1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
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 LIST_HEAD(rxe_dev_list);
49static DEFINE_SPINLOCK(dev_list_lock);
50
51struct rxe_dev *net_to_rxe(struct net_device *ndev)
52{
53 struct rxe_dev *rxe;
54 struct rxe_dev *found = NULL;
55
56 spin_lock_bh(&dev_list_lock);
57 list_for_each_entry(rxe, &rxe_dev_list, list) {
58 if (rxe->ndev == ndev) {
59 found = rxe;
60 break;
61 }
62 }
63 spin_unlock_bh(&dev_list_lock);
64
65 return found;
66}
67
68struct rxe_dev *get_rxe_by_name(const char *name)
69{
70 struct rxe_dev *rxe;
71 struct rxe_dev *found = NULL;
72
73 spin_lock_bh(&dev_list_lock);
74 list_for_each_entry(rxe, &rxe_dev_list, list) {
75 if (!strcmp(name, rxe->ib_dev.name)) {
76 found = rxe;
77 break;
78 }
79 }
80 spin_unlock_bh(&dev_list_lock);
81 return found;
82}
83
84
85struct rxe_recv_sockets recv_sockets;
86
87struct device *rxe_dma_device(struct rxe_dev *rxe)
88{
89 struct net_device *ndev;
90
91 ndev = rxe->ndev;
92
93 if (is_vlan_dev(ndev))
94 ndev = vlan_dev_real_dev(ndev);
95
96 return ndev->dev.parent;
97}
98
99int rxe_mcast_add(struct rxe_dev *rxe, union ib_gid *mgid)
100{
101 int err;
102 unsigned char ll_addr[ETH_ALEN];
103
104 ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr);
105 err = dev_mc_add(rxe->ndev, ll_addr);
106
107 return err;
108}
109
110int rxe_mcast_delete(struct rxe_dev *rxe, union ib_gid *mgid)
111{
112 int err;
113 unsigned char ll_addr[ETH_ALEN];
114
115 ipv6_eth_mc_map((struct in6_addr *)mgid->raw, ll_addr);
116 err = dev_mc_del(rxe->ndev, ll_addr);
117
118 return err;
119}
120
121static struct dst_entry *rxe_find_route4(struct net_device *ndev,
122 struct in_addr *saddr,
123 struct in_addr *daddr)
124{
125 struct rtable *rt;
126 struct flowi4 fl = { { 0 } };
127
128 memset(&fl, 0, sizeof(fl));
129 fl.flowi4_oif = ndev->ifindex;
130 memcpy(&fl.saddr, saddr, sizeof(*saddr));
131 memcpy(&fl.daddr, daddr, sizeof(*daddr));
132 fl.flowi4_proto = IPPROTO_UDP;
133
134 rt = ip_route_output_key(&init_net, &fl);
135 if (IS_ERR(rt)) {
136 pr_err_ratelimited("no route to %pI4\n", &daddr->s_addr);
137 return NULL;
138 }
139
140 return &rt->dst;
141}
142
143#if IS_ENABLED(CONFIG_IPV6)
144static struct dst_entry *rxe_find_route6(struct net_device *ndev,
145 struct in6_addr *saddr,
146 struct in6_addr *daddr)
147{
148 struct dst_entry *ndst;
149 struct flowi6 fl6 = { { 0 } };
150
151 memset(&fl6, 0, sizeof(fl6));
152 fl6.flowi6_oif = ndev->ifindex;
153 memcpy(&fl6.saddr, saddr, sizeof(*saddr));
154 memcpy(&fl6.daddr, daddr, sizeof(*daddr));
155 fl6.flowi6_proto = IPPROTO_UDP;
156
157 if (unlikely(ipv6_stub->ipv6_dst_lookup(sock_net(recv_sockets.sk6->sk),
158 recv_sockets.sk6->sk, &ndst, &fl6))) {
159 pr_err_ratelimited("no route to %pI6\n", daddr);
