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>
13 #include <linux/igmp.h>
15 #include <linux/in_route.h>
16 #include <linux/inetdevice.h>
17 #include <linux/jhash.h>
18 #include <linux/list.h>
19 #include <linux/kernel.h>
20 #include <linux/version.h>
21 #include <linux/workqueue.h>
22 #include <linux/rculist.h>
24 #include <net/dsfield.h>
27 #include <net/inet_ecn.h>
29 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
32 #include <net/route.h>
41 #include "vport-generic.h"
42 #include "vport-internal_dev.h"
44 #ifdef NEED_CACHE_TIMEOUT
46 * On kernels where we can't quickly detect changes in the rest of the system
47 * we use an expiration time to invalidate the cache. A shorter expiration
48 * reduces the length of time that we may potentially blackhole packets while
49 * a longer time increases performance by reducing the frequency that the
50 * cache needs to be rebuilt. A variety of factors may cause the cache to be
51 * invalidated before the expiration time but this is the maximum. The time
52 * is expressed in jiffies.
54 #define MAX_CACHE_EXP HZ
58 * Interval to check for and remove caches that are no longer valid. Caches
59 * are checked for validity before they are used for packet encapsulation and
60 * old caches are removed at that time. However, if no packets are sent through
61 * the tunnel then the cache will never be destroyed. Since it holds
62 * references to a number of system objects, the cache will continue to use
63 * system resources by not allowing those objects to be destroyed. The cache
64 * cleaner is periodically run to free invalid caches. It does not
65 * significantly affect system performance. A lower interval will release
66 * resources faster but will itself consume resources by requiring more frequent
67 * checks. A longer interval may result in messages being printed to the kernel
68 * message buffer about unreleased resources. The interval is expressed in
71 #define CACHE_CLEANER_INTERVAL (5 * HZ)
73 #define CACHE_DATA_ALIGN 16
74 #define PORT_TABLE_SIZE 1024
76 static struct hlist_head *port_table __read_mostly;
77 static int port_table_count;
79 static void cache_cleaner(struct work_struct *work);
80 static DECLARE_DELAYED_WORK(cache_cleaner_wq, cache_cleaner);
83 * These are just used as an optimization: they don't require any kind of
84 * synchronization because we could have just as easily read the value before
85 * the port change happened.
87 static unsigned int key_local_remote_ports __read_mostly;
88 static unsigned int key_remote_ports __read_mostly;
89 static unsigned int local_remote_ports __read_mostly;
90 static unsigned int remote_ports __read_mostly;
92 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
93 #define rt_dst(rt) (rt->dst)
95 #define rt_dst(rt) (rt->u.dst)
98 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,1,0)
99 static struct hh_cache *rt_hh(struct rtable *rt)
101 struct neighbour *neigh = dst_get_neighbour(&rt->dst);
102 if (!neigh || !(neigh->nud_state & NUD_CONNECTED) ||
108 #define rt_hh(rt) (rt_dst(rt).hh)
111 static inline struct vport *tnl_vport_to_vport(const struct tnl_vport *tnl_vport)
113 return vport_from_priv(tnl_vport);
116 /* This is analogous to rtnl_dereference for the tunnel cache. It checks that
117 * cache_lock is held, so it is only for update side code.
119 static inline struct tnl_cache *cache_dereference(struct tnl_vport *tnl_vport)
121 return rcu_dereference_protected(tnl_vport->cache,
122 lockdep_is_held(&tnl_vport->cache_lock));
125 static inline void schedule_cache_cleaner(void)
127 schedule_delayed_work(&cache_cleaner_wq, CACHE_CLEANER_INTERVAL);
130 static void free_cache(struct tnl_cache *cache)
135 flow_put(cache->flow);
136 ip_rt_put(cache->rt);
140 static void free_config_rcu(struct rcu_head *rcu)
142 struct tnl_mutable_config *c = container_of(rcu, struct tnl_mutable_config, rcu);
146 static void free_cache_rcu(struct rcu_head *rcu)
148 struct tnl_cache *c = container_of(rcu, struct tnl_cache, rcu);
152 /* Frees the portion of 'mutable' that requires RTNL and thus can't happen
153 * within an RCU callback. Fortunately this part doesn't require waiting for
154 * an RCU grace period.
