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>
40 #include "vport-generic.h"
41 #include "vport-internal_dev.h"
43 #ifdef NEED_CACHE_TIMEOUT
45 * On kernels where we can't quickly detect changes in the rest of the system
46 * we use an expiration time to invalidate the cache. A shorter expiration
47 * reduces the length of time that we may potentially blackhole packets while
48 * a longer time increases performance by reducing the frequency that the
49 * cache needs to be rebuilt. A variety of factors may cause the cache to be
50 * invalidated before the expiration time but this is the maximum. The time
51 * is expressed in jiffies.
53 #define MAX_CACHE_EXP HZ
57 * Interval to check for and remove caches that are no longer valid. Caches
58 * are checked for validity before they are used for packet encapsulation and
59 * old caches are removed at that time. However, if no packets are sent through
60 * the tunnel then the cache will never be destroyed. Since it holds
61 * references to a number of system objects, the cache will continue to use
62 * system resources by not allowing those objects to be destroyed. The cache
63 * cleaner is periodically run to free invalid caches. It does not
64 * significantly affect system performance. A lower interval will release
65 * resources faster but will itself consume resources by requiring more frequent
66 * checks. A longer interval may result in messages being printed to the kernel
67 * message buffer about unreleased resources. The interval is expressed in
70 #define CACHE_CLEANER_INTERVAL (5 * HZ)
72 #define CACHE_DATA_ALIGN 16
73 #define PORT_TABLE_SIZE 1024
75 static struct hlist_head *port_table __read_mostly;
76 static int port_table_count;
78 static void cache_cleaner(struct work_struct *work);
79 static DECLARE_DELAYED_WORK(cache_cleaner_wq, cache_cleaner);
82 * These are just used as an optimization: they don't require any kind of
83 * synchronization because we could have just as easily read the value before
84 * the port change happened.
86 static unsigned int key_local_remote_ports __read_mostly;
87 static unsigned int key_remote_ports __read_mostly;
88 static unsigned int key_multicast_ports __read_mostly;
89 static unsigned int local_remote_ports __read_mostly;
90 static unsigned int remote_ports __read_mostly;
91 static unsigned int multicast_ports __read_mostly;
93 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
94 #define rt_dst(rt) (rt->dst)
96 #define rt_dst(rt) (rt->u.dst)
99 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,1,0)
100 static struct hh_cache *rt_hh(struct rtable *rt)
102 struct neighbour *neigh = dst_get_neighbour(&rt->dst);
103 if (!neigh || !(neigh->nud_state & NUD_CONNECTED) ||
109 #define rt_hh(rt) (rt_dst(rt).hh)
112 static struct vport *tnl_vport_to_vport(const struct tnl_vport *tnl_vport)
114 return vport_from_priv(tnl_vport);
117 /* This is analogous to rtnl_dereference for the tunnel cache. It checks that
118 * cache_lock is held, so it is only for update side code.
120 static struct tnl_cache *cache_dereference(struct tnl_vport *tnl_vport)
122 return rcu_dereference_protected(tnl_vport->cache,
123 lockdep_is_held(&tnl_vport->cache_lock));
126 static void schedule_cache_cleaner(void)
128 schedule_delayed_work(&cache_cleaner_wq, CACHE_CLEANER_INTERVAL);
131 static void free_cache(struct tnl_cache *cache)
136 flow_put(cache->flow);
137 ip_rt_put(cache->rt);
141 static void free_config_rcu(struct rcu_head *rcu)
143 struct tnl_mutable_config *c = container_of(rcu, struct tnl_mutable_config, rcu);
147 static void free_cache_rcu(struct rcu_head *rcu)
149 struct tnl_cache *c = container_of(rcu, struct tnl_cache, rcu);
153 /* Frees the portion of 'mutable' that requires RTNL and thus can't happen
154 * within an RCU callback. Fortunately this part doesn't require waiting for
155 * an RCU grace period.