160 goto put;
161 }
162
163 if (unlikely(ndst->error)) {
164 pr_err("no route to %pI6\n", daddr);
165 goto put;
166 }
167
168 return ndst;
169put:
170 dst_release(ndst);
171 return NULL;
172}
173
174#else
175
176static struct dst_entry *rxe_find_route6(struct net_device *ndev,
177 struct in6_addr *saddr,
178 struct in6_addr *daddr)
179{
180 return NULL;
181}
182
183#endif
184
185static struct dst_entry *rxe_find_route(struct rxe_dev *rxe,
186 struct rxe_qp *qp,
187 struct rxe_av *av)
188{
189 struct dst_entry *dst = NULL;
190
191 if (qp_type(qp) == IB_QPT_RC)
192 dst = sk_dst_get(qp->sk->sk);
193
194 if (!dst || !dst_check(dst, qp->dst_cookie)) {
195 if (dst)
196 dst_release(dst);
197
198 if (av->network_type == RDMA_NETWORK_IPV4) {
199 struct in_addr *saddr;
200 struct in_addr *daddr;
201
202 saddr = &av->sgid_addr._sockaddr_in.sin_addr;
203 daddr = &av->dgid_addr._sockaddr_in.sin_addr;
204 dst = rxe_find_route4(rxe->ndev, saddr, daddr);
205 } else if (av->network_type == RDMA_NETWORK_IPV6) {
206 struct in6_addr *saddr6;
207 struct in6_addr *daddr6;
208
209 saddr6 = &av->sgid_addr._sockaddr_in6.sin6_addr;
210 daddr6 = &av->dgid_addr._sockaddr_in6.sin6_addr;
211 dst = rxe_find_route6(rxe->ndev, saddr6, daddr6);
212#if IS_ENABLED(CONFIG_IPV6)
213 if (dst)
214 qp->dst_cookie =
215 rt6_get_cookie((struct rt6_info *)dst);
216#endif
217 }
218 }
219
220 return dst;
221}
222
223static int rxe_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
224{
225 struct udphdr *udph;
226 struct net_device *ndev = skb->dev;
227 struct rxe_dev *rxe = net_to_rxe(ndev);
228 struct rxe_pkt_info *pkt = SKB_TO_PKT(skb);
229
230 if (!rxe)
231 goto drop;
232
233 if (skb_linearize(skb)) {
234 pr_err("skb_linearize failed\n");
235 goto drop;
236 }
237
238 udph = udp_hdr(skb);
239 pkt->rxe = rxe;
240 pkt->port_num = 1;
241 pkt->hdr = (u8 *)(udph + 1);
242 pkt->mask = RXE_GRH_MASK;
243 pkt->paylen = be16_to_cpu(udph->len) - sizeof(*udph);
244
245 return rxe_rcv(skb);
246drop:
247 kfree_skb(skb);
248 return 0;
249}
250
251static struct socket *rxe_setup_udp_tunnel(struct net *net, __be16 port,
252 bool ipv6)
253{
254 int err;
255 struct socket *sock;
256 struct udp_port_cfg udp_cfg = { };
257 struct udp_tunnel_sock_cfg tnl_cfg = { };
258
259 if (ipv6) {
260 udp_cfg.family = AF_INET6;
261 udp_cfg.ipv6_v6only = 1;
262 } else {
263 udp_cfg.family = AF_INET;
264 }
265
266 udp_cfg.local_udp_port = port;
267
268
269 err = udp_sock_create(net, &udp_cfg, &sock);
270 if (err < 0) {
271 pr_err("failed to create udp socket. err = %d\n", err);
272 return ERR_PTR(err);
273 }
274
275 tnl_cfg.encap_type = 1;
276 tnl_cfg.encap_rcv = rxe_udp_encap_recv;
277
278
279 setup_udp_tunnel_sock(net, sock, &tnl_cfg);
280
281 return sock;
282}
283
