Loading...
1#include <linux/module.h>
2#include <linux/errno.h>
3#include <linux/socket.h>
4#include <linux/skbuff.h>
5#include <linux/ip.h>
6#include <linux/udp.h>
7#include <linux/types.h>
8#include <linux/kernel.h>
9#include <net/genetlink.h>
10#include <net/gue.h>
11#include <net/fou.h>
12#include <net/ip.h>
13#include <net/protocol.h>
14#include <net/udp.h>
15#include <net/udp_tunnel.h>
16#include <net/xfrm.h>
17#include <uapi/linux/fou.h>
18#include <uapi/linux/genetlink.h>
19
20struct fou {
21 struct socket *sock;
22 u8 protocol;
23 u8 flags;
24 __be16 port;
25 u8 family;
26 u16 type;
27 struct list_head list;
28 struct rcu_head rcu;
29};
30
31#define FOU_F_REMCSUM_NOPARTIAL BIT(0)
32
33struct fou_cfg {
34 u16 type;
35 u8 protocol;
36 u8 flags;
37 struct udp_port_cfg udp_config;
38};
39
40static unsigned int fou_net_id;
41
42struct fou_net {
43 struct list_head fou_list;
44 struct mutex fou_lock;
45};
46
47static inline struct fou *fou_from_sock(struct sock *sk)
48{
49 return sk->sk_user_data;
50}
51
52static int fou_recv_pull(struct sk_buff *skb, struct fou *fou, size_t len)
53{
54 /* Remove 'len' bytes from the packet (UDP header and
55 * FOU header if present).
56 */
57 if (fou->family == AF_INET)
58 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(skb)->tot_len) - len);
59 else
60 ipv6_hdr(skb)->payload_len =
61 htons(ntohs(ipv6_hdr(skb)->payload_len) - len);
62
63 __skb_pull(skb, len);
64 skb_postpull_rcsum(skb, udp_hdr(skb), len);
65 skb_reset_transport_header(skb);
66 return iptunnel_pull_offloads(skb);
67}
68
69static int fou_udp_recv(struct sock *sk, struct sk_buff *skb)
70{
71 struct fou *fou = fou_from_sock(sk);
72
73 if (!fou)
74 return 1;
75
76 if (fou_recv_pull(skb, fou, sizeof(struct udphdr)))
77 goto drop;
78
79 return -fou->protocol;
80
81drop:
82 kfree_skb(skb);
83 return 0;
84}
85
86static struct guehdr *gue_remcsum(struct sk_buff *skb, struct guehdr *guehdr,
87 void *data, size_t hdrlen, u8 ipproto,
88 bool nopartial)
89{
90 __be16 *pd = data;
91 size_t start = ntohs(pd[0]);
92 size_t offset = ntohs(pd[1]);
93 size_t plen = sizeof(struct udphdr) + hdrlen +
94 max_t(size_t, offset + sizeof(u16), start);
95
96 if (skb->remcsum_offload)
97 return guehdr;
98
99 if (!pskb_may_pull(skb, plen))
100 return NULL;
101 guehdr = (struct guehdr *)&udp_hdr(skb)[1];
102
103 skb_remcsum_process(skb, (void *)guehdr + hdrlen,
104 start, offset, nopartial);
105
106 return guehdr;
107}
108
109static int gue_control_message(struct sk_buff *skb, struct guehdr *guehdr)
110{
111 /* No support yet */
112 kfree_skb(skb);
113 return 0;
114}
115
116static int gue_udp_recv(struct sock *sk, struct sk_buff *skb)
117{
118 struct fou *fou = fou_from_sock(sk);
119 size_t len, optlen, hdrlen;
120 struct guehdr *guehdr;
121 void *data;
122 u16 doffset = 0;
123
124 if (!fou)
125 return 1;
126
127 len = sizeof(struct udphdr) + sizeof(struct guehdr);
128 if (!pskb_may_pull(skb, len))
129 goto drop;
130
131 guehdr = (struct guehdr *)&udp_hdr(skb)[1];
132
133 switch (guehdr->version) {
134 case 0: /* Full GUE header present */
135 break;
136
137 case 1: {
138 /* Direct encasulation of IPv4 or IPv6 */
139
140 int prot;
141
142 switch (((struct iphdr *)guehdr)->version) {
143 case 4:
144 prot = IPPROTO_IPIP;
145 break;
146 case 6:
147 prot = IPPROTO_IPV6;
148 break;
149 default:
150 goto drop;
151 }
152
153 if (fou_recv_pull(skb, fou, sizeof(struct udphdr)))
154 goto drop;
155
156 return -prot;
157 }
158
159 default: /* Undefined version */
160 goto drop;
161 }
162
163 optlen = guehdr->hlen << 2;
164 len += optlen;
165
166 if (!pskb_may_pull(skb, len))
167 goto drop;
168
169 /* guehdr may change after pull */
170 guehdr = (struct guehdr *)&udp_hdr(skb)[1];
171
172 hdrlen = sizeof(struct guehdr) + optlen;
173
174 if (guehdr->version != 0 || validate_gue_flags(guehdr, optlen))
175 goto drop;
176
177 hdrlen = sizeof(struct guehdr) + optlen;
178
179 if (fou->family == AF_INET)
180 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(skb)->tot_len) - len);
181 else
182 ipv6_hdr(skb)->payload_len =
183 htons(ntohs(ipv6_hdr(skb)->payload_len) - len);
184
185 /* Pull csum through the guehdr now . This can be used if
186 * there is a remote checksum offload.
187 */
188 skb_postpull_rcsum(skb, udp_hdr(skb), len);
189
190 data = &guehdr[1];
191
192 if (guehdr->flags & GUE_FLAG_PRIV) {
193 __be32 flags = *(__be32 *)(data + doffset);
194
195 doffset += GUE_LEN_PRIV;
196
197 if (flags & GUE_PFLAG_REMCSUM) {
198 guehdr = gue_remcsum(skb, guehdr, data + doffset,
199 hdrlen, guehdr->proto_ctype,
200 !!(fou->flags &
201 FOU_F_REMCSUM_NOPARTIAL));
202 if (!guehdr)
203 goto drop;
204
205 data = &guehdr[1];
206
207 doffset += GUE_PLEN_REMCSUM;
208 }
209 }
210
211 if (unlikely(guehdr->control))
212 return gue_control_message(skb, guehdr);
213
214 __skb_pull(skb, sizeof(struct udphdr) + hdrlen);
215 skb_reset_transport_header(skb);
216
217 if (iptunnel_pull_offloads(skb))
218 goto drop;
219
220 return -guehdr->proto_ctype;
221
222drop:
223 kfree_skb(skb);
224 return 0;
225}
226
227static struct sk_buff **fou_gro_receive(struct sock *sk,
228 struct sk_buff **head,
229 struct sk_buff *skb)
230{
231 const struct net_offload *ops;
232 struct sk_buff **pp = NULL;
233 u8 proto = fou_from_sock(sk)->protocol;
234 const struct net_offload **offloads;
235
236 /* We can clear the encap_mark for FOU as we are essentially doing
237 * one of two possible things. We are either adding an L4 tunnel
238 * header to the outer L3 tunnel header, or we are are simply
239 * treating the GRE tunnel header as though it is a UDP protocol
240 * specific header such as VXLAN or GENEVE.
