1 #include <linux/module.h>
2 #include <linux/errno.h>
3 #include <linux/socket.h>
4 #include <linux/skbuff.h>
7 #include <linux/types.h>
8 #include <linux/kernel.h>
9 #include <net/genetlink.h>
12 #include <net/protocol.h>
14 #include <net/udp_tunnel.h>
16 #include <uapi/linux/fou.h>
17 #include <uapi/linux/genetlink.h>
25 struct list_head list;
29 #define FOU_F_REMCSUM_NOPARTIAL BIT(0)
35 struct udp_port_cfg udp_config;
38 static unsigned int fou_net_id;
41 struct list_head fou_list;
42 struct mutex fou_lock;
45 static inline struct fou *fou_from_sock(struct sock *sk)
47 return sk->sk_user_data;
50 static int fou_recv_pull(struct sk_buff *skb, size_t len)
52 struct iphdr *iph = ip_hdr(skb);
54 /* Remove 'len' bytes from the packet (UDP header and
55 * FOU header if present).
57 iph->tot_len = htons(ntohs(iph->tot_len) - len);
59 skb_postpull_rcsum(skb, udp_hdr(skb), len);
60 skb_reset_transport_header(skb);
61 return iptunnel_pull_offloads(skb);
64 static int fou_udp_recv(struct sock *sk, struct sk_buff *skb)
66 struct fou *fou = fou_from_sock(sk);
71 if (fou_recv_pull(skb, sizeof(struct udphdr)))
74 return -fou->protocol;
81 static struct guehdr *gue_remcsum(struct sk_buff *skb, struct guehdr *guehdr,
82 void *data, size_t hdrlen, u8 ipproto,
86 size_t start = ntohs(pd[0]);
87 size_t offset = ntohs(pd[1]);
88 size_t plen = sizeof(struct udphdr) + hdrlen +
89 max_t(size_t, offset + sizeof(u16), start);
91 if (skb->remcsum_offload)
94 if (!pskb_may_pull(skb, plen))
96 guehdr = (struct guehdr *)&udp_hdr(skb)[1];
98 skb_remcsum_process(skb, (void *)guehdr + hdrlen,
99 start, offset, nopartial);
104 static int gue_control_message(struct sk_buff *skb, struct guehdr *guehdr)
111 static int gue_udp_recv(struct sock *sk, struct sk_buff *skb)
113 struct fou *fou = fou_from_sock(sk);
114 size_t len, optlen, hdrlen;
115 struct guehdr *guehdr;
122 len = sizeof(struct udphdr) + sizeof(struct guehdr);
123 if (!pskb_may_pull(skb, len))
126 guehdr = (struct guehdr *)&udp_hdr(skb)[1];
128 optlen = guehdr->hlen << 2;
131 if (!pskb_may_pull(skb, len))
134 /* guehdr may change after pull */
135 guehdr = (struct guehdr *)&udp_hdr(skb)[1];
137 hdrlen = sizeof(struct guehdr) + optlen;
139 if (guehdr->version != 0 || validate_gue_flags(guehdr, optlen))
142 hdrlen = sizeof(struct guehdr) + optlen;
144 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(skb)->tot_len) - len);
146 /* Pull csum through the guehdr now . This can be used if
147 * there is a remote checksum offload.
149 skb_postpull_rcsum(skb, udp_hdr(skb), len);
153 if (guehdr->flags & GUE_FLAG_PRIV) {
154 __be32 flags = *(__be32 *)(data + doffset);
156 doffset += GUE_LEN_PRIV;
158 if (flags & GUE_PFLAG_REMCSUM) {
159 guehdr = gue_remcsum(skb, guehdr, data + doffset,
160 hdrlen, guehdr->proto_ctype,
162 FOU_F_REMCSUM_NOPARTIAL));
168 doffset += GUE_PLEN_REMCSUM;
172 if (unlikely(guehdr->control))
173 return gue_control_message(skb, guehdr);
175 __skb_pull(skb, sizeof(struct udphdr) + hdrlen);
176 skb_reset_transport_header(skb);
178 if (iptunnel_pull_offloads(skb))
181 return -guehdr->proto_ctype;
188 static struct sk_buff **fou_gro_receive(struct sock *sk,
189 struct sk_buff **head,
192 const struct net_offload *ops;
193 struct sk_buff **pp = NULL;
194 u8 proto = fou_from_sock(sk)->protocol;
195 const struct net_offload **offloads;
197 /* We can clear the encap_mark for FOU as we are essentially doing
198 * one of two possible things. We are either adding an L4 tunnel
199 * header to the outer L3 tunnel header, or we are are simply
200 * treating the GRE tunnel header as though it is a UDP protocol
201 * specific header such as VXLAN or GENEVE.