156 static void free_mutable_rtnl(struct tnl_mutable_config *mutable)
159 if (ipv4_is_multicast(mutable->key.daddr) && mutable->mlink) {
160 struct in_device *in_dev;
161 in_dev = inetdev_by_index(&init_net, mutable->mlink);
163 ip_mc_dec_group(in_dev, mutable->key.daddr);
167 static void assign_config_rcu(struct vport *vport,
168 struct tnl_mutable_config *new_config)
170 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
171 struct tnl_mutable_config *old_config;
173 old_config = rtnl_dereference(tnl_vport->mutable);
174 rcu_assign_pointer(tnl_vport->mutable, new_config);
176 free_mutable_rtnl(old_config);
177 call_rcu(&old_config->rcu, free_config_rcu);
180 static void assign_cache_rcu(struct vport *vport, struct tnl_cache *new_cache)
182 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
183 struct tnl_cache *old_cache;
185 old_cache = cache_dereference(tnl_vport);
186 rcu_assign_pointer(tnl_vport->cache, new_cache);
189 call_rcu(&old_cache->rcu, free_cache_rcu);
192 static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable)
194 if (mutable->flags & TNL_F_IN_KEY_MATCH) {
195 if (mutable->key.saddr)
196 return &local_remote_ports;
198 return &remote_ports;
200 if (mutable->key.saddr)
201 return &key_local_remote_ports;
203 return &key_remote_ports;
207 static u32 port_hash(const struct port_lookup_key *key)
209 return jhash2((u32*)key, (PORT_KEY_LEN / sizeof(u32)), 0);
212 static inline struct hlist_head *find_bucket(u32 hash)
214 return &port_table[(hash & (PORT_TABLE_SIZE - 1))];
217 static void port_table_add_port(struct vport *vport)
219 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
220 const struct tnl_mutable_config *mutable;
223 if (port_table_count == 0)
224 schedule_cache_cleaner();
226 mutable = rtnl_dereference(tnl_vport->mutable);
227 hash = port_hash(&mutable->key);
228 hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
231 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
234 static void port_table_move_port(struct vport *vport,
235 struct tnl_mutable_config *new_mutable)
237 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
240 hash = port_hash(&new_mutable->key);
241 hlist_del_init_rcu(&tnl_vport->hash_node);
242 hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
244 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
245 assign_config_rcu(vport, new_mutable);
246 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
249 static void port_table_remove_port(struct vport *vport)
251 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
253 hlist_del_init_rcu(&tnl_vport->hash_node);
256 if (port_table_count == 0)
257 cancel_delayed_work_sync(&cache_cleaner_wq);
259 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
262 static struct vport *port_table_lookup(struct port_lookup_key *key,
263 const struct tnl_mutable_config **pmutable)
265 struct hlist_node *n;
266 struct hlist_head *bucket;
267 u32 hash = port_hash(key);
268 struct tnl_vport * tnl_vport;
270 bucket = find_bucket(hash);
272 hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node) {
273 struct tnl_mutable_config *mutable;
275 mutable = rcu_dereference_rtnl(tnl_vport->mutable);
276 if (!memcmp(&mutable->key, key, PORT_KEY_LEN)) {
278 return tnl_vport_to_vport(tnl_vport);
285 struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be64 key,
287 const struct tnl_mutable_config **mutable)
289 struct port_lookup_key lookup;
292 if (ipv4_is_multicast(saddr)) {
294 lookup.daddr = saddr;
295 if (key_remote_ports) {
296 lookup.tunnel_type = tunnel_type | TNL_T_KEY_EXACT;
298 vport = port_table_lookup(&lookup, mutable);
303 lookup.tunnel_type = tunnel_type | TNL_T_KEY_MATCH;
305 vport = port_table_lookup(&lookup, mutable);
312 lookup.saddr = saddr;
313 lookup.daddr = daddr;
315 /* First try for exact match on in_key. */
317 lookup.tunnel_type = tunnel_type | TNL_T_KEY_EXACT;
318 if (key_local_remote_ports) {
319 vport = port_table_lookup(&lookup, mutable);
323 if (key_remote_ports) {
325 vport = port_table_lookup(&lookup, mutable);
329 lookup.saddr = saddr;
332 /* Then try matches that wildcard in_key. */
334 lookup.tunnel_type = tunnel_type | TNL_T_KEY_MATCH;
335 if (local_remote_ports) {
336 vport = port_table_lookup(&lookup, mutable);
342 vport = port_table_lookup(&lookup, mutable);
350 static void ecn_decapsulate(struct sk_buff *skb, u8 tos)
352 if (unlikely(INET_ECN_is_ce(tos))) {
353 __be16 protocol = skb->protocol;
355 skb_set_network_header(skb, ETH_HLEN);
357 if (protocol == htons(ETH_P_8021Q)) {
358 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
361 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
362 skb_set_network_header(skb, VLAN_ETH_HLEN);
365 if (protocol == htons(ETH_P_IP)) {
366 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
367 + sizeof(struct iphdr))))
370 IP_ECN_set_ce(ip_hdr(skb));
372 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
373 else if (protocol == htons(ETH_P_IPV6)) {
374 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
375 + sizeof(struct ipv6hdr))))
378 IP6_ECN_set_ce(ipv6_hdr(skb));
385 * tnl_rcv - ingress point for generic tunnel code
387 * @vport: port this packet was received on
388 * @skb: received packet
389 * @tos: ToS from encapsulating IP packet, used to copy ECN bits
391 * Must be called with rcu_read_lock.
393 * Packets received by this function are in the following state:
394 * - skb->data points to the inner Ethernet header.
395 * - The inner Ethernet header is in the linear data area.
396 * - skb->csum does not include the inner Ethernet header.
397 * - The layer pointers are undefined.
