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 #include <linux/if_arp.h>
10 #include <linux/if_ether.h>
12 #include <linux/if_vlan.h>
14 #include <linux/in_route.h>
15 #include <linux/jhash.h>
16 #include <linux/list.h>
17 #include <linux/kernel.h>
18 #include <linux/version.h>
19 #include <linux/workqueue.h>
20 #include <linux/rculist.h>
22 #include <net/dsfield.h>
25 #include <net/inet_ecn.h>
27 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
30 #include <net/route.h>
39 #include "vport-generic.h"
40 #include "vport-internal_dev.h"
42 #ifdef NEED_CACHE_TIMEOUT
44 * On kernels where we can't quickly detect changes in the rest of the system
45 * we use an expiration time to invalidate the cache. A shorter expiration
46 * reduces the length of time that we may potentially blackhole packets while
47 * a longer time increases performance by reducing the frequency that the
48 * cache needs to be rebuilt. A variety of factors may cause the cache to be
49 * invalidated before the expiration time but this is the maximum. The time
50 * is expressed in jiffies.
52 #define MAX_CACHE_EXP HZ
56 * Interval to check for and remove caches that are no longer valid. Caches
57 * are checked for validity before they are used for packet encapsulation and
58 * old caches are removed at that time. However, if no packets are sent through
59 * the tunnel then the cache will never be destroyed. Since it holds
60 * references to a number of system objects, the cache will continue to use
61 * system resources by not allowing those objects to be destroyed. The cache
62 * cleaner is periodically run to free invalid caches. It does not
63 * significantly affect system performance. A lower interval will release
64 * resources faster but will itself consume resources by requiring more frequent
65 * checks. A longer interval may result in messages being printed to the kernel
66 * message buffer about unreleased resources. The interval is expressed in
69 #define CACHE_CLEANER_INTERVAL (5 * HZ)
71 #define CACHE_DATA_ALIGN 16
72 #define PORT_TABLE_SIZE 1024
74 static struct hlist_head *port_table __read_mostly;
75 static int port_table_count;
77 static void cache_cleaner(struct work_struct *work);
78 static DECLARE_DELAYED_WORK(cache_cleaner_wq, cache_cleaner);
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.
85 static unsigned int key_local_remote_ports __read_mostly;
86 static unsigned int key_remote_ports __read_mostly;
87 static unsigned int local_remote_ports __read_mostly;
88 static unsigned int remote_ports __read_mostly;
90 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
91 #define rt_dst(rt) (rt->dst)
93 #define rt_dst(rt) (rt->u.dst)
96 static inline struct vport *tnl_vport_to_vport(const struct tnl_vport *tnl_vport)
98 return vport_from_priv(tnl_vport);
101 /* This is analogous to rtnl_dereference for the tunnel cache. It checks that
102 * cache_lock is held, so it is only for update side code.
104 static inline struct tnl_cache *cache_dereference(struct tnl_vport *tnl_vport)
106 return rcu_dereference_protected(tnl_vport->cache,
107 lockdep_is_held(&tnl_vport->cache_lock));
110 static inline void schedule_cache_cleaner(void)
112 schedule_delayed_work(&cache_cleaner_wq, CACHE_CLEANER_INTERVAL);
115 static void free_cache(struct tnl_cache *cache)
120 flow_put(cache->flow);
121 ip_rt_put(cache->rt);
125 static void free_config_rcu(struct rcu_head *rcu)
127 struct tnl_mutable_config *c = container_of(rcu, struct tnl_mutable_config, rcu);
131 static void free_cache_rcu(struct rcu_head *rcu)
133 struct tnl_cache *c = container_of(rcu, struct tnl_cache, rcu);
137 static void assign_config_rcu(struct vport *vport,
138 struct tnl_mutable_config *new_config)
140 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
141 struct tnl_mutable_config *old_config;
143 old_config = rtnl_dereference(tnl_vport->mutable);
144 rcu_assign_pointer(tnl_vport->mutable, new_config);
145 call_rcu(&old_config->rcu, free_config_rcu);
148 static void assign_cache_rcu(struct vport *vport, struct tnl_cache *new_cache)
150 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
151 struct tnl_cache *old_cache;
153 old_cache = cache_dereference(tnl_vport);
154 rcu_assign_pointer(tnl_vport->cache, new_cache);
157 call_rcu(&old_cache->rcu, free_cache_rcu);
160 static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable)
162 if (mutable->flags & TNL_F_IN_KEY_MATCH) {
163 if (mutable->key.saddr)
164 return &local_remote_ports;
166 return &remote_ports;
168 if (mutable->key.saddr)
169 return &key_local_remote_ports;
171 return &key_remote_ports;
