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56 * Interval to check for and remove caches that are no longer valid.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 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,1,0)
97 static struct hh_cache *rt_hh(struct rtable *rt)
99 struct neighbour *neigh = dst_get_neighbour(&rt->dst);
100 if (!neigh || !(neigh->nud_state & NUD_CONNECTED) ||
106 #define rt_hh(rt) (rt_dst(rt).hh)
109 static inline struct vport *tnl_vport_to_vport(const struct tnl_vport *tnl_vport)
111 return vport_from_priv(tnl_vport);
114 /* This is analogous to rtnl_dereference for the tunnel cache. It checks that
115 * cache_lock is held, so it is only for update side code.
117 static inline struct tnl_cache *cache_dereference(struct tnl_vport *tnl_vport)
119 return rcu_dereference_protected(tnl_vport->cache,
120 lockdep_is_held(&tnl_vport->cache_lock));
123 static inline void schedule_cache_cleaner(void)
125 schedule_delayed_work(&cache_cleaner_wq, CACHE_CLEANER_INTERVAL);
128 static void free_cache(struct tnl_cache *cache)
133 flow_put(cache->flow);
134 ip_rt_put(cache->rt);
138 static void free_config_rcu(struct rcu_head *rcu)
140 struct tnl_mutable_config *c = container_of(rcu, struct tnl_mutable_config, rcu);
144 static void free_cache_rcu(struct rcu_head *rcu)
146 struct tnl_cache *c = container_of(rcu, struct tnl_cache, rcu);
150 static void assign_config_rcu(struct vport *vport,
151 struct tnl_mutable_config *new_config)
153 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
154 struct tnl_mutable_config *old_config;
156 old_config = rtnl_dereference(tnl_vport->mutable);
157 rcu_assign_pointer(tnl_vport->mutable, new_config);
158 call_rcu(&old_config->rcu, free_config_rcu);
161 static void assign_cache_rcu(struct vport *vport, struct tnl_cache *new_cache)
163 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
164 struct tnl_cache *old_cache;
166 old_cache = cache_dereference(tnl_vport);
167 rcu_assign_pointer(tnl_vport->cache, new_cache);
170 call_rcu(&old_cache->rcu, free_cache_rcu);
173 static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable)
175 if (mutable->flags & TNL_F_IN_KEY_MATCH) {
176 if (mutable->key.saddr)
177 return &local_remote_ports;
179 return &remote_ports;
181 if (mutable->key.saddr)
182 return &key_local_remote_ports;
184 return &key_remote_ports;
188 static u32 port_hash(const struct port_lookup_key *key)
190 return jhash2((u32*)key, (PORT_KEY_LEN / sizeof(u32)), 0);
193 static inline struct hlist_head *find_bucket(u32 hash)
195 return &port_table[(hash & (PORT_TABLE_SIZE - 1))];
198 static void port_table_add_port(struct vport *vport)
200 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
201 const struct tnl_mutable_config *mutable;
204 if (port_table_count == 0)
205 schedule_cache_cleaner();
207 mutable = rtnl_dereference(tnl_vport->mutable);
208 hash = port_hash(&mutable->key);
209 hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
212 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
215 static void port_table_move_port(struct vport *vport,
216 struct tnl_mutable_config *new_mutable)
218 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
221 hash = port_hash(&new_mutable->key);
222 hlist_del_init_rcu(&tnl_vport->hash_node);
223 hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
225 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
226 assign_config_rcu(vport, new_mutable);
227 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
230 static void port_table_remove_port(struct vport *vport)
232 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
234 hlist_del_init_rcu(&tnl_vport->hash_node);
237 if (port_table_count == 0)
238 cancel_delayed_work_sync(&cache_cleaner_wq);
240 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
243 static struct vport *port_table_lookup(struct port_lookup_key *key,
244 const struct tnl_mutable_config **pmutable)
246 struct hlist_node *n;
247 struct hlist_head *bucket;
248 u32 hash = port_hash(key);
249 struct tnl_vport * tnl_vport;
251 bucket = find_bucket(hash);
253 hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node) {
254 struct tnl_mutable_config *mutable;
256 mutable = rcu_dereference_rtnl(tnl_vport->mutable);
257 if (!memcmp(&mutable->key, key, PORT_KEY_LEN)) {
259 return tnl_vport_to_vport(tnl_vport);
266 struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be64 key,
268 const struct tnl_mutable_config **mutable)
270 struct port_lookup_key lookup;
273 lookup.saddr = saddr;
274 lookup.daddr = daddr;
276 /* First try for exact match on in_key. */
278 lookup.tunnel_type = tunnel_type | TNL_T_KEY_EXACT;
279 if (key_local_remote_ports) {
280 vport = port_table_lookup(&lookup, mutable);
284 if (key_remote_ports) {
286 vport = port_table_lookup(&lookup, mutable);
290 lookup.saddr = saddr;
293 /* Then try matches that wildcard in_key. */
295 lookup.tunnel_type = tunnel_type | TNL_T_KEY_MATCH;
296 if (local_remote_ports) {
297 vport = port_table_lookup(&lookup, mutable);
303 vport = port_table_lookup(&lookup, mutable);
311 static void ecn_decapsulate(struct sk_buff *skb, u8 tos)
313 if (unlikely(INET_ECN_is_ce(tos))) {
314 __be16 protocol = skb->protocol;
316 skb_set_network_header(skb, ETH_HLEN);
318 if (protocol == htons(ETH_P_8021Q)) {
319 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
322 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
323 skb_set_network_header(skb, VLAN_ETH_HLEN);
326 if (protocol == htons(ETH_P_IP)) {
327 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
328 + sizeof(struct iphdr))))
331 IP_ECN_set_ce(ip_hdr(skb));
333 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
334 else if (protocol == htons(ETH_P_IPV6)) {
335 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
336 + sizeof(struct ipv6hdr))))
339 IP6_ECN_set_ce(ipv6_hdr(skb));
346 * tnl_rcv - ingress point for generic tunnel code
348 * @vport: port this packet was received on
349 * @skb: received packet
350 * @tos: ToS from encapsulating IP packet, used to copy ECN bits
352 * Must be called with rcu_read_lock.