399 void tnl_rcv(struct vport *vport, struct sk_buff *skb, u8 tos)
403 skb_reset_mac_header(skb);
406 if (likely(ntohs(eh->h_proto) >= 1536))
407 skb->protocol = eh->h_proto;
409 skb->protocol = htons(ETH_P_802_2);
413 skb_clear_rxhash(skb);
416 ecn_decapsulate(skb, tos);
417 vlan_set_tci(skb, 0);
419 if (unlikely(compute_ip_summed(skb, false))) {
424 vport_receive(vport, skb);
427 static bool check_ipv4_address(__be32 addr)
429 if (ipv4_is_multicast(addr) || ipv4_is_lbcast(addr)
430 || ipv4_is_loopback(addr) || ipv4_is_zeronet(addr))
436 static bool ipv4_should_icmp(struct sk_buff *skb)
438 struct iphdr *old_iph = ip_hdr(skb);
440 /* Don't respond to L2 broadcast. */
441 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
444 /* Don't respond to L3 broadcast or invalid addresses. */
445 if (!check_ipv4_address(old_iph->daddr) ||
446 !check_ipv4_address(old_iph->saddr))
449 /* Only respond to the first fragment. */
450 if (old_iph->frag_off & htons(IP_OFFSET))
453 /* Don't respond to ICMP error messages. */
454 if (old_iph->protocol == IPPROTO_ICMP) {
455 u8 icmp_type, *icmp_typep;
457 icmp_typep = skb_header_pointer(skb, (u8 *)old_iph +
458 (old_iph->ihl << 2) +
459 offsetof(struct icmphdr, type) -
460 skb->data, sizeof(icmp_type),
466 if (*icmp_typep > NR_ICMP_TYPES
467 || (*icmp_typep <= ICMP_PARAMETERPROB
468 && *icmp_typep != ICMP_ECHOREPLY
469 && *icmp_typep != ICMP_ECHO))
476 static void ipv4_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
477 unsigned int mtu, unsigned int payload_length)
479 struct iphdr *iph, *old_iph = ip_hdr(skb);
480 struct icmphdr *icmph;
483 iph = (struct iphdr *)skb_put(nskb, sizeof(struct iphdr));
484 icmph = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr));
485 payload = skb_put(nskb, payload_length);
489 iph->ihl = sizeof(struct iphdr) >> 2;
490 iph->tos = (old_iph->tos & IPTOS_TOS_MASK) |
491 IPTOS_PREC_INTERNETCONTROL;
492 iph->tot_len = htons(sizeof(struct iphdr)
493 + sizeof(struct icmphdr)
495 get_random_bytes(&iph->id, sizeof(iph->id));
498 iph->protocol = IPPROTO_ICMP;
499 iph->daddr = old_iph->saddr;
500 iph->saddr = old_iph->daddr;
505 icmph->type = ICMP_DEST_UNREACH;
506 icmph->code = ICMP_FRAG_NEEDED;
507 icmph->un.gateway = htonl(mtu);
510 nskb->csum = csum_partial((u8 *)icmph, sizeof(struct icmphdr), 0);
511 nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_iph - skb->data,
512 payload, payload_length,
514 icmph->checksum = csum_fold(nskb->csum);
517 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
518 static bool ipv6_should_icmp(struct sk_buff *skb)
520 struct ipv6hdr *old_ipv6h = ipv6_hdr(skb);
522 int payload_off = (u8 *)(old_ipv6h + 1) - skb->data;
523 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
525 /* Check source address is valid. */
526 addr_type = ipv6_addr_type(&old_ipv6h->saddr);
527 if (addr_type & IPV6_ADDR_MULTICAST || addr_type == IPV6_ADDR_ANY)
530 /* Don't reply to unspecified addresses. */
531 if (ipv6_addr_type(&old_ipv6h->daddr) == IPV6_ADDR_ANY)
534 /* Don't respond to ICMP error messages. */
535 payload_off = ipv6_skip_exthdr(skb, payload_off, &nexthdr);
539 if (nexthdr == NEXTHDR_ICMP) {
540 u8 icmp_type, *icmp_typep;
542 icmp_typep = skb_header_pointer(skb, payload_off +
543 offsetof(struct icmp6hdr,
545 sizeof(icmp_type), &icmp_type);
547 if (!icmp_typep || !(*icmp_typep & ICMPV6_INFOMSG_MASK))
554 static void ipv6_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
555 unsigned int mtu, unsigned int payload_length)
557 struct ipv6hdr *ipv6h, *old_ipv6h = ipv6_hdr(skb);
558 struct icmp6hdr *icmp6h;
561 ipv6h = (struct ipv6hdr *)skb_put(nskb, sizeof(struct ipv6hdr));
562 icmp6h = (struct icmp6hdr *)skb_put(nskb, sizeof(struct icmp6hdr));
563 payload = skb_put(nskb, payload_length);
568 memset(&ipv6h->flow_lbl, 0, sizeof(ipv6h->flow_lbl));
569 ipv6h->payload_len = htons(sizeof(struct icmp6hdr)
571 ipv6h->nexthdr = NEXTHDR_ICMP;
572 ipv6h->hop_limit = IPV6_DEFAULT_HOPLIMIT;
573 ipv6_addr_copy(&ipv6h->daddr, &old_ipv6h->saddr);
574 ipv6_addr_copy(&ipv6h->saddr, &old_ipv6h->daddr);
577 icmp6h->icmp6_type = ICMPV6_PKT_TOOBIG;
578 icmp6h->icmp6_code = 0;
579 icmp6h->icmp6_cksum = 0;
580 icmp6h->icmp6_mtu = htonl(mtu);
582 nskb->csum = csum_partial((u8 *)icmp6h, sizeof(struct icmp6hdr), 0);
583 nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_ipv6h - skb->data,
584 payload, payload_length,
586 icmp6h->icmp6_cksum = csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
587 sizeof(struct icmp6hdr)
589 ipv6h->nexthdr, nskb->csum);
593 bool tnl_frag_needed(struct vport *vport, const struct tnl_mutable_config *mutable,
594 struct sk_buff *skb, unsigned int mtu, __be64 flow_key)
596 unsigned int eth_hdr_len = ETH_HLEN;
597 unsigned int total_length = 0, header_length = 0, payload_length;
598 struct ethhdr *eh, *old_eh = eth_hdr(skb);
599 struct sk_buff *nskb;
602 if (skb->protocol == htons(ETH_P_IP)) {
603 if (mtu < IP_MIN_MTU)
606 if (!ipv4_should_icmp(skb))
609 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
610 else if (skb->protocol == htons(ETH_P_IPV6)) {
611 if (mtu < IPV6_MIN_MTU)
615 * In theory we should do PMTUD on IPv6 multicast messages but
616 * we don't have an address to send from so just fragment.