157 static void free_mutable_rtnl(struct tnl_mutable_config *mutable)
160 if (ipv4_is_multicast(mutable->key.daddr) && mutable->mlink) {
161 struct in_device *in_dev;
162 in_dev = inetdev_by_index(&init_net, mutable->mlink);
164 ip_mc_dec_group(in_dev, mutable->key.daddr);
168 static void assign_config_rcu(struct vport *vport,
169 struct tnl_mutable_config *new_config)
171 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
172 struct tnl_mutable_config *old_config;
174 old_config = rtnl_dereference(tnl_vport->mutable);
175 rcu_assign_pointer(tnl_vport->mutable, new_config);
177 free_mutable_rtnl(old_config);
178 call_rcu(&old_config->rcu, free_config_rcu);
181 static void assign_cache_rcu(struct vport *vport, struct tnl_cache *new_cache)
183 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
184 struct tnl_cache *old_cache;
186 old_cache = cache_dereference(tnl_vport);
187 rcu_assign_pointer(tnl_vport->cache, new_cache);
190 call_rcu(&old_cache->rcu, free_cache_rcu);
193 static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable)
195 bool is_multicast = ipv4_is_multicast(mutable->key.daddr);
197 if (mutable->flags & TNL_F_IN_KEY_MATCH) {
198 if (mutable->key.saddr)
199 return &local_remote_ports;
200 else if (is_multicast)
201 return &multicast_ports;
203 return &remote_ports;
205 if (mutable->key.saddr)
206 return &key_local_remote_ports;
207 else if (is_multicast)
208 return &key_multicast_ports;
210 return &key_remote_ports;
214 static u32 port_hash(const struct port_lookup_key *key)
216 return jhash2((u32 *)key, (PORT_KEY_LEN / sizeof(u32)), 0);
219 static struct hlist_head *find_bucket(u32 hash)
221 return &port_table[(hash & (PORT_TABLE_SIZE - 1))];
224 static void port_table_add_port(struct vport *vport)
226 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
227 const struct tnl_mutable_config *mutable;
230 if (port_table_count == 0)
231 schedule_cache_cleaner();
233 mutable = rtnl_dereference(tnl_vport->mutable);
234 hash = port_hash(&mutable->key);
235 hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
238 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
241 static void port_table_move_port(struct vport *vport,
242 struct tnl_mutable_config *new_mutable)
244 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
247 hash = port_hash(&new_mutable->key);
248 hlist_del_init_rcu(&tnl_vport->hash_node);
249 hlist_add_head_rcu(&tnl_vport->hash_node, find_bucket(hash));
251 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
252 assign_config_rcu(vport, new_mutable);
253 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))++;
256 static void port_table_remove_port(struct vport *vport)
258 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
260 hlist_del_init_rcu(&tnl_vport->hash_node);
263 if (port_table_count == 0)
264 cancel_delayed_work_sync(&cache_cleaner_wq);
266 (*find_port_pool(rtnl_dereference(tnl_vport->mutable)))--;
269 static struct vport *port_table_lookup(struct port_lookup_key *key,
270 const struct tnl_mutable_config **pmutable)
272 struct hlist_node *n;
273 struct hlist_head *bucket;
274 u32 hash = port_hash(key);
275 struct tnl_vport *tnl_vport;
277 bucket = find_bucket(hash);
279 hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node) {
280 struct tnl_mutable_config *mutable;
282 mutable = rcu_dereference_rtnl(tnl_vport->mutable);
283 if (!memcmp(&mutable->key, key, PORT_KEY_LEN)) {
285 return tnl_vport_to_vport(tnl_vport);
292 struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be64 key,
294 const struct tnl_mutable_config **mutable)
296 struct port_lookup_key lookup;
298 bool is_multicast = ipv4_is_multicast(saddr);
300 lookup.saddr = saddr;
301 lookup.daddr = daddr;