354 * Packets received by this function are in the following state:
355 * - skb->data points to the inner Ethernet header.
356 * - The inner Ethernet header is in the linear data area.
357 * - skb->csum does not include the inner Ethernet header.
358 * - The layer pointers are undefined.
360 void tnl_rcv(struct vport *vport, struct sk_buff *skb, u8 tos)
364 skb_reset_mac_header(skb);
367 if (likely(ntohs(eh->h_proto) >= 1536))
368 skb->protocol = eh->h_proto;
370 skb->protocol = htons(ETH_P_802_2);
374 skb_clear_rxhash(skb);
377 ecn_decapsulate(skb, tos);
378 vlan_set_tci(skb, 0);
380 if (unlikely(compute_ip_summed(skb, false))) {
385 vport_receive(vport, skb);
388 static bool check_ipv4_address(__be32 addr)
390 if (ipv4_is_multicast(addr) || ipv4_is_lbcast(addr)
391 || ipv4_is_loopback(addr) || ipv4_is_zeronet(addr))
397 static bool ipv4_should_icmp(struct sk_buff *skb)
399 struct iphdr *old_iph = ip_hdr(skb);
401 /* Don't respond to L2 broadcast. */
402 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
405 /* Don't respond to L3 broadcast or invalid addresses. */
406 if (!check_ipv4_address(old_iph->daddr) ||
407 !check_ipv4_address(old_iph->saddr))
410 /* Only respond to the first fragment. */
411 if (old_iph->frag_off & htons(IP_OFFSET))
414 /* Don't respond to ICMP error messages. */
415 if (old_iph->protocol == IPPROTO_ICMP) {
416 u8 icmp_type, *icmp_typep;
418 icmp_typep = skb_header_pointer(skb, (u8 *)old_iph +
419 (old_iph->ihl << 2) +
420 offsetof(struct icmphdr, type) -
421 skb->data, sizeof(icmp_type),
427 if (*icmp_typep > NR_ICMP_TYPES
428 || (*icmp_typep <= ICMP_PARAMETERPROB
429 && *icmp_typep != ICMP_ECHOREPLY
430 && *icmp_typep != ICMP_ECHO))
437 static void ipv4_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
438 unsigned int mtu, unsigned int payload_length)
440 struct iphdr *iph, *old_iph = ip_hdr(skb);
441 struct icmphdr *icmph;
444 iph = (struct iphdr *)skb_put(nskb, sizeof(struct iphdr));
445 icmph = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr));
446 payload = skb_put(nskb, payload_length);
450 iph->ihl = sizeof(struct iphdr) >> 2;
451 iph->tos = (old_iph->tos & IPTOS_TOS_MASK) |
452 IPTOS_PREC_INTERNETCONTROL;
453 iph->tot_len = htons(sizeof(struct iphdr)
454 + sizeof(struct icmphdr)
456 get_random_bytes(&iph->id, sizeof(iph->id));
459 iph->protocol = IPPROTO_ICMP;
460 iph->daddr = old_iph->saddr;
461 iph->saddr = old_iph->daddr;
466 icmph->type = ICMP_DEST_UNREACH;
467 icmph->code = ICMP_FRAG_NEEDED;
468 icmph->un.gateway = htonl(mtu);
471 nskb->csum = csum_partial((u8 *)icmph, sizeof(struct icmphdr), 0);
472 nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_iph - skb->data,
473 payload, payload_length,
475 icmph->checksum = csum_fold(nskb->csum);
478 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
479 static bool ipv6_should_icmp(struct sk_buff *skb)
481 struct ipv6hdr *old_ipv6h = ipv6_hdr(skb);
483 int payload_off = (u8 *)(old_ipv6h + 1) - skb->data;
484 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
486 /* Check source address is valid. */
487 addr_type = ipv6_addr_type(&old_ipv6h->saddr);
488 if (addr_type & IPV6_ADDR_MULTICAST || addr_type == IPV6_ADDR_ANY)
491 /* Don't reply to unspecified addresses. */
492 if (ipv6_addr_type(&old_ipv6h->daddr) == IPV6_ADDR_ANY)
495 /* Don't respond to ICMP error messages. */
496 payload_off = ipv6_skip_exthdr(skb, payload_off, &nexthdr);
500 if (nexthdr == NEXTHDR_ICMP) {
501 u8 icmp_type, *icmp_typep;
503 icmp_typep = skb_header_pointer(skb, payload_off +
504 offsetof(struct icmp6hdr,
506 sizeof(icmp_type), &icmp_type);
508 if (!icmp_typep || !(*icmp_typep & ICMPV6_INFOMSG_MASK))
