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
48static struct rxe_recv_sockets recv_sockets;
49
50struct device *rxe_dma_device(struct rxe_dev *rxe)
51{
52 struct net_device *ndev;
53
54 ndev = rxe->ndev;
55
56 if (is_vlan_dev(ndev))
57 ndev = vlan_dev_real_dev(ndev);
58
59 return ndev->dev.parent;
60}
61
62int rxe_mcast_add(struct rxe_dev *rxe, union ib_gid *mgid)
63{
64 int err;
65 unsigned char ll_addr[ETH_ALEN];
66
67 ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr);
68 err = dev_mc_add(rxe->ndev, ll_addr);
69
70 return err;
71}
72
73int rxe_mcast_delete(struct rxe_dev *rxe, union ib_gid *mgid)
74{
75 int err;
76 unsigned char ll_addr[ETH_ALEN];
77
78 ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr);
79 err = dev_mc_del(rxe->ndev, ll_addr);
80
81 return err;
82}
83
84static struct dst_entry *rxe_find_route4(struct net_device *ndev,
85 struct in_addr *saddr,
86 struct in_addr *daddr)
87{
88 struct rtable *rt;
89 struct flowi4 fl = { { 0 } };
90
91 memset(&fl, 0, sizeof(fl));
92 fl.flowi4_oif = ndev->ifindex;
93 memcpy(&fl.saddr, saddr, sizeof(*saddr));
94 memcpy(&fl.daddr, daddr, sizeof(*daddr));
95 fl.flowi4_proto = IPPROTO_UDP;
96
97 rt = ip_route_output_key(&init_net, &fl);
98 if (IS_ERR(rt)) {
99 pr_err_ratelimited("no route to %pI4\n", &daddr->s_addr);
100 return NULL;
101 }
102
103 return &rt->dst;
104}
105
106#if IS_ENABLED(CONFIG_IPV6)
107static struct dst_entry *rxe_find_route6(struct net_device *ndev,
108 struct in6_addr *saddr,
109 struct in6_addr *daddr)
110{
111 struct dst_entry *ndst;
112 struct flowi6 fl6 = { { 0 } };
113
114 memset(&fl6, 0, sizeof(fl6));
115 fl6.flowi6_oif = ndev->ifindex;
116 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
117 memcpy(&fl6.daddr, daddr, sizeof(*daddr));
118 fl6.flowi6_proto = IPPROTO_UDP;
119
120 if (unlikely(ipv6_stub->ipv6_dst_lookup(sock_net(recv_sockets.sk6->sk),
121 recv_sockets.sk6->sk, &ndst, &fl6))) {
122 pr_err_ratelimited("no route to %pI6\n", daddr);
123 goto put;
124 }
125
126 if (unlikely(ndst->error)) {
127 pr_err("no route to %pI6\n", daddr);
128 goto put;
129 }
130
131 return ndst;
132put:
133 dst_release(ndst);
134 return NULL;
135}
136
137#else
138
139static struct dst_entry *rxe_find_route6(struct net_device *ndev,
140 struct in6_addr *saddr,
141 struct in6_addr *daddr)
142{
143 return NULL;
144}
145
146#endif
147
148static struct dst_entry *rxe_find_route(struct net_device *ndev,
149 struct rxe_qp *qp,
150 struct rxe_av *av)
151{
152 struct dst_entry *dst = NULL;
153
154 if (qp_type(qp) == IB_QPT_RC)
155 dst = sk_dst_get(qp->sk->sk);
156
157 if (!dst || !dst_check(dst, qp->dst_cookie)) {
158 if (dst)
159 dst_release(dst);
160
161 if (av->network_type == RDMA_NETWORK_IPV4) {
162 struct in_addr *saddr;
163 struct in_addr *daddr;
164
165 saddr = &av->sgid_addr._sockaddr_in.sin_addr;
166 daddr = &av->dgid_addr._sockaddr_in.sin_addr;
167 dst = rxe_find_route4(ndev, saddr, daddr);
168 } else if (av->network_type == RDMA_NETWORK_IPV6) {
169 struct in6_addr *saddr6;
170 struct in6_addr *daddr6;
171
172 saddr6 = &av->sgid_addr._sockaddr_in6.sin6_addr;
