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 udp_offload udp_offloads;
26 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 void 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);
63 static int fou_udp_recv(struct sock *sk, struct sk_buff *skb)
65 struct fou *fou = fou_from_sock(sk);
70 fou_recv_pull(skb, sizeof(struct udphdr));
72 return -fou->protocol;
75 static struct guehdr *gue_remcsum(struct sk_buff *skb, struct guehdr *guehdr,
76 void *data, size_t hdrlen, u8 ipproto,
80 size_t start = ntohs(pd[0]);
81 size_t offset = ntohs(pd[1]);
82 size_t plen = sizeof(struct udphdr) + hdrlen +
83 max_t(size_t, offset + sizeof(u16), start);
85 if (skb->remcsum_offload)
88 if (!pskb_may_pull(skb, plen))
90 guehdr = (struct guehdr *)&udp_hdr(skb)[1];
92 skb_remcsum_process(skb, (void *)guehdr + hdrlen,
93 start, offset, nopartial);
98 static int gue_control_message(struct sk_buff *skb, struct guehdr *guehdr)
105 static int gue_udp_recv(struct sock *sk, struct sk_buff *skb)
107 struct fou *fou = fou_from_sock(sk);
108 size_t len, optlen, hdrlen;
109 struct guehdr *guehdr;
116 len = sizeof(struct udphdr) + sizeof(struct guehdr);
117 if (!pskb_may_pull(skb, len))
120 guehdr = (struct guehdr *)&udp_hdr(skb)[1];
122 optlen = guehdr->hlen << 2;
125 if (!pskb_may_pull(skb, len))
128 /* guehdr may change after pull */
129 guehdr = (struct guehdr *)&udp_hdr(skb)[1];
131 hdrlen = sizeof(struct guehdr) + optlen;
133 if (guehdr->version != 0 || validate_gue_flags(guehdr, optlen))
136 hdrlen = sizeof(struct guehdr) + optlen;
138 ip_hdr(skb)->tot_len = htons(ntohs(ip_hdr(skb)->tot_len) - len);
140 /* Pull csum through the guehdr now . This can be used if
141 * there is a remote checksum offload.
143 skb_postpull_rcsum(skb, udp_hdr(skb), len);
147 if (guehdr->flags & GUE_FLAG_PRIV) {
148 __be32 flags = *(__be32 *)(data + doffset);
150 doffset += GUE_LEN_PRIV;
152 if (flags & GUE_PFLAG_REMCSUM) {
153 guehdr = gue_remcsum(skb, guehdr, data + doffset,
154 hdrlen, guehdr->proto_ctype,
156 FOU_F_REMCSUM_NOPARTIAL));
162 doffset += GUE_PLEN_REMCSUM;
166 if (unlikely(guehdr->control))
167 return gue_control_message(skb, guehdr);
169 __skb_pull(skb, sizeof(struct udphdr) + hdrlen);
170 skb_reset_transport_header(skb);
172 return -guehdr->proto_ctype;
179 static struct sk_buff **fou_gro_receive(struct sk_buff **head,
181 struct udp_offload *uoff)
183 const struct net_offload *ops;
184 struct sk_buff **pp = NULL;
185 u8 proto = NAPI_GRO_CB(skb)->proto;
186 const struct net_offload **offloads;
189 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
190 ops = rcu_dereference(offloads[proto]);
191 if (!ops || !ops->callbacks.gro_receive)
194 pp = ops->callbacks.gro_receive(head, skb);
202 static int fou_gro_complete(struct sk_buff *skb, int nhoff,
203 struct udp_offload *uoff)
205 const struct net_offload *ops;
206 u8 proto = NAPI_GRO_CB(skb)->proto;
208 const struct net_offload **offloads;
210 udp_tunnel_gro_complete(skb, nhoff);
213 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
214 ops = rcu_dereference(offloads[proto]);
215 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
218 err = ops->callbacks.gro_complete(skb, nhoff);
226 static struct guehdr *gue_gro_remcsum(struct sk_buff *skb, unsigned int off,
227 struct guehdr *guehdr, void *data,
228 size_t hdrlen, struct gro_remcsum *grc,
232 size_t start = ntohs(pd[0]);
233 size_t offset = ntohs(pd[1]);
235 if (skb->remcsum_offload)
238 if (!NAPI_GRO_CB(skb)->csum_valid)
241 guehdr = skb_gro_remcsum_process(skb, (void *)guehdr, off, hdrlen,
242 start, offset, grc, nopartial);
244 skb->remcsum_offload = 1;
249 static struct sk_buff **gue_gro_receive(struct sk_buff **head,
251 struct udp_offload *uoff)
253 const struct net_offload **offloads;
254 const struct net_offload *ops;
255 struct sk_buff **pp = NULL;
257 struct guehdr *guehdr;
258 size_t len, optlen, hdrlen, off;
262 struct fou *fou = container_of(uoff, struct fou, udp_offloads);
263 struct gro_remcsum grc;
265 skb_gro_remcsum_init(&grc);
267 off = skb_gro_offset(skb);
268 len = off + sizeof(*guehdr);
270 guehdr = skb_gro_header_fast(skb, off);
271 if (skb_gro_header_hard(skb, len)) {
272 guehdr = skb_gro_header_slow(skb, len, off);
273 if (unlikely(!guehdr))
277 optlen = guehdr->hlen << 2;
280 if (skb_gro_header_hard(skb, len)) {
281 guehdr = skb_gro_header_slow(skb, len, off);
282 if (unlikely(!guehdr))
286 if (unlikely(guehdr->control) || guehdr->version != 0 ||
287 validate_gue_flags(guehdr, optlen))
290 hdrlen = sizeof(*guehdr) + optlen;
292 /* Adjust NAPI_GRO_CB(skb)->csum to account for guehdr,
293 * this is needed if there is a remote checkcsum offload.