175 static u32 port_hash(const struct port_lookup_key *key)
177 return jhash2((u32*)key, (PORT_KEY_LEN / sizeof(u32)), 0);
180 static inline struct hlist_head *find_bucket(u32 hash)
182 return &port_table[(hash & (PORT_TABLE_SIZE - 1))];
185 static void port_table_add_port(struct vport *vport)
187 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
188 const struct tnl_mutable_config *mutable;
191 if (port_table_count == 0)
192 schedule_cache_cleaner();
194 mutable = rtnl_dereference(tnl_vport->mutable);
195 hash = port_hash(&mutable->key);
196 hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
199 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
202 static void port_table_move_port(struct vport *vport,
203 struct tnl_mutable_config *new_mutable)
205 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
208 hash = port_hash(&new_mutable->key);
209 hlist_del_init_rcu(&tnl_vport->hash_node);
210 hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
212 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
213 assign_config_rcu(vport, new_mutable);
214 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
217 static void port_table_remove_port(struct vport *vport)
219 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
221 hlist_del_init_rcu(&tnl_vport->hash_node);
224 if (port_table_count == 0)
225 cancel_delayed_work_sync(&cache_cleaner_wq);
227 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
230 static struct vport *port_table_lookup(struct port_lookup_key *key,
231 const struct tnl_mutable_config **pmutable)
233 struct hlist_node *n;
234 struct hlist_head *bucket;
235 u32 hash = port_hash(key);
236 struct tnl_vport * tnl_vport;
238 bucket = find_bucket(hash);
240 hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node) {
241 struct tnl_mutable_config *mutable;
243 mutable = rcu_dereference_rtnl(tnl_vport->mutable);
244 if (!memcmp(&mutable->key, key, PORT_KEY_LEN)) {
246 return tnl_vport_to_vport(tnl_vport);
253 struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be64 key,
255 const struct tnl_mutable_config **mutable)
257 struct port_lookup_key lookup;
260 lookup.saddr = saddr;
261 lookup.daddr = daddr;
263 /* First try for exact match on in_key. */
265 lookup.tunnel_type = tunnel_type | TNL_T_KEY_EXACT;
266 if (key_local_remote_ports) {
267 vport = port_table_lookup(&lookup, mutable);
271 if (key_remote_ports) {
273 vport = port_table_lookup(&lookup, mutable);
277 lookup.saddr = saddr;
280 /* Then try matches that wildcard in_key. */
282 lookup.tunnel_type = tunnel_type | TNL_T_KEY_MATCH;
283 if (local_remote_ports) {
284 vport = port_table_lookup(&lookup, mutable);
290 vport = port_table_lookup(&lookup, mutable);
298 static void ecn_decapsulate(struct sk_buff *skb, u8 tos)
300 if (unlikely(INET_ECN_is_ce(tos))) {
301 __be16 protocol = skb->protocol;
303 skb_set_network_header(skb, ETH_HLEN);
305 if (protocol == htons(ETH_P_8021Q)) {
306 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
309 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
310 skb_set_network_header(skb, VLAN_ETH_HLEN);
313 if (protocol == htons(ETH_P_IP)) {
314 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
315 + sizeof(struct iphdr))))
318 IP_ECN_set_ce(ip_hdr(skb));
320 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
321 else if (protocol == htons(ETH_P_IPV6)) {
322 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
323 + sizeof(struct ipv6hdr))))
326 IP6_ECN_set_ce(ipv6_hdr(skb));
333 * tnl_rcv - ingress point for generic tunnel code
335 * @vport: port this packet was received on
336 * @skb: received packet
337 * @tos: ToS from encapsulating IP packet, used to copy ECN bits
339 * Must be called with rcu_read_lock.
341 * Packets received by this function are in the following state:
342 * - skb->data points to the inner Ethernet header.
343 * - The inner Ethernet header is in the linear data area.
344 * - skb->csum does not include the inner Ethernet header.
345 * - The layer pointers are undefined.
347 void tnl_rcv(struct vport *vport, struct sk_buff *skb, u8 tos)
351 skb_reset_mac_header(skb);
354 if (likely(ntohs(eh->h_proto) >= 1536))
355 skb->protocol = eh->h_proto;
357 skb->protocol = htons(ETH_P_802_2);
361 skb_clear_rxhash(skb);
364 ecn_decapsulate(skb, tos);
365 vlan_set_tci(skb, 0);
367 if (unlikely(compute_ip_summed(skb, false))) {
372 vport_receive(vport, skb);
375 static bool check_ipv4_address(__be32 addr)
377 if (ipv4_is_multicast(addr) || ipv4_is_lbcast(addr)
378 || ipv4_is_loopback(addr) || ipv4_is_zeronet(addr))
384 static bool ipv4_should_icmp(struct sk_buff *skb)
386 struct iphdr *old_iph = ip_hdr(skb);