618 if (ipv6_addr_type(&ipv6_hdr(skb)->daddr) & IPV6_ADDR_MULTICAST)
621 if (!ipv6_should_icmp(skb))
629 if (old_eh->h_proto == htons(ETH_P_8021Q))
630 eth_hdr_len = VLAN_ETH_HLEN;
632 payload_length = skb->len - eth_hdr_len;
633 if (skb->protocol == htons(ETH_P_IP)) {
634 header_length = sizeof(struct iphdr) + sizeof(struct icmphdr);
635 total_length = min_t(unsigned int, header_length +
636 payload_length, 576);
638 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
640 header_length = sizeof(struct ipv6hdr) +
641 sizeof(struct icmp6hdr);
642 total_length = min_t(unsigned int, header_length +
643 payload_length, IPV6_MIN_MTU);
647 payload_length = total_length - header_length;
649 nskb = dev_alloc_skb(NET_IP_ALIGN + eth_hdr_len + header_length +
654 skb_reserve(nskb, NET_IP_ALIGN);
656 /* Ethernet / VLAN */
657 eh = (struct ethhdr *)skb_put(nskb, eth_hdr_len);
658 memcpy(eh->h_dest, old_eh->h_source, ETH_ALEN);
659 memcpy(eh->h_source, mutable->eth_addr, ETH_ALEN);
660 nskb->protocol = eh->h_proto = old_eh->h_proto;
661 if (old_eh->h_proto == htons(ETH_P_8021Q)) {
662 struct vlan_ethhdr *vh = (struct vlan_ethhdr *)eh;
664 vh->h_vlan_TCI = vlan_eth_hdr(skb)->h_vlan_TCI;
665 vh->h_vlan_encapsulated_proto = skb->protocol;
667 vlan_set_tci(nskb, vlan_get_tci(skb));
668 skb_reset_mac_header(nskb);
671 if (skb->protocol == htons(ETH_P_IP))
672 ipv4_build_icmp(skb, nskb, mtu, payload_length);
673 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
675 ipv6_build_icmp(skb, nskb, mtu, payload_length);
679 * Assume that flow based keys are symmetric with respect to input
680 * and output and use the key that we were going to put on the
681 * outgoing packet for the fake received packet. If the keys are
682 * not symmetric then PMTUD needs to be disabled since we won't have
683 * any way of synthesizing packets.
685 if ((mutable->flags & (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION)) ==
686 (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION))
687 OVS_CB(nskb)->tun_id = flow_key;
689 if (unlikely(compute_ip_summed(nskb, false))) {
694 vport_receive(vport, nskb);
699 static bool check_mtu(struct sk_buff *skb,
701 const struct tnl_mutable_config *mutable,
702 const struct rtable *rt, __be16 *frag_offp)
704 bool df_inherit = mutable->flags & TNL_F_DF_INHERIT;
705 bool pmtud = mutable->flags & TNL_F_PMTUD;
706 __be16 frag_off = mutable->flags & TNL_F_DF_DEFAULT ? htons(IP_DF) : 0;
708 unsigned int packet_length = skb->len - ETH_HLEN;
710 /* Allow for one level of tagging in the packet length. */
711 if (!vlan_tx_tag_present(skb) &&
712 eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
713 packet_length -= VLAN_HLEN;
718 /* The tag needs to go in packet regardless of where it
719 * currently is, so subtract it from the MTU.
721 if (vlan_tx_tag_present(skb) ||
722 eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
723 vlan_header = VLAN_HLEN;
725 mtu = dst_mtu(&rt_dst(rt))
727 - mutable->tunnel_hlen
731 if (skb->protocol == htons(ETH_P_IP)) {
732 struct iphdr *iph = ip_hdr(skb);
735 frag_off = iph->frag_off & htons(IP_DF);
737 if (pmtud && iph->frag_off & htons(IP_DF)) {
738 mtu = max(mtu, IP_MIN_MTU);
740 if (packet_length > mtu &&
741 tnl_frag_needed(vport, mutable, skb, mtu,
742 OVS_CB(skb)->tun_id))
746 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
747 else if (skb->protocol == htons(ETH_P_IPV6)) {
748 /* IPv6 requires end hosts to do fragmentation
749 * if the packet is above the minimum MTU.