295 skb_gro_postpull_rcsum(skb, guehdr, hdrlen);
299 if (guehdr->flags & GUE_FLAG_PRIV) {
300 __be32 flags = *(__be32 *)(data + doffset);
302 doffset += GUE_LEN_PRIV;
304 if (flags & GUE_PFLAG_REMCSUM) {
305 guehdr = gue_gro_remcsum(skb, off, guehdr,
306 data + doffset, hdrlen, &grc,
308 FOU_F_REMCSUM_NOPARTIAL));
315 doffset += GUE_PLEN_REMCSUM;
319 skb_gro_pull(skb, hdrlen);
323 for (p = *head; p; p = p->next) {
324 const struct guehdr *guehdr2;
326 if (!NAPI_GRO_CB(p)->same_flow)
329 guehdr2 = (struct guehdr *)(p->data + off);
331 /* Compare base GUE header to be equal (covers
332 * hlen, version, proto_ctype, and flags.
334 if (guehdr->word != guehdr2->word) {
335 NAPI_GRO_CB(p)->same_flow = 0;
339 /* Compare optional fields are the same. */
340 if (guehdr->hlen && memcmp(&guehdr[1], &guehdr2[1],
341 guehdr->hlen << 2)) {
342 NAPI_GRO_CB(p)->same_flow = 0;
348 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
349 ops = rcu_dereference(offloads[guehdr->proto_ctype]);
350 if (WARN_ON_ONCE(!ops || !ops->callbacks.gro_receive))
353 pp = ops->callbacks.gro_receive(head, skb);
358 NAPI_GRO_CB(skb)->flush |= flush;
359 skb_gro_remcsum_cleanup(skb, &grc);
364 static int gue_gro_complete(struct sk_buff *skb, int nhoff,
365 struct udp_offload *uoff)
367 const struct net_offload **offloads;
368 struct guehdr *guehdr = (struct guehdr *)(skb->data + nhoff);
369 const struct net_offload *ops;
370 unsigned int guehlen;
374 proto = guehdr->proto_ctype;
376 guehlen = sizeof(*guehdr) + (guehdr->hlen << 2);
379 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
380 ops = rcu_dereference(offloads[proto]);
381 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
384 err = ops->callbacks.gro_complete(skb, nhoff + guehlen);
391 static int fou_add_to_port_list(struct net *net, struct fou *fou)
393 struct fou_net *fn = net_generic(net, fou_net_id);
396 mutex_lock(&fn->fou_lock);
397 list_for_each_entry(fout, &fn->fou_list, list) {
398 if (fou->port == fout->port) {
399 mutex_unlock(&fn->fou_lock);
404 list_add(&fou->list, &fn->fou_list);
405 mutex_unlock(&fn->fou_lock);
410 static void fou_release(struct fou *fou)
412 struct socket *sock = fou->sock;
413 struct sock *sk = sock->sk;
415 if (sk->sk_family == AF_INET)
416 udp_del_offload(&fou->udp_offloads);
417 list_del(&fou->list);
418 udp_tunnel_sock_release(sock);
423 static int fou_encap_init(struct sock *sk, struct fou *fou, struct fou_cfg *cfg)
425 udp_sk(sk)->encap_rcv = fou_udp_recv;
426 fou->protocol = cfg->protocol;
427 fou->udp_offloads.callbacks.gro_receive = fou_gro_receive;
428 fou->udp_offloads.callbacks.gro_complete = fou_gro_complete;
429 fou->udp_offloads.port = cfg->udp_config.local_udp_port;
430 fou->udp_offloads.ipproto = cfg->protocol;
435 static int gue_encap_init(struct sock *sk, struct fou *fou, struct fou_cfg *cfg)
437 udp_sk(sk)->encap_rcv = gue_udp_recv;
438 fou->udp_offloads.callbacks.gro_receive = gue_gro_receive;
439 fou->udp_offloads.callbacks.gro_complete = gue_gro_complete;
440 fou->udp_offloads.port = cfg->udp_config.local_udp_port;
445 static int fou_create(struct net *net, struct fou_cfg *cfg,