388 /* Don't respond to L2 broadcast. */
389 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
392 /* Don't respond to L3 broadcast or invalid addresses. */
393 if (!check_ipv4_address(old_iph->daddr) ||
394 !check_ipv4_address(old_iph->saddr))
397 /* Only respond to the first fragment. */
398 if (old_iph->frag_off & htons(IP_OFFSET))
401 /* Don't respond to ICMP error messages. */
402 if (old_iph->protocol == IPPROTO_ICMP) {
403 u8 icmp_type, *icmp_typep;
405 icmp_typep = skb_header_pointer(skb, (u8 *)old_iph +
406 (old_iph->ihl << 2) +
407 offsetof(struct icmphdr, type) -
408 skb->data, sizeof(icmp_type),
414 if (*icmp_typep > NR_ICMP_TYPES
415 || (*icmp_typep <= ICMP_PARAMETERPROB
416 && *icmp_typep != ICMP_ECHOREPLY
417 && *icmp_typep != ICMP_ECHO))
424 static void ipv4_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
425 unsigned int mtu, unsigned int payload_length)
427 struct iphdr *iph, *old_iph = ip_hdr(skb);
428 struct icmphdr *icmph;
431 iph = (struct iphdr *)skb_put(nskb, sizeof(struct iphdr));
432 icmph = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr));
433 payload = skb_put(nskb, payload_length);
437 iph->ihl = sizeof(struct iphdr) >> 2;
438 iph->tos = (old_iph->tos & IPTOS_TOS_MASK) |
439 IPTOS_PREC_INTERNETCONTROL;
440 iph->tot_len = htons(sizeof(struct iphdr)
441 + sizeof(struct icmphdr)
443 get_random_bytes(&iph->id, sizeof(iph->id));
446 iph->protocol = IPPROTO_ICMP;
447 iph->daddr = old_iph->saddr;
448 iph->saddr = old_iph->daddr;
453 icmph->type = ICMP_DEST_UNREACH;
454 icmph->code = ICMP_FRAG_NEEDED;
455 icmph->un.gateway = htonl(mtu);
458 nskb->csum = csum_partial((u8 *)icmph, sizeof(struct icmphdr), 0);
459 nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_iph - skb->data,
460 payload, payload_length,
462 icmph->checksum = csum_fold(nskb->csum);
465 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
466 static bool ipv6_should_icmp(struct sk_buff *skb)
468 struct ipv6hdr *old_ipv6h = ipv6_hdr(skb);
470 int payload_off = (u8 *)(old_ipv6h + 1) - skb->data;
471 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
473 /* Check source address is valid. */
474 addr_type = ipv6_addr_type(&old_ipv6h->saddr);
475 if (addr_type & IPV6_ADDR_MULTICAST || addr_type == IPV6_ADDR_ANY)
478 /* Don't reply to unspecified addresses. */
479 if (ipv6_addr_type(&old_ipv6h->daddr) == IPV6_ADDR_ANY)
482 /* Don't respond to ICMP error messages. */
483 payload_off = ipv6_skip_exthdr(skb, payload_off, &nexthdr);
487 if (nexthdr == NEXTHDR_ICMP) {
488 u8 icmp_type, *icmp_typep;
490 icmp_typep = skb_header_pointer(skb, payload_off +
491 offsetof(struct icmp6hdr,
493 sizeof(icmp_type), &icmp_type);
495 if (!icmp_typep || !(*icmp_typep & ICMPV6_INFOMSG_MASK))
502 static void ipv6_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
503 unsigned int mtu, unsigned int payload_length)
505 struct ipv6hdr *ipv6h, *old_ipv6h = ipv6_hdr(skb);
506 struct icmp6hdr *icmp6h;
509 ipv6h = (struct ipv6hdr *)skb_put(nskb, sizeof(struct ipv6hdr));
510 icmp6h = (struct icmp6hdr *)skb_put(nskb, sizeof(struct icmp6hdr));
511 payload = skb_put(nskb, payload_length);
516 memset(&ipv6h->flow_lbl, 0, sizeof(ipv6h->flow_lbl));
517 ipv6h->payload_len = htons(sizeof(struct icmp6hdr)
519 ipv6h->nexthdr = NEXTHDR_ICMP;
520 ipv6h->hop_limit = IPV6_DEFAULT_HOPLIMIT;
521 ipv6_addr_copy(&ipv6h->daddr, &old_ipv6h->saddr);
522 ipv6_addr_copy(&ipv6h->saddr, &old_ipv6h->daddr);
525 icmp6h->icmp6_type = ICMPV6_PKT_TOOBIG;
526 icmp6h->icmp6_code = 0;
527 icmp6h->icmp6_cksum = 0;
528 icmp6h->icmp6_mtu = htonl(mtu);
530 nskb->csum = csum_partial((u8 *)icmp6h, sizeof(struct icmp6hdr), 0);
531 nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_ipv6h - skb->data,
532 payload, payload_length,
534 icmp6h->icmp6_cksum = csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
535 sizeof(struct icmp6hdr)
537 ipv6h->nexthdr, nskb->csum);
541 bool tnl_frag_needed(struct vport *vport, const struct tnl_mutable_config *mutable,
542 struct sk_buff *skb, unsigned int mtu, __be64 flow_key)
544 unsigned int eth_hdr_len = ETH_HLEN;
545 unsigned int total_length = 0, header_length = 0, payload_length;
546 struct ethhdr *eh, *old_eh = eth_hdr(skb);
547 struct sk_buff *nskb;
550 if (skb->protocol == htons(ETH_P_IP)) {
551 if (mtu < IP_MIN_MTU)
554 if (!ipv4_should_icmp(skb))
557 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
558 else if (skb->protocol == htons(ETH_P_IPV6)) {
559 if (mtu < IPV6_MIN_MTU)
563 * In theory we should do PMTUD on IPv6 multicast messages but
564 * we don't have an address to send from so just fragment.
