2 * Copyright (c) 2010 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 #include <linux/if_arp.h>
10 #include <linux/if_ether.h>
12 #include <linux/if_tunnel.h>
13 #include <linux/if_vlan.h>
15 #include <linux/in_route.h>
16 #include <linux/jhash.h>
17 #include <linux/kernel.h>
18 #include <linux/version.h>
20 #include <net/dsfield.h>
23 #include <net/inet_ecn.h>
25 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
28 #include <net/protocol.h>
29 #include <net/route.h>
34 #include "openvswitch/gre.h"
37 #include "vport-generic.h"
39 /* The absolute minimum fragment size. Note that there are many other
40 * definitions of the minimum MTU. */
43 /* The GRE header is composed of a series of sections: a base and then a variable
44 * number of options. */
45 #define GRE_HEADER_SECTION 4
47 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
48 #define rt_dst(rt) (rt->dst)
50 #define rt_dst(rt) (rt->u.dst)
58 struct mutable_config {
61 unsigned char eth_addr[ETH_ALEN];
63 struct gre_port_config port_config;
65 int tunnel_hlen; /* Tunnel header length. */
70 struct tbl_node tbl_node;
74 /* Protected by RCU. */
75 struct mutable_config *mutable;
78 /* Protected by RCU. */
79 static struct tbl *port_table;
81 /* These are just used as an optimization: they don't require any kind of
82 * synchronization because we could have just as easily read the value before
83 * the port change happened. */
84 static unsigned int key_local_remote_ports;
85 static unsigned int key_remote_ports;
86 static unsigned int local_remote_ports;
87 static unsigned int remote_ports;
89 static inline struct gre_vport *gre_vport_priv(const struct vport *vport)
91 return vport_priv(vport);
94 static inline struct vport *gre_vport_to_vport(const struct gre_vport *gre_vport)
96 return vport_from_priv(gre_vport);
99 static inline struct gre_vport *gre_vport_table_cast(const struct tbl_node *node)
101 return container_of(node, struct gre_vport, tbl_node);
105 static void free_config(struct rcu_head *rcu)
107 struct mutable_config *c = container_of(rcu, struct mutable_config, rcu);
111 static void assign_config_rcu(struct vport *vport,
112 struct mutable_config *new_config)
114 struct gre_vport *gre_vport = gre_vport_priv(vport);
115 struct mutable_config *old_config;
117 old_config = rcu_dereference(gre_vport->mutable);
118 rcu_assign_pointer(gre_vport->mutable, new_config);
119 call_rcu(&old_config->rcu, free_config);
122 static unsigned int *find_port_pool(const struct mutable_config *mutable)
124 if (mutable->port_config.flags & GRE_F_IN_KEY_MATCH) {
125 if (mutable->port_config.saddr)
126 return &local_remote_ports;
128 return &remote_ports;
130 if (mutable->port_config.saddr)
131 return &key_local_remote_ports;
133 return &key_remote_ports;
144 struct port_lookup_key {
145 u32 vals[4]; /* Contains enum lookup_key keys. */
146 const struct mutable_config *mutable;
149 /* Modifies 'target' to store the rcu_dereferenced pointer that was used to do
150 * the comparision. */
151 static int port_cmp(const struct tbl_node *node, void *target)
153 const struct gre_vport *gre_vport = gre_vport_table_cast(node);
154 struct port_lookup_key *lookup = target;
156 lookup->mutable = rcu_dereference(gre_vport->mutable);
158 return ((lookup->mutable->port_config.flags & GRE_F_IN_KEY_MATCH) ==
159 lookup->vals[LOOKUP_KEY_MATCH]) &&
160 lookup->mutable->port_config.daddr == lookup->vals[LOOKUP_DADDR] &&
161 lookup->mutable->port_config.in_key == lookup->vals[LOOKUP_KEY] &&
162 lookup->mutable->port_config.saddr == lookup->vals[LOOKUP_SADDR];
165 static u32 port_hash(struct port_lookup_key *lookup)
167 return jhash2(lookup->vals, ARRAY_SIZE(lookup->vals), 0);
170 static int add_port(struct vport *vport)
172 struct gre_vport *gre_vport = gre_vport_priv(vport);
173 struct port_lookup_key lookup;
177 struct tbl *new_table;
179 new_table = tbl_create(0);
183 rcu_assign_pointer(port_table, new_table);
185 } else if (tbl_count(port_table) > tbl_n_buckets(port_table)) {
