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 struct mutable_config {
50 unsigned char eth_addr[ETH_ALEN];
52 struct gre_port_config port_config;
54 int tunnel_hlen; /* Tunnel header length. */
58 struct tbl_node tbl_node;
62 /* Protected by RCU. */
63 struct mutable_config *mutable;
66 /* Protected by RCU. */
67 static struct tbl *port_table;
69 /* These are just used as an optimization: they don't require any kind of
70 * synchronization because we could have just as easily read the value before
71 * the port change happened. */
72 static unsigned int key_local_remote_ports;
73 static unsigned int key_remote_ports;
74 static unsigned int local_remote_ports;
75 static unsigned int remote_ports;
77 static inline struct gre_vport *
78 gre_vport_priv(const struct vport *vport)
80 return vport_priv(vport);
83 static inline struct vport *
84 gre_vport_to_vport(const struct gre_vport *gre_vport)
86 return vport_from_priv(gre_vport);
89 static inline struct gre_vport *
90 gre_vport_table_cast(const struct tbl_node *node)
92 return container_of(node, struct gre_vport, tbl_node);
97 free_config(struct rcu_head *rcu)
99 struct mutable_config *c = container_of(rcu, struct mutable_config, rcu);
104 assign_config_rcu(struct vport *vport, struct mutable_config *new_config)
106 struct gre_vport *gre_vport = gre_vport_priv(vport);
107 struct mutable_config *old_config;
109 old_config = rcu_dereference(gre_vport->mutable);
110 rcu_assign_pointer(gre_vport->mutable, new_config);
111 call_rcu(&old_config->rcu, free_config);
114 static unsigned int *
115 find_port_pool(const struct mutable_config *mutable)
117 if (mutable->port_config.flags & GRE_F_IN_KEY_MATCH) {
118 if (mutable->port_config.saddr)
119 return &local_remote_ports;
121 return &remote_ports;
123 if (mutable->port_config.saddr)
124 return &key_local_remote_ports;
126 return &key_remote_ports;
137 struct port_lookup_key {
138 u32 vals[4]; /* Contains enum lookup_key keys. */
139 const struct mutable_config *mutable;
142 /* Modifies 'target' to store the rcu_dereferenced pointer that was used to do
143 * the comparision. */
145 port_cmp(const struct tbl_node *node, void *target)
147 const struct gre_vport *gre_vport = gre_vport_table_cast(node);
148 struct port_lookup_key *lookup = target;
150 lookup->mutable = rcu_dereference(gre_vport->mutable);
152 return ((lookup->mutable->port_config.flags & GRE_F_IN_KEY_MATCH) ==
153 lookup->vals[LOOKUP_KEY_MATCH]) &&
154 lookup->mutable->port_config.daddr == lookup->vals[LOOKUP_DADDR] &&
155 lookup->mutable->port_config.in_key == lookup->vals[LOOKUP_KEY] &&
156 lookup->mutable->port_config.saddr == lookup->vals[LOOKUP_SADDR];
160 port_hash(struct port_lookup_key *lookup)
162 return jhash2(lookup->vals, ARRAY_SIZE(lookup->vals), 0);
166 add_port(struct vport *vport)
168 struct gre_vport *gre_vport = gre_vport_priv(vport);
169 struct port_lookup_key lookup;
173 struct tbl *new_table;
175 new_table = tbl_create(0);
179 rcu_assign_pointer(port_table, new_table);
181 } else if (tbl_count(port_table) > tbl_n_buckets(port_table)) {
182 struct tbl *old_table = port_table;
183 struct tbl *new_table;
185 new_table = tbl_expand(old_table);
186 if (IS_ERR(new_table))
187 return PTR_ERR(new_table);
189 rcu_assign_pointer(port_table, new_table);
190 tbl_deferred_destroy(old_table, NULL);
193 lookup.vals[LOOKUP_SADDR] = gre_vport->mutable->port_config.saddr;
194 lookup.vals[LOOKUP_DADDR] = gre_vport->mutable->port_config.daddr;
195 lookup.vals[LOOKUP_KEY] = gre_vport->mutable->port_config.in_key;
196 lookup.vals[LOOKUP_KEY_MATCH] = gre_vport->mutable->port_config.flags & GRE_F_IN_KEY_MATCH;
198 err = tbl_insert(port_table, &gre_vport->tbl_node, port_hash(&lookup));
202 (*find_port_pool(gre_vport->mutable))++;
208 del_port(struct vport *vport)
210 struct gre_vport *gre_vport = gre_vport_priv(vport);
213 err = tbl_remove(port_table, &gre_vport->tbl_node);