566 if (ipv6_addr_type(&ipv6_hdr(skb)->daddr) & IPV6_ADDR_MULTICAST)
569 if (!ipv6_should_icmp(skb))
577 if (old_eh->h_proto == htons(ETH_P_8021Q))
578 eth_hdr_len = VLAN_ETH_HLEN;
580 payload_length = skb->len - eth_hdr_len;
581 if (skb->protocol == htons(ETH_P_IP)) {
582 header_length = sizeof(struct iphdr) + sizeof(struct icmphdr);
583 total_length = min_t(unsigned int, header_length +
584 payload_length, 576);
586 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
588 header_length = sizeof(struct ipv6hdr) +
589 sizeof(struct icmp6hdr);
590 total_length = min_t(unsigned int, header_length +
591 payload_length, IPV6_MIN_MTU);
595 payload_length = total_length - header_length;
597 nskb = dev_alloc_skb(NET_IP_ALIGN + eth_hdr_len + header_length +
602 skb_reserve(nskb, NET_IP_ALIGN);
604 /* Ethernet / VLAN */
605 eh = (struct ethhdr *)skb_put(nskb, eth_hdr_len);
606 memcpy(eh->h_dest, old_eh->h_source, ETH_ALEN);
607 memcpy(eh->h_source, mutable->eth_addr, ETH_ALEN);
608 nskb->protocol = eh->h_proto = old_eh->h_proto;
609 if (old_eh->h_proto == htons(ETH_P_8021Q)) {
610 struct vlan_ethhdr *vh = (struct vlan_ethhdr *)eh;
612 vh->h_vlan_TCI = vlan_eth_hdr(skb)->h_vlan_TCI;
613 vh->h_vlan_encapsulated_proto = skb->protocol;
615 vlan_set_tci(nskb, vlan_get_tci(skb));
616 skb_reset_mac_header(nskb);
619 if (skb->protocol == htons(ETH_P_IP))
620 ipv4_build_icmp(skb, nskb, mtu, payload_length);
621 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
623 ipv6_build_icmp(skb, nskb, mtu, payload_length);
627 * Assume that flow based keys are symmetric with respect to input
628 * and output and use the key that we were going to put on the
629 * outgoing packet for the fake received packet. If the keys are
630 * not symmetric then PMTUD needs to be disabled since we won't have
631 * any way of synthesizing packets.
633 if ((mutable->flags & (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION)) ==
634 (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION))
635 OVS_CB(nskb)->tun_id = flow_key;
637 if (unlikely(compute_ip_summed(nskb, false))) {
642 vport_receive(vport, nskb);
647 static bool check_mtu(struct sk_buff *skb,
649 const struct tnl_mutable_config *mutable,
650 const struct rtable *rt, __be16 *frag_offp)
652 bool df_inherit = mutable->flags & TNL_F_DF_INHERIT;
653 bool pmtud = mutable->flags & TNL_F_PMTUD;
654 __be16 frag_off = mutable->flags & TNL_F_DF_DEFAULT ? htons(IP_DF) : 0;
656 unsigned int packet_length = skb->len - ETH_HLEN;
658 /* Allow for one level of tagging in the packet length. */
659 if (!vlan_tx_tag_present(skb) &&
660 eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
661 packet_length -= VLAN_HLEN;
666 /* The tag needs to go in packet regardless of where it
667 * currently is, so subtract it from the MTU.
669 if (vlan_tx_tag_present(skb) ||
670 eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
671 vlan_header = VLAN_HLEN;
673 mtu = dst_mtu(&rt_dst(rt))
675 - mutable->tunnel_hlen
679 if (skb->protocol == htons(ETH_P_IP)) {
680 struct iphdr *iph = ip_hdr(skb);
683 frag_off = iph->frag_off & htons(IP_DF);
685 if (pmtud && iph->frag_off & htons(IP_DF)) {
686 mtu = max(mtu, IP_MIN_MTU);
688 if (packet_length > mtu &&
689 tnl_frag_needed(vport, mutable, skb, mtu,
690 OVS_CB(skb)->tun_id))
694 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
695 else if (skb->protocol == htons(ETH_P_IPV6)) {
696 /* IPv6 requires end hosts to do fragmentation
697 * if the packet is above the minimum MTU.