241 */
242 NAPI_GRO_CB(skb)->encap_mark = 0;
243
244 /* Flag this frame as already having an outer encap header */
245 NAPI_GRO_CB(skb)->is_fou = 1;
246
247 rcu_read_lock();
248 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
249 ops = rcu_dereference(offloads[proto]);
250 if (!ops || !ops->callbacks.gro_receive)
251 goto out_unlock;
252
253 pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
254
255out_unlock:
256 rcu_read_unlock();
257
258 return pp;
259}
260
261static int fou_gro_complete(struct sock *sk, struct sk_buff *skb,
262 int nhoff)
263{
264 const struct net_offload *ops;
265 u8 proto = fou_from_sock(sk)->protocol;
266 int err = -ENOSYS;
267 const struct net_offload **offloads;
268
269 rcu_read_lock();
270 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
271 ops = rcu_dereference(offloads[proto]);
272 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
273 goto out_unlock;
274
275 err = ops->callbacks.gro_complete(skb, nhoff);
276
277 skb_set_inner_mac_header(skb, nhoff);
278
279out_unlock:
280 rcu_read_unlock();
281
282 return err;
283}
284
285static struct guehdr *gue_gro_remcsum(struct sk_buff *skb, unsigned int off,
286 struct guehdr *guehdr, void *data,
287 size_t hdrlen, struct gro_remcsum *grc,
288 bool nopartial)
289{
290 __be16 *pd = data;
291 size_t start = ntohs(pd[0]);
292 size_t offset = ntohs(pd[1]);
293
294 if (skb->remcsum_offload)
295 return guehdr;
296
297 if (!NAPI_GRO_CB(skb)->csum_valid)
298 return NULL;
299
300 guehdr = skb_gro_remcsum_process(skb, (void *)guehdr, off, hdrlen,
301 start, offset, grc, nopartial);
302
303 skb->remcsum_offload = 1;
304
305 return guehdr;
306}
307
308static struct sk_buff **gue_gro_receive(struct sock *sk,
309 struct sk_buff **head,
310 struct sk_buff *skb)
311{
312 const struct net_offload **offloads;
313 const struct net_offload *ops;
314 struct sk_buff **pp = NULL;
315 struct sk_buff *p;
316 struct guehdr *guehdr;
317 size_t len, optlen, hdrlen, off;
318 void *data;
319 u16 doffset = 0;
320 int flush = 1;
321 struct fou *fou = fou_from_sock(sk);
322 struct gro_remcsum grc;
323 u8 proto;
324
325 skb_gro_remcsum_init(&grc);
326
327 off = skb_gro_offset(skb);
328 len = off + sizeof(*guehdr);
329
330 guehdr = skb_gro_header_fast(skb, off);
331 if (skb_gro_header_hard(skb, len)) {
332 guehdr = skb_gro_header_slow(skb, len, off);
333 if (unlikely(!guehdr))
334 goto out;
335 }
336
337 switch (guehdr->version) {
338 case 0:
339 break;
340 case 1:
341 switch (((struct iphdr *)guehdr)->version) {
342 case 4:
343 proto = IPPROTO_IPIP;
344 break;
345 case 6:
346 proto = IPPROTO_IPV6;
347 break;
348 default:
349 goto out;
350 }
351 goto next_proto;
352 default:
353 goto out;
354 }
355
356 optlen = guehdr->hlen << 2;
357 len += optlen;
358
359 if (skb_gro_header_hard(skb, len)) {
360 guehdr = skb_gro_header_slow(skb, len, off);
361 if (unlikely(!guehdr))
362 goto out;
363 }
364
365 if (unlikely(guehdr->control) || guehdr->version != 0 ||
366 validate_gue_flags(guehdr, optlen))
367 goto out;
368
369 hdrlen = sizeof(*guehdr) + optlen;
370
371 /* Adjust NAPI_GRO_CB(skb)->csum to account for guehdr,
372 * this is needed if there is a remote checkcsum offload.
373 */
374 skb_gro_postpull_rcsum(skb, guehdr, hdrlen);
375
376 data = &guehdr[1];
377
378 if (guehdr->flags & GUE_FLAG_PRIV) {
379 __be32 flags = *(__be32 *)(data + doffset);
380
381 doffset += GUE_LEN_PRIV;
382
383 if (flags & GUE_PFLAG_REMCSUM) {
384 guehdr = gue_gro_remcsum(skb, off, guehdr,
385 data + doffset, hdrlen, &grc,
386 !!(fou->flags &
387 FOU_F_REMCSUM_NOPARTIAL));
388
389 if (!guehdr)
390 goto out;
391
392 data = &guehdr[1];
393
394 doffset += GUE_PLEN_REMCSUM;
395 }
396 }
397
398 skb_gro_pull(skb, hdrlen);
399
400 for (p = *head; p; p = p->next) {
401 const struct guehdr *guehdr2;
402
403 if (!NAPI_GRO_CB(p)->same_flow)
404 continue;
405
406 guehdr2 = (struct guehdr *)(p->data + off);
407
408 /* Compare base GUE header to be equal (covers
409 * hlen, version, proto_ctype, and flags.
410 */
411 if (guehdr->word != guehdr2->word) {
412 NAPI_GRO_CB(p)->same_flow = 0;
413 continue;
414 }
415
416 /* Compare optional fields are the same. */
417 if (guehdr->hlen && memcmp(&guehdr[1], &guehdr2[1],
418 guehdr->hlen << 2)) {
419 NAPI_GRO_CB(p)->same_flow = 0;
420 continue;
421 }
422 }
423
424 proto = guehdr->proto_ctype;
425
426next_proto:
427
428 /* We can clear the encap_mark for GUE as we are essentially doing
429 * one of two possible things. We are either adding an L4 tunnel
430 * header to the outer L3 tunnel header, or we are are simply
431 * treating the GRE tunnel header as though it is a UDP protocol
432 * specific header such as VXLAN or GENEVE.
433 */
434 NAPI_GRO_CB(skb)->encap_mark = 0;
435
436 /* Flag this frame as already having an outer encap header */
437 NAPI_GRO_CB(skb)->is_fou = 1;
438
439 rcu_read_lock();
440 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
441 ops = rcu_dereference(offloads[proto]);
442 if (WARN_ON_ONCE(!ops || !ops->callbacks.gro_receive))
443 goto out_unlock;
444
445 pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
446 flush = 0;
447
448out_unlock:
449 rcu_read_unlock();
450out:
451 NAPI_GRO_CB(skb)->flush |= flush;
452 skb_gro_remcsum_cleanup(skb, &grc);
453 skb->remcsum_offload = 0;
454
455 return pp;
456}
457
458static int gue_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
459{
460 const struct net_offload **offloads;
461 struct guehdr *guehdr = (struct guehdr *)(skb->data + nhoff);
462 const struct net_offload *ops;
463 unsigned int guehlen = 0;
464 u8 proto;
465 int err = -ENOENT;
466
467 switch (guehdr->version) {
468 case 0:
469 proto = guehdr->proto_ctype;
470 guehlen = sizeof(*guehdr) + (guehdr->hlen << 2);
471 break;
472 case 1:
473 switch (((struct iphdr *)guehdr)->version) {
474 case 4:
475 proto = IPPROTO_IPIP;
476 break;
477 case 6:
478 proto = IPPROTO_IPV6;
479 break;
480 default:
481 return err;
482 }
483 break;
484 default:
485 return err;
486 }
487
488 rcu_read_lock();
489 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
490 ops = rcu_dereference(offloads[proto]);
491 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
492 goto out_unlock;
493
494 err = ops->callbacks.gro_complete(skb, nhoff + guehlen);
495
496 skb_set_inner_mac_header(skb, nhoff + guehlen);
497
498out_unlock:
499 rcu_read_unlock();
500 return err;
501}
502
503static int fou_add_to_port_list(struct net *net, struct fou *fou)
504{
505 struct fou_net *fn = net_generic(net, fou_net_id);
506 struct fou *fout;
507
508 mutex_lock(&fn->fou_lock);
509 list_for_each_entry(fout, &fn->fou_list, list) {
510 if (fou->port == fout->port &&
511 fou->family == fout->family) {
512 mutex_unlock(&fn->fou_lock);
513 return -EALREADY;
514 }
515 }
516
517 list_add(&fou->list, &fn->fou_list);
518 mutex_unlock(&fn->fou_lock);
519
520 return 0;
521}
522
523static void fou_release(struct fou *fou)
524{
525 struct socket *sock = fou->sock;
526
527 list_del(&fou->list);
528 udp_tunnel_sock_release(sock);
529
530 kfree_rcu(fou, rcu);
531}
532
533static int fou_create(struct net *net, struct fou_cfg *cfg,
534 struct socket **sockp)
535{
536 struct socket *sock = NULL;
537 struct fou *fou = NULL;
538 struct sock *sk;
539 struct udp_tunnel_sock_cfg tunnel_cfg;
540 int err;
541
542 /* Open UDP socket */
543 err = udp_sock_create(net, &cfg->udp_config, &sock);
544 if (err < 0)
545 goto error;
546
547 /* Allocate FOU port structure */