751 if (df_inherit && packet_length > IPV6_MIN_MTU)
752 frag_off = htons(IP_DF);
755 mtu = max(mtu, IPV6_MIN_MTU);
757 if (packet_length > mtu &&
758 tnl_frag_needed(vport, mutable, skb, mtu,
759 OVS_CB(skb)->tun_id))
765 *frag_offp = frag_off;
769 static void create_tunnel_header(const struct vport *vport,
770 const struct tnl_mutable_config *mutable,
771 const struct rtable *rt, void *header)
773 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
774 struct iphdr *iph = header;
777 iph->ihl = sizeof(struct iphdr) >> 2;
778 iph->frag_off = htons(IP_DF);
779 iph->protocol = tnl_vport->tnl_ops->ipproto;
780 iph->tos = mutable->tos;
781 iph->daddr = rt->rt_dst;
782 iph->saddr = rt->rt_src;
783 iph->ttl = mutable->ttl;
785 iph->ttl = ip4_dst_hoplimit(&rt_dst(rt));
787 tnl_vport->tnl_ops->build_header(vport, mutable, iph + 1);
790 static inline void *get_cached_header(const struct tnl_cache *cache)
792 return (void *)cache + ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN);
795 static inline bool check_cache_valid(const struct tnl_cache *cache,
796 const struct tnl_mutable_config *mutable)
803 hh = rt_hh(cache->rt);
805 #ifdef NEED_CACHE_TIMEOUT
806 time_before(jiffies, cache->expiration) &&
809 atomic_read(&init_net.ipv4.rt_genid) == cache->rt->rt_genid &&
812 hh->hh_lock.sequence == cache->hh_seq &&
814 mutable->seq == cache->mutable_seq &&
815 (!is_internal_dev(rt_dst(cache->rt).dev) ||
816 (cache->flow && !cache->flow->dead));
819 static void __cache_cleaner(struct tnl_vport *tnl_vport)
821 const struct tnl_mutable_config *mutable =
822 rcu_dereference(tnl_vport->mutable);
823 const struct tnl_cache *cache = rcu_dereference(tnl_vport->cache);
825 if (cache && !check_cache_valid(cache, mutable) &&
826 spin_trylock_bh(&tnl_vport->cache_lock)) {
827 assign_cache_rcu(tnl_vport_to_vport(tnl_vport), NULL);
828 spin_unlock_bh(&tnl_vport->cache_lock);
832 static void cache_cleaner(struct work_struct *work)
836 schedule_cache_cleaner();
839 for (i = 0; i < PORT_TABLE_SIZE; i++) {
840 struct hlist_node *n;
841 struct hlist_head *bucket;
842 struct tnl_vport *tnl_vport;
844 bucket = &port_table[i];
845 hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node)
846 __cache_cleaner(tnl_vport);
851 static inline void create_eth_hdr(struct tnl_cache *cache,
854 void *cache_data = get_cached_header(cache);
861 hh_seq = read_seqbegin(&hh->hh_lock);
862 hh_off = HH_DATA_ALIGN(hh->hh_len) - hh->hh_len;
863 memcpy(cache_data, (void *)hh->hh_data + hh_off, hh->hh_len);
864 cache->hh_len = hh->hh_len;
865 } while (read_seqretry(&hh->hh_lock, hh_seq));
867 cache->hh_seq = hh_seq;
869 read_lock(&hh->hh_lock);
870 hh_off = HH_DATA_ALIGN(hh->hh_len) - hh->hh_len;
871 memcpy(cache_data, (void *)hh->hh_data + hh_off, hh->hh_len);
872 cache->hh_len = hh->hh_len;
873 read_unlock(&hh->hh_lock);
877 static struct tnl_cache *build_cache(struct vport *vport,
878 const struct tnl_mutable_config *mutable,
881 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
882 struct tnl_cache *cache;
887 if (!(mutable->flags & TNL_F_HDR_CACHE))
891 * If there is no entry in the ARP cache or if this device does not
892 * support hard header caching just fall back to the IP stack.
900 * If lock is contended fall back to directly building the header.
901 * We're not going to help performance by sitting here spinning.
903 if (!spin_trylock(&tnl_vport->cache_lock))
906 cache = cache_dereference(tnl_vport);
907 if (check_cache_valid(cache, mutable))
912 cache_len = LL_RESERVED_SPACE(rt_dst(rt).dev) + mutable->tunnel_hlen;
914 cache = kzalloc(ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN) +
915 cache_len, GFP_ATOMIC);
919 create_eth_hdr(cache, hh);
920 cache_data = get_cached_header(cache) + cache->hh_len;
921 cache->len = cache->hh_len + mutable->tunnel_hlen;
923 create_tunnel_header(vport, mutable, rt, cache_data);
925 cache->mutable_seq = mutable->seq;
927 #ifdef NEED_CACHE_TIMEOUT
928 cache->expiration = jiffies + tnl_vport->cache_exp_interval;
931 if (is_internal_dev(rt_dst(rt).dev)) {
932 struct sw_flow_key flow_key;
933 struct vport *dst_vport;
937 struct sw_flow *flow;
939 dst_vport = internal_dev_get_vport(rt_dst(rt).dev);
943 skb = alloc_skb(cache->len, GFP_ATOMIC);
947 __skb_put(skb, cache->len);
948 memcpy(skb->data, get_cached_header(cache), cache->len);
950 err = flow_extract(skb, dst_vport->port_no, &flow_key,
957 flow = flow_tbl_lookup(rcu_dereference(dst_vport->dp->table),
958 &flow_key, flow_key_len);
966 assign_cache_rcu(vport, cache);
969 spin_unlock(&tnl_vport->cache_lock);
974 static struct rtable *__find_route(const struct tnl_mutable_config *mutable,
977 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
978 struct flowi fl = { .nl_u = { .ip4_u =
979 { .daddr = mutable->key.daddr,
980 .saddr = mutable->key.saddr,
985 if (unlikely(ip_route_output_key(&init_net, &rt, &fl)))
986 return ERR_PTR(-EADDRNOTAVAIL);
990 struct flowi4 fl = { .daddr = mutable->key.daddr,
991 .saddr = mutable->key.saddr,
993 .flowi4_proto = ipproto };
995 return ip_route_output_key(&init_net, &fl);
999 static struct rtable *find_route(struct vport *vport,
1000 const struct tnl_mutable_config *mutable,
1001 u8 tos, struct tnl_cache **cache)
1003 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1004 struct tnl_cache *cur_cache = rcu_dereference(tnl_vport->cache);
1009 if (likely(tos == mutable->tos && check_cache_valid(cur_cache, mutable))) {
1011 return cur_cache->rt;
1015 rt = __find_route(mutable, tnl_vport->tnl_ops->ipproto, tos);
1019 if (likely(tos == mutable->tos))
1020 *cache = build_cache(vport, mutable, rt);
1026 static inline bool need_linearize(const struct sk_buff *skb)
1030 if (unlikely(skb_shinfo(skb)->frag_list))
1034 * Generally speaking we should linearize if there are paged frags.