284void rxe_release_udp_tunnel(struct socket *sk)
285{
286 if (sk)
287 udp_tunnel_sock_release(sk);
288}
289
290static void prepare_udp_hdr(struct sk_buff *skb, __be16 src_port,
291 __be16 dst_port)
292{
293 struct udphdr *udph;
294
295 __skb_push(skb, sizeof(*udph));
296 skb_reset_transport_header(skb);
297 udph = udp_hdr(skb);
298
299 udph->dest = dst_port;
300 udph->source = src_port;
301 udph->len = htons(skb->len);
302 udph->check = 0;
303}
304
305static void prepare_ipv4_hdr(struct dst_entry *dst, struct sk_buff *skb,
306 __be32 saddr, __be32 daddr, __u8 proto,
307 __u8 tos, __u8 ttl, __be16 df, bool xnet)
308{
309 struct iphdr *iph;
310
311 skb_scrub_packet(skb, xnet);
312
313 skb_clear_hash(skb);
314 skb_dst_set(skb, dst_clone(dst));
315 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
316
317 skb_push(skb, sizeof(struct iphdr));
318 skb_reset_network_header(skb);
319
320 iph = ip_hdr(skb);
321
322 iph->version = IPVERSION;
323 iph->ihl = sizeof(struct iphdr) >> 2;
324 iph->frag_off = df;
325 iph->protocol = proto;
326 iph->tos = tos;
327 iph->daddr = daddr;
328 iph->saddr = saddr;
329 iph->ttl = ttl;
330 __ip_select_ident(dev_net(dst->dev), iph,
331 skb_shinfo(skb)->gso_segs ?: 1);
332 iph->tot_len = htons(skb->len);
333 ip_send_check(iph);
334}
335
336static void prepare_ipv6_hdr(struct dst_entry *dst, struct sk_buff *skb,
337 struct in6_addr *saddr, struct in6_addr *daddr,
338 __u8 proto, __u8 prio, __u8 ttl)
339{
340 struct ipv6hdr *ip6h;
341
342 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
343 IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED
344 | IPSKB_REROUTED);
345 skb_dst_set(skb, dst_clone(dst));
346
347 __skb_push(skb, sizeof(*ip6h));
348 skb_reset_network_header(skb);
349 ip6h = ipv6_hdr(skb);
350 ip6_flow_hdr(ip6h, prio, htonl(0));
351 ip6h->payload_len = htons(skb->len);
352 ip6h->nexthdr = proto;
353 ip6h->hop_limit = ttl;
354 ip6h->daddr = *daddr;
355 ip6h->saddr = *saddr;
356 ip6h->payload_len = htons(skb->len - sizeof(*ip6h));
357}
358
359static int prepare4(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
360 struct sk_buff *skb, struct rxe_av *av)
361{
362 struct rxe_qp *qp = pkt->qp;
363 struct dst_entry *dst;
364 bool xnet = false;
365 __be16 df = htons(IP_DF);
366 struct in_addr *saddr = &av->sgid_addr._sockaddr_in.sin_addr;
367 struct in_addr *daddr = &av->dgid_addr._sockaddr_in.sin_addr;
368
369 dst = rxe_find_route(rxe, qp, av);
370 if (!dst) {
371 pr_err("Host not reachable\n");
372 return -EHOSTUNREACH;
373 }
374
375 if (!memcmp(saddr, daddr, sizeof(*daddr)))
376 pkt->mask |= RXE_LOOPBACK_MASK;
377
378 prepare_udp_hdr(skb, htons(RXE_ROCE_V2_SPORT),
379 htons(ROCE_V2_UDP_DPORT));
380
381 prepare_ipv4_hdr(dst, skb, saddr->s_addr, daddr->s_addr, IPPROTO_UDP,