203 NAPI_GRO_CB(skb)->encap_mark = 0;
205 /* Flag this frame as already having an outer encap header */
206 NAPI_GRO_CB(skb)->is_fou = 1;
209 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
210 ops = rcu_dereference(offloads[proto]);
211 if (!ops || !ops->callbacks.gro_receive)
214 pp = ops->callbacks.gro_receive(head, skb);
222 static int fou_gro_complete(struct sock *sk, struct sk_buff *skb,
225 const struct net_offload *ops;
226 u8 proto = fou_from_sock(sk)->protocol;
228 const struct net_offload **offloads;
231 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
232 ops = rcu_dereference(offloads[proto]);
233 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
236 err = ops->callbacks.gro_complete(skb, nhoff);
238 skb_set_inner_mac_header(skb, nhoff);
246 static struct guehdr *gue_gro_remcsum(struct sk_buff *skb, unsigned int off,
247 struct guehdr *guehdr, void *data,
248 size_t hdrlen, struct gro_remcsum *grc,
252 size_t start = ntohs(pd[0]);
253 size_t offset = ntohs(pd[1]);
255 if (skb->remcsum_offload)
258 if (!NAPI_GRO_CB(skb)->csum_valid)
261 guehdr = skb_gro_remcsum_process(skb, (void *)guehdr, off, hdrlen,
262 start, offset, grc, nopartial);
264 skb->remcsum_offload = 1;
269 static struct sk_buff **gue_gro_receive(struct sock *sk,
270 struct sk_buff **head,
273 const struct net_offload **offloads;
274 const struct net_offload *ops;
275 struct sk_buff **pp = NULL;
277 struct guehdr *guehdr;
278 size_t len, optlen, hdrlen, off;
282 struct fou *fou = fou_from_sock(sk);
283 struct gro_remcsum grc;
285 skb_gro_remcsum_init(&grc);
287 off = skb_gro_offset(skb);
288 len = off + sizeof(*guehdr);
290 guehdr = skb_gro_header_fast(skb, off);
291 if (skb_gro_header_hard(skb, len)) {
292 guehdr = skb_gro_header_slow(skb, len, off);
293 if (unlikely(!guehdr))
297 optlen = guehdr->hlen << 2;
300 if (skb_gro_header_hard(skb, len)) {
301 guehdr = skb_gro_header_slow(skb, len, off);
302 if (unlikely(!guehdr))
306 if (unlikely(guehdr->control) || guehdr->version != 0 ||
307 validate_gue_flags(guehdr, optlen))
310 hdrlen = sizeof(*guehdr) + optlen;
312 /* Adjust NAPI_GRO_CB(skb)->csum to account for guehdr,
313 * this is needed if there is a remote checkcsum offload.
315 skb_gro_postpull_rcsum(skb, guehdr, hdrlen);
319 if (guehdr->flags & GUE_FLAG_PRIV) {
320 __be32 flags = *(__be32 *)(data + doffset);
322 doffset += GUE_LEN_PRIV;
324 if (flags & GUE_PFLAG_REMCSUM) {
325 guehdr = gue_gro_remcsum(skb, off, guehdr,
326 data + doffset, hdrlen, &grc,
328 FOU_F_REMCSUM_NOPARTIAL));
335 doffset += GUE_PLEN_REMCSUM;
339 skb_gro_pull(skb, hdrlen);
341 for (p = *head; p; p = p->next) {
342 const struct guehdr *guehdr2;
344 if (!NAPI_GRO_CB(p)->same_flow)
347 guehdr2 = (struct guehdr *)(p->data + off);
349 /* Compare base GUE header to be equal (covers
350 * hlen, version, proto_ctype, and flags.