173 daddr6 = &av->dgid_addr._sockaddr_in6.sin6_addr;
174 dst = rxe_find_route6(ndev, saddr6, daddr6);
175#if IS_ENABLED(CONFIG_IPV6)
176 if (dst)
177 qp->dst_cookie =
178 rt6_get_cookie((struct rt6_info *)dst);
179#endif
180 }
181
182 if (dst && (qp_type(qp) == IB_QPT_RC)) {
183 dst_hold(dst);
184 sk_dst_set(qp->sk->sk, dst);
185 }
186 }
187 return dst;
188}
189
190static int rxe_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
191{
192 struct udphdr *udph;
193 struct net_device *ndev = skb->dev;
194 struct net_device *rdev = ndev;
195 struct rxe_dev *rxe = rxe_get_dev_from_net(ndev);
196 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
197
198 if (!rxe && is_vlan_dev(rdev)) {
199 rdev = vlan_dev_real_dev(ndev);
200 rxe = rxe_get_dev_from_net(rdev);
201 }
202 if (!rxe)
203 goto drop;
204
205 if (skb_linearize(skb)) {
206 pr_err("skb_linearize failed\n");
207 ib_device_put(&rxe->ib_dev);
208 goto drop;
209 }
210
211 udph = udp_hdr(skb);
212 pkt->rxe = rxe;
213 pkt->port_num = 1;
214 pkt->hdr = (u8 *)(udph + 1);
215 pkt->mask = RXE_GRH_MASK;
216 pkt->paylen = be16_to_cpu(udph->len) - sizeof(*udph);
217
218 rxe_rcv(skb);
219
220 /*
221 * FIXME: this is in the wrong place, it needs to be done when pkt is
222 * destroyed
223 */
224 ib_device_put(&rxe->ib_dev);
225
226 return 0;
227drop:
228 kfree_skb(skb);
229
230 return 0;
231}
232
233static struct socket *rxe_setup_udp_tunnel(struct net *net, __be16 port,
234 bool ipv6)
235{
236 int err;
237 struct socket *sock;
238 struct udp_port_cfg udp_cfg = { };
239 struct udp_tunnel_sock_cfg tnl_cfg = { };
240
241 if (ipv6) {
242 udp_cfg.family = AF_INET6;
243 udp_cfg.ipv6_v6only = 1;
244 } else {
245 udp_cfg.family = AF_INET;
246 }
247
248 udp_cfg.local_udp_port = port;
249
250 /* Create UDP socket */
251 err = udp_sock_create(net, &udp_cfg, &sock);
252 if (err < 0) {
253 pr_err("failed to create udp socket. err = %d\n", err);
254 return ERR_PTR(err);
255 }
256
257 tnl_cfg.encap_type = 1;
258 tnl_cfg.encap_rcv = rxe_udp_encap_recv;
259
260 /* Setup UDP tunnel */
261 setup_udp_tunnel_sock(net, sock, &tnl_cfg);
262
263 return sock;
264}
265
266static void rxe_release_udp_tunnel(struct socket *sk)
267{
268 if (sk)
269 udp_tunnel_sock_release(sk);
270}
271
272static void prepare_udp_hdr(struct sk_buff *skb, __be16 src_port,
273 __be16 dst_port)
274{
275 struct udphdr *udph;
276
277 __skb_push(skb, sizeof(*udph));
278 skb_reset_transport_header(skb);
279 udph = udp_hdr(skb);
280
281 udph->dest = dst_port;
282 udph->source = src_port;
283 udph->len = htons(skb->len);
284 udph->check = 0;
285}
286
287static void prepare_ipv4_hdr(struct dst_entry *dst, struct sk_buff *skb,
288 __be32 saddr, __be32 daddr, __u8 proto,
289 __u8 tos, __u8 ttl, __be16 df, bool xnet)
290{
291 struct iphdr *iph;
292
293 skb_scrub_packet(skb, xnet);
294
295 skb_clear_hash(skb);
296 skb_dst_set(skb, dst_clone(dst));
297 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
298
299 skb_push(skb, sizeof(struct iphdr));
300 skb_reset_network_header(skb);
301
302 iph = ip_hdr(skb);
303
304 iph->version = IPVERSION;
305 iph->ihl = sizeof(struct iphdr) >> 2;
306 iph->frag_off = df;