382 av->grh.traffic_class, av->grh.hop_limit, df, xnet);
383
384 if (qp_type(qp) == IB_QPT_RC)
385 sk_dst_set(qp->sk->sk, dst);
386 else
387 dst_release(dst);
388
389 return 0;
390}
391
392static int prepare6(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
393 struct sk_buff *skb, struct rxe_av *av)
394{
395 struct rxe_qp *qp = pkt->qp;
396 struct dst_entry *dst;
397 struct in6_addr *saddr = &av->sgid_addr._sockaddr_in6.sin6_addr;
398 struct in6_addr *daddr = &av->dgid_addr._sockaddr_in6.sin6_addr;
399
400 dst = rxe_find_route(rxe, qp, av);
401 if (!dst) {
402 pr_err("Host not reachable\n");
403 return -EHOSTUNREACH;
404 }
405
406 if (!memcmp(saddr, daddr, sizeof(*daddr)))
407 pkt->mask |= RXE_LOOPBACK_MASK;
408
409 prepare_udp_hdr(skb, htons(RXE_ROCE_V2_SPORT),
410 htons(ROCE_V2_UDP_DPORT));
411
412 prepare_ipv6_hdr(dst, skb, saddr, daddr, IPPROTO_UDP,
413 av->grh.traffic_class,
414 av->grh.hop_limit);
415
416 if (qp_type(qp) == IB_QPT_RC)
417 sk_dst_set(qp->sk->sk, dst);
418 else
419 dst_release(dst);
420
421 return 0;
422}
423
424int rxe_prepare(struct rxe_dev *rxe, struct rxe_pkt_info *pkt,
425 struct sk_buff *skb, u32 *crc)
426{
427 int err = 0;
428 struct rxe_av *av = rxe_get_av(pkt);
429
430 if (av->network_type == RDMA_NETWORK_IPV4)
431 err = prepare4(rxe, pkt, skb, av);
432 else if (av->network_type == RDMA_NETWORK_IPV6)
433 err = prepare6(rxe, pkt, skb, av);
434
435 *crc = rxe_icrc_hdr(pkt, skb);
436
437 return err;
438}
439
440static void rxe_skb_tx_dtor(struct sk_buff *skb)
441{
442 struct sock *sk = skb->sk;
443 struct rxe_qp *qp = sk->sk_user_data;
444 int skb_out = atomic_dec_return(&qp->skb_out);
445
446 if (unlikely(qp->need_req_skb &&
447 skb_out < RXE_INFLIGHT_SKBS_PER_QP_LOW))
448 rxe_run_task(&qp->req.task, 1);
449
450 rxe_drop_ref(qp);
451}
452
453int rxe_send(struct rxe_dev *rxe, struct rxe_pkt_info *pkt, struct sk_buff *skb)
454{
455 struct sk_buff *nskb;
456 struct rxe_av *av;
457 int err;
458
459 av = rxe_get_av(pkt);
460
461 nskb = skb_clone(skb, GFP_ATOMIC);
462 if (!nskb)
463 return -ENOMEM;
464
465 nskb->destructor = rxe_skb_tx_dtor;
466 nskb->sk = pkt->qp->sk->sk;
467
468 rxe_add_ref(pkt->qp);
469 atomic_inc(&pkt->qp->skb_out);
470
471 if (av->network_type == RDMA_NETWORK_IPV4) {
472 err = ip_local_out(dev_net(skb_dst(skb)->dev), nskb->sk, nskb);
473 } else if (av->network_type == RDMA_NETWORK_IPV6) {
474 err = ip6_local_out(dev_net(skb_dst(skb)->dev), nskb->sk, nskb);
475 } else {
476 pr_err("Unknown layer 3 protocol: %d\n", av->network_type);
477 atomic_dec(&pkt->qp->skb_out);
478 rxe_drop_ref(pkt->qp);
479 kfree_skb(nskb);
480 return -EINVAL;
481 }
482
483 if (unlikely(net_xmit_eval(err))) {
484 pr_debug("error sending packet: %d\n", err);
485 return -EAGAIN;
486 }
487
488 kfree_skb(skb);