303 /* First try for exact match on in_key. */
305 lookup.tunnel_type = tunnel_type | TNL_T_KEY_EXACT;
306 if (!is_multicast && key_local_remote_ports) {
307 vport = port_table_lookup(&lookup, mutable);
311 if (key_remote_ports) {
313 vport = port_table_lookup(&lookup, mutable);
317 lookup.saddr = saddr;
320 /* Then try matches that wildcard in_key. */
322 lookup.tunnel_type = tunnel_type | TNL_T_KEY_MATCH;
323 if (!is_multicast && local_remote_ports) {
324 vport = port_table_lookup(&lookup, mutable);
330 vport = port_table_lookup(&lookup, mutable);
337 lookup.daddr = saddr;
338 if (key_multicast_ports) {
339 lookup.tunnel_type = tunnel_type | TNL_T_KEY_EXACT;
341 vport = port_table_lookup(&lookup, mutable);
345 if (multicast_ports) {
346 lookup.tunnel_type = tunnel_type | TNL_T_KEY_MATCH;
348 vport = port_table_lookup(&lookup, mutable);
357 static void ecn_decapsulate(struct sk_buff *skb, u8 tos)
359 if (unlikely(INET_ECN_is_ce(tos))) {
360 __be16 protocol = skb->protocol;
362 skb_set_network_header(skb, ETH_HLEN);
364 if (protocol == htons(ETH_P_8021Q)) {
365 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
368 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
369 skb_set_network_header(skb, VLAN_ETH_HLEN);
372 if (protocol == htons(ETH_P_IP)) {
373 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
374 + sizeof(struct iphdr))))
377 IP_ECN_set_ce(ip_hdr(skb));
379 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
380 else if (protocol == htons(ETH_P_IPV6)) {
381 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
382 + sizeof(struct ipv6hdr))))
385 IP6_ECN_set_ce(ipv6_hdr(skb));
392 * tnl_rcv - ingress point for generic tunnel code
394 * @vport: port this packet was received on
395 * @skb: received packet
396 * @tos: ToS from encapsulating IP packet, used to copy ECN bits
398 * Must be called with rcu_read_lock.
400 * Packets received by this function are in the following state:
401 * - skb->data points to the inner Ethernet header.
402 * - The inner Ethernet header is in the linear data area.
403 * - skb->csum does not include the inner Ethernet header.
404 * - The layer pointers are undefined.
406 void tnl_rcv(struct vport *vport, struct sk_buff *skb, u8 tos)
410 skb_reset_mac_header(skb);
413 if (likely(ntohs(eh->h_proto) >= 1536))
414 skb->protocol = eh->h_proto;
416 skb->protocol = htons(ETH_P_802_2);
420 skb_clear_rxhash(skb);
423 ecn_decapsulate(skb, tos);
424 vlan_set_tci(skb, 0);
426 if (unlikely(compute_ip_summed(skb, false))) {
431 vport_receive(vport, skb);
434 static bool check_ipv4_address(__be32 addr)
436 if (ipv4_is_multicast(addr) || ipv4_is_lbcast(addr)
437 || ipv4_is_loopback(addr) || ipv4_is_zeronet(addr))
443 static bool ipv4_should_icmp(struct sk_buff *skb)
445 struct iphdr *old_iph = ip_hdr(skb);
447 /* Don't respond to L2 broadcast. */
448 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
451 /* Don't respond to L3 broadcast or invalid addresses. */
452 if (!check_ipv4_address(old_iph->daddr) ||
453 !check_ipv4_address(old_iph->saddr))
456 /* Only respond to the first fragment. */
457 if (old_iph->frag_off & htons(IP_OFFSET))
460 /* Don't respond to ICMP error messages. */
461 if (old_iph->protocol == IPPROTO_ICMP) {
462 u8 icmp_type, *icmp_typep;
464 icmp_typep = skb_header_pointer(skb, (u8 *)old_iph +
465 (old_iph->ihl << 2) +
466 offsetof(struct icmphdr, type) -
467 skb->data, sizeof(icmp_type),
473 if (*icmp_typep > NR_ICMP_TYPES
474 || (*icmp_typep <= ICMP_PARAMETERPROB
475 && *icmp_typep != ICMP_ECHOREPLY
476 && *icmp_typep != ICMP_ECHO))
483 static void ipv4_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
484 unsigned int mtu, unsigned int payload_length)