548 fou = kzalloc(sizeof(*fou), GFP_KERNEL);
549 if (!fou) {
550 err = -ENOMEM;
551 goto error;
552 }
553
554 sk = sock->sk;
555
556 fou->port = cfg->udp_config.local_udp_port;
557 fou->family = cfg->udp_config.family;
558 fou->flags = cfg->flags;
559 fou->type = cfg->type;
560 fou->sock = sock;
561
562 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
563 tunnel_cfg.encap_type = 1;
564 tunnel_cfg.sk_user_data = fou;
565 tunnel_cfg.encap_destroy = NULL;
566
567 /* Initial for fou type */
568 switch (cfg->type) {
569 case FOU_ENCAP_DIRECT:
570 tunnel_cfg.encap_rcv = fou_udp_recv;
571 tunnel_cfg.gro_receive = fou_gro_receive;
572 tunnel_cfg.gro_complete = fou_gro_complete;
573 fou->protocol = cfg->protocol;
574 break;
575 case FOU_ENCAP_GUE:
576 tunnel_cfg.encap_rcv = gue_udp_recv;
577 tunnel_cfg.gro_receive = gue_gro_receive;
578 tunnel_cfg.gro_complete = gue_gro_complete;
579 break;
580 default:
581 err = -EINVAL;
582 goto error;
583 }
584
585 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
586
587 sk->sk_allocation = GFP_ATOMIC;
588
589 err = fou_add_to_port_list(net, fou);
590 if (err)
591 goto error;
592
593 if (sockp)
594 *sockp = sock;
595
596 return 0;
597
598error:
599 kfree(fou);
600 if (sock)
601 udp_tunnel_sock_release(sock);
602
603 return err;
604}
605
606static int fou_destroy(struct net *net, struct fou_cfg *cfg)
607{
608 struct fou_net *fn = net_generic(net, fou_net_id);
609 __be16 port = cfg->udp_config.local_udp_port;
610 u8 family = cfg->udp_config.family;
611 int err = -EINVAL;
612 struct fou *fou;
613
614 mutex_lock(&fn->fou_lock);
615 list_for_each_entry(fou, &fn->fou_list, list) {
616 if (fou->port == port && fou->family == family) {
617 fou_release(fou);
618 err = 0;
619 break;
620 }
621 }
622 mutex_unlock(&fn->fou_lock);
623
624 return err;
625}
626
627static struct genl_family fou_nl_family;
628
629static const struct nla_policy fou_nl_policy[FOU_ATTR_MAX + 1] = {
630 [FOU_ATTR_PORT] = { .type = NLA_U16, },
631 [FOU_ATTR_AF] = { .type = NLA_U8, },
632 [FOU_ATTR_IPPROTO] = { .type = NLA_U8, },
633 [FOU_ATTR_TYPE] = { .type = NLA_U8, },
634 [FOU_ATTR_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG, },
635};
636
637static int parse_nl_config(struct genl_info *info,
638 struct fou_cfg *cfg)
639{
640 memset(cfg, 0, sizeof(*cfg));
641
642 cfg->udp_config.family = AF_INET;
643
644 if (info->attrs[FOU_ATTR_AF]) {
645 u8 family = nla_get_u8(info->attrs[FOU_ATTR_AF]);
646
647 switch (family) {
648 case AF_INET:
649 break;
650 case AF_INET6:
651 cfg->udp_config.ipv6_v6only = 1;
652 break;
653 default:
654 return -EAFNOSUPPORT;
655 }
656
657 cfg->udp_config.family = family;
658 }
659
660 if (info->attrs[FOU_ATTR_PORT]) {
661 __be16 port = nla_get_be16(info->attrs[FOU_ATTR_PORT]);
662
663 cfg->udp_config.local_udp_port = port;
664 }
665
666 if (info->attrs[FOU_ATTR_IPPROTO])
667 cfg->protocol = nla_get_u8(info->attrs[FOU_ATTR_IPPROTO]);
668
669 if (info->attrs[FOU_ATTR_TYPE])
670 cfg->type = nla_get_u8(info->attrs[FOU_ATTR_TYPE]);
671
672 if (info->attrs[FOU_ATTR_REMCSUM_NOPARTIAL])
673 cfg->flags |= FOU_F_REMCSUM_NOPARTIAL;
674
675 return 0;
676}
677
678static int fou_nl_cmd_add_port(struct sk_buff *skb, struct genl_info *info)
679{
680 struct net *net = genl_info_net(info);
681 struct fou_cfg cfg;
682 int err;
683
684 err = parse_nl_config(info, &cfg);
685 if (err)
686 return err;
687
688 return fou_create(net, &cfg, NULL);
689}
690
691static int fou_nl_cmd_rm_port(struct sk_buff *skb, struct genl_info *info)
692{
693 struct net *net = genl_info_net(info);
694 struct fou_cfg cfg;
695 int err;
696
697 err = parse_nl_config(info, &cfg);
698 if (err)
699 return err;
700
701 return fou_destroy(net, &cfg);
702}
703
704static int fou_fill_info(struct fou *fou, struct sk_buff *msg)
705{
706 if (nla_put_u8(msg, FOU_ATTR_AF, fou->sock->sk->sk_family) ||
707 nla_put_be16(msg, FOU_ATTR_PORT, fou->port) ||
708 nla_put_u8(msg, FOU_ATTR_IPPROTO, fou->protocol) ||
709 nla_put_u8(msg, FOU_ATTR_TYPE, fou->type))
710 return -1;
711
712 if (fou->flags & FOU_F_REMCSUM_NOPARTIAL)
713 if (nla_put_flag(msg, FOU_ATTR_REMCSUM_NOPARTIAL))
714 return -1;
715 return 0;
716}
717
718static int fou_dump_info(struct fou *fou, u32 portid, u32 seq,
719 u32 flags, struct sk_buff *skb, u8 cmd)
720{
721 void *hdr;
722
723 hdr = genlmsg_put(skb, portid, seq, &fou_nl_family, flags, cmd);
724 if (!hdr)
725 return -ENOMEM;
726
727 if (fou_fill_info(fou, skb) < 0)
728 goto nla_put_failure;
729
730 genlmsg_end(skb, hdr);
731 return 0;
732
733nla_put_failure:
734 genlmsg_cancel(skb, hdr);
735 return -EMSGSIZE;
736}
737
738static int fou_nl_cmd_get_port(struct sk_buff *skb, struct genl_info *info)
739{
740 struct net *net = genl_info_net(info);
741 struct fou_net *fn = net_generic(net, fou_net_id);
742 struct sk_buff *msg;
743 struct fou_cfg cfg;
744 struct fou *fout;
745 __be16 port;
746 u8 family;
747 int ret;
748
749 ret = parse_nl_config(info, &cfg);
750 if (ret)
751 return ret;
752 port = cfg.udp_config.local_udp_port;
753 if (port == 0)
754 return -EINVAL;
755
756 family = cfg.udp_config.family;
757 if (family != AF_INET && family != AF_INET6)
758 return -EINVAL;
759
760 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
761 if (!msg)
762 return -ENOMEM;
763
764 ret = -ESRCH;
765 mutex_lock(&fn->fou_lock);
766 list_for_each_entry(fout, &fn->fou_list, list) {
767 if (port == fout->port && family == fout->family) {
768 ret = fou_dump_info(fout, info->snd_portid,
769 info->snd_seq, 0, msg,
770 info->genlhdr->cmd);
771 break;
772 }
773 }
774 mutex_unlock(&fn->fou_lock);
775 if (ret < 0)
776 goto out_free;
777
778 return genlmsg_reply(msg, info);
779
780out_free:
781 nlmsg_free(msg);
782 return ret;
783}
784
785static int fou_nl_dump(struct sk_buff *skb, struct netlink_callback *cb)
786{
787 struct net *net = sock_net(skb->sk);
788 struct fou_net *fn = net_generic(net, fou_net_id);
789 struct fou *fout;
790 int idx = 0, ret;
791
792 mutex_lock(&fn->fou_lock);
793 list_for_each_entry(fout, &fn->fou_list, list) {
794 if (idx++ < cb->args[0])
795 continue;
796 ret = fou_dump_info(fout, NETLINK_CB(cb->skb).portid,
797 cb->nlh->nlmsg_seq, NLM_F_MULTI,
798 skb, FOU_CMD_GET);
799 if (ret)
800 break;
801 }
802 mutex_unlock(&fn->fou_lock);
803
804 cb->args[0] = idx;
805 return skb->len;
806}
807
808static const struct genl_ops fou_nl_ops[] = {
809 {
810 .cmd = FOU_CMD_ADD,
811 .doit = fou_nl_cmd_add_port,
812 .policy = fou_nl_policy,
813 .flags = GENL_ADMIN_PERM,
814 },
815 {
816 .cmd = FOU_CMD_DEL,
817 .doit = fou_nl_cmd_rm_port,
818 .policy = fou_nl_policy,
819 .flags = GENL_ADMIN_PERM,
820 },
821 {
822 .cmd = FOU_CMD_GET,
823 .doit = fou_nl_cmd_get_port,
824 .dumpit = fou_nl_dump,
825 .policy = fou_nl_policy,
826 },
827};
828
829static struct genl_family fou_nl_family __ro_after_init = {
830 .hdrsize = 0,
831 .name = FOU_GENL_NAME,
832 .version = FOU_GENL_VERSION,
833 .maxattr = FOU_ATTR_MAX,
834 .netnsok = true,
835 .module = THIS_MODULE,
836 .ops = fou_nl_ops,
837 .n_ops = ARRAY_SIZE(fou_nl_ops),
838};
839
840size_t fou_encap_hlen(struct ip_tunnel_encap *e)
841{
842 return sizeof(struct udphdr);
843}
844EXPORT_SYMBOL(fou_encap_hlen);
845
846size_t gue_encap_hlen(struct ip_tunnel_encap *e)
847{
848 size_t len;
849 bool need_priv = false;
850
851 len = sizeof(struct udphdr) + sizeof(struct guehdr);
852
853 if (e->flags & TUNNEL_ENCAP_FLAG_REMCSUM) {
854 len += GUE_PLEN_REMCSUM;
855 need_priv = true;