489 return 0;
490}
491
492int rxe_loopback(struct sk_buff *skb)
493{
494 return rxe_rcv(skb);
495}
496
497static inline int addr_same(struct rxe_dev *rxe, struct rxe_av *av)
498{
499 return rxe->port.port_guid == av->grh.dgid.global.interface_id;
500}
501
502struct sk_buff *rxe_init_packet(struct rxe_dev *rxe, struct rxe_av *av,
503 int paylen, struct rxe_pkt_info *pkt)
504{
505 unsigned int hdr_len;
506 struct sk_buff *skb;
507
508 if (av->network_type == RDMA_NETWORK_IPV4)
509 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
510 sizeof(struct iphdr);
511 else
512 hdr_len = ETH_HLEN + sizeof(struct udphdr) +
513 sizeof(struct ipv6hdr);
514
515 skb = alloc_skb(paylen + hdr_len + LL_RESERVED_SPACE(rxe->ndev),
516 GFP_ATOMIC);
517 if (unlikely(!skb))
518 return NULL;
519
520 skb_reserve(skb, hdr_len + LL_RESERVED_SPACE(rxe->ndev));
521
522 skb->dev = rxe->ndev;
523 if (av->network_type == RDMA_NETWORK_IPV4)
524 skb->protocol = htons(ETH_P_IP);
525 else
526 skb->protocol = htons(ETH_P_IPV6);
527
528 pkt->rxe = rxe;
529 pkt->port_num = 1;
530 pkt->hdr = skb_put(skb, paylen);
531 pkt->mask |= RXE_GRH_MASK;
532
533 memset(pkt->hdr, 0, paylen);
534
535 return skb;
536}
537
538
539
540
541
542const char *rxe_parent_name(struct rxe_dev *rxe, unsigned int port_num)
543{
544 return rxe->ndev->name;
545}
546
547enum rdma_link_layer rxe_link_layer(struct rxe_dev *rxe, unsigned int port_num)
548{
549 return IB_LINK_LAYER_ETHERNET;
550}
551
552struct rxe_dev *rxe_net_add(struct net_device *ndev)
553{
554 int err;
555 struct rxe_dev *rxe = NULL;
556
557 rxe = (struct rxe_dev *)ib_alloc_device(sizeof(*rxe));
558 if (!rxe)
559 return NULL;
560
561 rxe->ndev = ndev;
562
563 err = rxe_add(rxe, ndev->mtu);
564 if (err) {
565 ib_dealloc_device(&rxe->ib_dev);
566 return NULL;
567 }
568
569 spin_lock_bh(&dev_list_lock);
570 list_add_tail(&rxe->list, &rxe_dev_list);
571 spin_unlock_bh(&dev_list_lock);
572 return rxe;
573}
574
575void rxe_remove_all(void)
576{
577 spin_lock_bh(&dev_list_lock);
578 while (!list_empty(&rxe_dev_list)) {
579 struct rxe_dev *rxe =
580 list_first_entry(&rxe_dev_list, struct rxe_dev, list);
581
582 list_del(&rxe->list);
583 spin_unlock_bh(&dev_list_lock);
584 rxe_remove(rxe);
585 spin_lock_bh(&dev_list_lock);
586 }
587 spin_unlock_bh(&dev_list_lock);
588}
589EXPORT_SYMBOL(rxe_remove_all);
590
591static void rxe_port_event(struct rxe_dev *rxe,
592 enum ib_event_type event)
593{
594 struct ib_event ev;
595
596 ev.device = &rxe->ib_dev;
597 ev.element.port_num = 1;
598 ev.event = event;
599
600 ib_dispatch_event(&ev);
601}
602
603
604void rxe_port_up(struct rxe_dev *rxe)
605{
606 struct rxe_port *port;
607
608 port = &rxe->port;
609 port->attr.state = IB_PORT_ACTIVE;
610 port->attr.phys_state = IB_PHYS_STATE_LINK_UP;
611