515 static void ipv6_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
516 unsigned int mtu, unsigned int payload_length)
518 struct ipv6hdr *ipv6h, *old_ipv6h = ipv6_hdr(skb);
519 struct icmp6hdr *icmp6h;
522 ipv6h = (struct ipv6hdr *)skb_put(nskb, sizeof(struct ipv6hdr));
523 icmp6h = (struct icmp6hdr *)skb_put(nskb, sizeof(struct icmp6hdr));
524 payload = skb_put(nskb, payload_length);
529 memset(&ipv6h->flow_lbl, 0, sizeof(ipv6h->flow_lbl));
530 ipv6h->payload_len = htons(sizeof(struct icmp6hdr)
532 ipv6h->nexthdr = NEXTHDR_ICMP;
533 ipv6h->hop_limit = IPV6_DEFAULT_HOPLIMIT;
534 ipv6_addr_copy(&ipv6h->daddr, &old_ipv6h->saddr);
535 ipv6_addr_copy(&ipv6h->saddr, &old_ipv6h->daddr);
538 icmp6h->icmp6_type = ICMPV6_PKT_TOOBIG;
539 icmp6h->icmp6_code = 0;
540 icmp6h->icmp6_cksum = 0;
541 icmp6h->icmp6_mtu = htonl(mtu);
543 nskb->csum = csum_partial((u8 *)icmp6h, sizeof(struct icmp6hdr), 0);
544 nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_ipv6h - skb->data,
545 payload, payload_length,
547 icmp6h->icmp6_cksum = csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
548 sizeof(struct icmp6hdr)
550 ipv6h->nexthdr, nskb->csum);
554 bool tnl_frag_needed(struct vport *vport, const struct tnl_mutable_config *mutable,
555 struct sk_buff *skb, unsigned int mtu, __be64 flow_key)
557 unsigned int eth_hdr_len = ETH_HLEN;
558 unsigned int total_length = 0, header_length = 0, payload_length;
559 struct ethhdr *eh, *old_eh = eth_hdr(skb);
560 struct sk_buff *nskb;
563 if (skb->protocol == htons(ETH_P_IP)) {
564 if (mtu < IP_MIN_MTU)
567 if (!ipv4_should_icmp(skb))
570 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
571 else if (skb->protocol == htons(ETH_P_IPV6)) {
572 if (mtu < IPV6_MIN_MTU)
576 * In theory we should do PMTUD on IPv6 multicast messages but
577 * we don't have an address to send from so just fragment.
579 if (ipv6_addr_type(&ipv6_hdr(skb)->daddr) & IPV6_ADDR_MULTICAST)
582 if (!ipv6_should_icmp(skb))
590 if (old_eh->h_proto == htons(ETH_P_8021Q))
591 eth_hdr_len = VLAN_ETH_HLEN;
593 payload_length = skb->len - eth_hdr_len;
594 if (skb->protocol == htons(ETH_P_IP)) {
595 header_length = sizeof(struct iphdr) + sizeof(struct icmphdr);
596 total_length = min_t(unsigned int, header_length +
597 payload_length, 576);
599 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
601 header_length = sizeof(struct ipv6hdr) +
602 sizeof(struct icmp6hdr);
603 total_length = min_t(unsigned int, header_length +
604 payload_length, IPV6_MIN_MTU);
608 payload_length = total_length - header_length;
610 nskb = dev_alloc_skb(NET_IP_ALIGN + eth_hdr_len + header_length +
615 skb_reserve(nskb, NET_IP_ALIGN);
617 /* Ethernet / VLAN */
618 eh = (struct ethhdr *)skb_put(nskb, eth_hdr_len);
619 memcpy(eh->h_dest, old_eh->h_source, ETH_ALEN);
620 memcpy(eh->h_source, mutable->eth_addr, ETH_ALEN);
621 nskb->protocol = eh->h_proto = old_eh->h_proto;
622 if (old_eh->h_proto == htons(ETH_P_8021Q)) {
623 struct vlan_ethhdr *vh = (struct vlan_ethhdr *)eh;
625 vh->h_vlan_TCI = vlan_eth_hdr(skb)->h_vlan_TCI;
626 vh->h_vlan_encapsulated_proto = skb->protocol;
628 vlan_set_tci(nskb, vlan_get_tci(skb));
629 skb_reset_mac_header(nskb);
632 if (skb->protocol == htons(ETH_P_IP))
633 ipv4_build_icmp(skb, nskb, mtu, payload_length);
634 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
636 ipv6_build_icmp(skb, nskb, mtu, payload_length);
640 * Assume that flow based keys are symmetric with respect to input
641 * and output and use the key that we were going to put on the
642 * outgoing packet for the fake received packet. If the keys are
643 * not symmetric then PMTUD needs to be disabled since we won't have
644 * any way of synthesizing packets.