352 if (guehdr->word != guehdr2->word) {
353 NAPI_GRO_CB(p)->same_flow = 0;
357 /* Compare optional fields are the same. */
358 if (guehdr->hlen && memcmp(&guehdr[1], &guehdr2[1],
359 guehdr->hlen << 2)) {
360 NAPI_GRO_CB(p)->same_flow = 0;
365 /* We can clear the encap_mark for GUE as we are essentially doing
366 * one of two possible things. We are either adding an L4 tunnel
367 * header to the outer L3 tunnel header, or we are are simply
368 * treating the GRE tunnel header as though it is a UDP protocol
369 * specific header such as VXLAN or GENEVE.
371 NAPI_GRO_CB(skb)->encap_mark = 0;
373 /* Flag this frame as already having an outer encap header */
374 NAPI_GRO_CB(skb)->is_fou = 1;
377 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
378 ops = rcu_dereference(offloads[guehdr->proto_ctype]);
379 if (WARN_ON_ONCE(!ops || !ops->callbacks.gro_receive))
382 pp = ops->callbacks.gro_receive(head, skb);
388 NAPI_GRO_CB(skb)->flush |= flush;
389 skb_gro_remcsum_cleanup(skb, &grc);
394 static int gue_gro_complete(struct sock *sk, struct sk_buff *skb, int nhoff)
396 const struct net_offload **offloads;
397 struct guehdr *guehdr = (struct guehdr *)(skb->data + nhoff);
398 const struct net_offload *ops;
399 unsigned int guehlen;
403 proto = guehdr->proto_ctype;
405 guehlen = sizeof(*guehdr) + (guehdr->hlen << 2);
408 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
409 ops = rcu_dereference(offloads[proto]);
410 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
413 err = ops->callbacks.gro_complete(skb, nhoff + guehlen);
415 skb_set_inner_mac_header(skb, nhoff + guehlen);
422 static int fou_add_to_port_list(struct net *net, struct fou *fou)
424 struct fou_net *fn = net_generic(net, fou_net_id);
427 mutex_lock(&fn->fou_lock);
428 list_for_each_entry(fout, &fn->fou_list, list) {
429 if (fou->port == fout->port) {
430 mutex_unlock(&fn->fou_lock);
435 list_add(&fou->list, &fn->fou_list);
436 mutex_unlock(&fn->fou_lock);
441 static void fou_release(struct fou *fou)
443 struct socket *sock = fou->sock;
445 list_del(&fou->list);
446 udp_tunnel_sock_release(sock);
451 static int fou_encap_init(struct sock *sk, struct fou *fou, struct fou_cfg *cfg)
453 udp_sk(sk)->encap_rcv = fou_udp_recv;
454 udp_sk(sk)->gro_receive = fou_gro_receive;
455 udp_sk(sk)->gro_complete = fou_gro_complete;
456 fou_from_sock(sk)->protocol = cfg->protocol;
461 static int gue_encap_init(struct sock *sk, struct fou *fou, struct fou_cfg *cfg)
463 udp_sk(sk)->encap_rcv = gue_udp_recv;
464 udp_sk(sk)->gro_receive = gue_gro_receive;
465 udp_sk(sk)->gro_complete = gue_gro_complete;
470 static int fou_create(struct net *net, struct fou_cfg *cfg,
471 struct socket **sockp)
473 struct socket *sock = NULL;
474 struct fou *fou = NULL;
478 /* Open UDP socket */
479 err = udp_sock_create(net, &cfg->udp_config, &sock);
483 /* Allocate FOU port structure */
484 fou = kzalloc(sizeof(*fou), GFP_KERNEL);
492 fou->flags = cfg->flags;
493 fou->port = cfg->udp_config.local_udp_port;
