv_sockets.sk6 = NULL;
628 pr_err("Failed to create IPv6 UDP tunnel\n");
629 return -1;
630 }
631#endif
632 return 0;
633}
634
635void rxe_net_exit(void)
636{
637 rxe_release_udp_tunnel(recv_sockets.sk6);
638 rxe_release_udp_tunnel(recv_sockets.sk4);
639 unregister_netdevice_notifier(&rxe_net_notifier);
640}
641
642int rxe_net_init(void)
643{
644 int err;
645
646 recv_sockets.sk6 = NULL;
647
648 err = rxe_net_ipv4_init();
649 if (err)
650 return err;
651 err = rxe_net_ipv6_init();
652 if (err)
653 goto err_out;
654 err = register_netdevice_notifier(&rxe_net_notifier);
655 if (err) {
656 pr_err("Failed to register netdev notifier\n");
657 goto err_out;
658 }
659 return 0;
660err_out:
661 rxe_net_exit();
662 return err;
663}
664