217 (*find_port_pool(gre_vport->mutable))--;
222 #define FIND_PORT_KEY (1 << 0)
223 #define FIND_PORT_MATCH (1 << 1)
224 #define FIND_PORT_ANY (FIND_PORT_KEY | FIND_PORT_MATCH)
226 static struct vport *
227 find_port(__be32 saddr, __be32 daddr, __be32 key, int port_type,
228 const struct mutable_config **mutable)
230 struct port_lookup_key lookup;
231 struct tbl *table = rcu_dereference(port_table);
232 struct tbl_node *tbl_node;
237 lookup.vals[LOOKUP_SADDR] = saddr;
238 lookup.vals[LOOKUP_DADDR] = daddr;
240 if (port_type & FIND_PORT_KEY) {
241 lookup.vals[LOOKUP_KEY] = key;
242 lookup.vals[LOOKUP_KEY_MATCH] = 0;
244 if (key_local_remote_ports) {
245 tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
250 if (key_remote_ports) {
251 lookup.vals[LOOKUP_SADDR] = 0;
253 tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
257 lookup.vals[LOOKUP_SADDR] = saddr;
261 if (port_type & FIND_PORT_MATCH) {
262 lookup.vals[LOOKUP_KEY] = 0;
263 lookup.vals[LOOKUP_KEY_MATCH] = GRE_F_IN_KEY_MATCH;
265 if (local_remote_ports) {
266 tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
272 lookup.vals[LOOKUP_SADDR] = 0;
274 tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
283 *mutable = lookup.mutable;
284 return gre_vport_to_vport(gre_vport_table_cast(tbl_node));
288 check_ipv4_address(__be32 addr)
290 if (ipv4_is_multicast(addr) || ipv4_is_lbcast(addr)
291 || ipv4_is_loopback(addr) || ipv4_is_zeronet(addr))
298 ipv4_should_icmp(struct sk_buff *skb)
300 struct iphdr *old_iph = ip_hdr(skb);
302 /* Don't respond to L2 broadcast. */
303 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
306 /* Don't respond to L3 broadcast or invalid addresses. */
307 if (!check_ipv4_address(old_iph->daddr) ||
308 !check_ipv4_address(old_iph->saddr))
311 /* Only respond to the first fragment. */
312 if (old_iph->frag_off & htons(IP_OFFSET))
315 /* Don't respond to ICMP error messages. */
316 if (old_iph->protocol == IPPROTO_ICMP) {
317 u8 icmp_type, *icmp_typep;
319 icmp_typep = skb_header_pointer(skb, (u8 *)old_iph +
320 (old_iph->ihl << 2) +
321 offsetof(struct icmphdr, type) -
322 skb->data, sizeof(icmp_type),
328 if (*icmp_typep > NR_ICMP_TYPES
329 || (*icmp_typep <= ICMP_PARAMETERPROB
330 && *icmp_typep != ICMP_ECHOREPLY
331 && *icmp_typep != ICMP_ECHO))
339 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)
382 ipv6_should_icmp(struct sk_buff *skb)
384 struct ipv6hdr *old_ipv6h = ipv6_hdr(skb);
386 int payload_off = (u8 *)(old_ipv6h + 1) - skb->data;
387 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
389 /* Check source address is valid. */
390 addr_type = ipv6_addr_type(&old_ipv6h->saddr);
391 if (addr_type & IPV6_ADDR_MULTICAST || addr_type == IPV6_ADDR_ANY)
394 /* Don't reply to unspecified addresses. */
395 if (ipv6_addr_type(&old_ipv6h->daddr) == IPV6_ADDR_ANY)
398 /* Don't respond to ICMP error messages. */
399 payload_off = ipv6_skip_exthdr(skb, payload_off, &nexthdr);
403 if (nexthdr == NEXTHDR_ICMP) {
404 u8 icmp_type, *icmp_typep;
406 icmp_typep = skb_header_pointer(skb, payload_off +
407 offsetof(struct icmp6hdr,
409 sizeof(icmp_type), &icmp_type);
411 if (!icmp_typep || !(*icmp_typep & ICMPV6_INFOMSG_MASK))
419 ipv6_build_icmp(struct sk_buff *skb, struct sk_buff *nskb, unsigned int mtu,
420 unsigned int payload_length)
422 struct ipv6hdr *ipv6h, *old_ipv6h = ipv6_hdr(skb);
423 struct icmp6hdr *icmp6h;
426 ipv6h = (struct ipv6hdr *)skb_put(nskb, sizeof(struct ipv6hdr));
427 icmp6h = (struct icmp6hdr *)skb_put(nskb, sizeof(struct icmp6hdr));
428 payload = skb_put(nskb, payload_length);
433 memset(&ipv6h->flow_lbl, 0, sizeof(ipv6h->flow_lbl));
434 ipv6h->payload_len = htons(sizeof(struct icmp6hdr)
436 ipv6h->nexthdr = NEXTHDR_ICMP;
437 ipv6h->hop_limit = IPV6_DEFAULT_HOPLIMIT;
438 ipv6_addr_copy(&ipv6h->daddr, &old_ipv6h->saddr);
439 ipv6_addr_copy(&ipv6h->saddr, &old_ipv6h->daddr);