699 if (df_inherit && packet_length > IPV6_MIN_MTU)
700 frag_off = htons(IP_DF);
703 mtu = max(mtu, IPV6_MIN_MTU);
705 if (packet_length > mtu &&
706 tnl_frag_needed(vport, mutable, skb, mtu,
707 OVS_CB(skb)->tun_id))
713 *frag_offp = frag_off;
717 static void create_tunnel_header(const struct vport *vport,
718 const struct tnl_mutable_config *mutable,
719 const struct rtable *rt, void *header)
721 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
722 struct iphdr *iph = header;
725 iph->ihl = sizeof(struct iphdr) >> 2;
726 iph->frag_off = htons(IP_DF);
727 iph->protocol = tnl_vport->tnl_ops->ipproto;
728 iph->tos = mutable->tos;
729 iph->daddr = rt->rt_dst;
730 iph->saddr = rt->rt_src;
731 iph->ttl = mutable->ttl;
733 iph->ttl = ip4_dst_hoplimit(&rt_dst(rt));
735 tnl_vport->tnl_ops->build_header(vport, mutable, iph + 1);
738 static inline void *get_cached_header(const struct tnl_cache *cache)
740 return (void *)cache + ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN);
743 static inline bool check_cache_valid(const struct tnl_cache *cache,
744 const struct tnl_mutable_config *mutable)
747 #ifdef NEED_CACHE_TIMEOUT
748 time_before(jiffies, cache->expiration) &&
751 atomic_read(&init_net.ipv4.rt_genid) == cache->rt->rt_genid &&
754 rt_dst(cache->rt).hh->hh_lock.sequence == cache->hh_seq &&
756 mutable->seq == cache->mutable_seq &&
757 (!is_internal_dev(rt_dst(cache->rt).dev) ||
758 (cache->flow && !cache->flow->dead));
761 static void __cache_cleaner(struct tnl_vport *tnl_vport)
763 const struct tnl_mutable_config *mutable =
764 rcu_dereference(tnl_vport->mutable);
765 const struct tnl_cache *cache = rcu_dereference(tnl_vport->cache);
767 if (cache && !check_cache_valid(cache, mutable) &&
768 spin_trylock_bh(&tnl_vport->cache_lock)) {
769 assign_cache_rcu(tnl_vport_to_vport(tnl_vport), NULL);
770 spin_unlock_bh(&tnl_vport->cache_lock);
774 static void cache_cleaner(struct work_struct *work)
778 schedule_cache_cleaner();
781 for (i = 0; i < PORT_TABLE_SIZE; i++) {
782 struct hlist_node *n;
783 struct hlist_head *bucket;
784 struct tnl_vport *tnl_vport;
786 bucket = &port_table[i];
787 hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node)
788 __cache_cleaner(tnl_vport);
793 static inline void create_eth_hdr(struct tnl_cache *cache,
794 const struct rtable *rt)
796 void *cache_data = get_cached_header(cache);
797 int hh_len = rt_dst(rt).hh->hh_len;
798 int hh_off = HH_DATA_ALIGN(rt_dst(rt).hh->hh_len) - hh_len;
804 hh_seq = read_seqbegin(&rt_dst(rt).hh->hh_lock);
805 memcpy(cache_data, (void *)rt_dst(rt).hh->hh_data + hh_off, hh_len);
806 } while (read_seqretry(&rt_dst(rt).hh->hh_lock, hh_seq));
808 cache->hh_seq = hh_seq;
810 read_lock(&rt_dst(rt).hh->hh_lock);
811 memcpy(cache_data, (void *)rt_dst(rt).hh->hh_data + hh_off, hh_len);
812 read_unlock(&rt_dst(rt).hh->hh_lock);
816 static struct tnl_cache *build_cache(struct vport *vport,
817 const struct tnl_mutable_config *mutable,
820 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
821 struct tnl_cache *cache;
825 if (!(mutable->flags & TNL_F_HDR_CACHE))
829 * If there is no entry in the ARP cache or if this device does not
830 * support hard header caching just fall back to the IP stack.
836 * If lock is contended fall back to directly building the header.
837 * We're not going to help performance by sitting here spinning.
839 if (!spin_trylock(&tnl_vport->cache_lock))
842 cache = cache_dereference(tnl_vport);
843 if (check_cache_valid(cache, mutable))
848 cache_len = rt_dst(rt).hh->hh_len + mutable->tunnel_hlen;
850 cache = kzalloc(ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN) +
851 cache_len, GFP_ATOMIC);
855 cache->len = cache_len;
857 create_eth_hdr(cache, rt);
858 cache_data = get_cached_header(cache) + rt_dst(rt).hh->hh_len;
860 create_tunnel_header(vport, mutable, rt, cache_data);
862 cache->mutable_seq = mutable->seq;
864 #ifdef NEED_CACHE_TIMEOUT
865 cache->expiration = jiffies + tnl_vport->cache_exp_interval;
868 if (is_internal_dev(rt_dst(rt).dev)) {
869 struct sw_flow_key flow_key;
870 struct vport *dst_vport;
875 struct sw_flow *flow;
877 dst_vport = internal_dev_get_vport(rt_dst(rt).dev);
881 skb = alloc_skb(cache->len, GFP_ATOMIC);
885 __skb_put(skb, cache->len);
886 memcpy(skb->data, get_cached_header(cache), cache->len);
888 err = flow_extract(skb, dst_vport->port_no, &flow_key,
889 &flow_key_len, &is_frag);
895 flow = flow_tbl_lookup(rcu_dereference(dst_vport->dp->table),
896 &flow_key, flow_key_len);
904 assign_cache_rcu(vport, cache);
907 spin_unlock(&tnl_vport->cache_lock);
912 static struct rtable *find_route(struct vport *vport,
913 const struct tnl_mutable_config *mutable,
914 u8 tos, struct tnl_cache **cache)
916 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
917 struct tnl_cache *cur_cache = rcu_dereference(tnl_vport->cache);
922 if (likely(tos == mutable->tos && check_cache_valid(cur_cache, mutable))) {
924 return cur_cache->rt;
927 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
928 struct flowi fl = { .nl_u = { .ip4_u =
929 { .daddr = mutable->key.daddr,
930 .saddr = mutable->key.saddr,
932 .proto = tnl_vport->tnl_ops->ipproto };
934 if (unlikely(ip_route_output_key(&init_net, &rt, &fl)))
937 struct flowi4 fl = { .daddr = mutable->key.daddr,
938 .saddr = mutable->key.saddr,
940 .flowi4_proto = tnl_vport->tnl_ops->ipproto };
942 rt = ip_route_output_key(&init_net, &fl);
947 if (likely(tos == mutable->tos))
948 *cache = build_cache(vport, mutable, rt);
954 static inline bool need_linearize(const struct sk_buff *skb)
958 if (unlikely(skb_shinfo(skb)->frag_list))
962 * Generally speaking we should linearize if there are paged frags.
