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>
19 static DEFINE_SPINLOCK(fou_lock);
20 static LIST_HEAD(fou_list);
26 struct udp_offload udp_offloads;
27 struct list_head list;
33 struct udp_port_cfg udp_config;
36 static inline struct fou *fou_from_sock(struct sock *sk)
38 return sk->sk_user_data;
41 static void fou_recv_pull(struct sk_buff *skb, size_t len)
43 struct iphdr *iph = ip_hdr(skb);
45 /* Remove 'len' bytes from the packet (UDP header and
46 * FOU header if present).
48 iph->tot_len = htons(ntohs(iph->tot_len) - len);
50 skb_postpull_rcsum(skb, udp_hdr(skb), len);
51 skb_reset_transport_header(skb);
54 static int fou_udp_recv(struct sock *sk, struct sk_buff *skb)
56 struct fou *fou = fou_from_sock(sk);
61 fou_recv_pull(skb, sizeof(struct udphdr));
63 return -fou->protocol;
66 static int gue_control_message(struct sk_buff *skb, struct guehdr *guehdr)
73 static int gue_udp_recv(struct sock *sk, struct sk_buff *skb)
75 struct fou *fou = fou_from_sock(sk);
76 size_t len, optlen, hdrlen;
77 struct guehdr *guehdr;
83 len = sizeof(struct udphdr) + sizeof(struct guehdr);
84 if (!pskb_may_pull(skb, len))
87 guehdr = (struct guehdr *)&udp_hdr(skb)[1];
89 optlen = guehdr->hlen << 2;
92 if (!pskb_may_pull(skb, len))
95 /* guehdr may change after pull */
96 guehdr = (struct guehdr *)&udp_hdr(skb)[1];
98 hdrlen = sizeof(struct guehdr) + optlen;
100 if (guehdr->version != 0 || validate_gue_flags(guehdr, optlen))
103 /* Pull UDP and GUE headers */
104 fou_recv_pull(skb, len);
108 if (guehdr->flags & GUE_FLAG_PRIV) {
109 data += GUE_LEN_PRIV;
111 /* Process private flags */
114 if (unlikely(guehdr->control))
115 return gue_control_message(skb, guehdr);
117 return -guehdr->proto_ctype;
124 static struct sk_buff **fou_gro_receive(struct sk_buff **head,
127 const struct net_offload *ops;
128 struct sk_buff **pp = NULL;
129 u8 proto = NAPI_GRO_CB(skb)->proto;
130 const struct net_offload **offloads;
133 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
134 ops = rcu_dereference(offloads[proto]);
135 if (!ops || !ops->callbacks.gro_receive)
138 pp = ops->callbacks.gro_receive(head, skb);
146 static int fou_gro_complete(struct sk_buff *skb, int nhoff)
148 const struct net_offload *ops;
149 u8 proto = NAPI_GRO_CB(skb)->proto;
151 const struct net_offload **offloads;
154 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
155 ops = rcu_dereference(offloads[proto]);
156 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
159 err = ops->callbacks.gro_complete(skb, nhoff);
167 static struct sk_buff **gue_gro_receive(struct sk_buff **head,
170 const struct net_offload **offloads;
171 const struct net_offload *ops;
172 struct sk_buff **pp = NULL;
174 struct guehdr *guehdr;
175 size_t len, optlen, hdrlen, off;
179 off = skb_gro_offset(skb);
180 len = off + sizeof(*guehdr);
182 guehdr = skb_gro_header_fast(skb, off);
183 if (skb_gro_header_hard(skb, len)) {
184 guehdr = skb_gro_header_slow(skb, len, off);
185 if (unlikely(!guehdr))
189 optlen = guehdr->hlen << 2;
192 if (skb_gro_header_hard(skb, len)) {
193 guehdr = skb_gro_header_slow(skb, len, off);
194 if (unlikely(!guehdr))
198 if (unlikely(guehdr->control) || guehdr->version != 0 ||
199 validate_gue_flags(guehdr, optlen))
202 hdrlen = sizeof(*guehdr) + optlen;
204 skb_gro_pull(skb, hdrlen);
206 /* Adjusted NAPI_GRO_CB(skb)->csum after skb_gro_pull()*/
207 skb_gro_postpull_rcsum(skb, guehdr, hdrlen);
211 if (guehdr->flags & GUE_FLAG_PRIV) {
212 data += GUE_LEN_PRIV;
214 /* Process private flags */
219 for (p = *head; p; p = p->next) {
220 const struct guehdr *guehdr2;
222 if (!NAPI_GRO_CB(p)->same_flow)
225 guehdr2 = (struct guehdr *)(p->data + off);
227 /* Compare base GUE header to be equal (covers
228 * hlen, version, proto_ctype, and flags.