495 /* Initial for fou type */
497 case FOU_ENCAP_DIRECT:
498 err = fou_encap_init(sk, fou, cfg);
503 err = gue_encap_init(sk, fou, cfg);
512 fou->type = cfg->type;
514 udp_sk(sk)->encap_type = 1;
517 sk->sk_user_data = fou;
520 inet_inc_convert_csum(sk);
522 sk->sk_allocation = GFP_ATOMIC;
524 err = fou_add_to_port_list(net, fou);
536 udp_tunnel_sock_release(sock);
541 static int fou_destroy(struct net *net, struct fou_cfg *cfg)
543 struct fou_net *fn = net_generic(net, fou_net_id);
544 __be16 port = cfg->udp_config.local_udp_port;
548 mutex_lock(&fn->fou_lock);
549 list_for_each_entry(fou, &fn->fou_list, list) {
550 if (fou->port == port) {
556 mutex_unlock(&fn->fou_lock);
561 static struct genl_family fou_nl_family = {
562 .id = GENL_ID_GENERATE,
564 .name = FOU_GENL_NAME,
565 .version = FOU_GENL_VERSION,
566 .maxattr = FOU_ATTR_MAX,
570 static struct nla_policy fou_nl_policy[FOU_ATTR_MAX + 1] = {
571 [FOU_ATTR_PORT] = { .type = NLA_U16, },
572 [FOU_ATTR_AF] = { .type = NLA_U8, },
573 [FOU_ATTR_IPPROTO] = { .type = NLA_U8, },
574 [FOU_ATTR_TYPE] = { .type = NLA_U8, },
575 [FOU_ATTR_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG, },
578 static int parse_nl_config(struct genl_info *info,
581 memset(cfg, 0, sizeof(*cfg));
583 cfg->udp_config.family = AF_INET;
585 if (info->attrs[FOU_ATTR_AF]) {
586 u8 family = nla_get_u8(info->attrs[FOU_ATTR_AF]);
588 if (family != AF_INET)
591 cfg->udp_config.family = family;
594 if (info->attrs[FOU_ATTR_PORT]) {
595 __be16 port = nla_get_be16(info->attrs[FOU_ATTR_PORT]);
597 cfg->udp_config.local_udp_port = port;
600 if (info->attrs[FOU_ATTR_IPPROTO])
601 cfg->protocol = nla_get_u8(info->attrs[FOU_ATTR_IPPROTO]);
603 if (info->attrs[FOU_ATTR_TYPE])
604 cfg->type = nla_get_u8(info->attrs[FOU_ATTR_TYPE]);
606 if (info->attrs[FOU_ATTR_REMCSUM_NOPARTIAL])
607 cfg->flags |= FOU_F_REMCSUM_NOPARTIAL;
612 static int fou_nl_cmd_add_port(struct sk_buff *skb, struct genl_info *info)
614 struct net *net = genl_info_net(info);
618 err = parse_nl_config(info, &cfg);
622 return fou_create(net, &cfg, NULL);
625 static int fou_nl_cmd_rm_port(struct sk_buff *skb, struct genl_info *info)
627 struct net *net = genl_info_net(info);
631 err = parse_nl_config(info, &cfg);
635 return fou_destroy(net, &cfg);
638 static int fou_fill_info(struct fou *fou, struct sk_buff *msg)
640 if (nla_put_u8(msg, FOU_ATTR_AF, fou->sock->sk->sk_family) ||
641 nla_put_be16(msg, FOU_ATTR_PORT, fou->port) ||
642 nla_put_u8(msg, FOU_ATTR_IPPROTO, fou->protocol) ||
643 nla_put_u8(msg, FOU_ATTR_TYPE, fou->type))
646 if (fou->flags & FOU_F_REMCSUM_NOPARTIAL)
647 if (nla_put_flag(msg, FOU_ATTR_REMCSUM_NOPARTIAL))
652 static int fou_dump_info(struct fou *fou, u32 portid, u32 seq,
653 u32 flags, struct sk_buff *skb, u8 cmd)
657 hdr = genlmsg_put(skb, portid, seq, &fou_nl_family, flags, cmd);
661 if (fou_fill_info(fou, skb) < 0)
662 goto nla_put_failure;
664 genlmsg_end(skb, hdr);
668 genlmsg_cancel(skb, hdr);