1035 * However, if all of the refcounts are 1 we know nobody else can
1036 * change them from underneath us and we can skip the linearization.
1038 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1039 if (unlikely(page_count(skb_shinfo(skb)->frags[i].page) > 1))
1045 static struct sk_buff *handle_offloads(struct sk_buff *skb,
1046 const struct tnl_mutable_config *mutable,
1047 const struct rtable *rt)
1052 min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
1053 + mutable->tunnel_hlen
1054 + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1056 if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
1057 int head_delta = SKB_DATA_ALIGN(min_headroom -
1060 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
1066 forward_ip_summed(skb, true);
1068 if (skb_is_gso(skb)) {
1069 struct sk_buff *nskb;
1071 nskb = skb_gso_segment(skb, 0);
1074 err = PTR_ERR(nskb);
1080 } else if (get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
1081 /* Pages aren't locked and could change at any time.
1082 * If this happens after we compute the checksum, the
1083 * checksum will be wrong. We linearize now to avoid
1086 if (unlikely(need_linearize(skb))) {
1087 err = __skb_linearize(skb);
1092 err = skb_checksum_help(skb);
1097 set_ip_summed(skb, OVS_CSUM_NONE);
1104 return ERR_PTR(err);
1107 static int send_frags(struct sk_buff *skb,
1108 const struct tnl_mutable_config *mutable)
1114 struct sk_buff *next = skb->next;
1115 int frag_len = skb->len - mutable->tunnel_hlen;
1119 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
1121 err = ip_local_out(skb);
1123 if (unlikely(net_xmit_eval(err)))
1125 sent_len += frag_len;
1132 * There's no point in continuing to send fragments once one has been
1133 * dropped so just free the rest. This may help improve the congestion
1134 * that caused the first packet to be dropped.
1136 tnl_free_linked_skbs(skb);
1140 int tnl_send(struct vport *vport, struct sk_buff *skb)
1142 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1143 const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
1145 enum vport_err_type err = VPORT_E_TX_ERROR;
1147 struct dst_entry *unattached_dst = NULL;
1148 struct tnl_cache *cache;
1150 __be16 frag_off = 0;
1155 /* Validate the protocol headers before we try to use them. */
1156 if (skb->protocol == htons(ETH_P_8021Q) &&
1157 !vlan_tx_tag_present(skb)) {
1158 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
1161 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
1162 skb_set_network_header(skb, VLAN_ETH_HLEN);
1165 if (skb->protocol == htons(ETH_P_IP)) {
1166 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1167 + sizeof(struct iphdr))))
1170 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1171 else if (skb->protocol == htons(ETH_P_IPV6)) {
1172 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1173 + sizeof(struct ipv6hdr))))
1179 if (skb->protocol == htons(ETH_P_IP))
1180 inner_tos = ip_hdr(skb)->tos;
1181 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1182 else if (skb->protocol == htons(ETH_P_IPV6))
1183 inner_tos = ipv6_get_dsfield(ipv6_hdr(skb));
1188 if (mutable->flags & TNL_F_TOS_INHERIT)
1193 tos = INET_ECN_encapsulate(tos, inner_tos);
1196 rt = find_route(vport, mutable, tos, &cache);
1199 if (unlikely(!cache))
1200 unattached_dst = &rt_dst(rt);
1206 skb_clear_rxhash(skb);
1209 skb = handle_offloads(skb, mutable, rt);
1214 if (unlikely(!check_mtu(skb, vport, mutable, rt, &frag_off))) {
1215 err = VPORT_E_TX_DROPPED;
1220 * If we are over the MTU, allow the IP stack to handle fragmentation.
1221 * Fragmentation is a slow path anyways.
1223 if (unlikely(skb->len + mutable->tunnel_hlen > dst_mtu(&rt_dst(rt)) &&
1225 unattached_dst = &rt_dst(rt);
1226 dst_hold(unattached_dst);
1233 ttl = ip4_dst_hoplimit(&rt_dst(rt));
1235 if (mutable->flags & TNL_F_TTL_INHERIT) {
1236 if (skb->protocol == htons(ETH_P_IP))
1237 ttl = ip_hdr(skb)->ttl;
1238 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1239 else if (skb->protocol == htons(ETH_P_IPV6))
1240 ttl = ipv6_hdr(skb)->hop_limit;
1246 struct sk_buff *next_skb = skb->next;
1249 if (unlikely(vlan_deaccel_tag(skb)))
1252 if (likely(cache)) {
1253 skb_push(skb, cache->len);
1254 memcpy(skb->data, get_cached_header(cache), cache->len);
1255 skb_reset_mac_header(skb);