486 struct iphdr *iph, *old_iph = ip_hdr(skb);
487 struct icmphdr *icmph;
490 iph = (struct iphdr *)skb_put(nskb, sizeof(struct iphdr));
491 icmph = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr));
492 payload = skb_put(nskb, payload_length);
496 iph->ihl = sizeof(struct iphdr) >> 2;
497 iph->tos = (old_iph->tos & IPTOS_TOS_MASK) |
498 IPTOS_PREC_INTERNETCONTROL;
499 iph->tot_len = htons(sizeof(struct iphdr)
500 + sizeof(struct icmphdr)
502 get_random_bytes(&iph->id, sizeof(iph->id));
505 iph->protocol = IPPROTO_ICMP;
506 iph->daddr = old_iph->saddr;
507 iph->saddr = old_iph->daddr;
512 icmph->type = ICMP_DEST_UNREACH;
513 icmph->code = ICMP_FRAG_NEEDED;
514 icmph->un.gateway = htonl(mtu);
517 nskb->csum = csum_partial((u8 *)icmph, sizeof(struct icmphdr), 0);
518 nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_iph - skb->data,
519 payload, payload_length,
521 icmph->checksum = csum_fold(nskb->csum);
524 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
525 static bool ipv6_should_icmp(struct sk_buff *skb)
527 struct ipv6hdr *old_ipv6h = ipv6_hdr(skb);
529 int payload_off = (u8 *)(old_ipv6h + 1) - skb->data;
530 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
532 /* Check source address is valid. */
533 addr_type = ipv6_addr_type(&old_ipv6h->saddr);
534 if (addr_type & IPV6_ADDR_MULTICAST || addr_type == IPV6_ADDR_ANY)
537 /* Don't reply to unspecified addresses. */
538 if (ipv6_addr_type(&old_ipv6h->daddr) == IPV6_ADDR_ANY)
541 /* Don't respond to ICMP error messages. */
542 payload_off = ipv6_skip_exthdr(skb, payload_off, &nexthdr);
546 if (nexthdr == NEXTHDR_ICMP) {
547 u8 icmp_type, *icmp_typep;
549 icmp_typep = skb_header_pointer(skb, payload_off +
550 offsetof(struct icmp6hdr,
552 sizeof(icmp_type), &icmp_type);
554 if (!icmp_typep || !(*icmp_typep & ICMPV6_INFOMSG_MASK))
561 static void ipv6_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
562 unsigned int mtu, unsigned int payload_length)
564 struct ipv6hdr *ipv6h, *old_ipv6h = ipv6_hdr(skb);
565 struct icmp6hdr *icmp6h;
568 ipv6h = (struct ipv6hdr *)skb_put(nskb, sizeof(struct ipv6hdr));
569 icmp6h = (struct icmp6hdr *)skb_put(nskb, sizeof(struct icmp6hdr));
570 payload = skb_put(nskb, payload_length);
575 memset(&ipv6h->flow_lbl, 0, sizeof(ipv6h->flow_lbl));
576 ipv6h->payload_len = htons(sizeof(struct icmp6hdr)
578 ipv6h->nexthdr = NEXTHDR_ICMP;
579 ipv6h->hop_limit = IPV6_DEFAULT_HOPLIMIT;
580 ipv6_addr_copy(&ipv6h->daddr, &old_ipv6h->saddr);
581 ipv6_addr_copy(&ipv6h->saddr, &old_ipv6h->daddr);
584 icmp6h->icmp6_type = ICMPV6_PKT_TOOBIG;
585 icmp6h->icmp6_code = 0;
586 icmp6h->icmp6_cksum = 0;
587 icmp6h->icmp6_mtu = htonl(mtu);
589 nskb->csum = csum_partial((u8 *)icmp6h, sizeof(struct icmp6hdr), 0);
590 nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_ipv6h - skb->data,
591 payload, payload_length,
593 icmp6h->icmp6_cksum = csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
594 sizeof(struct icmp6hdr)
596 ipv6h->nexthdr, nskb->csum);
600 bool tnl_frag_needed(struct vport *vport,
601 const struct tnl_mutable_config *mutable,
602 struct sk_buff *skb, unsigned int mtu, __be64 flow_key)
604 unsigned int eth_hdr_len = ETH_HLEN;
605 unsigned int total_length = 0, header_length = 0, payload_length;
606 struct ethhdr *eh, *old_eh = eth_hdr(skb);
607 struct sk_buff *nskb;
610 if (skb->protocol == htons(ETH_P_IP)) {
611 if (mtu < IP_MIN_MTU)
614 if (!ipv4_should_icmp(skb))
617 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
618 else if (skb->protocol == htons(ETH_P_IPV6)) {
619 if (mtu < IPV6_MIN_MTU)
623 * In theory we should do PMTUD on IPv6 multicast messages but
624 * we don't have an address to send from so just fragment.
