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 list_head *head,
229 struct sk_buff *skb)
230{
231 u8 proto = fou_from_sock(sk)->protocol;
232 const struct net_offload **offloads;
233 const struct net_offload *ops;
234 struct sk_buff *pp = NULL;
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 list_head *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 list_for_each_entry(p, head, list) {
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 skb_gro_flush_final_remcsum(skb, pp, flush, &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
862int __fou_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
863 u8 *protocol, __be16 *sport, int type)
864{
865 int err;
866
867 err = iptunnel_handle_offloads(skb, type);
868 if (err)
869 return err;
870
871 *sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
872 skb, 0, 0, false);
873
874 return 0;
875}
876EXPORT_SYMBOL(__fou_build_header);
877
878int __gue_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
879 u8 *protocol, __be16 *sport, int type)
880{
881 struct guehdr *guehdr;
882 size_t hdrlen, optlen = 0;
883 void *data;
884 bool need_priv = false;
885 int err;
886
887 if ((e->flags & TUNNEL_ENCAP_FLAG_REMCSUM) &&
888 skb->ip_summed == CHECKSUM_PARTIAL) {
889 optlen += GUE_PLEN_REMCSUM;
890 type |= SKB_GSO_TUNNEL_REMCSUM;
891 need_priv = true;
892 }
893
894 optlen += need_priv ? GUE_LEN_PRIV : 0;
895
896 err = iptunnel_handle_offloads(skb, type);
897 if (err)
898 return err;
899
900 /* Get source port (based on flow hash) before skb_push */
901 *sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
902 skb, 0, 0, false);
903
904 hdrlen = sizeof(struct guehdr) + optlen;
905
906 skb_push(skb, hdrlen);
907
908 guehdr = (struct guehdr *)skb->data;
909
910 guehdr->control = 0;
911 guehdr->version = 0;
912 guehdr->hlen = optlen >> 2;
913 guehdr->flags = 0;
914 guehdr->proto_ctype = *protocol;
915
916 data = &guehdr[1];
917
918 if (need_priv) {
919 __be32 *flags = data;
920
921 guehdr->flags |= GUE_FLAG_PRIV;
922 *flags = 0;
923 data += GUE_LEN_PRIV;
924
925 if (type & SKB_GSO_TUNNEL_REMCSUM) {
926 u16 csum_start = skb_checksum_start_offset(skb);
927 __be16 *pd = data;
928
929 if (csum_start < hdrlen)
930 return -EINVAL;
931
932 csum_start -= hdrlen;
933 pd[0] = htons(csum_start);
934 pd[1] = htons(csum_start + skb->csum_offset);
935
936 if (!skb_is_gso(skb)) {
937 skb->ip_summed = CHECKSUM_NONE;
938 skb->encapsulation = 0;
939 }
940
941 *flags |= GUE_PFLAG_REMCSUM;
942 data += GUE_PLEN_REMCSUM;
943 }
944
945 }
946
947 return 0;
948}
949EXPORT_SYMBOL(__gue_build_header);
950
951#ifdef CONFIG_NET_FOU_IP_TUNNELS
952
953static void fou_build_udp(struct sk_buff *skb, struct ip_tunnel_encap *e,
954 struct flowi4 *fl4, u8 *protocol, __be16 sport)
955{
956 struct udphdr *uh;
957
958 skb_push(skb, sizeof(struct udphdr));
959 skb_reset_transport_header(skb);
960
961 uh = udp_hdr(skb);
962
963 uh->dest = e->dport;
964 uh->source = sport;
965 uh->len = htons(skb->len);
966 udp_set_csum(!(e->flags & TUNNEL_ENCAP_FLAG_CSUM), skb,
967 fl4->saddr, fl4->daddr, skb->len);
968
969 *protocol = IPPROTO_UDP;
970}
971
972static int fou_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
973 u8 *protocol, struct flowi4 *fl4)
974{
975 int type = e->flags & TUNNEL_ENCAP_FLAG_CSUM ? SKB_GSO_UDP_TUNNEL_CSUM :
976 SKB_GSO_UDP_TUNNEL;
977 __be16 sport;