442 icmp6h->icmp6_type = ICMPV6_PKT_TOOBIG;
443 icmp6h->icmp6_code = 0;
444 icmp6h->icmp6_cksum = 0;
445 icmp6h->icmp6_mtu = htonl(mtu);
447 nskb->csum = csum_partial((u8 *)icmp6h, sizeof(struct icmp6hdr), 0);
448 nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_ipv6h - skb->data,
449 payload, payload_length,
451 icmp6h->icmp6_cksum = csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
452 sizeof(struct icmp6hdr)
454 ipv6h->nexthdr, nskb->csum);
459 send_frag_needed(struct vport *vport, const struct mutable_config *mutable,
460 struct sk_buff *skb, unsigned int mtu, __be32 flow_key)
462 unsigned int eth_hdr_len = ETH_HLEN;
463 unsigned int total_length = 0, header_length = 0, payload_length;
464 struct ethhdr *eh, *old_eh = eth_hdr(skb);
465 struct sk_buff *nskb;
468 if (skb->protocol == htons(ETH_P_IP)) {
469 if (mtu < IP_MIN_MTU)
472 if (!ipv4_should_icmp(skb))
475 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
476 else if (skb->protocol == htons(ETH_P_IPV6)) {
477 if (mtu < IPV6_MIN_MTU)
480 /* In theory we should do PMTUD on IPv6 multicast messages but
481 * we don't have an address to send from so just fragment. */
482 if (ipv6_addr_type(&ipv6_hdr(skb)->daddr) & IPV6_ADDR_MULTICAST)
485 if (!ipv6_should_icmp(skb))
493 if (old_eh->h_proto == htons(ETH_P_8021Q))
494 eth_hdr_len = VLAN_ETH_HLEN;
496 payload_length = skb->len - eth_hdr_len;
497 if (skb->protocol == htons(ETH_P_IP)) {
498 header_length = sizeof(struct iphdr) + sizeof(struct icmphdr);
499 total_length = min_t(unsigned int, header_length +
500 payload_length, 576);
502 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
504 header_length = sizeof(struct ipv6hdr) +
505 sizeof(struct icmp6hdr);
506 total_length = min_t(unsigned int, header_length +
507 payload_length, IPV6_MIN_MTU);
511 total_length = min(total_length, mutable->mtu);
512 payload_length = total_length - header_length;
514 nskb = dev_alloc_skb(NET_IP_ALIGN + eth_hdr_len + header_length +
519 skb_reserve(nskb, NET_IP_ALIGN);
521 /* Ethernet / VLAN */
522 eh = (struct ethhdr *)skb_put(nskb, eth_hdr_len);
523 memcpy(eh->h_dest, old_eh->h_source, ETH_ALEN);
524 memcpy(eh->h_source, mutable->eth_addr, ETH_ALEN);
525 nskb->protocol = eh->h_proto = old_eh->h_proto;
526 if (old_eh->h_proto == htons(ETH_P_8021Q)) {
527 struct vlan_ethhdr *vh = (struct vlan_ethhdr *)eh;
529 vh->h_vlan_TCI = vlan_eth_hdr(skb)->h_vlan_TCI;
530 vh->h_vlan_encapsulated_proto = skb->protocol;
532 skb_reset_mac_header(nskb);
535 if (skb->protocol == htons(ETH_P_IP))
536 ipv4_build_icmp(skb, nskb, mtu, payload_length);
537 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
539 ipv6_build_icmp(skb, nskb, mtu, payload_length);
542 /* Assume that flow based keys are symmetric with respect to input
543 * and output and use the key that we were going to put on the
544 * outgoing packet for the fake received packet. If the keys are
545 * not symmetric then PMTUD needs to be disabled since we won't have
546 * any way of synthesizing packets. */
547 if (mutable->port_config.flags & GRE_F_IN_KEY_MATCH &&
548 mutable->port_config.flags & GRE_F_OUT_KEY_ACTION)
549 OVS_CB(nskb)->tun_id = flow_key;
551 compute_ip_summed(nskb, false);
552 vport_receive(vport, nskb);
557 static struct sk_buff *
558 check_headroom(struct sk_buff *skb, int headroom)
560 if (skb_headroom(skb) < headroom || skb_header_cloned(skb)) {
561 struct sk_buff *nskb = skb_realloc_headroom(skb, max(headroom, 64));
564 return ERR_PTR(-ENOMEM);
567 set_skb_csum_bits(skb, nskb);
570 skb_set_owner_w(nskb, skb->sk);
580 create_gre_header(struct sk_buff *skb, const struct mutable_config *mutable)
582 struct iphdr *iph = ip_hdr(skb);
583 __be16 *flags = (__be16 *)(iph + 1);
584 __be16 *protocol = flags + 1;
585 __be32 *options = (__be32 *)((u8 *)iph + mutable->tunnel_hlen
586 - GRE_HEADER_SECTION);
588 *protocol = htons(ETH_P_TEB);