646 if ((mutable->flags & (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION)) ==
647 (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION))
648 OVS_CB(nskb)->tun_id = flow_key;
650 if (unlikely(compute_ip_summed(nskb, false))) {
655 vport_receive(vport, nskb);
660 static bool check_mtu(struct sk_buff *skb,
662 const struct tnl_mutable_config *mutable,
663 const struct rtable *rt, __be16 *frag_offp)
665 bool df_inherit = mutable->flags & TNL_F_DF_INHERIT;
666 bool pmtud = mutable->flags & TNL_F_PMTUD;
667 __be16 frag_off = mutable->flags & TNL_F_DF_DEFAULT ? htons(IP_DF) : 0;
669 unsigned int packet_length = skb->len - ETH_HLEN;
671 /* Allow for one level of tagging in the packet length. */
672 if (!vlan_tx_tag_present(skb) &&
673 eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
674 packet_length -= VLAN_HLEN;
679 /* The tag needs to go in packet regardless of where it
680 * currently is, so subtract it from the MTU.
682 if (vlan_tx_tag_present(skb) ||
683 eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
684 vlan_header = VLAN_HLEN;
686 mtu = dst_mtu(&rt_dst(rt))
688 - mutable->tunnel_hlen
692 if (skb->protocol == htons(ETH_P_IP)) {
693 struct iphdr *iph = ip_hdr(skb);
696 frag_off = iph->frag_off & htons(IP_DF);
698 if (pmtud && iph->frag_off & htons(IP_DF)) {
699 mtu = max(mtu, IP_MIN_MTU);
701 if (packet_length > mtu &&
702 tnl_frag_needed(vport, mutable, skb, mtu,
703 OVS_CB(skb)->tun_id))
707 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
708 else if (skb->protocol == htons(ETH_P_IPV6)) {
709 /* IPv6 requires end hosts to do fragmentation
710 * if the packet is above the minimum MTU.
712 if (df_inherit && packet_length > IPV6_MIN_MTU)
713 frag_off = htons(IP_DF);
716 mtu = max(mtu, IPV6_MIN_MTU);
718 if (packet_length > mtu &&
719 tnl_frag_needed(vport, mutable, skb, mtu,
720 OVS_CB(skb)->tun_id))
726 *frag_offp = frag_off;
730 static void create_tunnel_header(const struct vport *vport,
731 const struct tnl_mutable_config *mutable,
732 const struct rtable *rt, void *header)
734 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
735 struct iphdr *iph = header;
738 iph->ihl = sizeof(struct iphdr) >> 2;
739 iph->frag_off = htons(IP_DF);
740 iph->protocol = tnl_vport->tnl_ops->ipproto;
741 iph->tos = mutable->tos;
742 iph->daddr = rt->rt_dst;
743 iph->saddr = rt->rt_src;
744 iph->ttl = mutable->ttl;
746 iph->ttl = ip4_dst_hoplimit(&rt_dst(rt));
748 tnl_vport->tnl_ops->build_header(vport, mutable, iph + 1);
751 static inline void *get_cached_header(const struct tnl_cache *cache)
753 return (void *)cache + ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN);
756 static inline bool check_cache_valid(const struct tnl_cache *cache,
757 const struct tnl_mutable_config *mutable)
764 hh = rt_hh(cache->rt);
766 #ifdef NEED_CACHE_TIMEOUT
767 time_before(jiffies, cache->expiration) &&
770 atomic_read(&init_net.ipv4.rt_genid) == cache->rt->rt_genid &&
773 hh->hh_lock.sequence == cache->hh_seq &&
775 mutable->seq == cache->mutable_seq &&
776 (!is_internal_dev(rt_dst(cache->rt).dev) ||
777 (cache->flow && !cache->flow->dead));
780 static void __cache_cleaner(struct tnl_vport *tnl_vport)
782 const struct tnl_mutable_config *mutable =
783 rcu_dereference(tnl_vport->mutable);
784 const struct tnl_cache *cache = rcu_dereference(tnl_vport->cache);
786 if (cache && !check_cache_valid(cache, mutable) &&
787 spin_trylock_bh(&tnl_vport->cache_lock)) {
788 assign_cache_rcu(tnl_vport_to_vport(tnl_vport), NULL);
789 spin_unlock_bh(&tnl_vport->cache_lock);
793 static void cache_cleaner(struct work_struct *work)
797 schedule_cache_cleaner();
800 for (i = 0; i < PORT_TABLE_SIZE; i++) {
801 struct hlist_node *n;
802 struct hlist_head *bucket;
803 struct tnl_vport *tnl_vport;
805 bucket = &port_table[i];
806 hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node)
807 __cache_cleaner(tnl_vport);
812 static inline void create_eth_hdr(struct tnl_cache *cache,
815 void *cache_data = get_cached_header(cache);
822 hh_seq = read_seqbegin(&hh->hh_lock);
823 hh_off = HH_DATA_ALIGN(hh->hh_len) - hh->hh_len;
824 memcpy(cache_data, (void *)hh->hh_data + hh_off, hh->hh_len);
825 cache->hh_len = hh->hh_len;
826 } while (read_seqretry(&hh->hh_lock, hh_seq));
828 cache->hh_seq = hh_seq;
830 read_lock(&hh->hh_lock);
831 hh_off = HH_DATA_ALIGN(hh->hh_len) - hh->hh_len;
832 memcpy(cache_data, (void *)hh->hh_data + hh_off, hh->hh_len);
833 cache->hh_len = hh->hh_len;
834 read_unlock(&hh->hh_lock);
838 static struct tnl_cache *build_cache(struct vport *vport,
839 const struct tnl_mutable_config *mutable,
842 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
843 struct tnl_cache *cache;
848 if (!(mutable->flags & TNL_F_HDR_CACHE))
852 * If there is no entry in the ARP cache or if this device does not
853 * support hard header caching just fall back to the IP stack.