446 struct socket **sockp)
448 struct socket *sock = NULL;
449 struct fou *fou = NULL;
453 /* Open UDP socket */
454 err = udp_sock_create(net, &cfg->udp_config, &sock);
458 /* Allocate FOU port structure */
459 fou = kzalloc(sizeof(*fou), GFP_KERNEL);
467 fou->flags = cfg->flags;
468 fou->port = cfg->udp_config.local_udp_port;
470 /* Initial for fou type */
472 case FOU_ENCAP_DIRECT:
473 err = fou_encap_init(sk, fou, cfg);
478 err = gue_encap_init(sk, fou, cfg);
487 fou->type = cfg->type;
489 udp_sk(sk)->encap_type = 1;
492 sk->sk_user_data = fou;
495 inet_inc_convert_csum(sk);
497 sk->sk_allocation = GFP_ATOMIC;
499 if (cfg->udp_config.family == AF_INET) {
500 err = udp_add_offload(&fou->udp_offloads);
505 err = fou_add_to_port_list(net, fou);
517 udp_tunnel_sock_release(sock);
522 static int fou_destroy(struct net *net, struct fou_cfg *cfg)
524 struct fou_net *fn = net_generic(net, fou_net_id);
525 __be16 port = cfg->udp_config.local_udp_port;
529 mutex_lock(&fn->fou_lock);
530 list_for_each_entry(fou, &fn->fou_list, list) {
531 if (fou->port == port) {
537 mutex_unlock(&fn->fou_lock);
542 static struct genl_family fou_nl_family = {
543 .id = GENL_ID_GENERATE,
545 .name = FOU_GENL_NAME,
546 .version = FOU_GENL_VERSION,
547 .maxattr = FOU_ATTR_MAX,
551 static struct nla_policy fou_nl_policy[FOU_ATTR_MAX + 1] = {
552 [FOU_ATTR_PORT] = { .type = NLA_U16, },
553 [FOU_ATTR_AF] = { .type = NLA_U8, },
554 [FOU_ATTR_IPPROTO] = { .type = NLA_U8, },
555 [FOU_ATTR_TYPE] = { .type = NLA_U8, },
556 [FOU_ATTR_REMCSUM_NOPARTIAL] = { .type = NLA_FLAG, },
559 static int parse_nl_config(struct genl_info *info,
562 memset(cfg, 0, sizeof(*cfg));
564 cfg->udp_config.family = AF_INET;
566 if (info->attrs[FOU_ATTR_AF]) {
567 u8 family = nla_get_u8(info->attrs[FOU_ATTR_AF]);
569 if (family != AF_INET)
572 cfg->udp_config.family = family;
575 if (info->attrs[FOU_ATTR_PORT]) {
576 __be16 port = nla_get_be16(info->attrs[FOU_ATTR_PORT]);
578 cfg->udp_config.local_udp_port = port;
581 if (info->attrs[FOU_ATTR_IPPROTO])
582 cfg->protocol = nla_get_u8(info->attrs[FOU_ATTR_IPPROTO]);
584 if (info->attrs[FOU_ATTR_TYPE])
585 cfg->type = nla_get_u8(info->attrs[FOU_ATTR_TYPE]);
587 if (info->attrs[FOU_ATTR_REMCSUM_NOPARTIAL])
588 cfg->flags |= FOU_F_REMCSUM_NOPARTIAL;
593 static int fou_nl_cmd_add_port(struct sk_buff *skb, struct genl_info *info)
595 struct net *net = genl_info_net(info);
599 err = parse_nl_config(info, &cfg);
603 return fou_create(net, &cfg, NULL);
606 static int fou_nl_cmd_rm_port(struct sk_buff *skb, struct genl_info *info)
608 struct net *net = genl_info_net(info);
612 err = parse_nl_config(info, &cfg);
616 return fou_destroy(net, &cfg);
619 static int fou_fill_info(struct fou *fou, struct sk_buff *msg)
621 if (nla_put_u8(msg, FOU_ATTR_AF, fou->sock->sk->sk_family) ||
622 nla_put_be16(msg, FOU_ATTR_PORT, fou->port) ||
623 nla_put_u8(msg, FOU_ATTR_IPPROTO, fou->protocol) ||
624 nla_put_u8(msg, FOU_ATTR_TYPE, fou->type))
627 if (fou->flags & FOU_F_REMCSUM_NOPARTIAL)
628 if (nla_put_flag(msg, FOU_ATTR_REMCSUM_NOPARTIAL))
633 static int fou_dump_info(struct fou *fou, u32 portid, u32 seq,