856 }
857
858 len += need_priv ? GUE_LEN_PRIV : 0;
859
860 return len;
861}
862EXPORT_SYMBOL(gue_encap_hlen);
863
864int __fou_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
865 u8 *protocol, __be16 *sport, int type)
866{
867 int err;
868
869 err = iptunnel_handle_offloads(skb, type);
870 if (err)
871 return err;
872
873 *sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
874 skb, 0, 0, false);
875
876 return 0;
877}
878EXPORT_SYMBOL(__fou_build_header);
879
880int __gue_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
881 u8 *protocol, __be16 *sport, int type)
882{
883 struct guehdr *guehdr;
884 size_t hdrlen, optlen = 0;
885 void *data;
886 bool need_priv = false;
887 int err;
888
889 if ((e->flags & TUNNEL_ENCAP_FLAG_REMCSUM) &&
890 skb->ip_summed == CHECKSUM_PARTIAL) {
891 optlen += GUE_PLEN_REMCSUM;
892 type |= SKB_GSO_TUNNEL_REMCSUM;
893 need_priv = true;
894 }
895
896 optlen += need_priv ? GUE_LEN_PRIV : 0;
897
898 err = iptunnel_handle_offloads(skb, type);
899 if (err)
900 return err;
901
902 /* Get source port (based on flow hash) before skb_push */
903 *sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
904 skb, 0, 0, false);
905
906 hdrlen = sizeof(struct guehdr) + optlen;
907
908 skb_push(skb, hdrlen);
909
910 guehdr = (struct guehdr *)skb->data;
911
912 guehdr->control = 0;
913 guehdr->version = 0;
914 guehdr->hlen = optlen >> 2;
915 guehdr->flags = 0;
916 guehdr->proto_ctype = *protocol;
917
918 data = &guehdr[1];
919
920 if (need_priv) {
921 __be32 *flags = data;
922
923 guehdr->flags |= GUE_FLAG_PRIV;
924 *flags = 0;
925 data += GUE_LEN_PRIV;
926
927 if (type & SKB_GSO_TUNNEL_REMCSUM) {
928 u16 csum_start = skb_checksum_start_offset(skb);
929 __be16 *pd = data;
930
931 if (csum_start < hdrlen)
932 return -EINVAL;
933
934 csum_start -= hdrlen;
935 pd[0] = htons(csum_start);
936 pd[1] = htons(csum_start + skb->csum_offset);
937
938 if (!skb_is_gso(skb)) {
939 skb->ip_summed = CHECKSUM_NONE;
940 skb->encapsulation = 0;
941 }
942
943 *flags |= GUE_PFLAG_REMCSUM;
944 data += GUE_PLEN_REMCSUM;
945 }
946
947 }
948
949 return 0;
950}
951EXPORT_SYMBOL(__gue_build_header);
952
953#ifdef CONFIG_NET_FOU_IP_TUNNELS
954
955static void fou_build_udp(struct sk_buff *skb, struct ip_tunnel_encap *e,
956 struct flowi4 *fl4, u8 *protocol, __be16 sport)
957{
958 struct udphdr *uh;
959
960 skb_push(skb, sizeof(struct udphdr));
961 skb_reset_transport_header(skb);
962
963 uh = udp_hdr(skb);
964
965 uh->dest = e->dport;
966 uh->source = sport;
967 uh->len = htons(skb->len);
968 udp_set_csum(!(e->flags & TUNNEL_ENCAP_FLAG_CSUM), skb,
969 fl4->saddr, fl4->daddr, skb->len);
970
971 *protocol = IPPROTO_UDP;
972}
973
974static int fou_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
975 u8 *protocol, struct flowi4 *fl4)
976{
977 int type = e->flags & TUNNEL_ENCAP_FLAG_CSUM ? SKB_GSO_UDP_TUNNEL_CSUM :
978 SKB_GSO_UDP_TUNNEL;
979 __be16 sport;
980 int err;
981
982 err = __fou_build_header(skb, e, protocol, &sport, type);
983 if (err)
984 return err;
985
986 fou_build_udp(skb, e, fl4, protocol, sport);
987
988 return 0;
989}
990
991static int gue_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
992 u8 *protocol, struct flowi4 *fl4)
993{
994 int type = e->flags & TUNNEL_ENCAP_FLAG_CSUM ? SKB_GSO_UDP_TUNNEL_CSUM :
995 SKB_GSO_UDP_TUNNEL;
996 __be16 sport;
997 int err;
998
999 err = __gue_build_header(skb, e, protocol, &sport, type);
1000 if (err)
1001 return err;
1002
1003 fou_build_udp(skb, e, fl4, protocol, sport);
1004
1005 return 0;
1006}
1007
1008
1009static const struct ip_tunnel_encap_ops fou_iptun_ops = {
1010 .encap_hlen = fou_encap_hlen,
1011 .build_header = fou_build_header,
1012};
1013
1014static const struct ip_tunnel_encap_ops gue_iptun_ops = {
1015 .encap_hlen = gue_encap_hlen,
1016 .build_header = gue_build_header,
1017};
1018
1019static int ip_tunnel_encap_add_fou_ops(void)
1020{
1021 int ret;
1022
1023 ret = ip_tunnel_encap_add_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
1024 if (ret < 0) {
1025 pr_err("can't add fou ops\n");
1026 return ret;
1027 }
1028
1029 ret = ip_tunnel_encap_add_ops(&gue_iptun_ops, TUNNEL_ENCAP_GUE);
1030 if (ret < 0) {
1031 pr_err("can't add gue ops\n");
1032 ip_tunnel_encap_del_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
1033 return ret;
1034 }
1035
1036 return 0;
1037}
1038
1039static void ip_tunnel_encap_del_fou_ops(void)
1040{
1041 ip_tunnel_encap_del_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
1042 ip_tunnel_encap_del_ops(&gue_iptun_ops, TUNNEL_ENCAP_GUE);
1043}
1044
1045#else
1046
1047static int ip_tunnel_encap_add_fou_ops(void)
1048{
1049 return 0;
1050}
1051
1052static void ip_tunnel_encap_del_fou_ops(void)
1053{
1054}
1055
1056#endif
1057
1058static __net_init int fou_init_net(struct net *net)
1059{
1060 struct fou_net *fn = net_generic(net, fou_net_id);
1061
1062 INIT_LIST_HEAD(&fn->fou_list);
1063 mutex_init(&fn->fou_lock);
1064 return 0;
1065}
1066
1067static __net_exit void fou_exit_net(struct net *net)
1068{
1069 struct fou_net *fn = net_generic(net, fou_net_id);
1070 struct fou *fou, *next;
1071
1072 /* Close all the FOU sockets */
1073 mutex_lock(&fn->fou_lock);
1074 list_for_each_entry_safe(fou, next, &fn->fou_list, list)
1075 fou_release(fou);
1076 mutex_unlock(&fn->fou_lock);
1077}
1078
1079static struct pernet_operations fou_net_ops = {
1080 .init = fou_init_net,
1081 .exit = fou_exit_net,
1082 .id = &fou_net_id,
1083 .size = sizeof(struct fou_net),
1084};
1085
1086static int __init fou_init(void)
1087{
1088 int ret;
1089
1090 ret = register_pernet_device(&fou_net_ops);
1091 if (ret)
1092 goto exit;
1093
1094 ret = genl_register_family(&fou_nl_family);
1095 if (ret < 0)
1096 goto unregister;
1097
1098 ret = ip_tunnel_encap_add_fou_ops();
1099 if (ret == 0)
1100 return 0;
1101
1102 genl_unregister_family(&fou_nl_family);
1103unregister:
1104 unregister_pernet_device(&fou_net_ops);
1105exit:
1106 return ret;
1107}
1108
1109static void __exit fou_fini(void)
1110{
1111 ip_tunnel_encap_del_fou_ops();
1112 genl_unregister_family(&fou_nl_family);
1113 unregister_pernet_device(&fou_net_ops);
1114}
1115
1116module_init(fou_init);
1117module_exit(fou_fini);
1118MODULE_AUTHOR("Tom Herbert <therbert@google.com>");
1119MODULE_LICENSE("GPL");
1#include <linux/module.h>
2#include <linux/errno.h>
3#include <linux/socket.h>
4#include <linux/skbuff.h>
5#include <linux/ip.h>
6#include <linux/udp.h>
7#include <linux/types.h>
8#include <linux/kernel.h>
9#include <net/genetlink.h>
10#include <net/gue.h>
11#include <net/ip.h>
12#include <net/protocol.h>
13#include <net/udp.h>
14#include <net/udp_tunnel.h>
15#include <net/xfrm.h>
16#include <uapi/linux/fou.h>
17#include <uapi/linux/genetlink.h>
18
19struct fou {
20 struct socket *sock;
21 u8 protocol;
22 u8 flags;
23 __be16 port;
24 u8 family;
25 u16 type;
26 struct list_head list;
27 struct rcu_head rcu;
28};
29
30#define FOU_F_REMCSUM_NOPARTIAL BIT(0)
31
32struct fou_cfg {
33 u16 type;
34 u8 protocol;
35 u8 flags;
36 struct udp_port_cfg udp_config;
37};
38
39static unsigned int fou_net_id;
40
41struct fou_net {
42 struct list_head fou_list;
43 struct mutex fou_lock;
44};
45
46static inline struct fou *fou_from_sock(struct sock *sk)
47{
48 return sk->sk_user_data;
49}
50
51static int fou_recv_pull(struct sk_buff *skb, struct fou *fou, size_t len)
52{
53 /* Remove 'len' bytes from the packet (UDP header and
54 * FOU header if present).
