2 * Copyright (c) 2010, 2011 Nicira Networks.
3 * Distributed under the terms of the GNU GPL version 2.
5 * Significant portions of this file may be copied from parts of the Linux
6 * kernel, by Linus Torvalds and others.
9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
12 #include <linux/skbuff.h>
14 #include <linux/if_tunnel.h>
15 #include <linux/if_vlan.h>
20 #include <net/protocol.h>
24 #include "vport-generic.h"
27 * The GRE header is composed of a series of sections: a base and then a variable
30 #define GRE_HEADER_SECTION 4
37 static int gre_hdr_len(const struct tnl_mutable_config *mutable)
41 len = GRE_HEADER_SECTION;
43 if (mutable->flags & TNL_F_CSUM)
44 len += GRE_HEADER_SECTION;
46 if (mutable->out_key || mutable->flags & TNL_F_OUT_KEY_ACTION)
47 len += GRE_HEADER_SECTION;
52 /* Returns the least-significant 32 bits of a __be64. */
53 static __be32 be64_get_low32(__be64 x)
56 return (__force __be32)x;
58 return (__force __be32)((__force u64)x >> 32);
62 static void gre_build_header(const struct vport *vport,
63 const struct tnl_mutable_config *mutable,
66 struct gre_base_hdr *greh = header;
67 __be32 *options = (__be32 *)(greh + 1);
69 greh->protocol = htons(ETH_P_TEB);
72 if (mutable->flags & TNL_F_CSUM) {
73 greh->flags |= GRE_CSUM;
78 if (mutable->out_key || mutable->flags & TNL_F_OUT_KEY_ACTION)
79 greh->flags |= GRE_KEY;
82 *options = be64_get_low32(mutable->out_key);
85 static struct sk_buff *gre_update_header(const struct vport *vport,
86 const struct tnl_mutable_config *mutable,
87 struct dst_entry *dst,
90 __be32 *options = (__be32 *)(skb_network_header(skb) + mutable->tunnel_hlen
91 - GRE_HEADER_SECTION);
93 /* Work backwards over the options so the checksum is last. */
94 if (mutable->flags & TNL_F_OUT_KEY_ACTION) {
95 *options = be64_get_low32(OVS_CB(skb)->tun_id);
99 if (mutable->flags & TNL_F_CSUM)
100 *(__sum16 *)options = csum_fold(skb_checksum(skb,
101 skb_transport_offset(skb),
102 skb->len - skb_transport_offset(skb),
105 * Allow our local IP stack to fragment the outer packet even if the
106 * DF bit is set as a last resort.
113 /* Zero-extends a __be32 into the least-significant 32 bits of a __be64. */
114 static __be64 be32_extend_to_be64(__be32 x)
117 return (__force __be64)x;
119 return (__force __be64)((__force u64)x << 32);
123 static int parse_header(struct iphdr *iph, __be16 *flags, __be64 *key)
125 /* IP and ICMP protocol handlers check that the IHL is valid. */
126 struct gre_base_hdr *greh = (struct gre_base_hdr *)((u8 *)iph + (iph->ihl << 2));
127 __be32 *options = (__be32 *)(greh + 1);
130 *flags = greh->flags;
132 if (unlikely(greh->flags & (GRE_VERSION | GRE_ROUTING)))
135 if (unlikely(greh->protocol != htons(ETH_P_TEB)))
138 hdr_len = GRE_HEADER_SECTION;
140 if (greh->flags & GRE_CSUM) {
141 hdr_len += GRE_HEADER_SECTION;
145 if (greh->flags & GRE_KEY) {
146 hdr_len += GRE_HEADER_SECTION;
148 *key = be32_extend_to_be64(*options);
153 if (unlikely(greh->flags & GRE_SEQ))
154 hdr_len += GRE_HEADER_SECTION;
159 /* Called with rcu_read_lock and BH disabled. */
160 static void gre_err(struct sk_buff *skb, u32 info)
163 const struct tnl_mutable_config *mutable;
164 const int type = icmp_hdr(skb)->type;
165 const int code = icmp_hdr(skb)->code;
166 int mtu = ntohs(icmp_hdr(skb)->un.frag.mtu);
171 int tunnel_hdr_len, tot_hdr_len;
172 unsigned int orig_mac_header;
173 unsigned int orig_nw_header;
175 if (type != ICMP_DEST_UNREACH || code != ICMP_FRAG_NEEDED)
179 * The mimimum size packet that we would actually be able to process:
180 * encapsulating IP header, minimum GRE header, Ethernet header,
183 if (!pskb_may_pull(skb, sizeof(struct iphdr) + GRE_HEADER_SECTION +
184 ETH_HLEN + sizeof(struct iphdr)))
187 iph = (struct iphdr *)skb->data;
189 tunnel_hdr_len = parse_header(iph, &flags, &key);
190 if (tunnel_hdr_len < 0)
193 vport = tnl_find_port(iph->saddr, iph->daddr, key,
194 TNL_T_PROTO_GRE | TNL_T_KEY_EITHER, &mutable);
199 * Packets received by this function were previously sent by us, so
200 * any comparisons should be to the output values, not the input.
201 * However, it's not really worth it to have a hash table based on
202 * output keys (especially since ICMP error handling of tunneled packets
203 * isn't that reliable anyways). Therefore, we do a lookup based on the
204 * out key as if it were the in key and then check to see if the input
205 * and output keys are the same.