230 if (guehdr->word != guehdr2->word) {
231 NAPI_GRO_CB(p)->same_flow = 0;
235 /* Compare optional fields are the same. */
236 if (guehdr->hlen && memcmp(&guehdr[1], &guehdr2[1],
237 guehdr->hlen << 2)) {
238 NAPI_GRO_CB(p)->same_flow = 0;
244 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
245 ops = rcu_dereference(offloads[guehdr->proto_ctype]);
246 if (WARN_ON(!ops || !ops->callbacks.gro_receive))
249 pp = ops->callbacks.gro_receive(head, skb);
254 NAPI_GRO_CB(skb)->flush |= flush;
259 static int gue_gro_complete(struct sk_buff *skb, int nhoff)
261 const struct net_offload **offloads;
262 struct guehdr *guehdr = (struct guehdr *)(skb->data + nhoff);
263 const struct net_offload *ops;
264 unsigned int guehlen;
268 proto = guehdr->proto_ctype;
270 guehlen = sizeof(*guehdr) + (guehdr->hlen << 2);
273 offloads = NAPI_GRO_CB(skb)->is_ipv6 ? inet6_offloads : inet_offloads;
274 ops = rcu_dereference(offloads[proto]);
275 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
278 err = ops->callbacks.gro_complete(skb, nhoff + guehlen);
285 static int fou_add_to_port_list(struct fou *fou)
289 spin_lock(&fou_lock);
290 list_for_each_entry(fout, &fou_list, list) {
291 if (fou->port == fout->port) {
292 spin_unlock(&fou_lock);
297 list_add(&fou->list, &fou_list);
298 spin_unlock(&fou_lock);
303 static void fou_release(struct fou *fou)
305 struct socket *sock = fou->sock;
306 struct sock *sk = sock->sk;
308 udp_del_offload(&fou->udp_offloads);
310 list_del(&fou->list);
312 /* Remove hooks into tunnel socket */
313 sk->sk_user_data = NULL;
320 static int fou_encap_init(struct sock *sk, struct fou *fou, struct fou_cfg *cfg)
322 udp_sk(sk)->encap_rcv = fou_udp_recv;
323 fou->protocol = cfg->protocol;
324 fou->udp_offloads.callbacks.gro_receive = fou_gro_receive;
325 fou->udp_offloads.callbacks.gro_complete = fou_gro_complete;
326 fou->udp_offloads.port = cfg->udp_config.local_udp_port;
327 fou->udp_offloads.ipproto = cfg->protocol;
332 static int gue_encap_init(struct sock *sk, struct fou *fou, struct fou_cfg *cfg)
334 udp_sk(sk)->encap_rcv = gue_udp_recv;
335 fou->udp_offloads.callbacks.gro_receive = gue_gro_receive;
336 fou->udp_offloads.callbacks.gro_complete = gue_gro_complete;
337 fou->udp_offloads.port = cfg->udp_config.local_udp_port;
342 static int fou_create(struct net *net, struct fou_cfg *cfg,
343 struct socket **sockp)
345 struct fou *fou = NULL;
347 struct socket *sock = NULL;
350 /* Open UDP socket */
351 err = udp_sock_create(net, &cfg->udp_config, &sock);
355 /* Allocate FOU port structure */
356 fou = kzalloc(sizeof(*fou), GFP_KERNEL);
364 fou->port = cfg->udp_config.local_udp_port;
366 /* Initial for fou type */
368 case FOU_ENCAP_DIRECT:
369 err = fou_encap_init(sk, fou, cfg);
374 err = gue_encap_init(sk, fou, cfg);
383 udp_sk(sk)->encap_type = 1;
386 sk->sk_user_data = fou;
389 udp_set_convert_csum(sk, true);
391 sk->sk_allocation = GFP_ATOMIC;
393 if (cfg->udp_config.family == AF_INET) {
394 err = udp_add_offload(&fou->udp_offloads);
399 err = fou_add_to_port_list(fou);
416 static int fou_destroy(struct net *net, struct fou_cfg *cfg)
419 u16 port = cfg->udp_config.local_udp_port;
422 spin_lock(&fou_lock);
423 list_for_each_entry(fou, &fou_list, list) {
424 if (fou->port == port) {
425 udp_del_offload(&fou->udp_offloads);
431 spin_unlock(&fou_lock);
436 static struct genl_family fou_nl_family = {
437 .id = GENL_ID_GENERATE,
439 .name = FOU_GENL_NAME,
440 .version = FOU_GENL_VERSION,
441 .maxattr = FOU_ATTR_MAX,
445 static struct nla_policy fou_nl_policy[FOU_ATTR_MAX + 1] = {
446 [FOU_ATTR_PORT] = { .type = NLA_U16, },
447 [FOU_ATTR_AF] = { .type = NLA_U8, },
448 [FOU_ATTR_IPPROTO] = { .type = NLA_U8, },
449 [FOU_ATTR_TYPE] = { .type = NLA_U8, },
452 static int parse_nl_config(struct genl_info *info,
455 memset(cfg, 0, sizeof(*cfg));