963 * However, if all of the refcounts are 1 we know nobody else can
964 * change them from underneath us and we can skip the linearization.
966 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
967 if (unlikely(page_count(skb_shinfo(skb)->frags[i].page) > 1))
973 static struct sk_buff *handle_offloads(struct sk_buff *skb,
974 const struct tnl_mutable_config *mutable,
975 const struct rtable *rt)
980 min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
981 + mutable->tunnel_hlen
982 + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
984 if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
985 int head_delta = SKB_DATA_ALIGN(min_headroom -
988 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
994 forward_ip_summed(skb, true);
996 if (skb_is_gso(skb)) {
997 struct sk_buff *nskb;
999 nskb = skb_gso_segment(skb, 0);
1002 err = PTR_ERR(nskb);
1008 } else if (get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
1009 /* Pages aren't locked and could change at any time.
1010 * If this happens after we compute the checksum, the
1011 * checksum will be wrong. We linearize now to avoid
1014 if (unlikely(need_linearize(skb))) {
1015 err = __skb_linearize(skb);
1020 err = skb_checksum_help(skb);
1025 set_ip_summed(skb, OVS_CSUM_NONE);
1032 return ERR_PTR(err);
1035 static int send_frags(struct sk_buff *skb,
1036 const struct tnl_mutable_config *mutable)
1042 struct sk_buff *next = skb->next;
1043 int frag_len = skb->len - mutable->tunnel_hlen;
1047 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
1049 err = ip_local_out(skb);
1051 if (unlikely(net_xmit_eval(err)))
1053 sent_len += frag_len;
1060 * There's no point in continuing to send fragments once one has been
1061 * dropped so just free the rest. This may help improve the congestion
1062 * that caused the first packet to be dropped.
1064 tnl_free_linked_skbs(skb);
1068 int tnl_send(struct vport *vport, struct sk_buff *skb)
1070 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1071 const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
1073 enum vport_err_type err = VPORT_E_TX_ERROR;
1075 struct dst_entry *unattached_dst = NULL;
1076 struct tnl_cache *cache;
1078 __be16 frag_off = 0;
1083 /* Validate the protocol headers before we try to use them. */
1084 if (skb->protocol == htons(ETH_P_8021Q) &&
1085 !vlan_tx_tag_present(skb)) {
1086 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
1089 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
1090 skb_set_network_header(skb, VLAN_ETH_HLEN);
1093 if (skb->protocol == htons(ETH_P_IP)) {
1094 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1095 + sizeof(struct iphdr))))
1098 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1099 else if (skb->protocol == htons(ETH_P_IPV6)) {
1100 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1101 + sizeof(struct ipv6hdr))))
1107 if (skb->protocol == htons(ETH_P_IP))
1108 inner_tos = ip_hdr(skb)->tos;
1109 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1110 else if (skb->protocol == htons(ETH_P_IPV6))
1111 inner_tos = ipv6_get_dsfield(ipv6_hdr(skb));
1116 if (mutable->flags & TNL_F_TOS_INHERIT)
1121 tos = INET_ECN_encapsulate(tos, inner_tos);
1124 rt = find_route(vport, mutable, tos, &cache);
1127 if (unlikely(!cache))
1128 unattached_dst = &rt_dst(rt);
1134 skb_clear_rxhash(skb);
1137 skb = handle_offloads(skb, mutable, rt);
1142 if (unlikely(!check_mtu(skb, vport, mutable, rt, &frag_off))) {
1143 err = VPORT_E_TX_DROPPED;
1148 * If we are over the MTU, allow the IP stack to handle fragmentation.
1149 * Fragmentation is a slow path anyways.