861 * If lock is contended fall back to directly building the header.
862 * We're not going to help performance by sitting here spinning.
864 if (!spin_trylock(&tnl_vport->cache_lock))
867 cache = cache_dereference(tnl_vport);
868 if (check_cache_valid(cache, mutable))
873 cache_len = LL_RESERVED_SPACE(rt_dst(rt).dev) + mutable->tunnel_hlen;
875 cache = kzalloc(ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN) +
876 cache_len, GFP_ATOMIC);
880 create_eth_hdr(cache, hh);
881 cache_data = get_cached_header(cache) + cache->hh_len;
882 cache->len = cache->hh_len + mutable->tunnel_hlen;
884 create_tunnel_header(vport, mutable, rt, cache_data);
886 cache->mutable_seq = mutable->seq;
888 #ifdef NEED_CACHE_TIMEOUT
889 cache->expiration = jiffies + tnl_vport->cache_exp_interval;
892 if (is_internal_dev(rt_dst(rt).dev)) {
893 struct sw_flow_key flow_key;
894 struct vport *dst_vport;
898 struct sw_flow *flow;
900 dst_vport = internal_dev_get_vport(rt_dst(rt).dev);
904 skb = alloc_skb(cache->len, GFP_ATOMIC);
908 __skb_put(skb, cache->len);
909 memcpy(skb->data, get_cached_header(cache), cache->len);
911 err = flow_extract(skb, dst_vport->port_no, &flow_key,
918 flow = flow_tbl_lookup(rcu_dereference(dst_vport->dp->table),
919 &flow_key, flow_key_len);
927 assign_cache_rcu(vport, cache);
930 spin_unlock(&tnl_vport->cache_lock);
935 static struct rtable *find_route(struct vport *vport,
936 const struct tnl_mutable_config *mutable,
937 u8 tos, struct tnl_cache **cache)
939 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
940 struct tnl_cache *cur_cache = rcu_dereference(tnl_vport->cache);
945 if (likely(tos == mutable->tos && check_cache_valid(cur_cache, mutable))) {
947 return cur_cache->rt;
950 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
951 struct flowi fl = { .nl_u = { .ip4_u =
952 { .daddr = mutable->key.daddr,
953 .saddr = mutable->key.saddr,
955 .proto = tnl_vport->tnl_ops->ipproto };
957 if (unlikely(ip_route_output_key(&init_net, &rt, &fl)))
960 struct flowi4 fl = { .daddr = mutable->key.daddr,
961 .saddr = mutable->key.saddr,
963 .flowi4_proto = tnl_vport->tnl_ops->ipproto };
965 rt = ip_route_output_key(&init_net, &fl);
970 if (likely(tos == mutable->tos))
971 *cache = build_cache(vport, mutable, rt);
977 static inline bool need_linearize(const struct sk_buff *skb)
981 if (unlikely(skb_shinfo(skb)->frag_list))
985 * Generally speaking we should linearize if there are paged frags.
986 * However, if all of the refcounts are 1 we know nobody else can
987 * change them from underneath us and we can skip the linearization.
989 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
990 if (unlikely(page_count(skb_shinfo(skb)->frags[i].page) > 1))
996 static struct sk_buff *handle_offloads(struct sk_buff *skb,
997 const struct tnl_mutable_config *mutable,
998 const struct rtable *rt)
1003 min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
1004 + mutable->tunnel_hlen
1005 + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1007 if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
1008 int head_delta = SKB_DATA_ALIGN(min_headroom -
1011 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
1017 forward_ip_summed(skb, true);
1019 if (skb_is_gso(skb)) {
1020 struct sk_buff *nskb;
1022 nskb = skb_gso_segment(skb, 0);
1025 err = PTR_ERR(nskb);
1031 } else if (get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
1032 /* Pages aren't locked and could change at any time.