591 /* Work backwards over the options so the checksum is last. */
592 if (mutable->port_config.out_key ||
593 mutable->port_config.flags & GRE_F_OUT_KEY_ACTION) {
596 if (mutable->port_config.flags & GRE_F_OUT_KEY_ACTION)
597 *options = OVS_CB(skb)->tun_id;
599 *options = mutable->port_config.out_key;
604 if (mutable->port_config.flags & GRE_F_OUT_CSUM) {
608 *(__sum16 *)options = csum_fold(skb_checksum(skb,
609 sizeof(struct iphdr),
610 skb->len - sizeof(struct iphdr),
616 check_checksum(struct sk_buff *skb)
618 struct iphdr *iph = ip_hdr(skb);
619 __be16 flags = *(__be16 *)(iph + 1);
622 if (flags & GRE_CSUM) {
623 switch (skb->ip_summed) {
624 case CHECKSUM_COMPLETE:
625 csum = csum_fold(skb->csum);
633 csum = __skb_checksum_complete(skb);
634 skb->ip_summed = CHECKSUM_COMPLETE;
643 parse_gre_header(struct iphdr *iph, __be16 *flags, __be32 *key)
645 /* IP and ICMP protocol handlers check that the IHL is valid. */
646 __be16 *flagsp = (__be16 *)((u8 *)iph + (iph->ihl << 2));
647 __be16 *protocol = flagsp + 1;
648 __be32 *options = (__be32 *)(protocol + 1);
653 if (*flags & (GRE_VERSION | GRE_ROUTING))
656 if (*protocol != htons(ETH_P_TEB))
659 hdr_len = GRE_HEADER_SECTION;
661 if (*flags & GRE_CSUM) {
662 hdr_len += GRE_HEADER_SECTION;
666 if (*flags & GRE_KEY) {
667 hdr_len += GRE_HEADER_SECTION;
674 if (*flags & GRE_SEQ)
675 hdr_len += GRE_HEADER_SECTION;
681 ecn_encapsulate(u8 tos, struct sk_buff *skb)
685 if (skb->protocol == htons(ETH_P_IP))
686 inner = ((struct iphdr *)skb_network_header(skb))->tos;
687 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
688 else if (skb->protocol == htons(ETH_P_IPV6))
689 inner = ipv6_get_dsfield((struct ipv6hdr *)skb_network_header(skb));
694 return INET_ECN_encapsulate(tos, inner);
698 ecn_decapsulate(u8 tos, struct sk_buff *skb)
700 if (INET_ECN_is_ce(tos)) {
701 __be16 protocol = skb->protocol;
702 unsigned int nw_header = skb_network_header(skb) - skb->data;
704 if (skb->protocol == htons(ETH_P_8021Q)) {
705 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
708 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
709 nw_header += VLAN_HLEN;
712 if (protocol == htons(ETH_P_IP)) {
713 if (unlikely(!pskb_may_pull(skb, nw_header
714 + sizeof(struct iphdr))))
717 IP_ECN_set_ce((struct iphdr *)(nw_header + skb->data));
719 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
720 else if (protocol == htons(ETH_P_IPV6)) {
721 if (unlikely(!pskb_may_pull(skb, nw_header
722 + sizeof(struct ipv6hdr))))
725 IP6_ECN_set_ce((struct ipv6hdr *)(nw_header
732 static struct sk_buff *
733 handle_gso(struct sk_buff *skb)
735 if (skb_is_gso(skb)) {
736 struct sk_buff *nskb = skb_gso_segment(skb, 0);
746 handle_csum_offload(struct sk_buff *skb)
748 if (skb->ip_summed == CHECKSUM_PARTIAL)
749 return skb_checksum_help(skb);
751 skb->ip_summed = CHECKSUM_NONE;
756 /* Called with rcu_read_lock. */
758 gre_err(struct sk_buff *skb, u32 info)
761 const struct mutable_config *mutable;
762 const int type = icmp_hdr(skb)->type;
763 const int code = icmp_hdr(skb)->code;
764 int mtu = ntohs(icmp_hdr(skb)->un.frag.mtu);
769 int tunnel_hdr_len, tot_hdr_len;
770 unsigned int orig_mac_header;
771 unsigned int orig_nw_header;
773 if (type != ICMP_DEST_UNREACH || code != ICMP_FRAG_NEEDED)
776 /* The mimimum size packet that we would actually be able to process:
777 * encapsulating IP header, minimum GRE header, Ethernet header,
778 * inner IPv4 header. */
779 if (!pskb_may_pull(skb, sizeof(struct iphdr) + GRE_HEADER_SECTION +
780 ETH_HLEN + sizeof(struct iphdr)))
783 iph = (struct iphdr *)skb->data;
785 tunnel_hdr_len = parse_gre_header(iph, &flags, &key);
786 if (tunnel_hdr_len < 0)
789 vport = find_port(iph->saddr, iph->daddr, key, FIND_PORT_ANY, &mutable);
793 /* Packets received by this function were previously sent by us, so
794 * any comparisons should be to the output values, not the input.