672 static int fou_nl_cmd_get_port(struct sk_buff *skb, struct genl_info *info)
674 struct net *net = genl_info_net(info);
675 struct fou_net *fn = net_generic(net, fou_net_id);
682 ret = parse_nl_config(info, &cfg);
685 port = cfg.udp_config.local_udp_port;
689 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
694 mutex_lock(&fn->fou_lock);
695 list_for_each_entry(fout, &fn->fou_list, list) {
696 if (port == fout->port) {
697 ret = fou_dump_info(fout, info->snd_portid,
698 info->snd_seq, 0, msg,
703 mutex_unlock(&fn->fou_lock);
707 return genlmsg_reply(msg, info);
714 static int fou_nl_dump(struct sk_buff *skb, struct netlink_callback *cb)
716 struct net *net = sock_net(skb->sk);
717 struct fou_net *fn = net_generic(net, fou_net_id);
721 mutex_lock(&fn->fou_lock);
722 list_for_each_entry(fout, &fn->fou_list, list) {
723 if (idx++ < cb->args[0])
725 ret = fou_dump_info(fout, NETLINK_CB(cb->skb).portid,
726 cb->nlh->nlmsg_seq, NLM_F_MULTI,
731 mutex_unlock(&fn->fou_lock);
737 static const struct genl_ops fou_nl_ops[] = {
740 .doit = fou_nl_cmd_add_port,
741 .policy = fou_nl_policy,
742 .flags = GENL_ADMIN_PERM,
746 .doit = fou_nl_cmd_rm_port,
747 .policy = fou_nl_policy,
748 .flags = GENL_ADMIN_PERM,
752 .doit = fou_nl_cmd_get_port,
753 .dumpit = fou_nl_dump,
754 .policy = fou_nl_policy,
758 size_t fou_encap_hlen(struct ip_tunnel_encap *e)
760 return sizeof(struct udphdr);
762 EXPORT_SYMBOL(fou_encap_hlen);
764 size_t gue_encap_hlen(struct ip_tunnel_encap *e)
767 bool need_priv = false;
769 len = sizeof(struct udphdr) + sizeof(struct guehdr);
771 if (e->flags & TUNNEL_ENCAP_FLAG_REMCSUM) {
772 len += GUE_PLEN_REMCSUM;
776 len += need_priv ? GUE_LEN_PRIV : 0;
780 EXPORT_SYMBOL(gue_encap_hlen);
782 static void fou_build_udp(struct sk_buff *skb, struct ip_tunnel_encap *e,
783 struct flowi4 *fl4, u8 *protocol, __be16 sport)
787 skb_push(skb, sizeof(struct udphdr));
788 skb_reset_transport_header(skb);
794 uh->len = htons(skb->len);
795 udp_set_csum(!(e->flags & TUNNEL_ENCAP_FLAG_CSUM), skb,
796 fl4->saddr, fl4->daddr, skb->len);
798 *protocol = IPPROTO_UDP;
801 int fou_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
802 u8 *protocol, struct flowi4 *fl4)
804 int type = e->flags & TUNNEL_ENCAP_FLAG_CSUM ? SKB_GSO_UDP_TUNNEL_CSUM :
809 err = iptunnel_handle_offloads(skb, type);
813 sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
815 fou_build_udp(skb, e, fl4, protocol, sport);
819 EXPORT_SYMBOL(fou_build_header);
821 int gue_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
822 u8 *protocol, struct flowi4 *fl4)
824 int type = e->flags & TUNNEL_ENCAP_FLAG_CSUM ? SKB_GSO_UDP_TUNNEL_CSUM :
826 struct guehdr *guehdr;
827 size_t hdrlen, optlen = 0;
830 bool need_priv = false;
833 if ((e->flags & TUNNEL_ENCAP_FLAG_REMCSUM) &&
834 skb->ip_summed == CHECKSUM_PARTIAL) {
835 optlen += GUE_PLEN_REMCSUM;
836 type |= SKB_GSO_TUNNEL_REMCSUM;
840 optlen += need_priv ? GUE_LEN_PRIV : 0;
842 err = iptunnel_handle_offloads(skb, type);