186 struct tbl *old_table = port_table;
187 struct tbl *new_table;
189 new_table = tbl_expand(old_table);
190 if (IS_ERR(new_table))
191 return PTR_ERR(new_table);
193 rcu_assign_pointer(port_table, new_table);
194 tbl_deferred_destroy(old_table, NULL);
197 lookup.vals[LOOKUP_SADDR] = gre_vport->mutable->port_config.saddr;
198 lookup.vals[LOOKUP_DADDR] = gre_vport->mutable->port_config.daddr;
199 lookup.vals[LOOKUP_KEY] = gre_vport->mutable->port_config.in_key;
200 lookup.vals[LOOKUP_KEY_MATCH] = gre_vport->mutable->port_config.flags & GRE_F_IN_KEY_MATCH;
202 err = tbl_insert(port_table, &gre_vport->tbl_node, port_hash(&lookup));
206 (*find_port_pool(gre_vport->mutable))++;
211 static int del_port(struct vport *vport)
213 struct gre_vport *gre_vport = gre_vport_priv(vport);
216 err = tbl_remove(port_table, &gre_vport->tbl_node);
220 (*find_port_pool(gre_vport->mutable))--;
225 #define FIND_PORT_KEY (1 << 0)
226 #define FIND_PORT_MATCH (1 << 1)
227 #define FIND_PORT_ANY (FIND_PORT_KEY | FIND_PORT_MATCH)
229 static struct vport *find_port(__be32 saddr, __be32 daddr, __be32 key,
231 const struct mutable_config **mutable)
233 struct port_lookup_key lookup;
234 struct tbl *table = rcu_dereference(port_table);
235 struct tbl_node *tbl_node;
240 lookup.vals[LOOKUP_SADDR] = saddr;
241 lookup.vals[LOOKUP_DADDR] = daddr;
243 if (port_type & FIND_PORT_KEY) {
244 lookup.vals[LOOKUP_KEY] = key;
245 lookup.vals[LOOKUP_KEY_MATCH] = 0;
247 if (key_local_remote_ports) {
248 tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
253 if (key_remote_ports) {
254 lookup.vals[LOOKUP_SADDR] = 0;
256 tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
260 lookup.vals[LOOKUP_SADDR] = saddr;
264 if (port_type & FIND_PORT_MATCH) {
265 lookup.vals[LOOKUP_KEY] = 0;
266 lookup.vals[LOOKUP_KEY_MATCH] = GRE_F_IN_KEY_MATCH;
268 if (local_remote_ports) {
269 tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
275 lookup.vals[LOOKUP_SADDR] = 0;
277 tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
286 *mutable = lookup.mutable;
287 return gre_vport_to_vport(gre_vport_table_cast(tbl_node));
290 static bool check_ipv4_address(__be32 addr)
292 if (ipv4_is_multicast(addr) || ipv4_is_lbcast(addr)
293 || ipv4_is_loopback(addr) || ipv4_is_zeronet(addr))
299 static bool ipv4_should_icmp(struct sk_buff *skb)
301 struct iphdr *old_iph = ip_hdr(skb);
303 /* Don't respond to L2 broadcast. */
304 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
307 /* Don't respond to L3 broadcast or invalid addresses. */
308 if (!check_ipv4_address(old_iph->daddr) ||
309 !check_ipv4_address(old_iph->saddr))
312 /* Only respond to the first fragment. */
313 if (old_iph->frag_off & htons(IP_OFFSET))
316 /* Don't respond to ICMP error messages. */
317 if (old_iph->protocol == IPPROTO_ICMP) {
318 u8 icmp_type, *icmp_typep;
320 icmp_typep = skb_header_pointer(skb, (u8 *)old_iph +
321 (old_iph->ihl << 2) +
322 offsetof(struct icmphdr, type) -
323 skb->data, sizeof(icmp_type),
329 if (*icmp_typep > NR_ICMP_TYPES
330 || (*icmp_typep <= ICMP_PARAMETERPROB
331 && *icmp_typep != ICMP_ECHOREPLY
332 && *icmp_typep != ICMP_ECHO))
339 static void ipv4_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
340 unsigned int mtu, unsigned int payload_length)
342 struct iphdr *iph, *old_iph = ip_hdr(skb);
343 struct icmphdr *icmph;
346 iph = (struct iphdr *)skb_put(nskb, sizeof(struct iphdr));
347 icmph = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr));
348 payload = skb_put(nskb, payload_length);
352 iph->ihl = sizeof(struct iphdr) >> 2;
353 iph->tos = (old_iph->tos & IPTOS_TOS_MASK) |
354 IPTOS_PREC_INTERNETCONTROL;
355 iph->tot_len = htons(sizeof(struct iphdr)
356 + sizeof(struct icmphdr)
358 get_random_bytes(&iph->id, sizeof(iph->id));
361 iph->protocol = IPPROTO_ICMP;
362 iph->daddr = old_iph->saddr;
363 iph->saddr = old_iph->daddr;
368 icmph->type = ICMP_DEST_UNREACH;
369 icmph->code = ICMP_FRAG_NEEDED;
370 icmph->un.gateway = htonl(mtu);
373 nskb->csum = csum_partial((u8 *)icmph, sizeof(struct icmphdr), 0);