795 * However, it's not really worth it to have a hash table based on
796 * output keys (especially since ICMP error handling of tunneled packets
797 * isn't that reliable anyways). Therefore, we do a lookup based on the
798 * out key as if it were the in key and then check to see if the input
799 * and output keys are the same. */
800 if (mutable->port_config.in_key != mutable->port_config.out_key)
803 if (!!(mutable->port_config.flags & GRE_F_IN_KEY_MATCH) !=
804 !!(mutable->port_config.flags & GRE_F_OUT_KEY_ACTION))
807 if ((mutable->port_config.flags & GRE_F_OUT_CSUM) && !(flags & GRE_CSUM))
810 tunnel_hdr_len += iph->ihl << 2;
812 orig_mac_header = skb_mac_header(skb) - skb->data;
813 orig_nw_header = skb_network_header(skb) - skb->data;
814 skb_set_mac_header(skb, tunnel_hdr_len);
816 tot_hdr_len = tunnel_hdr_len + ETH_HLEN;
818 skb->protocol = eth_hdr(skb)->h_proto;
819 if (skb->protocol == htons(ETH_P_8021Q)) {
820 tot_hdr_len += VLAN_HLEN;
821 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
824 skb_set_network_header(skb, tot_hdr_len);
827 if (skb->protocol == htons(ETH_P_IP))
828 tot_hdr_len += sizeof(struct iphdr);
829 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
830 else if (skb->protocol == htons(ETH_P_IPV6))
831 tot_hdr_len += sizeof(struct ipv6hdr);
836 if (!pskb_may_pull(skb, tot_hdr_len))
839 if (skb->protocol == htons(ETH_P_IP)) {
840 if (mtu < IP_MIN_MTU) {
841 if (ntohs(ip_hdr(skb)->tot_len) >= IP_MIN_MTU)
848 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
849 else if (skb->protocol == htons(ETH_P_IPV6)) {
850 if (mtu < IPV6_MIN_MTU) {
851 unsigned int packet_length = sizeof(struct ipv6hdr) +
852 ntohs(ipv6_hdr(skb)->payload_len);
854 if (packet_length >= IPV6_MIN_MTU
855 || ntohs(ipv6_hdr(skb)->payload_len) == 0)
863 __pskb_pull(skb, tunnel_hdr_len);
864 send_frag_needed(vport, mutable, skb, mtu, key);
865 skb_push(skb, tunnel_hdr_len);
868 skb_set_mac_header(skb, orig_mac_header);
869 skb_set_network_header(skb, orig_nw_header);
870 skb->protocol = htons(ETH_P_IP);
873 /* Called with rcu_read_lock. */
875 gre_rcv(struct sk_buff *skb)
878 const struct mutable_config *mutable;
884 if (!pskb_may_pull(skb, GRE_HEADER_SECTION + ETH_HLEN))
887 if (!check_checksum(skb))
892 hdr_len = parse_gre_header(iph, &flags, &key);
896 vport = find_port(iph->daddr, iph->saddr, key, FIND_PORT_ANY, &mutable);
898 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
902 if ((mutable->port_config.flags & GRE_F_IN_CSUM) && !(flags & GRE_CSUM)) {
903 vport_record_error(vport, VPORT_E_RX_CRC);
907 if (!pskb_pull(skb, hdr_len) || !pskb_may_pull(skb, ETH_HLEN)) {
908 vport_record_error(vport, VPORT_E_RX_ERROR);
912 skb->pkt_type = PACKET_HOST;
913 skb->protocol = eth_type_trans(skb, skb->dev);
914 skb_postpull_rcsum(skb, skb_transport_header(skb), hdr_len + ETH_HLEN);
919 skb_reset_network_header(skb);
921 ecn_decapsulate(iph->tos, skb);
923 if (mutable->port_config.flags & GRE_F_IN_KEY_MATCH)
924 OVS_CB(skb)->tun_id = key;
926 OVS_CB(skb)->tun_id = 0;
928 skb_push(skb, ETH_HLEN);
929 compute_ip_summed(skb, false);
931 vport_receive(vport, skb);
941 build_packet(struct vport *vport, const struct mutable_config *mutable,
942 struct iphdr *iph, struct rtable *rt, int max_headroom, int mtu,
946 struct iphdr *new_iph;
947 int orig_len = skb->len;
948 __be16 frag_off = iph->frag_off;
950 skb = check_headroom(skb, max_headroom);
951 if (unlikely(IS_ERR(skb)))
954 err = handle_csum_offload(skb);
958 if (skb->protocol == htons(ETH_P_IP)) {