1256 skb_set_network_header(skb, cache->hh_len);
1259 skb_push(skb, mutable->tunnel_hlen);
1260 create_tunnel_header(vport, mutable, rt, skb->data);
1261 skb_reset_network_header(skb);
1264 skb_dst_set(skb, dst_clone(unattached_dst));
1266 skb_dst_set(skb, unattached_dst);
1267 unattached_dst = NULL;
1270 skb_set_transport_header(skb, skb_network_offset(skb) + sizeof(struct iphdr));
1275 iph->frag_off = frag_off;
1276 ip_select_ident(iph, &rt_dst(rt), NULL);
1278 skb = tnl_vport->tnl_ops->update_header(vport, mutable, &rt_dst(rt), skb);
1282 if (likely(cache)) {
1283 int orig_len = skb->len - cache->len;
1284 struct vport *cache_vport = internal_dev_get_vport(rt_dst(rt).dev);
1286 skb->protocol = htons(ETH_P_IP);
1288 iph->tot_len = htons(skb->len - skb_network_offset(skb));
1292 if (unlikely(compute_ip_summed(skb, true))) {
1297 OVS_CB(skb)->flow = cache->flow;
1298 vport_receive(cache_vport, skb);
1299 sent_len += orig_len;
1303 skb->dev = rt_dst(rt).dev;
1304 xmit_err = dev_queue_xmit(skb);
1306 if (likely(net_xmit_eval(xmit_err) == 0))
1307 sent_len += orig_len;
1310 sent_len += send_frags(skb, mutable);
1316 if (unlikely(sent_len == 0))
1317 vport_record_error(vport, VPORT_E_TX_DROPPED);
1322 tnl_free_linked_skbs(skb);
1324 vport_record_error(vport, err);
1326 dst_release(unattached_dst);
1330 static const struct nla_policy tnl_policy[OVS_TUNNEL_ATTR_MAX + 1] = {
1331 [OVS_TUNNEL_ATTR_FLAGS] = { .type = NLA_U32 },
1332 [OVS_TUNNEL_ATTR_DST_IPV4] = { .type = NLA_U32 },
1333 [OVS_TUNNEL_ATTR_SRC_IPV4] = { .type = NLA_U32 },
1334 [OVS_TUNNEL_ATTR_OUT_KEY] = { .type = NLA_U64 },
1335 [OVS_TUNNEL_ATTR_IN_KEY] = { .type = NLA_U64 },
1336 [OVS_TUNNEL_ATTR_TOS] = { .type = NLA_U8 },
1337 [OVS_TUNNEL_ATTR_TTL] = { .type = NLA_U8 },
1340 /* Sets OVS_TUNNEL_ATTR_* fields in 'mutable', which must initially be zeroed. */
1341 static int tnl_set_config(struct nlattr *options, const struct tnl_ops *tnl_ops,
1342 const struct vport *cur_vport,
1343 struct tnl_mutable_config *mutable)
1345 const struct vport *old_vport;
1346 const struct tnl_mutable_config *old_mutable;
1347 struct nlattr *a[OVS_TUNNEL_ATTR_MAX + 1];
1353 err = nla_parse_nested(a, OVS_TUNNEL_ATTR_MAX, options, tnl_policy);
1357 if (!a[OVS_TUNNEL_ATTR_FLAGS] || !a[OVS_TUNNEL_ATTR_DST_IPV4])
1360 mutable->flags = nla_get_u32(a[OVS_TUNNEL_ATTR_FLAGS]) & TNL_F_PUBLIC;
1362 mutable->key.daddr = nla_get_be32(a[OVS_TUNNEL_ATTR_DST_IPV4]);
1363 if (a[OVS_TUNNEL_ATTR_SRC_IPV4]) {
1364 if (ipv4_is_multicast(mutable->key.daddr))
1366 mutable->key.saddr = nla_get_be32(a[OVS_TUNNEL_ATTR_SRC_IPV4]);
1369 if (a[OVS_TUNNEL_ATTR_TOS]) {
1370 mutable->tos = nla_get_u8(a[OVS_TUNNEL_ATTR_TOS]);
1371 if (mutable->tos != RT_TOS(mutable->tos))
1375 if (a[OVS_TUNNEL_ATTR_TTL])
1376 mutable->ttl = nla_get_u8(a[OVS_TUNNEL_ATTR_TTL]);
1378 mutable->key.tunnel_type = tnl_ops->tunnel_type;
1379 if (!a[OVS_TUNNEL_ATTR_IN_KEY]) {
1380 mutable->key.tunnel_type |= TNL_T_KEY_MATCH;
1381 mutable->flags |= TNL_F_IN_KEY_MATCH;
1383 mutable->key.tunnel_type |= TNL_T_KEY_EXACT;
1384 mutable->key.in_key = nla_get_be64(a[OVS_TUNNEL_ATTR_IN_KEY]);
1387 if (!a[OVS_TUNNEL_ATTR_OUT_KEY])
1388 mutable->flags |= TNL_F_OUT_KEY_ACTION;
1390 mutable->out_key = nla_get_be64(a[OVS_TUNNEL_ATTR_OUT_KEY]);
1392 mutable->tunnel_hlen = tnl_ops->hdr_len(mutable);
1393 if (mutable->tunnel_hlen < 0)
1394 return mutable->tunnel_hlen;
1396 mutable->tunnel_hlen += sizeof(struct iphdr);
1398 old_vport = port_table_lookup(&mutable->key, &old_mutable);
1399 if (old_vport && old_vport != cur_vport)
1403 if (ipv4_is_multicast(mutable->key.daddr)) {
1404 struct net_device *dev;
1407 rt = __find_route(mutable, tnl_ops->ipproto, mutable->tos);
1409 return -EADDRNOTAVAIL;
1410 dev = rt_dst(rt).dev;
1412 if (__in_dev_get_rtnl(dev) == NULL)
1413 return -EADDRNOTAVAIL;
1414 mutable->mlink = dev->ifindex;
1415 ip_mc_inc_group(__in_dev_get_rtnl(dev), mutable->key.daddr);
1421 struct vport *tnl_create(const struct vport_parms *parms,
1422 const struct vport_ops *vport_ops,
1423 const struct tnl_ops *tnl_ops)
1425 struct vport *vport;
1426 struct tnl_vport *tnl_vport;
1427 struct tnl_mutable_config *mutable;
1428 int initial_frag_id;
1431 vport = vport_alloc(sizeof(struct tnl_vport), vport_ops, parms);
1432 if (IS_ERR(vport)) {
1433 err = PTR_ERR(vport);