1033 * If this happens after we compute the checksum, the
1034 * checksum will be wrong. We linearize now to avoid
1037 if (unlikely(need_linearize(skb))) {
1038 err = __skb_linearize(skb);
1043 err = skb_checksum_help(skb);
1048 set_ip_summed(skb, OVS_CSUM_NONE);
1055 return ERR_PTR(err);
1058 static int send_frags(struct sk_buff *skb,
1059 const struct tnl_mutable_config *mutable)
1065 struct sk_buff *next = skb->next;
1066 int frag_len = skb->len - mutable->tunnel_hlen;
1070 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
1072 err = ip_local_out(skb);
1074 if (unlikely(net_xmit_eval(err)))
1076 sent_len += frag_len;
1083 * There's no point in continuing to send fragments once one has been
1084 * dropped so just free the rest. This may help improve the congestion
1085 * that caused the first packet to be dropped.
1087 tnl_free_linked_skbs(skb);
1091 int tnl_send(struct vport *vport, struct sk_buff *skb)
1093 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1094 const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
1096 enum vport_err_type err = VPORT_E_TX_ERROR;
1098 struct dst_entry *unattached_dst = NULL;
1099 struct tnl_cache *cache;
1101 __be16 frag_off = 0;
1106 /* Validate the protocol headers before we try to use them. */
1107 if (skb->protocol == htons(ETH_P_8021Q) &&
1108 !vlan_tx_tag_present(skb)) {
1109 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
1112 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
1113 skb_set_network_header(skb, VLAN_ETH_HLEN);
1116 if (skb->protocol == htons(ETH_P_IP)) {
1117 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1118 + sizeof(struct iphdr))))
1121 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1122 else if (skb->protocol == htons(ETH_P_IPV6)) {
1123 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1124 + sizeof(struct ipv6hdr))))
1130 if (skb->protocol == htons(ETH_P_IP))
1131 inner_tos = ip_hdr(skb)->tos;
1132 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1133 else if (skb->protocol == htons(ETH_P_IPV6))
1134 inner_tos = ipv6_get_dsfield(ipv6_hdr(skb));
1139 if (mutable->flags & TNL_F_TOS_INHERIT)
1144 tos = INET_ECN_encapsulate(tos, inner_tos);
1147 rt = find_route(vport, mutable, tos, &cache);
1150 if (unlikely(!cache))
1151 unattached_dst = &rt_dst(rt);
1157 skb_clear_rxhash(skb);
1160 skb = handle_offloads(skb, mutable, rt);
1165 if (unlikely(!check_mtu(skb, vport, mutable, rt, &frag_off))) {
1166 err = VPORT_E_TX_DROPPED;
1171 * If we are over the MTU, allow the IP stack to handle fragmentation.
1172 * Fragmentation is a slow path anyways.
1174 if (unlikely(skb->len + mutable->tunnel_hlen > dst_mtu(&rt_dst(rt)) &&
1176 unattached_dst = &rt_dst(rt);
1177 dst_hold(unattached_dst);
1184 ttl = ip4_dst_hoplimit(&rt_dst(rt));
1186 if (mutable->flags & TNL_F_TTL_INHERIT) {
1187 if (skb->protocol == htons(ETH_P_IP))
1188 ttl = ip_hdr(skb)->ttl;
1189 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1190 else if (skb->protocol == htons(ETH_P_IPV6))
1191 ttl = ipv6_hdr(skb)->hop_limit;
1197 struct sk_buff *next_skb = skb->next;
1200 if (unlikely(vlan_deaccel_tag(skb)))
1203 if (likely(cache)) {
1204 skb_push(skb, cache->len);
1205 memcpy(skb->data, get_cached_header(cache), cache->len);
1206 skb_reset_mac_header(skb);
1207 skb_set_network_header(skb, cache->hh_len);
1210 skb_push(skb, mutable->tunnel_hlen);
1211 create_tunnel_header(vport, mutable, rt, skb->data);
1212 skb_reset_network_header(skb);
1215 skb_dst_set(skb, dst_clone(unattached_dst));
1217 skb_dst_set(skb, unattached_dst);
1218 unattached_dst = NULL;
1221 skb_set_transport_header(skb, skb_network_offset(skb) + sizeof(struct iphdr));
1226 iph->frag_off = frag_off;
1227 ip_select_ident(iph, &rt_dst(rt), NULL);