959 struct iphdr *old_iph = ip_hdr(skb);
961 if ((old_iph->frag_off & htons(IP_DF)) &&
962 mtu < ntohs(old_iph->tot_len)) {
963 if (send_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id))
968 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
969 else if (skb->protocol == htons(ETH_P_IPV6)) {
970 unsigned int packet_length = skb->len - ETH_HLEN
971 - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0);
973 /* IPv6 requires PMTUD if the packet is above the minimum MTU. */
974 if (packet_length > IPV6_MIN_MTU)
975 frag_off = htons(IP_DF);
977 if (mtu < packet_length) {
978 if (send_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id))
984 skb_reset_transport_header(skb);
985 new_iph = (struct iphdr *)skb_push(skb, mutable->tunnel_hlen);
986 skb_reset_network_header(skb);
988 memcpy(new_iph, iph, sizeof(struct iphdr));
989 new_iph->frag_off = frag_off;
990 ip_select_ident(new_iph, &rt->u.dst, NULL);
992 create_gre_header(skb, mutable);
994 /* Allow our local IP stack to fragment the outer packet even if the
995 * DF bit is set as a last resort. */
998 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
999 IPCB(skb)->flags = 0;
1001 err = ip_local_out(skb);
1002 if (likely(net_xmit_eval(err) == 0))
1005 vport_record_error(vport, VPORT_E_TX_ERROR);
1012 vport_record_error(vport, VPORT_E_TX_DROPPED);
1018 gre_send(struct vport *vport, struct sk_buff *skb)
1020 struct gre_vport *gre_vport = gre_vport_priv(vport);
1021 const struct mutable_config *mutable = rcu_dereference(gre_vport->mutable);
1023 struct iphdr *old_iph;
1030 /* Validate the protocol headers before we try to use them. */
1031 if (skb->protocol == htons(ETH_P_8021Q)) {
1032 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
1035 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
1036 skb_set_network_header(skb, VLAN_ETH_HLEN);
1039 if (skb->protocol == htons(ETH_P_IP)) {
1040 if (unlikely(!pskb_may_pull(skb, skb_network_header(skb)
1041 + sizeof(struct iphdr) - skb->data)))
1044 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1045 else if (skb->protocol == htons(ETH_P_IPV6)) {
1046 if (unlikely(!pskb_may_pull(skb, skb_network_header(skb)
1047 + sizeof(struct ipv6hdr) - skb->data)))
1051 old_iph = ip_hdr(skb);
1053 iph.tos = mutable->port_config.tos;
1054 if (mutable->port_config.flags & GRE_F_TOS_INHERIT) {
1055 if (skb->protocol == htons(ETH_P_IP))
1056 iph.tos = old_iph->tos;
1057 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1058 else if (skb->protocol == htons(ETH_P_IPV6))
1059 iph.tos = ipv6_get_dsfield(ipv6_hdr(skb));
1062 iph.tos = ecn_encapsulate(iph.tos, skb);
1065 struct flowi fl = { .nl_u = { .ip4_u =
1066 { .daddr = mutable->port_config.daddr,
1067 .saddr = mutable->port_config.saddr,
1068 .tos = RT_TOS(iph.tos) } },
1069 .proto = IPPROTO_GRE };
1071 if (ip_route_output_key(&init_net, &rt, &fl))
1075 iph.ttl = mutable->port_config.ttl;
1076 if (mutable->port_config.flags & GRE_F_TTL_INHERIT) {
1077 if (skb->protocol == htons(ETH_P_IP))
1078 iph.ttl = old_iph->ttl;
1079 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1080 else if (skb->protocol == htons(ETH_P_IPV6))
1081 iph.ttl = ipv6_hdr(skb)->hop_limit;
1085 iph.ttl = dst_metric(&rt->u.dst, RTAX_HOPLIMIT);
1087 iph.frag_off = (mutable->port_config.flags & GRE_F_PMTUD) ? htons(IP_DF) : 0;
1089 mtu = dst_mtu(&rt->u.dst)
1091 - mutable->tunnel_hlen
1092 - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0);
1096 if (skb->protocol == htons(ETH_P_IP)) {