978 int err;
979
980 err = __fou_build_header(skb, e, protocol, &sport, type);
981 if (err)
982 return err;
983
984 fou_build_udp(skb, e, fl4, protocol, sport);
985
986 return 0;
987}
988
989static int gue_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
990 u8 *protocol, struct flowi4 *fl4)
991{
992 int type = e->flags & TUNNEL_ENCAP_FLAG_CSUM ? SKB_GSO_UDP_TUNNEL_CSUM :
993 SKB_GSO_UDP_TUNNEL;
994 __be16 sport;
995 int err;
996
997 err = __gue_build_header(skb, e, protocol, &sport, type);
998 if (err)
999 return err;
1000
1001 fou_build_udp(skb, e, fl4, protocol, sport);
1002
1003 return 0;
1004}
1005
1006
1007static const struct ip_tunnel_encap_ops fou_iptun_ops = {
1008 .encap_hlen = fou_encap_hlen,
1009 .build_header = fou_build_header,
1010};
1011
1012static const struct ip_tunnel_encap_ops gue_iptun_ops = {
1013 .encap_hlen = gue_encap_hlen,
1014 .build_header = gue_build_header,
1015};
1016
1017static int ip_tunnel_encap_add_fou_ops(void)
1018{
1019 int ret;
1020
1021 ret = ip_tunnel_encap_add_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
1022 if (ret < 0) {
1023 pr_err("can't add fou ops\n");
1024 return ret;
1025 }
1026
1027 ret = ip_tunnel_encap_add_ops(&gue_iptun_ops, TUNNEL_ENCAP_GUE);
1028 if (ret < 0) {
1029 pr_err("can't add gue ops\n");
1030 ip_tunnel_encap_del_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
1031 return ret;
1032 }
1033
1034 return 0;
1035}
1036
1037static void ip_tunnel_encap_del_fou_ops(void)
1038{
1039 ip_tunnel_encap_del_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
1040 ip_tunnel_encap_del_ops(&gue_iptun_ops, TUNNEL_ENCAP_GUE);
1041}
1042
1043#else
1044
1045static int ip_tunnel_encap_add_fou_ops(void)
1046{
1047 return 0;
1048}
1049
1050static void ip_tunnel_encap_del_fou_ops(void)
1051{
1052}
1053
1054#endif
1055
1056static __net_init int fou_init_net(struct net *net)
1057{
1058 struct fou_net *fn = net_generic(net, fou_net_id);
1059
1060 INIT_LIST_HEAD(&fn->fou_list);
1061 mutex_init(&fn->fou_lock);
1062 return 0;
1063}
1064
1065static __net_exit void fou_exit_net(struct net *net)
1066{
1067 struct fou_net *fn = net_generic(net, fou_net_id);
1068 struct fou *fou, *next;
1069
1070 /* Close all the FOU sockets */
1071 mutex_lock(&fn->fou_lock);
1072 list_for_each_entry_safe(fou, next, &fn->fou_list, list)
1073 fou_release(fou);
1074 mutex_unlock(&fn->fou_lock);
1075}
1076
1077static struct pernet_operations fou_net_ops = {
1078 .init = fou_init_net,
1079 .exit = fou_exit_net,
1080 .id = &fou_net_id,
1081 .size = sizeof(struct fou_net),
1082};
1083
1084static int __init fou_init(void)
1085{
1086 int ret;
1087
1088 ret = register_pernet_device(&fou_net_ops);
1089 if (ret)
1090 goto exit;
1091
1092 ret = genl_register_family(&fou_nl_family);
1093 if (ret < 0)
1094 goto unregister;
1095
1096 ret = ip_tunnel_encap_add_fou_ops();
1097 if (ret == 0)
1098 return 0;
1099
1100 genl_unregister_family(&fou_nl_family);
1101unregister:
1102 unregister_pernet_device(&fou_net_ops);
1103exit:
1104 return ret;
1105}
1106
1107static void __exit fou_fini(void)
1108{
1109 ip_tunnel_encap_del_fou_ops();
1110 genl_unregister_family(&fou_nl_family);
1111 unregister_pernet_device(&fou_net_ops);
1112}
1113
1114module_init(fou_init);
1115module_exit(fou_fini);
1116MODULE_AUTHOR("Tom Herbert <therbert@google.com>");
1117MODULE_LICENSE("GPL");
1118