457 cfg->udp_config.family = AF_INET;
459 if (info->attrs[FOU_ATTR_AF]) {
460 u8 family = nla_get_u8(info->attrs[FOU_ATTR_AF]);
462 if (family != AF_INET && family != AF_INET6)
465 cfg->udp_config.family = family;
468 if (info->attrs[FOU_ATTR_PORT]) {
469 u16 port = nla_get_u16(info->attrs[FOU_ATTR_PORT]);
471 cfg->udp_config.local_udp_port = port;
474 if (info->attrs[FOU_ATTR_IPPROTO])
475 cfg->protocol = nla_get_u8(info->attrs[FOU_ATTR_IPPROTO]);
477 if (info->attrs[FOU_ATTR_TYPE])
478 cfg->type = nla_get_u8(info->attrs[FOU_ATTR_TYPE]);
483 static int fou_nl_cmd_add_port(struct sk_buff *skb, struct genl_info *info)
488 err = parse_nl_config(info, &cfg);
492 return fou_create(&init_net, &cfg, NULL);
495 static int fou_nl_cmd_rm_port(struct sk_buff *skb, struct genl_info *info)
499 parse_nl_config(info, &cfg);
501 return fou_destroy(&init_net, &cfg);
504 static const struct genl_ops fou_nl_ops[] = {
507 .doit = fou_nl_cmd_add_port,
508 .policy = fou_nl_policy,
509 .flags = GENL_ADMIN_PERM,
513 .doit = fou_nl_cmd_rm_port,
514 .policy = fou_nl_policy,
515 .flags = GENL_ADMIN_PERM,
519 static void fou_build_udp(struct sk_buff *skb, struct ip_tunnel_encap *e,
520 struct flowi4 *fl4, u8 *protocol, __be16 sport)
524 skb_push(skb, sizeof(struct udphdr));
525 skb_reset_transport_header(skb);
531 uh->len = htons(skb->len);
533 udp_set_csum(!(e->flags & TUNNEL_ENCAP_FLAG_CSUM), skb,
534 fl4->saddr, fl4->daddr, skb->len);
536 *protocol = IPPROTO_UDP;
539 int fou_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
540 u8 *protocol, struct flowi4 *fl4)
542 bool csum = !!(e->flags & TUNNEL_ENCAP_FLAG_CSUM);
543 int type = csum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
546 skb = iptunnel_handle_offloads(skb, csum, type);
551 sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
553 fou_build_udp(skb, e, fl4, protocol, sport);
557 EXPORT_SYMBOL(fou_build_header);
559 int gue_build_header(struct sk_buff *skb, struct ip_tunnel_encap *e,
560 u8 *protocol, struct flowi4 *fl4)
562 bool csum = !!(e->flags & TUNNEL_ENCAP_FLAG_CSUM);
563 int type = csum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
564 struct guehdr *guehdr;
565 size_t hdrlen, optlen = 0;
568 bool need_priv = false;
570 if ((e->flags & TUNNEL_ENCAP_FLAG_REMCSUM) &&
571 skb->ip_summed == CHECKSUM_PARTIAL) {
573 optlen += GUE_PLEN_REMCSUM;
574 type |= SKB_GSO_TUNNEL_REMCSUM;
578 optlen += need_priv ? GUE_LEN_PRIV : 0;
580 skb = iptunnel_handle_offloads(skb, csum, type);
585 /* Get source port (based on flow hash) before skb_push */
586 sport = e->sport ? : udp_flow_src_port(dev_net(skb->dev),
589 hdrlen = sizeof(struct guehdr) + optlen;
591 skb_push(skb, hdrlen);
593 guehdr = (struct guehdr *)skb->data;
597 guehdr->hlen = optlen >> 2;
599 guehdr->proto_ctype = *protocol;
604 __be32 *flags = data;
606 guehdr->flags |= GUE_FLAG_PRIV;
608 data += GUE_LEN_PRIV;
610 if (type & SKB_GSO_TUNNEL_REMCSUM) {
611 u16 csum_start = skb_checksum_start_offset(skb);
614 if (csum_start < hdrlen)
617 csum_start -= hdrlen;
618 pd[0] = htons(csum_start);
619 pd[1] = htons(csum_start + skb->csum_offset);
621 if (!skb_is_gso(skb)) {
622 skb->ip_summed = CHECKSUM_NONE;
623 skb->encapsulation = 0;
626 *flags |= GUE_PFLAG_REMCSUM;
627 data += GUE_PLEN_REMCSUM;
632 fou_build_udp(skb, e, fl4, protocol, sport);
636 EXPORT_SYMBOL(gue_build_header);
638 static int __init fou_init(void)
642 ret = genl_register_family_with_ops(&fou_nl_family,
648 static void __exit fou_fini(void)
650 struct fou *fou, *next;
652 genl_unregister_family(&fou_nl_family);
654 /* Close all the FOU sockets */
656 spin_lock(&fou_lock);
657 list_for_each_entry_safe(fou, next, &fou_list, list)
659 spin_unlock(&fou_lock);
662 module_init(fou_init);
663 module_exit(fou_fini);
664 MODULE_AUTHOR("Tom Herbert <therbert@google.com>");
665 MODULE_LICENSE("GPL");