1437 tnl_vport = tnl_vport_priv(vport);
1439 strcpy(tnl_vport->name, parms->name);
1440 tnl_vport->tnl_ops = tnl_ops;
1442 mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1445 goto error_free_vport;
1448 vport_gen_rand_ether_addr(mutable->eth_addr);
1450 get_random_bytes(&initial_frag_id, sizeof(int));
1451 atomic_set(&tnl_vport->frag_id, initial_frag_id);
1453 err = tnl_set_config(parms->options, tnl_ops, NULL, mutable);
1455 goto error_free_mutable;
1457 spin_lock_init(&tnl_vport->cache_lock);
1459 #ifdef NEED_CACHE_TIMEOUT
1460 tnl_vport->cache_exp_interval = MAX_CACHE_EXP -
1461 (net_random() % (MAX_CACHE_EXP / 2));
1464 rcu_assign_pointer(tnl_vport->mutable, mutable);
1466 port_table_add_port(vport);
1470 free_mutable_rtnl(mutable);
1475 return ERR_PTR(err);
1478 int tnl_set_options(struct vport *vport, struct nlattr *options)
1480 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1481 const struct tnl_mutable_config *old_mutable;
1482 struct tnl_mutable_config *mutable;
1485 mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1491 /* Copy fields whose values should be retained. */
1492 old_mutable = rtnl_dereference(tnl_vport->mutable);
1493 mutable->seq = old_mutable->seq + 1;
1494 memcpy(mutable->eth_addr, old_mutable->eth_addr, ETH_ALEN);
1496 /* Parse the others configured by userspace. */
1497 err = tnl_set_config(options, tnl_vport->tnl_ops, vport, mutable);
1501 if (port_hash(&mutable->key) != port_hash(&old_mutable->key))
1502 port_table_move_port(vport, mutable);
1504 assign_config_rcu(vport, mutable);
1509 free_mutable_rtnl(mutable);
1515 int tnl_get_options(const struct vport *vport, struct sk_buff *skb)
1517 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1518 const struct tnl_mutable_config *mutable = rcu_dereference_rtnl(tnl_vport->mutable);
1520 NLA_PUT_U32(skb, OVS_TUNNEL_ATTR_FLAGS, mutable->flags & TNL_F_PUBLIC);
1521 NLA_PUT_BE32(skb, OVS_TUNNEL_ATTR_DST_IPV4, mutable->key.daddr);
1523 if (!(mutable->flags & TNL_F_IN_KEY_MATCH))
1524 NLA_PUT_BE64(skb, OVS_TUNNEL_ATTR_IN_KEY, mutable->key.in_key);
1525 if (!(mutable->flags & TNL_F_OUT_KEY_ACTION))
1526 NLA_PUT_BE64(skb, OVS_TUNNEL_ATTR_OUT_KEY, mutable->out_key);
1527 if (mutable->key.saddr)
1528 NLA_PUT_BE32(skb, OVS_TUNNEL_ATTR_SRC_IPV4, mutable->key.saddr);
1530 NLA_PUT_U8(skb, OVS_TUNNEL_ATTR_TOS, mutable->tos);
1532 NLA_PUT_U8(skb, OVS_TUNNEL_ATTR_TTL, mutable->ttl);
1540 static void free_port_rcu(struct rcu_head *rcu)
1542 struct tnl_vport *tnl_vport = container_of(rcu,
1543 struct tnl_vport, rcu);
1545 free_cache((struct tnl_cache __force *)tnl_vport->cache);
1546 kfree((struct tnl_mutable __force *)tnl_vport->mutable);
1547 vport_free(tnl_vport_to_vport(tnl_vport));
1550 void tnl_destroy(struct vport *vport)
1552 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1553 struct tnl_mutable_config *mutable;
1555 mutable = rtnl_dereference(tnl_vport->mutable);
1556 port_table_remove_port(vport);
1557 free_mutable_rtnl(mutable);
1558 call_rcu(&tnl_vport->rcu, free_port_rcu);
1561 int tnl_set_addr(struct vport *vport, const unsigned char *addr)
1563 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1564 struct tnl_mutable_config *old_mutable, *mutable;
1566 old_mutable = rtnl_dereference(tnl_vport->mutable);
1567 mutable = kmemdup(old_mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL);
1571 old_mutable->mlink = 0;
1573 memcpy(mutable->eth_addr, addr, ETH_ALEN);
1574 assign_config_rcu(vport, mutable);
1579 const char *tnl_get_name(const struct vport *vport)
1581 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1582 return tnl_vport->name;
1585 const unsigned char *tnl_get_addr(const struct vport *vport)
1587 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1588 return rcu_dereference_rtnl(tnl_vport->mutable)->eth_addr;
1591 void tnl_free_linked_skbs(struct sk_buff *skb)
1594 struct sk_buff *next = skb->next;
1604 port_table = kmalloc(PORT_TABLE_SIZE * sizeof(struct hlist_head *),
1609 for (i = 0; i < PORT_TABLE_SIZE; i++)
1610 INIT_HLIST_HEAD(&port_table[i]);
1619 for (i = 0; i < PORT_TABLE_SIZE; i++) {
1620 struct tnl_vport * tnl_vport;
1621 struct hlist_head *hash_head;
1622 struct hlist_node *n;
1624 hash_head = &port_table[i];
1625 hlist_for_each_entry(tnl_vport, n, hash_head, hash_node) {