846 /* Get source port (based on flow hash) before skb_push */
847 sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
850 hdrlen = sizeof(struct guehdr) + optlen;
852 skb_push(skb, hdrlen);
854 guehdr = (struct guehdr *)skb->data;
858 guehdr->hlen = optlen >> 2;
860 guehdr->proto_ctype = *protocol;
865 __be32 *flags = data;
867 guehdr->flags |= GUE_FLAG_PRIV;
869 data += GUE_LEN_PRIV;
871 if (type & SKB_GSO_TUNNEL_REMCSUM) {
872 u16 csum_start = skb_checksum_start_offset(skb);
875 if (csum_start < hdrlen)
878 csum_start -= hdrlen;
879 pd[0] = htons(csum_start);
880 pd[1] = htons(csum_start + skb->csum_offset);
882 if (!skb_is_gso(skb)) {
883 skb->ip_summed = CHECKSUM_NONE;
884 skb->encapsulation = 0;
887 *flags |= GUE_PFLAG_REMCSUM;
888 data += GUE_PLEN_REMCSUM;
893 fou_build_udp(skb, e, fl4, protocol, sport);
897 EXPORT_SYMBOL(gue_build_header);
899 #ifdef CONFIG_NET_FOU_IP_TUNNELS
901 static const struct ip_tunnel_encap_ops fou_iptun_ops = {
902 .encap_hlen = fou_encap_hlen,
903 .build_header = fou_build_header,
906 static const struct ip_tunnel_encap_ops gue_iptun_ops = {
907 .encap_hlen = gue_encap_hlen,
908 .build_header = gue_build_header,
911 static int ip_tunnel_encap_add_fou_ops(void)
915 ret = ip_tunnel_encap_add_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
917 pr_err("can't add fou ops\n");
921 ret = ip_tunnel_encap_add_ops(&gue_iptun_ops, TUNNEL_ENCAP_GUE);
923 pr_err("can't add gue ops\n");
924 ip_tunnel_encap_del_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
931 static void ip_tunnel_encap_del_fou_ops(void)
933 ip_tunnel_encap_del_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
934 ip_tunnel_encap_del_ops(&gue_iptun_ops, TUNNEL_ENCAP_GUE);
939 static int ip_tunnel_encap_add_fou_ops(void)
944 static void ip_tunnel_encap_del_fou_ops(void)
950 static __net_init int fou_init_net(struct net *net)
952 struct fou_net *fn = net_generic(net, fou_net_id);
954 INIT_LIST_HEAD(&fn->fou_list);
955 mutex_init(&fn->fou_lock);
959 static __net_exit void fou_exit_net(struct net *net)
961 struct fou_net *fn = net_generic(net, fou_net_id);
962 struct fou *fou, *next;
964 /* Close all the FOU sockets */
965 mutex_lock(&fn->fou_lock);
966 list_for_each_entry_safe(fou, next, &fn->fou_list, list)
968 mutex_unlock(&fn->fou_lock);
971 static struct pernet_operations fou_net_ops = {
972 .init = fou_init_net,
973 .exit = fou_exit_net,
975 .size = sizeof(struct fou_net),
978 static int __init fou_init(void)
982 ret = register_pernet_device(&fou_net_ops);
986 ret = genl_register_family_with_ops(&fou_nl_family,
991 ret = ip_tunnel_encap_add_fou_ops();
995 genl_unregister_family(&fou_nl_family);
997 unregister_pernet_device(&fou_net_ops);
1002 static void __exit fou_fini(void)
1004 ip_tunnel_encap_del_fou_ops();
1005 genl_unregister_family(&fou_nl_family);
1006 unregister_pernet_device(&fou_net_ops);
1009 module_init(fou_init);
1010 module_exit(fou_fini);
1011 MODULE_AUTHOR("Tom Herbert <therbert@google.com>");
1012 MODULE_LICENSE("GPL");