1229 skb = tnl_vport->tnl_ops->update_header(vport, mutable, &rt_dst(rt), skb);
1233 if (likely(cache)) {
1234 int orig_len = skb->len - cache->len;
1235 struct vport *cache_vport = internal_dev_get_vport(rt_dst(rt).dev);
1237 skb->protocol = htons(ETH_P_IP);
1239 iph->tot_len = htons(skb->len - skb_network_offset(skb));
1243 if (unlikely(compute_ip_summed(skb, true))) {
1248 OVS_CB(skb)->flow = cache->flow;
1249 vport_receive(cache_vport, skb);
1250 sent_len += orig_len;
1254 skb->dev = rt_dst(rt).dev;
1255 xmit_err = dev_queue_xmit(skb);
1257 if (likely(net_xmit_eval(xmit_err) == 0))
1258 sent_len += orig_len;
1261 sent_len += send_frags(skb, mutable);
1267 if (unlikely(sent_len == 0))
1268 vport_record_error(vport, VPORT_E_TX_DROPPED);
1273 tnl_free_linked_skbs(skb);
1275 vport_record_error(vport, err);
1277 dst_release(unattached_dst);
1281 static const struct nla_policy tnl_policy[OVS_TUNNEL_ATTR_MAX + 1] = {
1282 [OVS_TUNNEL_ATTR_FLAGS] = { .type = NLA_U32 },
1283 [OVS_TUNNEL_ATTR_DST_IPV4] = { .type = NLA_U32 },
1284 [OVS_TUNNEL_ATTR_SRC_IPV4] = { .type = NLA_U32 },
1285 [OVS_TUNNEL_ATTR_OUT_KEY] = { .type = NLA_U64 },
1286 [OVS_TUNNEL_ATTR_IN_KEY] = { .type = NLA_U64 },
1287 [OVS_TUNNEL_ATTR_TOS] = { .type = NLA_U8 },
1288 [OVS_TUNNEL_ATTR_TTL] = { .type = NLA_U8 },
1291 /* Sets OVS_TUNNEL_ATTR_* fields in 'mutable', which must initially be zeroed. */
1292 static int tnl_set_config(struct nlattr *options, const struct tnl_ops *tnl_ops,
1293 const struct vport *cur_vport,
1294 struct tnl_mutable_config *mutable)
1296 const struct vport *old_vport;
1297 const struct tnl_mutable_config *old_mutable;
1298 struct nlattr *a[OVS_TUNNEL_ATTR_MAX + 1];
1304 err = nla_parse_nested(a, OVS_TUNNEL_ATTR_MAX, options, tnl_policy);
1308 if (!a[OVS_TUNNEL_ATTR_FLAGS] || !a[OVS_TUNNEL_ATTR_DST_IPV4])
1311 mutable->flags = nla_get_u32(a[OVS_TUNNEL_ATTR_FLAGS]) & TNL_F_PUBLIC;
1313 if (a[OVS_TUNNEL_ATTR_SRC_IPV4])
1314 mutable->key.saddr = nla_get_be32(a[OVS_TUNNEL_ATTR_SRC_IPV4]);
1315 mutable->key.daddr = nla_get_be32(a[OVS_TUNNEL_ATTR_DST_IPV4]);
1317 if (a[OVS_TUNNEL_ATTR_TOS]) {
1318 mutable->tos = nla_get_u8(a[OVS_TUNNEL_ATTR_TOS]);
1319 if (mutable->tos != RT_TOS(mutable->tos))
1323 if (a[OVS_TUNNEL_ATTR_TTL])
1324 mutable->ttl = nla_get_u8(a[OVS_TUNNEL_ATTR_TTL]);
1326 mutable->key.tunnel_type = tnl_ops->tunnel_type;
1327 if (!a[OVS_TUNNEL_ATTR_IN_KEY]) {
1328 mutable->key.tunnel_type |= TNL_T_KEY_MATCH;
1329 mutable->flags |= TNL_F_IN_KEY_MATCH;
1331 mutable->key.tunnel_type |= TNL_T_KEY_EXACT;
1332 mutable->key.in_key = nla_get_be64(a[OVS_TUNNEL_ATTR_IN_KEY]);
1335 if (!a[OVS_TUNNEL_ATTR_OUT_KEY])
1336 mutable->flags |= TNL_F_OUT_KEY_ACTION;
1338 mutable->out_key = nla_get_be64(a[OVS_TUNNEL_ATTR_OUT_KEY]);
1340 mutable->tunnel_hlen = tnl_ops->hdr_len(mutable);
1341 if (mutable->tunnel_hlen < 0)
1342 return mutable->tunnel_hlen;
1344 mutable->tunnel_hlen += sizeof(struct iphdr);
1346 old_vport = port_table_lookup(&mutable->key, &old_mutable);
1347 if (old_vport && old_vport != cur_vport)
1353 struct vport *tnl_create(const struct vport_parms *parms,
1354 const struct vport_ops *vport_ops,
1355 const struct tnl_ops *tnl_ops)
1357 struct vport *vport;
1358 struct tnl_vport *tnl_vport;
1359 struct tnl_mutable_config *mutable;
1360 int initial_frag_id;
1363 vport = vport_alloc(sizeof(struct tnl_vport), vport_ops, parms);
1364 if (IS_ERR(vport)) {
1365 err = PTR_ERR(vport);
1369 tnl_vport = tnl_vport_priv(vport);
1371 strcpy(tnl_vport->name, parms->name);
1372 tnl_vport->tnl_ops = tnl_ops;
1374 mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1377 goto error_free_vport;
1380 vport_gen_rand_ether_addr(mutable->eth_addr);
1382 get_random_bytes(&initial_frag_id, sizeof(int));
1383 atomic_set(&tnl_vport->frag_id, initial_frag_id);