1151 if (unlikely(skb->len + mutable->tunnel_hlen > dst_mtu(&rt_dst(rt)) &&
1153 unattached_dst = &rt_dst(rt);
1154 dst_hold(unattached_dst);
1161 ttl = ip4_dst_hoplimit(&rt_dst(rt));
1163 if (mutable->flags & TNL_F_TTL_INHERIT) {
1164 if (skb->protocol == htons(ETH_P_IP))
1165 ttl = ip_hdr(skb)->ttl;
1166 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1167 else if (skb->protocol == htons(ETH_P_IPV6))
1168 ttl = ipv6_hdr(skb)->hop_limit;
1174 struct sk_buff *next_skb = skb->next;
1177 if (unlikely(vlan_deaccel_tag(skb)))
1180 if (likely(cache)) {
1181 skb_push(skb, cache->len);
1182 memcpy(skb->data, get_cached_header(cache), cache->len);
1183 skb_reset_mac_header(skb);
1184 skb_set_network_header(skb, rt_dst(rt).hh->hh_len);
1187 skb_push(skb, mutable->tunnel_hlen);
1188 create_tunnel_header(vport, mutable, rt, skb->data);
1189 skb_reset_network_header(skb);
1192 skb_dst_set(skb, dst_clone(unattached_dst));
1194 skb_dst_set(skb, unattached_dst);
1195 unattached_dst = NULL;
1198 skb_set_transport_header(skb, skb_network_offset(skb) + sizeof(struct iphdr));
1203 iph->frag_off = frag_off;
1204 ip_select_ident(iph, &rt_dst(rt), NULL);
1206 skb = tnl_vport->tnl_ops->update_header(vport, mutable, &rt_dst(rt), skb);
1210 if (likely(cache)) {
1211 int orig_len = skb->len - cache->len;
1212 struct vport *cache_vport = internal_dev_get_vport(rt_dst(rt).dev);
1214 skb->protocol = htons(ETH_P_IP);
1216 iph->tot_len = htons(skb->len - skb_network_offset(skb));
1220 if (unlikely(compute_ip_summed(skb, true))) {
1225 OVS_CB(skb)->flow = cache->flow;
1226 vport_receive(cache_vport, skb);
1227 sent_len += orig_len;
1231 skb->dev = rt_dst(rt).dev;
1232 xmit_err = dev_queue_xmit(skb);
1234 if (likely(net_xmit_eval(xmit_err) == 0))
1235 sent_len += orig_len;
1238 sent_len += send_frags(skb, mutable);
1244 if (unlikely(sent_len == 0))
1245 vport_record_error(vport, VPORT_E_TX_DROPPED);
1250 tnl_free_linked_skbs(skb);
1252 vport_record_error(vport, err);
1254 dst_release(unattached_dst);
1258 static const struct nla_policy tnl_policy[OVS_TUNNEL_ATTR_MAX + 1] = {
1259 [OVS_TUNNEL_ATTR_FLAGS] = { .type = NLA_U32 },
1260 [OVS_TUNNEL_ATTR_DST_IPV4] = { .type = NLA_U32 },
1261 [OVS_TUNNEL_ATTR_SRC_IPV4] = { .type = NLA_U32 },
1262 [OVS_TUNNEL_ATTR_OUT_KEY] = { .type = NLA_U64 },
1263 [OVS_TUNNEL_ATTR_IN_KEY] = { .type = NLA_U64 },
1264 [OVS_TUNNEL_ATTR_TOS] = { .type = NLA_U8 },
1265 [OVS_TUNNEL_ATTR_TTL] = { .type = NLA_U8 },
1268 /* Sets OVS_TUNNEL_ATTR_* fields in 'mutable', which must initially be zeroed. */
1269 static int tnl_set_config(struct nlattr *options, const struct tnl_ops *tnl_ops,
1270 const struct vport *cur_vport,
1271 struct tnl_mutable_config *mutable)
1273 const struct vport *old_vport;
1274 const struct tnl_mutable_config *old_mutable;
1275 struct nlattr *a[OVS_TUNNEL_ATTR_MAX + 1];
1281 err = nla_parse_nested(a, OVS_TUNNEL_ATTR_MAX, options, tnl_policy);
1285 if (!a[OVS_TUNNEL_ATTR_FLAGS] || !a[OVS_TUNNEL_ATTR_DST_IPV4])
1288 mutable->flags = nla_get_u32(a[OVS_TUNNEL_ATTR_FLAGS]) & TNL_F_PUBLIC;
1290 if (a[OVS_TUNNEL_ATTR_SRC_IPV4])
1291 mutable->key.saddr = nla_get_be32(a[OVS_TUNNEL_ATTR_SRC_IPV4]);
1292 mutable->key.daddr = nla_get_be32(a[OVS_TUNNEL_ATTR_DST_IPV4]);
1294 if (a[OVS_TUNNEL_ATTR_TOS]) {
1295 mutable->tos = nla_get_u8(a[OVS_TUNNEL_ATTR_TOS]);
1296 if (mutable->tos != RT_TOS(mutable->tos))
1300 if (a[OVS_TUNNEL_ATTR_TTL])
1301 mutable->ttl = nla_get_u8(a[OVS_TUNNEL_ATTR_TTL]);
1303 mutable->key.tunnel_type = tnl_ops->tunnel_type;
1304 if (!a[OVS_TUNNEL_ATTR_IN_KEY]) {
1305 mutable->key.tunnel_type |= TNL_T_KEY_MATCH;
1306 mutable->flags |= TNL_F_IN_KEY_MATCH;
1308 mutable->key.tunnel_type |= TNL_T_KEY_EXACT;
1309 mutable->key.in_key = nla_get_be64(a[OVS_TUNNEL_ATTR_IN_KEY]);
1312 if (!a[OVS_TUNNEL_ATTR_OUT_KEY])
1313 mutable->flags |= TNL_F_OUT_KEY_ACTION;
1315 mutable->out_key = nla_get_be64(a[OVS_TUNNEL_ATTR_OUT_KEY]);
1317 mutable->tunnel_hlen = tnl_ops->hdr_len(mutable);
1318 if (mutable->tunnel_hlen < 0)
1319 return mutable->tunnel_hlen;
1321 mutable->tunnel_hlen += sizeof(struct iphdr);
1323 old_vport = port_table_lookup(&mutable->key, &old_mutable);
1324 if (old_vport && old_vport != cur_vport)
1330 struct vport *tnl_create(const struct vport_parms *parms,
1331 const struct vport_ops *vport_ops,
1332 const struct tnl_ops *tnl_ops)