307 iph->protocol = proto;
308 iph->tos = tos;
309 iph->daddr = daddr;
310 iph->saddr = saddr;
311 iph->ttl = ttl;
312 __ip_select_ident(dev_net(dst->dev), iph,
313 skb_shinfo(skb)->gso_segs ?: 1);
314 iph->tot_len = htons(skb->len);
315 ip_send_check(iph);
316}
317
318static void prepare_ipv6_hdr(struct dst_entry *dst, struct sk_buff *skb,
319 struct in6_addr *saddr, struct in6_addr *daddr,
320 __u8 proto, __u8 prio, __u8 ttl)
321{
322 struct ipv6hdr *ip6h;
323
324 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
325 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED
326 | IPSKB_REROUTED);
327 skb_dst_set(skb, dst_clone(dst));
328
329 __skb_push(skb, sizeof(*ip6h));
330 skb_reset_network_header(skb);
331 ip6h = ipv6_hdr(skb);
332 ip6_flow_hdr(ip6h, prio, htonl(0));
333 ip6h->payload_len = htons(skb->len);
334 ip6h->nexthdr = proto;
335 ip6h->hop_limit = ttl;
336 ip6h->daddr = *daddr;
337 ip6h->saddr = *saddr;
338 ip6h->payload_len = htons(skb->len - sizeof(*ip6h));
339}
340
341static int prepare4(struct rxe_pkt_info *pkt, struct sk_buff *skb,
342 struct rxe_av *av)
343{
344 struct rxe_qp *qp = pkt->qp;
345 struct dst_entry *dst;
346 bool xnet = false;
347 __be16 df = htons(IP_DF);
348 struct in_addr *saddr = &av->sgid_addr._sockaddr_in.sin_addr;
349 struct in_addr *daddr = &av->dgid_addr._sockaddr_in.sin_addr;
350
351 dst = rxe_find_route(skb->dev, qp, av);
352 if (!dst) {
353 pr_err("Host not reachable\n");
354 return -EHOSTUNREACH;
355 }
356
357 prepare_udp_hdr(skb, cpu_to_be16(qp->src_port),
358 cpu_to_be16(ROCE_V2_UDP_DPORT));
359
360 prepare_ipv4_hdr(dst, skb, saddr->s_addr, daddr->s_addr, IPPROTO_UDP,
361 av->grh.traffic_class, av->grh.hop_limit, df, xnet);
362
363 dst_release(dst);
364 return 0;
365}
366
367static int prepare6(struct rxe_pkt_info *pkt, struct sk_buff *skb,
368 struct rxe_av *av)
369{
370 struct rxe_qp *qp = pkt->qp;
371 struct dst_entry *dst;
372 struct in6_addr *saddr = &av->sgid_addr._sockaddr_in6.sin6_addr;
373 struct in6_addr *daddr = &av->dgid_addr._sockaddr_in6.sin6_addr;
374
375 dst = rxe_find_route(skb->dev, qp, av);
376 if (!dst) {
377 pr_err("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_ipv6_hdr(dst, skb, saddr, daddr, IPPROTO_UDP,
385 av->grh.traffic_class,
386 av->grh.hop_limit);
387
388 dst_release(dst);
389 return 0;
390}
391
392int rxe_prepare(struct rxe_pkt_info *pkt, struct sk_buff *skb, u32 *crc)
393{
394 int err = 0;
395 struct rxe_av *av = rxe_get_av(pkt);
396
397 if (av->network_type == RDMA_NETWORK_IPV4)
398 err = prepare4(pkt, skb, av);
399 else if (av->network_type == RDMA_NETWORK_IPV6)
400 err = prepare6(pkt, skb, av);
401
402 *crc = rxe_icrc_hdr(pkt, skb);
403
404 if (ether_addr_equal(skb->dev->dev_addr, av->dmac))
405 pkt->mask |= RXE_LOOPBACK_MASK;
406
407 return err;
408}
409
410static void rxe_skb_tx_dtor(struct sk_buff *skb)
411{
412 struct sock *sk = skb->sk;
413 struct rxe_qp *qp = sk->sk_user_data;
414 int skb_out = atomic_dec_return(&qp->skb_out);
415
416 if (unlikely(qp->need_req_skb &&
417 skb_out < RXE_INFLIGHT_SKBS_PER_QP_LOW))
418 rxe_run_task(&qp->req.task, 1);
419
420 rxe_drop_ref(qp);