1097 iph.frag_off |= old_iph->frag_off & htons(IP_DF);
1098 mtu = max(mtu, IP_MIN_MTU);
1100 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1101 else if (skb->protocol == htons(ETH_P_IPV6))
1102 mtu = max(mtu, IPV6_MIN_MTU);
1106 iph.ihl = sizeof(struct iphdr) >> 2;
1107 iph.protocol = IPPROTO_GRE;
1108 iph.daddr = rt->rt_dst;
1109 iph.saddr = rt->rt_src;
1114 skb_dst_set(skb, &rt->u.dst);
1116 /* If we are doing GSO on a pskb it is better to make sure that the
1117 * headroom is correct now. We will only have to copy the portion in
1118 * the linear data area and GSO will preserve headroom when it creates
1119 * the segments. This is particularly beneficial on Xen where we get
1120 * lots of GSO pskbs. Conversely, we delay copying if it is just to
1121 * get our own writable clone because GSO may do the copy for us. */
1122 max_headroom = LL_RESERVED_SPACE(rt->u.dst.dev) + rt->u.dst.header_len
1123 + mutable->tunnel_hlen;
1125 if (skb_headroom(skb) < max_headroom) {
1126 skb = check_headroom(skb, max_headroom);
1127 if (unlikely(IS_ERR(skb))) {
1128 vport_record_error(vport, VPORT_E_TX_DROPPED);
1133 forward_ip_summed(skb);
1135 if (unlikely(vswitch_skb_checksum_setup(skb)))
1138 skb = handle_gso(skb);
1139 if (unlikely(IS_ERR(skb))) {
1140 vport_record_error(vport, VPORT_E_TX_DROPPED);
1144 /* Process GSO segments. Try to do any work for the entire packet that
1145 * doesn't involve actually writing to it before this point. */
1148 struct sk_buff *next_skb = skb->next;
1151 orig_len += build_packet(vport, mutable, &iph, rt, max_headroom, mtu, skb);
1160 vport_record_error(vport, VPORT_E_TX_ERROR);
1165 static struct net_protocol gre_protocol_handlers = {
1167 .err_handler = gre_err,
1175 err = inet_add_protocol(&gre_protocol_handlers, IPPROTO_GRE);
1177 printk(KERN_WARNING "openvswitch: cannot register gre protocol handler\n");
1185 tbl_destroy(port_table, NULL);
1186 inet_del_protocol(&gre_protocol_handlers, IPPROTO_GRE);
1190 set_config(const struct vport *cur_vport, struct mutable_config *mutable,
1191 const void __user *uconfig)
1193 const struct vport *old_vport;
1194 const struct mutable_config *old_mutable;
1197 if (copy_from_user(&mutable->port_config, uconfig, sizeof(struct gre_port_config)))
1200 if (mutable->port_config.daddr == 0)
1203 if (mutable->port_config.flags & GRE_F_IN_KEY_MATCH) {
1204 port_type = FIND_PORT_MATCH;
1205 mutable->port_config.in_key = 0;
1207 port_type = FIND_PORT_KEY;
1209 old_vport = find_port(mutable->port_config.saddr,
1210 mutable->port_config.daddr,
1211 mutable->port_config.in_key, port_type,
1214 if (old_vport && old_vport != cur_vport)
1217 if (mutable->port_config.flags & GRE_F_OUT_KEY_ACTION)
1218 mutable->port_config.out_key = 0;
1220 mutable->tunnel_hlen = sizeof(struct iphdr) + GRE_HEADER_SECTION;
1222 if (mutable->port_config.flags & GRE_F_OUT_CSUM)
1223 mutable->tunnel_hlen += GRE_HEADER_SECTION;
1225 if (mutable->port_config.out_key ||
1226 mutable->port_config.flags & GRE_F_OUT_KEY_ACTION)
1227 mutable->tunnel_hlen += GRE_HEADER_SECTION;
1232 static struct vport *
1233 gre_create(const char *name, const void __user *config)
1235 struct vport *vport;
1236 struct gre_vport *gre_vport;
1239 vport = vport_alloc(sizeof(struct gre_vport), &gre_vport_ops);
1240 if (IS_ERR(vport)) {
1241 err = PTR_ERR(vport);
1245 gre_vport = gre_vport_priv(vport);
1247 strcpy(gre_vport->name, name);
1249 gre_vport->mutable = kmalloc(sizeof(struct mutable_config), GFP_KERNEL);
1250 if (!gre_vport->mutable) {
1252 goto error_free_vport;