374 nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_iph - skb->data,
375 payload, payload_length,
377 icmph->checksum = csum_fold(nskb->csum);
380 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
381 static bool ipv6_should_icmp(struct sk_buff *skb)
383 struct ipv6hdr *old_ipv6h = ipv6_hdr(skb);
385 int payload_off = (u8 *)(old_ipv6h + 1) - skb->data;
386 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
388 /* Check source address is valid. */
389 addr_type = ipv6_addr_type(&old_ipv6h->saddr);
390 if (addr_type & IPV6_ADDR_MULTICAST || addr_type == IPV6_ADDR_ANY)
393 /* Don't reply to unspecified addresses. */
394 if (ipv6_addr_type(&old_ipv6h->daddr) == IPV6_ADDR_ANY)
397 /* Don't respond to ICMP error messages. */
398 payload_off = ipv6_skip_exthdr(skb, payload_off, &nexthdr);
402 if (nexthdr == NEXTHDR_ICMP) {
403 u8 icmp_type, *icmp_typep;
405 icmp_typep = skb_header_pointer(skb, payload_off +
406 offsetof(struct icmp6hdr,
408 sizeof(icmp_type), &icmp_type);
410 if (!icmp_typep || !(*icmp_typep & ICMPV6_INFOMSG_MASK))
417 static void ipv6_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
418 unsigned int mtu, unsigned int payload_length)
420 struct ipv6hdr *ipv6h, *old_ipv6h = ipv6_hdr(skb);
421 struct icmp6hdr *icmp6h;
424 ipv6h = (struct ipv6hdr *)skb_put(nskb, sizeof(struct ipv6hdr));
425 icmp6h = (struct icmp6hdr *)skb_put(nskb, sizeof(struct icmp6hdr));
426 payload = skb_put(nskb, payload_length);
431 memset(&ipv6h->flow_lbl, 0, sizeof(ipv6h->flow_lbl));
432 ipv6h->payload_len = htons(sizeof(struct icmp6hdr)
434 ipv6h->nexthdr = NEXTHDR_ICMP;
435 ipv6h->hop_limit = IPV6_DEFAULT_HOPLIMIT;
436 ipv6_addr_copy(&ipv6h->daddr, &old_ipv6h->saddr);
437 ipv6_addr_copy(&ipv6h->saddr, &old_ipv6h->daddr);
440 icmp6h->icmp6_type = ICMPV6_PKT_TOOBIG;
441 icmp6h->icmp6_code = 0;
442 icmp6h->icmp6_cksum = 0;
443 icmp6h->icmp6_mtu = htonl(mtu);
445 nskb->csum = csum_partial((u8 *)icmp6h, sizeof(struct icmp6hdr), 0);
446 nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_ipv6h - skb->data,
447 payload, payload_length,
449 icmp6h->icmp6_cksum = csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
450 sizeof(struct icmp6hdr)
452 ipv6h->nexthdr, nskb->csum);
456 static bool send_frag_needed(struct vport *vport,
457 const struct mutable_config *mutable,
458 struct sk_buff *skb, unsigned int mtu,
461 unsigned int eth_hdr_len = ETH_HLEN;
462 unsigned int total_length = 0, header_length = 0, payload_length;
463 struct ethhdr *eh, *old_eh = eth_hdr(skb);
464 struct sk_buff *nskb;
467 if (skb->protocol == htons(ETH_P_IP)) {
468 if (mtu < IP_MIN_MTU)
471 if (!ipv4_should_icmp(skb))
474 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
475 else if (skb->protocol == htons(ETH_P_IPV6)) {
476 if (mtu < IPV6_MIN_MTU)
479 /* In theory we should do PMTUD on IPv6 multicast messages but
480 * we don't have an address to send from so just fragment. */
481 if (ipv6_addr_type(&ipv6_hdr(skb)->daddr) & IPV6_ADDR_MULTICAST)
484 if (!ipv6_should_icmp(skb))
492 if (old_eh->h_proto == htons(ETH_P_8021Q))
493 eth_hdr_len = VLAN_ETH_HLEN;
495 payload_length = skb->len - eth_hdr_len;
496 if (skb->protocol == htons(ETH_P_IP)) {
497 header_length = sizeof(struct iphdr) + sizeof(struct icmphdr);
498 total_length = min_t(unsigned int, header_length +
499 payload_length, 576);
501 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
503 header_length = sizeof(struct ipv6hdr) +
504 sizeof(struct icmp6hdr);
505 total_length = min_t(unsigned int, header_length +
506 payload_length, IPV6_MIN_MTU);
510 total_length = min(total_length, mutable->mtu);
511 payload_length = total_length - header_length;
513 nskb = dev_alloc_skb(NET_IP_ALIGN + eth_hdr_len + header_length +
518 skb_reserve(nskb, NET_IP_ALIGN);
520 /* Ethernet / VLAN */
521 eh = (struct ethhdr *)skb_put(nskb, eth_hdr_len);
522 memcpy(eh->h_dest, old_eh->h_source, ETH_ALEN);
523 memcpy(eh->h_source, mutable->eth_addr, ETH_ALEN);
524 nskb->protocol = eh->h_proto = old_eh->h_proto;