207 if (mutable->in_key != mutable->out_key)
210 if (!!(mutable->flags & TNL_F_IN_KEY_MATCH) !=
211 !!(mutable->flags & TNL_F_OUT_KEY_ACTION))
214 if ((mutable->flags & TNL_F_CSUM) && !(flags & GRE_CSUM))
217 tunnel_hdr_len += iph->ihl << 2;
219 orig_mac_header = skb_mac_header(skb) - skb->data;
220 orig_nw_header = skb_network_header(skb) - skb->data;
221 skb_set_mac_header(skb, tunnel_hdr_len);
223 tot_hdr_len = tunnel_hdr_len + ETH_HLEN;
225 skb->protocol = eth_hdr(skb)->h_proto;
226 if (skb->protocol == htons(ETH_P_8021Q)) {
227 tot_hdr_len += VLAN_HLEN;
228 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
231 skb_set_network_header(skb, tot_hdr_len);
234 if (skb->protocol == htons(ETH_P_IP))
235 tot_hdr_len += sizeof(struct iphdr);
236 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
237 else if (skb->protocol == htons(ETH_P_IPV6))
238 tot_hdr_len += sizeof(struct ipv6hdr);
243 if (!pskb_may_pull(skb, tot_hdr_len))
246 if (skb->protocol == htons(ETH_P_IP)) {
247 if (mtu < IP_MIN_MTU) {
248 if (ntohs(ip_hdr(skb)->tot_len) >= IP_MIN_MTU)
255 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
256 else if (skb->protocol == htons(ETH_P_IPV6)) {
257 if (mtu < IPV6_MIN_MTU) {
258 unsigned int packet_length = sizeof(struct ipv6hdr) +
259 ntohs(ipv6_hdr(skb)->payload_len);
261 if (packet_length >= IPV6_MIN_MTU
262 || ntohs(ipv6_hdr(skb)->payload_len) == 0)
270 __skb_pull(skb, tunnel_hdr_len);
271 tnl_frag_needed(vport, mutable, skb, mtu, key);
272 __skb_push(skb, tunnel_hdr_len);
275 skb_set_mac_header(skb, orig_mac_header);
276 skb_set_network_header(skb, orig_nw_header);
277 skb->protocol = htons(ETH_P_IP);
280 static bool check_checksum(struct sk_buff *skb)
282 struct iphdr *iph = ip_hdr(skb);
283 struct gre_base_hdr *greh = (struct gre_base_hdr *)(iph + 1);
286 if (greh->flags & GRE_CSUM) {
287 switch (skb->ip_summed) {
288 case CHECKSUM_COMPLETE:
289 csum = csum_fold(skb->csum);
297 csum = __skb_checksum_complete(skb);
298 skb->ip_summed = CHECKSUM_COMPLETE;
306 /* Called with rcu_read_lock and BH disabled. */
307 static int gre_rcv(struct sk_buff *skb)
310 const struct tnl_mutable_config *mutable;
316 if (unlikely(!pskb_may_pull(skb, sizeof(struct gre_base_hdr) + ETH_HLEN)))
319 if (unlikely(!check_checksum(skb)))
322 hdr_len = parse_header(ip_hdr(skb), &flags, &key);
323 if (unlikely(hdr_len < 0))
326 if (unlikely(!pskb_may_pull(skb, hdr_len + ETH_HLEN)))
330 vport = tnl_find_port(iph->daddr, iph->saddr, key,
331 TNL_T_PROTO_GRE | TNL_T_KEY_EITHER, &mutable);
332 if (unlikely(!vport)) {
333 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
337 if (mutable->flags & TNL_F_IN_KEY_MATCH)
338 OVS_CB(skb)->tun_id = key;
340 OVS_CB(skb)->tun_id = 0;
342 __skb_pull(skb, hdr_len);
343 skb_postpull_rcsum(skb, skb_transport_header(skb), hdr_len + ETH_HLEN);
353 static const struct tnl_ops gre_tnl_ops = {
354 .tunnel_type = TNL_T_PROTO_GRE,
355 .ipproto = IPPROTO_GRE,
356 .hdr_len = gre_hdr_len,
357 .build_header = gre_build_header,
358 .update_header = gre_update_header,
361 static struct vport *gre_create(const struct vport_parms *parms)
363 return tnl_create(parms, &gre_vport_ops, &gre_tnl_ops);
366 static const struct net_protocol gre_protocol_handlers = {
368 .err_handler = gre_err,
371 static int gre_init(void)
375 err = inet_add_protocol(&gre_protocol_handlers, IPPROTO_GRE);
377 pr_warn("cannot register gre protocol handler\n");
382 static void gre_exit(void)
384 inet_del_protocol(&gre_protocol_handlers, IPPROTO_GRE);
387 const struct vport_ops gre_vport_ops = {
388 .type = ODP_VPORT_TYPE_GRE,
389 .flags = VPORT_F_GEN_STATS | VPORT_F_TUN_ID,
392 .create = gre_create,
393 .destroy = tnl_destroy,
394 .set_addr = tnl_set_addr,
395 .get_name = tnl_get_name,
396 .get_addr = tnl_get_addr,
397 .get_options = tnl_get_options,
398 .set_options = tnl_set_options,
399 .get_dev_flags = vport_gen_get_dev_flags,
400 .is_running = vport_gen_is_running,
401 .get_operstate = vport_gen_get_operstate,