1385 err = tnl_set_config(parms->options, tnl_ops, NULL, mutable);
1387 goto error_free_mutable;
1389 spin_lock_init(&tnl_vport->cache_lock);
1391 #ifdef NEED_CACHE_TIMEOUT
1392 tnl_vport->cache_exp_interval = MAX_CACHE_EXP -
1393 (net_random() % (MAX_CACHE_EXP / 2));
1396 rcu_assign_pointer(tnl_vport->mutable, mutable);
1398 port_table_add_port(vport);
1406 return ERR_PTR(err);
1409 int tnl_set_options(struct vport *vport, struct nlattr *options)
1411 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1412 const struct tnl_mutable_config *old_mutable;
1413 struct tnl_mutable_config *mutable;
1416 mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1422 /* Copy fields whose values should be retained. */
1423 old_mutable = rtnl_dereference(tnl_vport->mutable);
1424 mutable->seq = old_mutable->seq + 1;
1425 memcpy(mutable->eth_addr, old_mutable->eth_addr, ETH_ALEN);
1427 /* Parse the others configured by userspace. */
1428 err = tnl_set_config(options, tnl_vport->tnl_ops, vport, mutable);
1432 if (port_hash(&mutable->key) != port_hash(&old_mutable->key))
1433 port_table_move_port(vport, mutable);
1435 assign_config_rcu(vport, mutable);
1445 int tnl_get_options(const struct vport *vport, struct sk_buff *skb)
1447 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1448 const struct tnl_mutable_config *mutable = rcu_dereference_rtnl(tnl_vport->mutable);
1450 NLA_PUT_U32(skb, OVS_TUNNEL_ATTR_FLAGS, mutable->flags & TNL_F_PUBLIC);
1451 NLA_PUT_BE32(skb, OVS_TUNNEL_ATTR_DST_IPV4, mutable->key.daddr);
1453 if (!(mutable->flags & TNL_F_IN_KEY_MATCH))
1454 NLA_PUT_BE64(skb, OVS_TUNNEL_ATTR_IN_KEY, mutable->key.in_key);
1455 if (!(mutable->flags & TNL_F_OUT_KEY_ACTION))
1456 NLA_PUT_BE64(skb, OVS_TUNNEL_ATTR_OUT_KEY, mutable->out_key);
1457 if (mutable->key.saddr)
1458 NLA_PUT_BE32(skb, OVS_TUNNEL_ATTR_SRC_IPV4, mutable->key.saddr);
1460 NLA_PUT_U8(skb, OVS_TUNNEL_ATTR_TOS, mutable->tos);
1462 NLA_PUT_U8(skb, OVS_TUNNEL_ATTR_TTL, mutable->ttl);
1470 static void free_port_rcu(struct rcu_head *rcu)
1472 struct tnl_vport *tnl_vport = container_of(rcu,
1473 struct tnl_vport, rcu);
1475 free_cache((struct tnl_cache __force *)tnl_vport->cache);
1476 kfree((struct tnl_mutable __force *)tnl_vport->mutable);
1477 vport_free(tnl_vport_to_vport(tnl_vport));
1480 void tnl_destroy(struct vport *vport)
1482 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1483 const struct tnl_mutable_config *mutable;
1485 mutable = rtnl_dereference(tnl_vport->mutable);
1486 port_table_remove_port(vport);
1487 call_rcu(&tnl_vport->rcu, free_port_rcu);
1490 int tnl_set_addr(struct vport *vport, const unsigned char *addr)
1492 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1493 struct tnl_mutable_config *mutable;
1495 mutable = kmemdup(rtnl_dereference(tnl_vport->mutable),
1496 sizeof(struct tnl_mutable_config), GFP_KERNEL);
1500 memcpy(mutable->eth_addr, addr, ETH_ALEN);
1501 assign_config_rcu(vport, mutable);
1506 const char *tnl_get_name(const struct vport *vport)
1508 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1509 return tnl_vport->name;
1512 const unsigned char *tnl_get_addr(const struct vport *vport)
1514 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1515 return rcu_dereference_rtnl(tnl_vport->mutable)->eth_addr;
1518 void tnl_free_linked_skbs(struct sk_buff *skb)
1521 struct sk_buff *next = skb->next;
1531 port_table = kmalloc(PORT_TABLE_SIZE * sizeof(struct hlist_head *),
1536 for (i = 0; i < PORT_TABLE_SIZE; i++)
1537 INIT_HLIST_HEAD(&port_table[i]);
1546 for (i = 0; i < PORT_TABLE_SIZE; i++) {
1547 struct tnl_vport * tnl_vport;
1548 struct hlist_head *hash_head;
1549 struct hlist_node *n;
1551 hash_head = &port_table[i];
1552 hlist_for_each_entry(tnl_vport, n, hash_head, hash_node) {