612 rxe_port_event(rxe, IB_EVENT_PORT_ACTIVE);
613 pr_info("set %s active\n", rxe->ib_dev.name);
614}
615
616
617void rxe_port_down(struct rxe_dev *rxe)
618{
619 struct rxe_port *port;
620
621 port = &rxe->port;
622 port->attr.state = IB_PORT_DOWN;
623 port->attr.phys_state = IB_PHYS_STATE_LINK_DOWN;
624
625 rxe_port_event(rxe, IB_EVENT_PORT_ERR);
626 pr_info("set %s down\n", rxe->ib_dev.name);
627}
628
629static int rxe_notify(struct notifier_block *not_blk,
630 unsigned long event,
631 void *arg)
632{
633 struct net_device *ndev = netdev_notifier_info_to_dev(arg);
634 struct rxe_dev *rxe = net_to_rxe(ndev);
635
636 if (!rxe)
637 goto out;
638
639 switch (event) {
640 case NETDEV_UNREGISTER:
641 list_del(&rxe->list);
642 rxe_remove(rxe);
643 break;
644 case NETDEV_UP:
645 rxe_port_up(rxe);
646 break;
647 case NETDEV_DOWN:
648 rxe_port_down(rxe);
649 break;
650 case NETDEV_CHANGEMTU:
651 pr_info("%s changed mtu to %d\n", ndev->name, ndev->mtu);
652 rxe_set_mtu(rxe, ndev->mtu);
653 break;
654 case NETDEV_CHANGE:
655 if (netif_running(ndev) && netif_carrier_ok(ndev))
656 rxe_port_up(rxe);
657 else
658 rxe_port_down(rxe);
659 break;
660 case NETDEV_REBOOT:
661 case NETDEV_GOING_DOWN:
662 case NETDEV_CHANGEADDR:
663 case NETDEV_CHANGENAME:
664 case NETDEV_FEAT_CHANGE:
665 default:
666 pr_info("ignoring netdev event = %ld for %s\n",
667 event, ndev->name);
668 break;
669 }
670out:
671 return NOTIFY_OK;
672}
673
674struct notifier_block rxe_net_notifier = {
675 .notifier_call = rxe_notify,
676};
677
678static int rxe_net_ipv4_init(void)
679{
680 recv_sockets.sk4 = rxe_setup_udp_tunnel(&init_net,
681 htons(ROCE_V2_UDP_DPORT), false);
682 if (IS_ERR(recv_sockets.sk4)) {
683 recv_sockets.sk4 = NULL;
684 pr_err("Failed to create IPv4 UDP tunnel\n");
685 return -1;
686 }
687
688 return 0;
689}
690
691static int rxe_net_ipv6_init(void)
692{
693#if IS_ENABLED(CONFIG_IPV6)
694
695 recv_sockets.sk6 = rxe_setup_udp_tunnel(&init_net,
696 htons(ROCE_V2_UDP_DPORT), true);
697 if (IS_ERR(recv_sockets.sk6)) {
698 recv_sockets.sk6 = NULL;
699 pr_err("Failed to create IPv6 UDP tunnel\n");
700 return -1;
701 }
702#endif
703 return 0;
704}
705
706void rxe_net_exit(void)
707{
708 rxe_release_udp_tunnel(recv_sockets.sk6);
709 rxe_release_udp_tunnel(recv_sockets.sk4);
710 unregister_netdevice_notifier(&rxe_net_notifier);
711}
712
713int rxe_net_init(void)
714{
715 int err;
716
717 recv_sockets.sk6 = NULL;
718
719 err = rxe_net_ipv4_init();
720 if (err)
721 return err;
722 err = rxe_net_ipv6_init();
723 if (err)
724 goto err_out;
725 err = register_netdevice_notifier(&rxe_net_notifier);
726 if (err) {
727 pr_err("Failed to register netdev notifier\n");
728 goto err_out;
729 }
730 return 0;
731err_out:
732 rxe_net_exit();
733 return err;
734}
735