525 if (old_eh->h_proto == htons(ETH_P_8021Q)) {
526 struct vlan_ethhdr *vh = (struct vlan_ethhdr *)eh;
528 vh->h_vlan_TCI = vlan_eth_hdr(skb)->h_vlan_TCI;
529 vh->h_vlan_encapsulated_proto = skb->protocol;
531 skb_reset_mac_header(nskb);
534 if (skb->protocol == htons(ETH_P_IP))
535 ipv4_build_icmp(skb, nskb, mtu, payload_length);
536 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
538 ipv6_build_icmp(skb, nskb, mtu, payload_length);
541 /* Assume that flow based keys are symmetric with respect to input
542 * and output and use the key that we were going to put on the
543 * outgoing packet for the fake received packet. If the keys are
544 * not symmetric then PMTUD needs to be disabled since we won't have
545 * any way of synthesizing packets. */
546 if (mutable->port_config.flags & GRE_F_IN_KEY_MATCH &&
547 mutable->port_config.flags & GRE_F_OUT_KEY_ACTION)
548 OVS_CB(nskb)->tun_id = flow_key;
550 compute_ip_summed(nskb, false);
551 vport_receive(vport, nskb);
556 static struct sk_buff *check_headroom(struct sk_buff *skb, int headroom)
558 if (skb_headroom(skb) < headroom || skb_header_cloned(skb)) {
559 struct sk_buff *nskb = skb_realloc_headroom(skb, headroom + 16);
562 return ERR_PTR(-ENOMEM);
565 set_skb_csum_bits(skb, nskb);
568 skb_set_owner_w(nskb, skb->sk);
577 static void create_gre_header(struct sk_buff *skb,
578 const struct mutable_config *mutable)
580 struct iphdr *iph = ip_hdr(skb);
581 struct gre_base_hdr *greh = (struct gre_base_hdr *)(iph + 1);
582 __be32 *options = (__be32 *)((u8 *)iph + mutable->tunnel_hlen
583 - GRE_HEADER_SECTION);
585 greh->protocol = htons(ETH_P_TEB);
588 /* Work backwards over the options so the checksum is last. */
589 if (mutable->port_config.out_key ||
590 mutable->port_config.flags & GRE_F_OUT_KEY_ACTION) {
591 greh->flags |= GRE_KEY;
593 if (mutable->port_config.flags & GRE_F_OUT_KEY_ACTION)
594 *options = OVS_CB(skb)->tun_id;
596 *options = mutable->port_config.out_key;
601 if (mutable->port_config.flags & GRE_F_CSUM) {
602 greh->flags |= GRE_CSUM;
605 *(__sum16 *)options = csum_fold(skb_checksum(skb,
606 sizeof(struct iphdr),
607 skb->len - sizeof(struct iphdr),
612 static int check_checksum(struct sk_buff *skb)
614 struct iphdr *iph = ip_hdr(skb);
615 __be16 flags = *(__be16 *)(iph + 1);
618 if (flags & GRE_CSUM) {
619 switch (skb->ip_summed) {
620 case CHECKSUM_COMPLETE:
621 csum = csum_fold(skb->csum);
629 csum = __skb_checksum_complete(skb);
630 skb->ip_summed = CHECKSUM_COMPLETE;
638 static int parse_gre_header(struct iphdr *iph, __be16 *flags, __be32 *key)
640 /* IP and ICMP protocol handlers check that the IHL is valid. */
641 struct gre_base_hdr *greh = (struct gre_base_hdr *)((u8 *)iph + (iph->ihl << 2));
642 __be32 *options = (__be32 *)(greh + 1);
645 *flags = greh->flags;
647 if (greh->flags & (GRE_VERSION | GRE_ROUTING))
650 if (greh->protocol != htons(ETH_P_TEB))
653 hdr_len = GRE_HEADER_SECTION;
655 if (greh->flags & GRE_CSUM) {
656 hdr_len += GRE_HEADER_SECTION;
660 if (greh->flags & GRE_KEY) {
661 hdr_len += GRE_HEADER_SECTION;
668 if (greh->flags & GRE_SEQ)
669 hdr_len += GRE_HEADER_SECTION;
674 static inline u8 ecn_encapsulate(u8 tos, struct sk_buff *skb)
678 if (skb->protocol == htons(ETH_P_IP))
679 inner = ((struct iphdr *)skb_network_header(skb))->tos;
680 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
681 else if (skb->protocol == htons(ETH_P_IPV6))
682 inner = ipv6_get_dsfield((struct ipv6hdr *)skb_network_header(skb));
687 return INET_ECN_encapsulate(tos, inner);
690 static inline void ecn_decapsulate(u8 tos, struct sk_buff *skb)
692 if (INET_ECN_is_ce(tos)) {
693 __be16 protocol = skb->protocol;
694 unsigned int nw_header = skb_network_header(skb) - skb->data;
696 if (skb->protocol == htons(ETH_P_8021Q)) {
697 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
700 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
701 nw_header += VLAN_HLEN;
704 if (protocol == htons(ETH_P_IP)) {
705 if (unlikely(!pskb_may_pull(skb, nw_header