421}
422
423int rxe_send(struct rxe_pkt_info *pkt, struct sk_buff *skb)
424{
425 struct rxe_av *av;
426 int err;
427
428 av = rxe_get_av(pkt);
429
430 skb->destructor = rxe_skb_tx_dtor;
431 skb->sk = pkt->qp->sk->sk;
432
433 rxe_add_ref(pkt->qp);
434 atomic_inc(&pkt->qp->skb_out);
435
436 if (av->network_type == RDMA_NETWORK_IPV4) {
437 err = ip_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
438 } else if (av->network_type == RDMA_NETWORK_IPV6) {
439 err = ip6_local_out(dev_net(skb_dst(skb)->dev), skb->sk, skb);
440 } else {
441 pr_err("Unknown layer 3 protocol: %d\n", av->network_type);
442 atomic_dec(&pkt->qp->skb_out);
443 rxe_drop_ref(pkt->qp);
444 kfree_skb(skb);
445 return -EINVAL;
446 }
447
448 if (unlikely(net_xmit_eval(err))) {
449 pr_debug("error sending packet: %d\n", err);
450 return -EAGAIN;
451 }
452
453 return 0;
454}
455
456void rxe_loopback(struct sk_buff *skb)
457{
458 rxe_rcv(skb);
459}
460
461struct sk_buff *rxe_init_packet(struct rxe_dev *rxe, struct rxe_av *av,
462 int paylen, struct rxe_pkt_info *pkt)
463{
464 unsigned int hdr_len;
465 struct sk_buff *skb;
466 struct net_device *ndev;
467 const struct ib_gid_attr *attr;
468 const int port_num = 1;
469
470 attr = rdma_get_gid_attr(&rxe->ib_dev, port_num, av->grh.sgid_index);
471 if (IS_ERR(attr))
472 return NULL;
473 ndev = attr->ndev;
474
475 if (av->network_type == RDMA_NETWORK_IPV4)
476 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
477 sizeof(struct iphdr);
478 else
479 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
480 sizeof(struct ipv6hdr);
481
482 skb = alloc_skb(paylen + hdr_len + LL_RESERVED_SPACE(ndev),
483 GFP_ATOMIC);
484
485 if (unlikely(!skb))
486 goto out;
487
488 skb_reserve(skb, hdr_len + LL_RESERVED_SPACE(rxe->ndev));
489
490 skb->dev = ndev;
491 if (av->network_type == RDMA_NETWORK_IPV4)
492 skb->protocol = htons(ETH_P_IP);
493 else
494 skb->protocol = htons(ETH_P_IPV6);
495
496 pkt->rxe = rxe;
497 pkt->port_num = port_num;
498 pkt->hdr = skb_put_zero(skb, paylen);
499 pkt->mask |= RXE_GRH_MASK;
500
501out:
502 rdma_put_gid_attr(attr);
503 return skb;
504}
505
506/*
507 * this is required by rxe_cfg to match rxe devices in
508 * /sys/class/infiniband up with their underlying ethernet devices
509 */
510const char *rxe_parent_name(struct rxe_dev *rxe, unsigned int port_num)
511{
512 return rxe->ndev->name;
513}
514
515enum rdma_link_layer rxe_link_layer(struct rxe_dev *rxe, unsigned int port_num)
516{
517 return IB_LINK_LAYER_ETHERNET;
518}
519
520int rxe_net_add(const char *ibdev_name, struct net_device *ndev)
521{
522 int err;
523 struct rxe_dev *rxe = NULL;
524
525 rxe = ib_alloc_device(rxe_dev, ib_dev);
526 if (!rxe)
527 return -ENOMEM;
528
529 rxe->ndev = ndev;
530
531 err = rxe_add(rxe, ndev->mtu, ibdev_name);
532 if (err) {
533 ib_dealloc_device(&rxe->ib_dev);
534 return err;
535 }
536
537 return 0;
538}
539
540static void rxe_port_event(struct rxe_dev *rxe,
541 enum ib_event_type event)
542{
543 struct ib_event ev;
544
545 ev.device = &rxe->ib_dev;
546 ev.element.port_num = 1;
547 ev.event = event;
548
549 ib_dispatch_event(&ev);
550}
551