55 */
56 if (fou->family == AF_INET)
57 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(skb)->tot_len) - len);
58 else
59 ipv6_hdr(skb)->payload_len =
60 htons(ntohs(ipv6_hdr(skb)->payload_len) - len);
61
62 __skb_pull(skb, len);
63 skb_postpull_rcsum(skb, udp_hdr(skb), len);
64 skb_reset_transport_header(skb);
65 return iptunnel_pull_offloads(skb);
66}
67
68static int fou_udp_recv(struct sock *sk, struct sk_buff *skb)
69{
70 struct fou *fou = fou_from_sock(sk);
71
72 if (!fou)
73 return 1;
74
75 if (fou_recv_pull(skb, fou, sizeof(struct udphdr)))
76 goto drop;
77
78 return -fou->protocol;
79
80drop:
81 kfree_skb(skb);
82 return 0;
83}
84
85static struct guehdr *gue_remcsum(struct sk_buff *skb, struct guehdr *guehdr,
86 void *data, size_t hdrlen, u8 ipproto,
87 bool nopartial)
88{
89 __be16 *pd = data;
90 size_t start = ntohs(pd[0]);
91 size_t offset = ntohs(pd[1]);
92 size_t plen = sizeof(struct udphdr) + hdrlen +
93 max_t(size_t, offset + sizeof(u16), start);
94
95 if (skb->remcsum_offload)
96 return guehdr;
97
98 if (!pskb_may_pull(skb, plen))
99 return NULL;
100 guehdr = (struct guehdr *)&udp_hdr(skb)[1];
101
102 skb_remcsum_process(skb, (void *)guehdr + hdrlen,
103 start, offset, nopartial);
104
105 return guehdr;
106}
107
108static int gue_control_message(struct sk_buff *skb, struct guehdr *guehdr)
109{
110 /* No support yet */
111 kfree_skb(skb);
112 return 0;
113}
114
115static int gue_udp_recv(struct sock *sk, struct sk_buff *skb)
116{
117 struct fou *fou = fou_from_sock(sk);
118 size_t len, optlen, hdrlen;
119 struct guehdr *guehdr;
120 void *data;
121 u16 doffset = 0;
122
123 if (!fou)
124 return 1;
125
126 len = sizeof(struct udphdr) + sizeof(struct guehdr);
127 if (!pskb_may_pull(skb, len))
128 goto drop;
129
130 guehdr = (struct guehdr *)&udp_hdr(skb)[1];
131
132 switch (guehdr->version) {
133 case 0: /* Full GUE header present */
134 break;
135
136 case 1: {
137 /* Direct encasulation of IPv4 or IPv6 */
138
139 int prot;
140
141 switch (((struct iphdr *)guehdr)->version) {
142 case 4:
143 prot = IPPROTO_IPIP;
144 break;
145 case 6:
146 prot = IPPROTO_IPV6;
147 break;
148 default:
149 goto drop;
150 }
151
152 if (fou_recv_pull(skb, fou, sizeof(struct udphdr)))
153 goto drop;
154
155 return -prot;
156 }
157
158 default: /* Undefined version */
159 goto drop;
160 }
161
162 optlen = guehdr->hlen << 2;
163 len += optlen;
164
165 if (!pskb_may_pull(skb, len))
166 goto drop;
167
168 /* guehdr may change after pull */
169 guehdr = (struct guehdr *)&udp_hdr(skb)[1];
170
171 hdrlen = sizeof(struct guehdr) + optlen;
172
173 if (guehdr->version != 0 || validate_gue_flags(guehdr, optlen))
174 goto drop;
175
176 hdrlen = sizeof(struct guehdr) + optlen;
177
178 if (fou->family == AF_INET)
179 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(skb)->tot_len) - len);
180 else
181 ipv6_hdr(skb)->payload_len =
182 htons(ntohs(ipv6_hdr(skb)->payload_len) - len);
183
184 /* Pull csum through the guehdr now . This can be used if
185 * there is a remote checksum offload.
186 */
187 skb_postpull_rcsum(skb, udp_hdr(skb), len);
188
189 data = &guehdr[1];
190
191 if (guehdr->flags & GUE_FLAG_PRIV) {
192 __be32 flags = *(__be32 *)(data + doffset);
193
194 doffset += GUE_LEN_PRIV;
195
196 if (flags & GUE_PFLAG_REMCSUM) {
197 guehdr = gue_remcsum(skb, guehdr, data + doffset,
198 hdrlen, guehdr->proto_ctype,
199 !!(fou->flags &
200 FOU_F_REMCSUM_NOPARTIAL));
201 if (!guehdr)
202 goto drop;
203
204 data = &guehdr[1];
205
206 doffset += GUE_PLEN_REMCSUM;
207 }
208 }
209
210 if (unlikely(guehdr->control))
211 return gue_control_message(skb, guehdr);
212
213 __skb_pull(skb, sizeof(struct udphdr) + hdrlen);
214 skb_reset_transport_header(skb);
215
216 if (iptunnel_pull_offloads(skb))
217 goto drop;
218
219 return -guehdr->proto_ctype;
220
221drop:
222 kfree_skb(skb);
223 return 0;
224}
225
226static struct sk_buff **fou_gro_receive(struct sock *sk,
227 struct sk_buff **head,
228 struct sk_buff *skb)
229{
230 const struct net_offload *ops;
231 struct sk_buff **pp = NULL;
232 u8 proto = fou_from_sock(sk)->protocol;
233 const struct net_offload **offloads;
234
235 /* We can clear the encap_mark for FOU as we are essentially doing
236 * one of two possible things. We are either adding an L4 tunnel
237 * header to the outer L3 tunnel header, or we are are simply
238 * treating the GRE tunnel header as though it is a UDP protocol
239 * specific header such as VXLAN or GENEVE.