706 + sizeof(struct iphdr))))
709 IP_ECN_set_ce((struct iphdr *)(nw_header + skb->data));
711 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
712 else if (protocol == htons(ETH_P_IPV6)) {
713 if (unlikely(!pskb_may_pull(skb, nw_header
714 + sizeof(struct ipv6hdr))))
717 IP6_ECN_set_ce((struct ipv6hdr *)(nw_header
724 static struct sk_buff *handle_gso(struct sk_buff *skb)
726 if (skb_is_gso(skb)) {
727 struct sk_buff *nskb = skb_gso_segment(skb, 0);
736 static int handle_csum_offload(struct sk_buff *skb)
738 if (skb->ip_summed == CHECKSUM_PARTIAL)
739 return skb_checksum_help(skb);
741 skb->ip_summed = CHECKSUM_NONE;
746 /* Called with rcu_read_lock. */
747 static void gre_err(struct sk_buff *skb, u32 info)
750 const struct mutable_config *mutable;
751 const int type = icmp_hdr(skb)->type;
752 const int code = icmp_hdr(skb)->code;
753 int mtu = ntohs(icmp_hdr(skb)->un.frag.mtu);
758 int tunnel_hdr_len, tot_hdr_len;
759 unsigned int orig_mac_header;
760 unsigned int orig_nw_header;
762 if (type != ICMP_DEST_UNREACH || code != ICMP_FRAG_NEEDED)
765 /* The mimimum size packet that we would actually be able to process:
766 * encapsulating IP header, minimum GRE header, Ethernet header,
767 * inner IPv4 header. */
768 if (!pskb_may_pull(skb, sizeof(struct iphdr) + GRE_HEADER_SECTION +
769 ETH_HLEN + sizeof(struct iphdr)))
772 iph = (struct iphdr *)skb->data;
774 tunnel_hdr_len = parse_gre_header(iph, &flags, &key);
775 if (tunnel_hdr_len < 0)
778 vport = find_port(iph->saddr, iph->daddr, key, FIND_PORT_ANY, &mutable);
782 /* Packets received by this function were previously sent by us, so
783 * any comparisons should be to the output values, not the input.
784 * However, it's not really worth it to have a hash table based on
785 * output keys (especially since ICMP error handling of tunneled packets
786 * isn't that reliable anyways). Therefore, we do a lookup based on the
787 * out key as if it were the in key and then check to see if the input
788 * and output keys are the same. */
789 if (mutable->port_config.in_key != mutable->port_config.out_key)
792 if (!!(mutable->port_config.flags & GRE_F_IN_KEY_MATCH) !=
793 !!(mutable->port_config.flags & GRE_F_OUT_KEY_ACTION))
796 if ((mutable->port_config.flags & GRE_F_CSUM) && !(flags & GRE_CSUM))
799 tunnel_hdr_len += iph->ihl << 2;
801 orig_mac_header = skb_mac_header(skb) - skb->data;
802 orig_nw_header = skb_network_header(skb) - skb->data;
803 skb_set_mac_header(skb, tunnel_hdr_len);
805 tot_hdr_len = tunnel_hdr_len + ETH_HLEN;
807 skb->protocol = eth_hdr(skb)->h_proto;
808 if (skb->protocol == htons(ETH_P_8021Q)) {
809 tot_hdr_len += VLAN_HLEN;
810 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
813 skb_set_network_header(skb, tot_hdr_len);
816 if (skb->protocol == htons(ETH_P_IP))
817 tot_hdr_len += sizeof(struct iphdr);
818 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
819 else if (skb->protocol == htons(ETH_P_IPV6))
820 tot_hdr_len += sizeof(struct ipv6hdr);
825 if (!pskb_may_pull(skb, tot_hdr_len))
828 if (skb->protocol == htons(ETH_P_IP)) {
829 if (mtu < IP_MIN_MTU) {
830 if (ntohs(ip_hdr(skb)->tot_len) >= IP_MIN_MTU)
837 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
838 else if (skb->protocol == htons(ETH_P_IPV6)) {
839 if (mtu < IPV6_MIN_MTU) {
840 unsigned int packet_length = sizeof(struct ipv6hdr) +
841 ntohs(ipv6_hdr(skb)->payload_len);
843 if (packet_length >= IPV6_MIN_MTU
844 || ntohs(ipv6_hdr(skb)->payload_len) == 0)
852 __pskb_pull(skb, tunnel_hdr_len);
853 send_frag_needed(vport, mutable, skb, mtu, key);
854 skb_push(skb, tunnel_hdr_len);
857 skb_set_mac_header(skb, orig_mac_header);
858 skb_set_network_header(skb, orig_nw_header);
859 skb->protocol = htons(ETH_P_IP);
862 /* Called with rcu_read_lock. */
863 static int gre_rcv(struct sk_buff *skb)
866 const struct mutable_config *mutable;
872 if (!pskb_may_pull(skb, GRE_HEADER_SECTION + ETH_HLEN))
875 if (!check_checksum(skb))
880 hdr_len = parse_gre_header(iph, &flags, &key);