634 u32 flags, struct sk_buff *skb, u8 cmd)
638 hdr = genlmsg_put(skb, portid, seq, &fou_nl_family, flags, cmd);
642 if (fou_fill_info(fou, skb) < 0)
643 goto nla_put_failure;
645 genlmsg_end(skb, hdr);
649 genlmsg_cancel(skb, hdr);
653 static int fou_nl_cmd_get_port(struct sk_buff *skb, struct genl_info *info)
655 struct net *net = genl_info_net(info);
656 struct fou_net *fn = net_generic(net, fou_net_id);
663 ret = parse_nl_config(info, &cfg);
666 port = cfg.udp_config.local_udp_port;
670 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
675 mutex_lock(&fn->fou_lock);
676 list_for_each_entry(fout, &fn->fou_list, list) {
677 if (port == fout->port) {
678 ret = fou_dump_info(fout, info->snd_portid,
679 info->snd_seq, 0, msg,
684 mutex_unlock(&fn->fou_lock);
688 return genlmsg_reply(msg, info);
695 static int fou_nl_dump(struct sk_buff *skb, struct netlink_callback *cb)
697 struct net *net = sock_net(skb->sk);
698 struct fou_net *fn = net_generic(net, fou_net_id);
702 mutex_lock(&fn->fou_lock);
703 list_for_each_entry(fout, &fn->fou_list, list) {
704 if (idx++ < cb->args[0])
706 ret = fou_dump_info(fout, NETLINK_CB(cb->skb).portid,
707 cb->nlh->nlmsg_seq, NLM_F_MULTI,
712 mutex_unlock(&fn->fou_lock);
718 static const struct genl_ops fou_nl_ops[] = {
721 .doit = fou_nl_cmd_add_port,
722 .policy = fou_nl_policy,
723 .flags = GENL_ADMIN_PERM,
727 .doit = fou_nl_cmd_rm_port,
728 .policy = fou_nl_policy,
729 .flags = GENL_ADMIN_PERM,
733 .doit = fou_nl_cmd_get_port,
734 .dumpit = fou_nl_dump,
735 .policy = fou_nl_policy,
739 size_t fou_encap_hlen(struct ip_tunnel_encap *e)
741 return sizeof(struct udphdr);
743 EXPORT_SYMBOL(fou_encap_hlen);
745 size_t gue_encap_hlen(struct ip_tunnel_encap *e)
748 bool need_priv = false;
750 len = sizeof(struct udphdr) + sizeof(struct guehdr);
752 if (e->flags & TUNNEL_ENCAP_FLAG_REMCSUM) {
753 len += GUE_PLEN_REMCSUM;
757 len += need_priv ? GUE_LEN_PRIV : 0;
761 EXPORT_SYMBOL(gue_encap_hlen);
763 static void fou_build_udp(struct sk_buff *skb, struct ip_tunnel_encap *e,
764 struct flowi4 *fl4, u8 *protocol, __be16 sport)
768 skb_push(skb, sizeof(struct udphdr));
769 skb_reset_transport_header(skb);
775 uh->len = htons(skb->len);
777 udp_set_csum(!(e->flags & TUNNEL_ENCAP_FLAG_CSUM), skb,
778 fl4->saddr, fl4->daddr, skb->len);
780 *protocol = IPPROTO_UDP;
783 int fou_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
784 u8 *protocol, struct flowi4 *fl4)
786 bool csum = !!(e->flags & TUNNEL_ENCAP_FLAG_CSUM);
787 int type = csum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
790 skb = iptunnel_handle_offloads(skb, csum, type);
795 sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
797 fou_build_udp(skb, e, fl4, protocol, sport);
801 EXPORT_SYMBOL(fou_build_header);
803 int gue_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
804 u8 *protocol, struct flowi4 *fl4)
806 bool csum = !!(e->flags & TUNNEL_ENCAP_FLAG_CSUM);
807 int type = csum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
808 struct guehdr *guehdr;
809 size_t hdrlen, optlen = 0;
812 bool need_priv = false;
814 if ((e->flags & TUNNEL_ENCAP_FLAG_REMCSUM) &&