240 */
241 NAPI_GRO_CB(skb)->encap_mark = 0;
242
243 /* Flag this frame as already having an outer encap header */
244 NAPI_GRO_CB(skb)->is_fou = 1;
245
246 rcu_read_lock();
247 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
248 ops = rcu_dereference(offloads[proto]);
249 if (!ops || !ops->callbacks.gro_receive)
250 goto out_unlock;
251
252 pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
253
254out_unlock:
255 rcu_read_unlock();
256
257 return pp;
258}
259
260static int fou_gro_complete(struct sock *sk, struct sk_buff *skb,
261 int nhoff)
262{
263 const struct net_offload *ops;
264 u8 proto = fou_from_sock(sk)->protocol;
265 int err = -ENOSYS;
266 const struct net_offload **offloads;
267
268 rcu_read_lock();
269 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
270 ops = rcu_dereference(offloads[proto]);
271 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
272 goto out_unlock;
273
274 err = ops->callbacks.gro_complete(skb, nhoff);
275
276 skb_set_inner_mac_header(skb, nhoff);
277
278out_unlock:
279 rcu_read_unlock();
280
281 return err;
282}
283
284static struct guehdr *gue_gro_remcsum(struct sk_buff *skb, unsigned int off,
285 struct guehdr *guehdr, void *data,
286 size_t hdrlen, struct gro_remcsum *grc,
287 bool nopartial)
288{
289 __be16 *pd = data;
290 size_t start = ntohs(pd[0]);
291 size_t offset = ntohs(pd[1]);
292
293 if (skb->remcsum_offload)
294 return guehdr;
295
296 if (!NAPI_GRO_CB(skb)->csum_valid)
297 return NULL;
298
299 guehdr = skb_gro_remcsum_process(skb, (void *)guehdr, off, hdrlen,
300 start, offset, grc, nopartial);
301
302 skb->remcsum_offload = 1;
303
304 return guehdr;
305}
306
307static struct sk_buff **gue_gro_receive(struct sock *sk,
308 struct sk_buff **head,
309 struct sk_buff *skb)
310{
311 const struct net_offload **offloads;
312 const struct net_offload *ops;
313 struct sk_buff **pp = NULL;
314 struct sk_buff *p;
315 struct guehdr *guehdr;
316 size_t len, optlen, hdrlen, off;
317 void *data;
318 u16 doffset = 0;
319 int flush = 1;
320 struct fou *fou = fou_from_sock(sk);
321 struct gro_remcsum grc;
322 u8 proto;
323
324 skb_gro_remcsum_init(&grc);
325
326 off = skb_gro_offset(skb);
327 len = off + sizeof(*guehdr);
328
329 guehdr = skb_gro_header_fast(skb, off);
330 if (skb_gro_header_hard(skb, len)) {
331 guehdr = skb_gro_header_slow(skb, len, off);
332 if (unlikely(!guehdr))
333 goto out;
334 }
335
336 switch (guehdr->version) {
337 case 0:
338 break;
339 case 1:
340 switch (((struct iphdr *)guehdr)->version) {
341 case 4:
342 proto = IPPROTO_IPIP;
343 break;
344 case 6:
345 proto = IPPROTO_IPV6;
346 break;
347 default:
348 goto out;
349 }
350 goto next_proto;
351 default:
352 goto out;
353 }
354
355 optlen = guehdr->hlen << 2;
356 len += optlen;
357
358 if (skb_gro_header_hard(skb, len)) {
359 guehdr = skb_gro_header_slow(skb, len, off);
360 if (unlikely(!guehdr))
361 goto out;
362 }
363
364 if (unlikely(guehdr->control) || guehdr->version != 0 ||
365 validate_gue_flags(guehdr, optlen))
366 goto out;
367
368 hdrlen = sizeof(*guehdr) + optlen;
369
370 /* Adjust NAPI_GRO_CB(skb)->csum to account for guehdr,
371 * this is needed if there is a remote checkcsum offload.
372 */
373 skb_gro_postpull_rcsum(skb, guehdr, hdrlen);
374
375 data = &guehdr[1];
376
377 if (guehdr->flags & GUE_FLAG_PRIV) {
378 __be32 flags = *(__be32 *)(data + doffset);
379
380 doffset += GUE_LEN_PRIV;
381
382 if (flags & GUE_PFLAG_REMCSUM) {
383 guehdr = gue_gro_remcsum(skb, off, guehdr,
384 data + doffset, hdrlen, &grc,
385 !!(fou->flags &
386 FOU_F_REMCSUM_NOPARTIAL));
387
388 if (!guehdr)
389 goto out;
390
391 data = &guehdr[1];
392
393 doffset += GUE_PLEN_REMCSUM;
394 }
395 }
396
397 skb_gro_pull(skb, hdrlen);
398
399 for (p = *head; p; p = p->next) {
400 const struct guehdr *guehdr2;
401
402 if (!NAPI_GRO_CB(p)->same_flow)
403 continue;
404
405 guehdr2 = (struct guehdr *)(p->data + off);
406
407 /* Compare base GUE header to be equal (covers
408 * hlen, version, proto_ctype, and flags.
409 */
410 if (guehdr->word != guehdr2->word) {
411 NAPI_GRO_CB(p)->same_flow = 0;
412 continue;
413 }
414
415 /* Compare optional fields are the same. */
416 if (guehdr->hlen && memcmp(&guehdr[1], &guehdr2[1],
417 guehdr->hlen << 2)) {
418 NAPI_GRO_CB(p)->same_flow = 0;
419 continue;
420 }
421 }
422
423 proto = guehdr->proto_ctype;
424
425next_proto:
426
427 /* We can clear the encap_mark for GUE as we are essentially doing
428 * one of two possible things. We are either adding an L4 tunnel
429 * header to the outer L3 tunnel header, or we are are simply
430 * treating the GRE tunnel header as though it is a UDP protocol
431 * specific header such as VXLAN or GENEVE.
432 */
433 NAPI_GRO_CB(skb)->encap_mark = 0;
434
435 /* Flag this frame as already having an outer encap header */
436 NAPI_GRO_CB(skb)->is_fou = 1;
437
438 rcu_read_lock();
439 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
440 ops = rcu_dereference(offloads[proto]);
441 if (WARN_ON_ONCE(!ops || !ops->callbacks.gro_receive))
442 goto out_unlock;
443
444 pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
445 flush = 0;
446
447out_unlock:
448 rcu_read_unlock();
449out:
450 NAPI_GRO_CB(skb)->flush |= flush;
451 skb_gro_remcsum_cleanup(skb, &grc);
452
453 return pp;
454}
455
456static int gue_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
457{
458 const struct net_offload **offloads;
459 struct guehdr *guehdr = (struct guehdr *)(skb->data + nhoff);
460 const struct net_offload *ops;
461 unsigned int guehlen = 0;
462 u8 proto;
463 int err = -ENOENT;
464
465 switch (guehdr->version) {
466 case 0:
467 proto = guehdr->proto_ctype;
468 guehlen = sizeof(*guehdr) + (guehdr->hlen << 2);
469 break;
470 case 1:
471 switch (((struct iphdr *)guehdr)->version) {
472 case 4:
473 proto = IPPROTO_IPIP;
474 break;
475 case 6:
476 proto = IPPROTO_IPV6;
477 break;
478 default:
479 return err;
480 }
481 break;
482 default:
483 return err;
484 }
485
486 rcu_read_lock();
487 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
488 ops = rcu_dereference(offloads[proto]);
489 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
490 goto out_unlock;
491
492 err = ops->callbacks.gro_complete(skb, nhoff + guehlen);
493
494 skb_set_inner_mac_header(skb, nhoff + guehlen);
495
496out_unlock:
497 rcu_read_unlock();
498 return err;
499}
500
501static int fou_add_to_port_list(struct net *net, struct fou *fou)
502{
503 struct fou_net *fn = net_generic(net, fou_net_id);
504 struct fou *fout;
505
506 mutex_lock(&fn->fou_lock);
507 list_for_each_entry(fout, &fn->fou_list, list) {
508 if (fou->port == fout->port &&
509 fou->family == fout->family) {
510 mutex_unlock(&fn->fou_lock);
511 return -EALREADY;
512 }
513 }
514
515 list_add(&fou->list, &fn->fou_list);
516 mutex_unlock(&fn->fou_lock);
517
518 return 0;
519}
520
521static void fou_release(struct fou *fou)
522{
523 struct socket *sock = fou->sock;
524
525 list_del(&fou->list);
526 udp_tunnel_sock_release(sock);
527
528 kfree_rcu(fou, rcu);
529}
530
531static int fou_create(struct net *net, struct fou_cfg *cfg,