884 vport = find_port(iph->daddr, iph->saddr, key, FIND_PORT_ANY, &mutable);
886 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
890 if (!pskb_pull(skb, hdr_len) || !pskb_may_pull(skb, ETH_HLEN)) {
891 vport_record_error(vport, VPORT_E_RX_ERROR);
895 skb->pkt_type = PACKET_HOST;
896 skb->protocol = eth_type_trans(skb, skb->dev);
897 skb_postpull_rcsum(skb, skb_transport_header(skb), hdr_len + ETH_HLEN);
902 skb_reset_network_header(skb);
904 ecn_decapsulate(iph->tos, skb);
906 if (mutable->port_config.flags & GRE_F_IN_KEY_MATCH)
907 OVS_CB(skb)->tun_id = key;
909 OVS_CB(skb)->tun_id = 0;
911 skb_push(skb, ETH_HLEN);
912 compute_ip_summed(skb, false);
914 vport_receive(vport, skb);
923 static int build_packet(struct vport *vport, const struct mutable_config *mutable,
924 struct iphdr *iph, struct rtable *rt, int max_headroom,
925 int mtu, struct sk_buff *skb)
928 struct iphdr *new_iph;
929 int orig_len = skb->len;
930 __be16 frag_off = iph->frag_off;
932 skb = check_headroom(skb, max_headroom);
933 if (unlikely(IS_ERR(skb)))
936 err = handle_csum_offload(skb);
940 if (skb->protocol == htons(ETH_P_IP)) {
941 struct iphdr *old_iph = ip_hdr(skb);
943 if ((old_iph->frag_off & htons(IP_DF)) &&
944 mtu < ntohs(old_iph->tot_len)) {
945 if (send_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id))
950 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
951 else if (skb->protocol == htons(ETH_P_IPV6)) {
952 unsigned int packet_length = skb->len - ETH_HLEN
953 - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0);
955 /* IPv6 requires PMTUD if the packet is above the minimum MTU. */
956 if (packet_length > IPV6_MIN_MTU)
957 frag_off = htons(IP_DF);
959 if (mtu < packet_length) {
960 if (send_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id))
966 skb_reset_transport_header(skb);
967 new_iph = (struct iphdr *)skb_push(skb, mutable->tunnel_hlen);
968 skb_reset_network_header(skb);
970 memcpy(new_iph, iph, sizeof(struct iphdr));
971 new_iph->frag_off = frag_off;
972 ip_select_ident(new_iph, &rt_dst(rt), NULL);
974 create_gre_header(skb, mutable);
976 /* Allow our local IP stack to fragment the outer packet even if the
977 * DF bit is set as a last resort. */
980 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
981 IPCB(skb)->flags = 0;
983 err = ip_local_out(skb);
984 if (likely(net_xmit_eval(err) == 0))
987 vport_record_error(vport, VPORT_E_TX_ERROR);
994 vport_record_error(vport, VPORT_E_TX_DROPPED);
999 static int gre_send(struct vport *vport, struct sk_buff *skb)
1001 struct gre_vport *gre_vport = gre_vport_priv(vport);
1002 const struct mutable_config *mutable = rcu_dereference(gre_vport->mutable);
1004 struct iphdr *old_iph;
1011 /* Validate the protocol headers before we try to use them. */
1012 if (skb->protocol == htons(ETH_P_8021Q)) {
1013 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
1016 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
1017 skb_set_network_header(skb, VLAN_ETH_HLEN);
1020 if (skb->protocol == htons(ETH_P_IP)) {
1021 if (unlikely(!pskb_may_pull(skb, skb_network_header(skb)
1022 + sizeof(struct iphdr) - skb->data)))
1025 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1026 else if (skb->protocol == htons(ETH_P_IPV6)) {
1027 if (unlikely(!pskb_may_pull(skb, skb_network_header(skb)
1028 + sizeof(struct ipv6hdr) - skb->data)))
1032 old_iph = ip_hdr(skb);
1034 iph.tos = mutable->port_config.tos;
1035 if (mutable->port_config.flags & GRE_F_TOS_INHERIT) {
1036 if (skb->protocol == htons(ETH_P_IP))
1037 iph.tos = old_iph->tos;
1038 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1039 else if (skb->protocol == htons(ETH_P_IPV6))
1040 iph.tos = ipv6_get_dsfield(ipv6_hdr(skb));
1043 iph.tos = ecn_encapsulate(iph.tos, skb);
1046 struct flowi fl = { .nl_u = { .ip4_u =
1047 { .daddr = mutable->port_config.daddr,
1048 .saddr = mutable->port_config.saddr,
1049 .tos = RT_TOS(iph.tos) } },
1050 .proto = IPPROTO_GRE };
1052 if (ip_route_output_key(&init_net, &rt, &fl))