815 skb->ip_summed == CHECKSUM_PARTIAL) {
817 optlen += GUE_PLEN_REMCSUM;
818 type |= SKB_GSO_TUNNEL_REMCSUM;
822 optlen += need_priv ? GUE_LEN_PRIV : 0;
824 skb = iptunnel_handle_offloads(skb, csum, type);
829 /* Get source port (based on flow hash) before skb_push */
830 sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
833 hdrlen = sizeof(struct guehdr) + optlen;
835 skb_push(skb, hdrlen);
837 guehdr = (struct guehdr *)skb->data;
841 guehdr->hlen = optlen >> 2;
843 guehdr->proto_ctype = *protocol;
848 __be32 *flags = data;
850 guehdr->flags |= GUE_FLAG_PRIV;
852 data += GUE_LEN_PRIV;
854 if (type & SKB_GSO_TUNNEL_REMCSUM) {
855 u16 csum_start = skb_checksum_start_offset(skb);
858 if (csum_start < hdrlen)
861 csum_start -= hdrlen;
862 pd[0] = htons(csum_start);
863 pd[1] = htons(csum_start + skb->csum_offset);
865 if (!skb_is_gso(skb)) {
866 skb->ip_summed = CHECKSUM_NONE;
867 skb->encapsulation = 0;
870 *flags |= GUE_PFLAG_REMCSUM;
871 data += GUE_PLEN_REMCSUM;
876 fou_build_udp(skb, e, fl4, protocol, sport);
880 EXPORT_SYMBOL(gue_build_header);
882 #ifdef CONFIG_NET_FOU_IP_TUNNELS
884 static const struct ip_tunnel_encap_ops fou_iptun_ops = {
885 .encap_hlen = fou_encap_hlen,
886 .build_header = fou_build_header,
889 static const struct ip_tunnel_encap_ops gue_iptun_ops = {
890 .encap_hlen = gue_encap_hlen,
891 .build_header = gue_build_header,
894 static int ip_tunnel_encap_add_fou_ops(void)
898 ret = ip_tunnel_encap_add_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
900 pr_err("can't add fou ops\n");
904 ret = ip_tunnel_encap_add_ops(&gue_iptun_ops, TUNNEL_ENCAP_GUE);
906 pr_err("can't add gue ops\n");
907 ip_tunnel_encap_del_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
914 static void ip_tunnel_encap_del_fou_ops(void)
916 ip_tunnel_encap_del_ops(&fou_iptun_ops, TUNNEL_ENCAP_FOU);
917 ip_tunnel_encap_del_ops(&gue_iptun_ops, TUNNEL_ENCAP_GUE);
922 static int ip_tunnel_encap_add_fou_ops(void)
927 static void ip_tunnel_encap_del_fou_ops(void)
933 static __net_init int fou_init_net(struct net *net)
935 struct fou_net *fn = net_generic(net, fou_net_id);
937 INIT_LIST_HEAD(&fn->fou_list);
938 mutex_init(&fn->fou_lock);
942 static __net_exit void fou_exit_net(struct net *net)
944 struct fou_net *fn = net_generic(net, fou_net_id);
945 struct fou *fou, *next;
947 /* Close all the FOU sockets */
948 mutex_lock(&fn->fou_lock);
949 list_for_each_entry_safe(fou, next, &fn->fou_list, list)
951 mutex_unlock(&fn->fou_lock);
954 static struct pernet_operations fou_net_ops = {
955 .init = fou_init_net,
956 .exit = fou_exit_net,
958 .size = sizeof(struct fou_net),
961 static int __init fou_init(void)
965 ret = register_pernet_device(&fou_net_ops);
969 ret = genl_register_family_with_ops(&fou_nl_family,
974 ret = ip_tunnel_encap_add_fou_ops();
978 genl_unregister_family(&fou_nl_family);
980 unregister_pernet_device(&fou_net_ops);
985 static void __exit fou_fini(void)
987 ip_tunnel_encap_del_fou_ops();
988 genl_unregister_family(&fou_nl_family);
989 unregister_pernet_device(&fou_net_ops);
992 module_init(fou_init);
993 module_exit(fou_fini);
995 MODULE_LICENSE("GPL");