532 struct socket **sockp)
533{
534 struct socket *sock = NULL;
535 struct fou *fou = NULL;
536 struct sock *sk;
537 struct udp_tunnel_sock_cfg tunnel_cfg;
538 int err;
539
540 /* Open UDP socket */
541 err = udp_sock_create(net, &cfg->udp_config, &sock);
542 if (err < 0)
543 goto error;
544
545 /* Allocate FOU port structure */
546 fou = kzalloc(sizeof(*fou), GFP_KERNEL);
547 if (!fou) {
548 err = -ENOMEM;
549 goto error;
550 }
551
552 sk = sock->sk;
553
554 fou->port = cfg->udp_config.local_udp_port;
555 fou->family = cfg->udp_config.family;
556 fou->flags = cfg->flags;
557 fou->type = cfg->type;
558 fou->sock = sock;
559
560 memset(&tunnel_cfg, 0, sizeof(tunnel_cfg));
561 tunnel_cfg.encap_type = 1;
562 tunnel_cfg.sk_user_data = fou;
563 tunnel_cfg.encap_destroy = NULL;
564
565 /* Initial for fou type */
566 switch (cfg->type) {
567 case FOU_ENCAP_DIRECT:
568 tunnel_cfg.encap_rcv = fou_udp_recv;
569 tunnel_cfg.gro_receive = fou_gro_receive;
570 tunnel_cfg.gro_complete = fou_gro_complete;
571 fou->protocol = cfg->protocol;
572 break;
573 case FOU_ENCAP_GUE:
574 tunnel_cfg.encap_rcv = gue_udp_recv;
575 tunnel_cfg.gro_receive = gue_gro_receive;
576 tunnel_cfg.gro_complete = gue_gro_complete;
577 break;
578 default:
579 err = -EINVAL;
580 goto error;
581 }
582
583 setup_udp_tunnel_sock(net, sock, &tunnel_cfg);
584
585 sk->sk_allocation = GFP_ATOMIC;
586
587 err = fou_add_to_port_list(net, fou);
588 if (err)
589 goto error;
590
591 if (sockp)
592 *sockp = sock;
593
594 return 0;
595
596error:
597 kfree(fou);
598 if (sock)
599 udp_tunnel_sock_release(sock);
600
601 return err;
602}
603
604static int fou_destroy(struct net *net, struct fou_cfg *cfg)
605{
606 struct fou_net *fn = net_generic(net, fou_net_id);
607 __be16 port = cfg->udp_config.local_udp_port;
608 u8 family = cfg->udp_config.family;
609 int err = -EINVAL;
610 struct fou *fou;
611
612 mutex_lock(&fn->fou_lock);
613 list_for_each_entry(fou, &fn->fou_list, list) {
614 if (fou->port == port && fou->family == family) {
615 fou_release(fou);
616 err = 0;
617 break;
618 }
619 }
620 mutex_unlock(&fn->fou_lock);
621
622 return err;
623}
624
625static struct genl_family fou_nl_family;
626
627static const struct nla_policy fou_nl_policy[FOU_ATTR_MAX + 1] = {
628 [FOU_ATTR_PORT] = { .type = NLA_U16, },
629 [FOU_ATTR_AF] = { .type = NLA_U8, },
630 [FOU_ATTR_IPPROTO] = { .type = NLA_U8, },
631 [FOU_ATTR_TYPE] = { .type = NLA_U8, },
632 [FOU_ATTR_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG, },
633};
634
635static int parse_nl_config(struct genl_info *info,
636 struct fou_cfg *cfg)
637{
638 memset(cfg, 0, sizeof(*cfg));
639
640 cfg->udp_config.family = AF_INET;
641
642 if (info->attrs[FOU_ATTR_AF]) {
643 u8 family = nla_get_u8(info->attrs[FOU_ATTR_AF]);
644
645 switch (family) {
646 case AF_INET:
647 break;
648 case AF_INET6:
649 cfg->udp_config.ipv6_v6only = 1;
650 break;
651 default:
652 return -EAFNOSUPPORT;
653 }
654
655 cfg->udp_config.family = family;
656 }
657
658 if (info->attrs[FOU_ATTR_PORT]) {
659 __be16 port = nla_get_be16(info->attrs[FOU_ATTR_PORT]);
660
661 cfg->udp_config.local_udp_port = port;
662 }
663
664 if (info->attrs[FOU_ATTR_IPPROTO])
665 cfg->protocol = nla_get_u8(info->attrs[FOU_ATTR_IPPROTO]);
666
667 if (info->attrs[FOU_ATTR_TYPE])
668 cfg->type = nla_get_u8(info->attrs[FOU_ATTR_TYPE]);
669
670 if (info->attrs[FOU_ATTR_REMCSUM_NOPARTIAL])
671 cfg->flags |= FOU_F_REMCSUM_NOPARTIAL;
672
673 return 0;
674}
675
676static int fou_nl_cmd_add_port(struct sk_buff *skb, struct genl_info *info)
677{
678 struct net *net = genl_info_net(info);
679 struct fou_cfg cfg;
680 int err;
681
682 err = parse_nl_config(info, &cfg);
683 if (err)
684 return err;
685
686 return fou_create(net, &cfg, NULL);
687}
688
689static int fou_nl_cmd_rm_port(struct sk_buff *skb, struct genl_info *info)
690{
691 struct net *net = genl_info_net(info);
692 struct fou_cfg cfg;
693 int err;
694
695 err = parse_nl_config(info, &cfg);
696 if (err)
697 return err;
698
699 return fou_destroy(net, &cfg);
700}
701
702static int fou_fill_info(struct fou *fou, struct sk_buff *msg)
703{
704 if (nla_put_u8(msg, FOU_ATTR_AF, fou->sock->sk->sk_family) ||
705 nla_put_be16(msg, FOU_ATTR_PORT, fou->port) ||
706 nla_put_u8(msg, FOU_ATTR_IPPROTO, fou->protocol) ||
707 nla_put_u8(msg, FOU_ATTR_TYPE, fou->type))
708 return -1;
709
710 if (fou->flags & FOU_F_REMCSUM_NOPARTIAL)
711 if (nla_put_flag(msg, FOU_ATTR_REMCSUM_NOPARTIAL))
712 return -1;
713 return 0;
714}
715
716static int fou_dump_info(struct fou *fou, u32 portid, u32 seq,
717 u32 flags, struct sk_buff *skb, u8 cmd)
718{
719 void *hdr;
720
721 hdr = genlmsg_put(skb, portid, seq, &fou_nl_family, flags, cmd);
722 if (!hdr)
723 return -ENOMEM;
724
725 if (fou_fill_info(fou, skb) < 0)
726 goto nla_put_failure;
727
728 genlmsg_end(skb, hdr);
729 return 0;
730
731nla_put_failure:
732 genlmsg_cancel(skb, hdr);
733 return -EMSGSIZE;
734}
735
736static int fou_nl_cmd_get_port(struct sk_buff *skb, struct genl_info *info)
737{
738 struct net *net = genl_info_net(info);
739 struct fou_net *fn = net_generic(net, fou_net_id);
740 struct sk_buff *msg;
741 struct fou_cfg cfg;
742 struct fou *fout;
743 __be16 port;
744 u8 family;
745 int ret;
746
747 ret = parse_nl_config(info, &cfg);
748 if (ret)
749 return ret;
750 port = cfg.udp_config.local_udp_port;
751 if (port == 0)
752 return -EINVAL;
753
754 family = cfg.udp_config.family;
755 if (family != AF_INET && family != AF_INET6)
756 return -EINVAL;
757
758 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
759 if (!msg)
760 return -ENOMEM;
761
762 ret = -ESRCH;
763 mutex_lock(&fn->fou_lock);
764 list_for_each_entry(fout, &fn->fou_list, list) {
765 if (port == fout->port && family == fout->family) {
766 ret = fou_dump_info(fout, info->snd_portid,
767 info->snd_seq, 0, msg,
768 info->genlhdr->cmd);
769 break;
770 }
771 }
772 mutex_unlock(&fn->fou_lock);
773 if (ret < 0)
774 goto out_free;
775
776 return genlmsg_reply(msg, info);
777
778out_free:
779 nlmsg_free(msg);
780 return ret;
781}
782
783static int fou_nl_dump(struct sk_buff *skb, struct netlink_callback *cb)
784{
785 struct net *net = sock_net(skb->sk);
786 struct fou_net *fn = net_generic(net, fou_net_id);
787 struct fou *fout;
788 int idx = 0, ret;
789
790 mutex_lock(&fn->fou_lock);
791 list_for_each_entry(fout, &fn->fou_list, list) {
792 if (idx++ < cb->args[0])
793 continue;
794 ret = fou_dump_info(fout, NETLINK_CB(cb->skb).portid,
795 cb->nlh->nlmsg_seq, NLM_F_MULTI,
796 skb, FOU_CMD_GET);
797 if (ret)
798 break;
799 }
800 mutex_unlock(&fn->fou_lock);
801
802 cb->args[0] = idx;
803 return skb->len;
804}
805
806static const struct genl_ops fou_nl_ops[] = {
807 {
808 .cmd = FOU_CMD_ADD,
809 .doit = fou_nl_cmd_add_port,
810 .policy = fou_nl_policy,
811 .flags = GENL_ADMIN_PERM,
812 },
813 {
814 .cmd = FOU_CMD_DEL,
815 .doit = fou_nl_cmd_rm_port,
816 .policy = fou_nl_policy,
817 .flags = GENL_ADMIN_PERM,
818 },
819 {
820 .cmd = FOU_CMD_GET,
821 .doit = fou_nl_cmd_get_port,
822 .dumpit = fou_nl_dump,