1056 iph.ttl = mutable->port_config.ttl;
1057 if (mutable->port_config.flags & GRE_F_TTL_INHERIT) {
1058 if (skb->protocol == htons(ETH_P_IP))
1059 iph.ttl = old_iph->ttl;
1060 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1061 else if (skb->protocol == htons(ETH_P_IPV6))
1062 iph.ttl = ipv6_hdr(skb)->hop_limit;
1066 iph.ttl = dst_metric(&rt_dst(rt), RTAX_HOPLIMIT);
1068 iph.frag_off = (mutable->port_config.flags & GRE_F_PMTUD) ? htons(IP_DF) : 0;
1070 mtu = dst_mtu(&rt_dst(rt))
1072 - mutable->tunnel_hlen
1073 - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0);
1077 if (skb->protocol == htons(ETH_P_IP)) {
1078 iph.frag_off |= old_iph->frag_off & htons(IP_DF);
1079 mtu = max(mtu, IP_MIN_MTU);
1081 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1082 else if (skb->protocol == htons(ETH_P_IPV6))
1083 mtu = max(mtu, IPV6_MIN_MTU);
1087 iph.ihl = sizeof(struct iphdr) >> 2;
1088 iph.protocol = IPPROTO_GRE;
1089 iph.daddr = rt->rt_dst;
1090 iph.saddr = rt->rt_src;
1095 skb_dst_set(skb, &rt_dst(rt));
1097 /* If we are doing GSO on a pskb it is better to make sure that the
1098 * headroom is correct now. We will only have to copy the portion in
1099 * the linear data area and GSO will preserve headroom when it creates
1100 * the segments. This is particularly beneficial on Xen where we get
1101 * lots of GSO pskbs. Conversely, we delay copying if it is just to
1102 * get our own writable clone because GSO may do the copy for us. */
1103 max_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
1104 + mutable->tunnel_hlen;
1106 if (skb_headroom(skb) < max_headroom) {
1107 skb = check_headroom(skb, max_headroom);
1108 if (unlikely(IS_ERR(skb))) {
1109 vport_record_error(vport, VPORT_E_TX_DROPPED);
1114 forward_ip_summed(skb);
1116 if (unlikely(vswitch_skb_checksum_setup(skb)))
1119 skb = handle_gso(skb);
1120 if (unlikely(IS_ERR(skb))) {
1121 vport_record_error(vport, VPORT_E_TX_DROPPED);
1125 /* Process GSO segments. Try to do any work for the entire packet that
1126 * doesn't involve actually writing to it before this point. */
1129 struct sk_buff *next_skb = skb->next;
1132 orig_len += build_packet(vport, mutable, &iph, rt, max_headroom, mtu, skb);
1141 vport_record_error(vport, VPORT_E_TX_ERROR);
1146 static struct net_protocol gre_protocol_handlers = {
1148 .err_handler = gre_err,
1151 static int gre_init(void)
1155 err = inet_add_protocol(&gre_protocol_handlers, IPPROTO_GRE);
1157 printk(KERN_WARNING "openvswitch: cannot register gre protocol handler\n");
1162 static void gre_exit(void)
1164 tbl_destroy(port_table, NULL);
1165 inet_del_protocol(&gre_protocol_handlers, IPPROTO_GRE);
1168 static int set_config(const struct vport *cur_vport,
1169 struct mutable_config *mutable, const void __user *uconfig)
1171 const struct vport *old_vport;
1172 const struct mutable_config *old_mutable;
1175 if (copy_from_user(&mutable->port_config, uconfig, sizeof(struct gre_port_config)))
1178 if (mutable->port_config.daddr == 0)
1181 if (mutable->port_config.flags & GRE_F_IN_KEY_MATCH) {
1182 port_type = FIND_PORT_MATCH;
1183 mutable->port_config.in_key = 0;
1185 port_type = FIND_PORT_KEY;
1187 old_vport = find_port(mutable->port_config.saddr,
1188 mutable->port_config.daddr,
1189 mutable->port_config.in_key, port_type,
1192 if (old_vport && old_vport != cur_vport)
1195 if (mutable->port_config.flags & GRE_F_OUT_KEY_ACTION)
1196 mutable->port_config.out_key = 0;
1198 mutable->tunnel_hlen = sizeof(struct iphdr) + GRE_HEADER_SECTION;
1200 if (mutable->port_config.flags & GRE_F_CSUM)
1201 mutable->tunnel_hlen += GRE_HEADER_SECTION;
1203 if (mutable->port_config.out_key ||
1204 mutable->port_config.flags & GRE_F_OUT_KEY_ACTION)
1205 mutable->tunnel_hlen += GRE_HEADER_SECTION;
1210 static struct vport *gre_create(const char *name, const void __user *config)
1212 struct vport *vport;
1213 struct gre_vport *gre_vport;
1216 vport = vport_alloc(sizeof(struct gre_vport), &gre_vport_ops);
1217 if (IS_ERR(vport)) {
1218 err = PTR_ERR(vport);