1334 struct vport *vport;
1335 struct tnl_vport *tnl_vport;
1336 struct tnl_mutable_config *mutable;
1337 int initial_frag_id;
1340 vport = vport_alloc(sizeof(struct tnl_vport), vport_ops, parms);
1341 if (IS_ERR(vport)) {
1342 err = PTR_ERR(vport);
1346 tnl_vport = tnl_vport_priv(vport);
1348 strcpy(tnl_vport->name, parms->name);
1349 tnl_vport->tnl_ops = tnl_ops;
1351 mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1354 goto error_free_vport;
1357 vport_gen_rand_ether_addr(mutable->eth_addr);
1359 get_random_bytes(&initial_frag_id, sizeof(int));
1360 atomic_set(&tnl_vport->frag_id, initial_frag_id);
1362 err = tnl_set_config(parms->options, tnl_ops, NULL, mutable);
1364 goto error_free_mutable;
1366 spin_lock_init(&tnl_vport->cache_lock);
1368 #ifdef NEED_CACHE_TIMEOUT
1369 tnl_vport->cache_exp_interval = MAX_CACHE_EXP -
1370 (net_random() % (MAX_CACHE_EXP / 2));
1373 rcu_assign_pointer(tnl_vport->mutable, mutable);
1375 port_table_add_port(vport);
1383 return ERR_PTR(err);
1386 int tnl_set_options(struct vport *vport, struct nlattr *options)
1388 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1389 const struct tnl_mutable_config *old_mutable;
1390 struct tnl_mutable_config *mutable;
1393 mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1399 /* Copy fields whose values should be retained. */
1400 old_mutable = rtnl_dereference(tnl_vport->mutable);
1401 mutable->seq = old_mutable->seq + 1;
1402 memcpy(mutable->eth_addr, old_mutable->eth_addr, ETH_ALEN);
1404 /* Parse the others configured by userspace. */
1405 err = tnl_set_config(options, tnl_vport->tnl_ops, vport, mutable);
1409 if (port_hash(&mutable->key) != port_hash(&old_mutable->key))
1410 port_table_move_port(vport, mutable);
1420 int tnl_get_options(const struct vport *vport, struct sk_buff *skb)
1422 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1423 const struct tnl_mutable_config *mutable = rcu_dereference_rtnl(tnl_vport->mutable);
1425 NLA_PUT_U32(skb, OVS_TUNNEL_ATTR_FLAGS, mutable->flags & TNL_F_PUBLIC);
1426 NLA_PUT_BE32(skb, OVS_TUNNEL_ATTR_DST_IPV4, mutable->key.daddr);
1428 if (!(mutable->flags & TNL_F_IN_KEY_MATCH))
1429 NLA_PUT_BE64(skb, OVS_TUNNEL_ATTR_IN_KEY, mutable->key.in_key);
1430 if (!(mutable->flags & TNL_F_OUT_KEY_ACTION))
1431 NLA_PUT_BE64(skb, OVS_TUNNEL_ATTR_OUT_KEY, mutable->out_key);
1432 if (mutable->key.saddr)
1433 NLA_PUT_BE32(skb, OVS_TUNNEL_ATTR_SRC_IPV4, mutable->key.saddr);
1435 NLA_PUT_U8(skb, OVS_TUNNEL_ATTR_TOS, mutable->tos);
1437 NLA_PUT_U8(skb, OVS_TUNNEL_ATTR_TTL, mutable->ttl);
1445 static void free_port_rcu(struct rcu_head *rcu)
1447 struct tnl_vport *tnl_vport = container_of(rcu,
1448 struct tnl_vport, rcu);
1450 free_cache((struct tnl_cache __force *)tnl_vport->cache);
1451 kfree((struct tnl_mutable __force *)tnl_vport->mutable);
1452 vport_free(tnl_vport_to_vport(tnl_vport));
1455 void tnl_destroy(struct vport *vport)
1457 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1458 const struct tnl_mutable_config *mutable;
1460 mutable = rtnl_dereference(tnl_vport->mutable);
1461 port_table_remove_port(vport);
1462 call_rcu(&tnl_vport->rcu, free_port_rcu);
1465 int tnl_set_addr(struct vport *vport, const unsigned char *addr)
1467 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1468 struct tnl_mutable_config *mutable;
1470 mutable = kmemdup(rtnl_dereference(tnl_vport->mutable),
1471 sizeof(struct tnl_mutable_config), GFP_KERNEL);
1475 memcpy(mutable->eth_addr, addr, ETH_ALEN);
1476 assign_config_rcu(vport, mutable);
1481 const char *tnl_get_name(const struct vport *vport)
1483 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1484 return tnl_vport->name;
1487 const unsigned char *tnl_get_addr(const struct vport *vport)
1489 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1490 return rcu_dereference_rtnl(tnl_vport->mutable)->eth_addr;
1493 void tnl_free_linked_skbs(struct sk_buff *skb)
1496 struct sk_buff *next = skb->next;
1506 port_table = kmalloc(PORT_TABLE_SIZE * sizeof(struct hlist_head *),
1511 for (i = 0; i < PORT_TABLE_SIZE; i++)
1512 INIT_HLIST_HEAD(&port_table[i]);
1521 for (i = 0; i < PORT_TABLE_SIZE; i++) {
1522 struct tnl_vport * tnl_vport;
1523 struct hlist_head *hash_head;
1524 struct hlist_node *n;
1526 hash_head = &port_table[i];
1527 hlist_for_each_entry(tnl_vport, n, hash_head, hash_node) {