823 .policy = fou_nl_policy,
824 },
825};
826
827static struct genl_family fou_nl_family __ro_after_init = {
828 .hdrsize = 0,
829 .name = FOU_GENL_NAME,
830 .version = FOU_GENL_VERSION,
831 .maxattr = FOU_ATTR_MAX,
832 .netnsok = true,
833 .module = THIS_MODULE,
834 .ops = fou_nl_ops,
835 .n_ops = ARRAY_SIZE(fou_nl_ops),
836};
837
838size_t fou_encap_hlen(struct ip_tunnel_encap *e)
839{
840 return sizeof(struct udphdr);
841}
842EXPORT_SYMBOL(fou_encap_hlen);
843
844size_t gue_encap_hlen(struct ip_tunnel_encap *e)
845{
846 size_t len;
847 bool need_priv = false;
848
849 len = sizeof(struct udphdr) + sizeof(struct guehdr);
850
851 if (e->flags & TUNNEL_ENCAP_FLAG_REMCSUM) {
852 len += GUE_PLEN_REMCSUM;
853 need_priv = true;
854 }
855
856 len += need_priv ? GUE_LEN_PRIV : 0;
857
858 return len;
859}
860EXPORT_SYMBOL(gue_encap_hlen);
861
862static void fou_build_udp(struct sk_buff *skb, struct ip_tunnel_encap *e,
863 struct flowi4 *fl4, u8 *protocol, __be16 sport)
864{
865 struct udphdr *uh;
866
867 skb_push(skb, sizeof(struct udphdr));
868 skb_reset_transport_header(skb);
869
870 uh = udp_hdr(skb);
871
872 uh->dest = e->dport;
873 uh->source = sport;
874 uh->len = htons(skb->len);
875 udp_set_csum(!(e->flags & TUNNEL_ENCAP_FLAG_CSUM), skb,
876 fl4->saddr, fl4->daddr, skb->len);
877
878 *protocol = IPPROTO_UDP;
879}
880
881int __fou_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
882 u8 *protocol, __be16 *sport, int type)
883{
884 int err;
885
886 err = iptunnel_handle_offloads(skb, type);
887 if (err)
888 return err;
889
890 *sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
891 skb, 0, 0, false);
892
893 return 0;
894}
895EXPORT_SYMBOL(__fou_build_header);
896
897int fou_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
898 u8 *protocol, struct flowi4 *fl4)
899{
900 int type = e->flags & TUNNEL_ENCAP_FLAG_CSUM ? SKB_GSO_UDP_TUNNEL_CSUM :
901 SKB_GSO_UDP_TUNNEL;
902 __be16 sport;
903 int err;
904
905 err = __fou_build_header(skb, e, protocol, &sport, type);
906 if (err)
907 return err;
908
909 fou_build_udp(skb, e, fl4, protocol, sport);
910
911 return 0;
912}
913EXPORT_SYMBOL(fou_build_header);
914
915int __gue_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
916 u8 *protocol, __be16 *sport, int type)
917{
918 struct guehdr *guehdr;
919 size_t hdrlen, optlen = 0;
920 void *data;
921 bool need_priv = false;
922 int err;
923
924 if ((e->flags & TUNNEL_ENCAP_FLAG_REMCSUM) &&
925 skb->ip_summed == CHECKSUM_PARTIAL) {
926 optlen += GUE_PLEN_REMCSUM;
927 type |= SKB_GSO_TUNNEL_REMCSUM;
928 need_priv = true;
929 }
930
931 optlen += need_priv ? GUE_LEN_PRIV : 0;
932
933 err = iptunnel_handle_offloads(skb, type);
934 if (err)
935 return err;
936
937 /* Get source port (based on flow hash) before skb_push */
938 *sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
939 skb, 0, 0, false);
940
941 hdrlen = sizeof(struct guehdr) + optlen;
942
943 skb_push(skb, hdrlen);
944
945 guehdr = (struct guehdr *)skb->data;
946
947 guehdr->control = 0;
948 guehdr->version = 0;
949 guehdr->hlen = optlen >> 2;
950 guehdr->flags = 0;
951 guehdr->proto_ctype = *protocol;
952
953 data = &guehdr[1];
954
955 if (need_priv) {
956 __be32 *flags = data;
957
958 guehdr->flags |= GUE_FLAG_PRIV;
959 *flags = 0;
960 data += GUE_LEN_PRIV;
961
962 if (type & SKB_GSO_TUNNEL_REMCSUM) {
963 u16 csum_start = skb_checksum_start_offset(skb);
964 __be16 *pd = data;
965
966 if (csum_start < hdrlen)
967 return -EINVAL;
968
969 csum_start -= hdrlen;
970 pd[0] = htons(csum_start);
971 pd[1] = htons(csum_start + skb->csum_offset);
972
973 if (!skb_is_gso(skb)) {
974 skb->ip_summed = CHECKSUM_NONE;
975 skb->encapsulation = 0;
976 }
977
978 *flags |= GUE_PFLAG_REMCSUM;
979 data += GUE_PLEN_REMCSUM;
980 }
981
982 }
983
984 return 0;
985}
986EXPORT_SYMBOL(__gue_build_header);
987
988int gue_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
989 u8 *protocol, struct flowi4 *fl4)
990{
991 int type = e->flags & TUNNEL_ENCAP_FLAG_CSUM ? SKB_GSO_UDP_TUNNEL_CSUM :
992 SKB_GSO_UDP_TUNNEL;
993 __be16 sport;
994 int err;
995
996 err = __gue_build_header(skb, e, protocol, &sport, type);
997 if (err)
998 return err;
999
1000 fou_build_udp(skb, e, fl4, protocol, sport);
1001
1002 return 0;
1003}
1004EXPORT_SYMBOL(gue_build_header);
1005
1006#ifdef CONFIG_NET_FOU_IP_TUNNELS
1007
1008static const struct ip_tunnel_encap_ops fou_iptun_ops = {
1009 .encap_hlen = fou_encap_hlen,
1010 .build_header = fou_build_header,
1011};
1012
1013static const struct ip_tunnel_encap_ops gue_iptun_ops = {
1014 .encap_hlen = gue_encap_hlen,
1015 .build_header = gue_build_header,
1016};
1017
1018static int ip_tunnel_encap_add_fou_ops(void)
1019{
1020 int ret;
1021
1022 ret = ip_tunnel_encap_add_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
1023 if (ret < 0) {
1024 pr_err("can't add fou ops\n");
1025 return ret;
1026 }
1027
1028 ret = ip_tunnel_encap_add_ops(&gue_iptun_ops, TUNNEL_ENCAP_GUE);
1029 if (ret < 0) {
1030 pr_err("can't add gue ops\n");
1031 ip_tunnel_encap_del_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
1032 return ret;
1033 }
1034
1035 return 0;
1036}
1037
1038static void ip_tunnel_encap_del_fou_ops(void)
1039{
1040 ip_tunnel_encap_del_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
1041 ip_tunnel_encap_del_ops(&gue_iptun_ops, TUNNEL_ENCAP_GUE);
1042}
1043
1044#else
1045
1046static int ip_tunnel_encap_add_fou_ops(void)
1047{
1048 return 0;
1049}
1050
1051static void ip_tunnel_encap_del_fou_ops(void)
1052{
1053}
1054
1055#endif
1056
1057static __net_init int fou_init_net(struct net *net)
1058{
1059 struct fou_net *fn = net_generic(net, fou_net_id);
1060
1061 INIT_LIST_HEAD(&fn->fou_list);
1062 mutex_init(&fn->fou_lock);
1063 return 0;
1064}
1065
1066static __net_exit void fou_exit_net(struct net *net)
1067{
1068 struct fou_net *fn = net_generic(net, fou_net_id);
1069 struct fou *fou, *next;
1070
1071 /* Close all the FOU sockets */
1072 mutex_lock(&fn->fou_lock);
1073 list_for_each_entry_safe(fou, next, &fn->fou_list, list)
1074 fou_release(fou);
1075 mutex_unlock(&fn->fou_lock);
1076}
1077
1078static struct pernet_operations fou_net_ops = {
1079 .init = fou_init_net,
1080 .exit = fou_exit_net,
1081 .id = &fou_net_id,
1082 .size = sizeof(struct fou_net),
1083};
1084
1085static int __init fou_init(void)
1086{
1087 int ret;
1088
1089 ret = register_pernet_device(&fou_net_ops);
1090 if (ret)
1091 goto exit;
1092
1093 ret = genl_register_family(&fou_nl_family);
1094 if (ret < 0)
1095 goto unregister;
1096
1097 ret = ip_tunnel_encap_add_fou_ops();
1098 if (ret == 0)
1099 return 0;
1100
1101 genl_unregister_family(&fou_nl_family);
1102unregister:
1103 unregister_pernet_device(&fou_net_ops);
1104exit:
1105 return ret;
1106}
1107
1108static void __exit fou_fini(void)
1109{
1110 ip_tunnel_encap_del_fou_ops();
1111 genl_unregister_family(&fou_nl_family);
1112 unregister_pernet_device(&fou_net_ops);
1113}
1114
1115module_init(fou_init);
1116module_exit(fou_fini);
1117MODULE_AUTHOR("Tom Herbert <therbert@google.com>");
1118MODULE_LICENSE("GPL");