1222 gre_vport = gre_vport_priv(vport);
1224 strcpy(gre_vport->name, name);
1226 gre_vport->mutable = kmalloc(sizeof(struct mutable_config), GFP_KERNEL);
1227 if (!gre_vport->mutable) {
1229 goto error_free_vport;
1232 vport_gen_rand_ether_addr(gre_vport->mutable->eth_addr);
1233 gre_vport->mutable->mtu = ETH_DATA_LEN;
1235 err = set_config(NULL, gre_vport->mutable, config);
1237 goto error_free_mutable;
1239 err = add_port(vport);
1241 goto error_free_mutable;
1246 kfree(gre_vport->mutable);
1250 return ERR_PTR(err);
1253 static int gre_modify(struct vport *vport, const void __user *config)
1255 struct gre_vport *gre_vport = gre_vport_priv(vport);
1256 struct mutable_config *mutable;
1258 int update_hash = 0;
1260 mutable = kmemdup(gre_vport->mutable, sizeof(struct mutable_config), GFP_KERNEL);
1266 err = set_config(vport, mutable, config);
1270 /* Only remove the port from the hash table if something that would
1271 * affect the lookup has changed. */
1272 if (gre_vport->mutable->port_config.saddr != mutable->port_config.saddr ||
1273 gre_vport->mutable->port_config.daddr != mutable->port_config.daddr ||
1274 gre_vport->mutable->port_config.in_key != mutable->port_config.in_key ||
1275 (gre_vport->mutable->port_config.flags & GRE_F_IN_KEY_MATCH) !=
1276 (mutable->port_config.flags & GRE_F_IN_KEY_MATCH))
1280 /* This update is not atomic but the lookup uses the config, which
1281 * serves as an inherent double check. */
1283 err = del_port(vport);
1288 assign_config_rcu(vport, mutable);
1291 err = add_port(vport);
1304 static void free_port(struct rcu_head *rcu)
1306 struct gre_vport *gre_vport = container_of(rcu, struct gre_vport, rcu);
1308 kfree(gre_vport->mutable);
1309 vport_free(gre_vport_to_vport(gre_vport));
1312 static int gre_destroy(struct vport *vport)
1314 struct gre_vport *gre_vport = gre_vport_priv(vport);
1316 const struct mutable_config *old_mutable;
1318 /* Do a hash table lookup to make sure that the port exists. It should
1319 * exist but might not if a modify failed earlier. */
1320 if (gre_vport->mutable->port_config.flags & GRE_F_IN_KEY_MATCH)
1321 port_type = FIND_PORT_MATCH;
1323 port_type = FIND_PORT_KEY;
1325 if (vport == find_port(gre_vport->mutable->port_config.saddr,
1326 gre_vport->mutable->port_config.daddr,
1327 gre_vport->mutable->port_config.in_key, port_type, &old_mutable))
1330 call_rcu(&gre_vport->rcu, free_port);
1335 static int gre_set_mtu(struct vport *vport, int mtu)
1337 struct gre_vport *gre_vport = gre_vport_priv(vport);
1338 struct mutable_config *mutable;
1340 mutable = kmemdup(gre_vport->mutable, sizeof(struct mutable_config), GFP_KERNEL);
1345 assign_config_rcu(vport, mutable);
1350 static int gre_set_addr(struct vport *vport, const unsigned char *addr)
1352 struct gre_vport *gre_vport = gre_vport_priv(vport);
1353 struct mutable_config *mutable;
1355 mutable = kmemdup(gre_vport->mutable, sizeof(struct mutable_config), GFP_KERNEL);
1359 memcpy(mutable->eth_addr, addr, ETH_ALEN);
1360 assign_config_rcu(vport, mutable);
1366 static const char *gre_get_name(const struct vport *vport)
1368 const struct gre_vport *gre_vport = gre_vport_priv(vport);
1369 return gre_vport->name;
1372 static const unsigned char *gre_get_addr(const struct vport *vport)
1374 const struct gre_vport *gre_vport = gre_vport_priv(vport);
1375 return rcu_dereference(gre_vport->mutable)->eth_addr;
1378 static int gre_get_mtu(const struct vport *vport)
1380 const struct gre_vport *gre_vport = gre_vport_priv(vport);
1381 return rcu_dereference(gre_vport->mutable)->mtu;
1384 struct vport_ops gre_vport_ops = {
1386 .flags = VPORT_F_GEN_STATS | VPORT_F_TUN_ID,
1389 .create = gre_create,
1390 .modify = gre_modify,
1391 .destroy = gre_destroy,
1392 .set_mtu = gre_set_mtu,
1393 .set_addr = gre_set_addr,
1394 .get_name = gre_get_name,
1395 .get_addr = gre_get_addr,
1396 .get_dev_flags = vport_gen_get_dev_flags,
1397 .is_running = vport_gen_is_running,
1398 .get_operstate = vport_gen_get_operstate,
1399 .get_mtu = gre_get_mtu,