1255 vport_gen_rand_ether_addr(gre_vport->mutable->eth_addr);
1256 gre_vport->mutable->mtu = ETH_DATA_LEN;
1258 err = set_config(NULL, gre_vport->mutable, config);
1260 goto error_free_mutable;
1262 err = add_port(vport);
1264 goto error_free_mutable;
1269 kfree(gre_vport->mutable);
1273 return ERR_PTR(err);
1277 gre_modify(struct vport *vport, const void __user *config)
1279 struct gre_vport *gre_vport = gre_vport_priv(vport);
1280 struct mutable_config *mutable;
1282 int update_hash = 0;
1284 mutable = kmemdup(gre_vport->mutable, sizeof(struct mutable_config), GFP_KERNEL);
1290 err = set_config(vport, mutable, config);
1294 /* Only remove the port from the hash table if something that would
1295 * affect the lookup has changed. */
1296 if (gre_vport->mutable->port_config.saddr != mutable->port_config.saddr ||
1297 gre_vport->mutable->port_config.daddr != mutable->port_config.daddr ||
1298 gre_vport->mutable->port_config.in_key != mutable->port_config.in_key ||
1299 (gre_vport->mutable->port_config.flags & GRE_F_IN_KEY_MATCH) !=
1300 (mutable->port_config.flags & GRE_F_IN_KEY_MATCH))
1304 /* This update is not atomic but the lookup uses the config, which
1305 * serves as an inherent double check. */
1307 err = del_port(vport);
1312 assign_config_rcu(vport, mutable);
1315 err = add_port(vport);
1329 gre_destroy(struct vport *vport)
1331 struct gre_vport *gre_vport = gre_vport_priv(vport);
1333 const struct mutable_config *old_mutable;
1335 /* Do a hash table lookup to make sure that the port exists. It should
1336 * exist but might not if a modify failed earlier. */
1337 if (gre_vport->mutable->port_config.flags & GRE_F_IN_KEY_MATCH)
1338 port_type = FIND_PORT_MATCH;
1340 port_type = FIND_PORT_KEY;
1342 if (vport == find_port(gre_vport->mutable->port_config.saddr,
1343 gre_vport->mutable->port_config.daddr,
1344 gre_vport->mutable->port_config.in_key, port_type, &old_mutable))
1347 kfree(gre_vport->mutable);
1354 gre_set_mtu(struct vport *vport, int mtu)
1356 struct gre_vport *gre_vport = gre_vport_priv(vport);
1357 struct mutable_config *mutable;
1359 mutable = kmemdup(gre_vport->mutable, sizeof(struct mutable_config), GFP_KERNEL);
1364 assign_config_rcu(vport, mutable);
1370 gre_set_addr(struct vport *vport, const unsigned char *addr)
1372 struct gre_vport *gre_vport = gre_vport_priv(vport);
1373 struct mutable_config *mutable;
1375 mutable = kmemdup(gre_vport->mutable, sizeof(struct mutable_config), GFP_KERNEL);
1379 memcpy(mutable->eth_addr, addr, ETH_ALEN);
1380 assign_config_rcu(vport, mutable);
1387 gre_get_name(const struct vport *vport)
1389 const struct gre_vport *gre_vport = gre_vport_priv(vport);
1390 return gre_vport->name;
1393 static const unsigned char *
1394 gre_get_addr(const struct vport *vport)
1396 const struct gre_vport *gre_vport = gre_vport_priv(vport);
1397 return rcu_dereference(gre_vport->mutable)->eth_addr;
1401 gre_get_mtu(const struct vport *vport)
1403 const struct gre_vport *gre_vport = gre_vport_priv(vport);
1404 return rcu_dereference(gre_vport->mutable)->mtu;
1407 struct vport_ops gre_vport_ops = {
1409 .flags = VPORT_F_GEN_STATS | VPORT_F_TUN_ID,
1412 .create = gre_create,
1413 .modify = gre_modify,
1414 .destroy = gre_destroy,
1415 .set_mtu = gre_set_mtu,
1416 .set_addr = gre_set_addr,
1417 .get_name = gre_get_name,
1418 .get_addr = gre_get_addr,
1419 .get_dev_flags = vport_gen_get_dev_flags,
1420 .is_running = vport_gen_is_running,
1421 .get_operstate = vport_gen_get_operstate,
1422 .get_mtu = gre_get_mtu,