552/* Caller must hold net_info_lock */
553void rxe_port_up(struct rxe_dev *rxe)
554{
555 struct rxe_port *port;
556
557 port = &rxe->port;
558 port->attr.state = IB_PORT_ACTIVE;
559
560 rxe_port_event(rxe, IB_EVENT_PORT_ACTIVE);
561 dev_info(&rxe->ib_dev.dev, "set active\n");
562}
563
564/* Caller must hold net_info_lock */
565void rxe_port_down(struct rxe_dev *rxe)
566{
567 struct rxe_port *port;
568
569 port = &rxe->port;
570 port->attr.state = IB_PORT_DOWN;
571
572 rxe_port_event(rxe, IB_EVENT_PORT_ERR);
573 rxe_counter_inc(rxe, RXE_CNT_LINK_DOWNED);
574 dev_info(&rxe->ib_dev.dev, "set down\n");
575}
576
577void rxe_set_port_state(struct rxe_dev *rxe)
578{
579 if (netif_running(rxe->ndev) && netif_carrier_ok(rxe->ndev))
580 rxe_port_up(rxe);
581 else
582 rxe_port_down(rxe);
583}
584
585static int rxe_notify(struct notifier_block *not_blk,
586 unsigned long event,
587 void *arg)
588{
589 struct net_device *ndev = netdev_notifier_info_to_dev(arg);
590 struct rxe_dev *rxe = rxe_get_dev_from_net(ndev);
591
592 if (!rxe)
593 return NOTIFY_OK;
594
595 switch (event) {
596 case NETDEV_UNREGISTER:
597 ib_unregister_device_queued(&rxe->ib_dev);
598 break;
599 case NETDEV_UP:
600 rxe_port_up(rxe);
601 break;
602 case NETDEV_DOWN:
603 rxe_port_down(rxe);
604 break;
605 case NETDEV_CHANGEMTU:
606 pr_info("%s changed mtu to %d\n", ndev->name, ndev->mtu);
607 rxe_set_mtu(rxe, ndev->mtu);
608 break;
609 case NETDEV_CHANGE:
610 rxe_set_port_state(rxe);
611 break;
612 case NETDEV_REBOOT:
613 case NETDEV_GOING_DOWN:
614 case NETDEV_CHANGEADDR:
615 case NETDEV_CHANGENAME:
616 case NETDEV_FEAT_CHANGE:
617 default:
618 pr_info("ignoring netdev event = %ld for %s\n",
619 event, ndev->name);
620 break;
621 }
622
623 ib_device_put(&rxe->ib_dev);
624 return NOTIFY_OK;
625}
626
627static struct notifier_block rxe_net_notifier = {
628 .notifier_call = rxe_notify,
629};
630
631static int rxe_net_ipv4_init(void)
632{
633 recv_sockets.sk4 = rxe_setup_udp_tunnel(&init_net,
634 htons(ROCE_V2_UDP_DPORT), false);
635 if (IS_ERR(recv_sockets.sk4)) {
636 recv_sockets.sk4 = NULL;
637 pr_err("Failed to create IPv4 UDP tunnel\n");
638 return -1;
639 }
640
641 return 0;
642}
643
644static int rxe_net_ipv6_init(void)
645{
646#if IS_ENABLED(CONFIG_IPV6)
647
648 recv_sockets.sk6 = rxe_setup_udp_tunnel(&init_net,
649 htons(ROCE_V2_UDP_DPORT), true);
650 if (IS_ERR(recv_sockets.sk6)) {
651 recv_sockets.sk6 = NULL;
652 pr_err("Failed to create IPv6 UDP tunnel\n");
653 return -1;
654 }
655#endif
656 return 0;
657}
658
659void rxe_net_exit(void)
660{
661 rxe_release_udp_tunnel(recv_sockets.sk6);
662 rxe_release_udp_tunnel(recv_sockets.sk4);
663 unregister_netdevice_notifier(&rxe_net_notifier);
664}
665
666int rxe_net_init(void)
667{
668 int err;
669
670 recv_sockets.sk6 = NULL;
671
672 err = rxe_net_ipv4_init();
673 if (err)
674 return err;
675 err = rxe_net_ipv6_init();
676 if (err)
677 goto err_out;
678 err = register_netdevice_notifier(&rxe_net_notifier);
679 if (err) {
680 pr_err("Failed to register netdev notifier\n");
681 goto err_out;
682 }
683 return 0;
684err_out:
685 rxe_net_exit();
686 return err;
687}
688