626 if (ipv6_addr_type(&ipv6_hdr(skb)->daddr) & IPV6_ADDR_MULTICAST)
629 if (!ipv6_should_icmp(skb))
637 if (old_eh->h_proto == htons(ETH_P_8021Q))
638 eth_hdr_len = VLAN_ETH_HLEN;
640 payload_length = skb->len - eth_hdr_len;
641 if (skb->protocol == htons(ETH_P_IP)) {
642 header_length = sizeof(struct iphdr) + sizeof(struct icmphdr);
643 total_length = min_t(unsigned int, header_length +
644 payload_length, 576);
646 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
648 header_length = sizeof(struct ipv6hdr) +
649 sizeof(struct icmp6hdr);
650 total_length = min_t(unsigned int, header_length +
651 payload_length, IPV6_MIN_MTU);
655 payload_length = total_length - header_length;
657 nskb = dev_alloc_skb(NET_IP_ALIGN + eth_hdr_len + header_length +
662 skb_reserve(nskb, NET_IP_ALIGN);
664 /* Ethernet / VLAN */
665 eh = (struct ethhdr *)skb_put(nskb, eth_hdr_len);
666 memcpy(eh->h_dest, old_eh->h_source, ETH_ALEN);
667 memcpy(eh->h_source, mutable->eth_addr, ETH_ALEN);
668 nskb->protocol = eh->h_proto = old_eh->h_proto;
669 if (old_eh->h_proto == htons(ETH_P_8021Q)) {
670 struct vlan_ethhdr *vh = (struct vlan_ethhdr *)eh;
672 vh->h_vlan_TCI = vlan_eth_hdr(skb)->h_vlan_TCI;
673 vh->h_vlan_encapsulated_proto = skb->protocol;
675 vlan_set_tci(nskb, vlan_get_tci(skb));
676 skb_reset_mac_header(nskb);
679 if (skb->protocol == htons(ETH_P_IP))
680 ipv4_build_icmp(skb, nskb, mtu, payload_length);
681 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
683 ipv6_build_icmp(skb, nskb, mtu, payload_length);
687 * Assume that flow based keys are symmetric with respect to input
688 * and output and use the key that we were going to put on the
689 * outgoing packet for the fake received packet. If the keys are
690 * not symmetric then PMTUD needs to be disabled since we won't have
691 * any way of synthesizing packets.
693 if ((mutable->flags & (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION)) ==
694 (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION))
695 OVS_CB(nskb)->tun_id = flow_key;
697 if (unlikely(compute_ip_summed(nskb, false))) {
702 vport_receive(vport, nskb);
707 static bool check_mtu(struct sk_buff *skb,
709 const struct tnl_mutable_config *mutable,
710 const struct rtable *rt, __be16 *frag_offp)
712 bool df_inherit = mutable->flags & TNL_F_DF_INHERIT;
713 bool pmtud = mutable->flags & TNL_F_PMTUD;
714 __be16 frag_off = mutable->flags & TNL_F_DF_DEFAULT ? htons(IP_DF) : 0;
716 unsigned int packet_length = skb->len - ETH_HLEN;
718 /* Allow for one level of tagging in the packet length. */
719 if (!vlan_tx_tag_present(skb) &&
720 eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
721 packet_length -= VLAN_HLEN;
726 /* The tag needs to go in packet regardless of where it
727 * currently is, so subtract it from the MTU.
729 if (vlan_tx_tag_present(skb) ||
730 eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
731 vlan_header = VLAN_HLEN;
733 mtu = dst_mtu(&rt_dst(rt))
735 - mutable->tunnel_hlen
739 if (skb->protocol == htons(ETH_P_IP)) {
740 struct iphdr *iph = ip_hdr(skb);
743 frag_off = iph->frag_off & htons(IP_DF);
745 if (pmtud && iph->frag_off & htons(IP_DF)) {
746 mtu = max(mtu, IP_MIN_MTU);
748 if (packet_length > mtu &&
749 tnl_frag_needed(vport, mutable, skb, mtu,
750 OVS_CB(skb)->tun_id))
754 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
755 else if (skb->protocol == htons(ETH_P_IPV6)) {
756 /* IPv6 requires end hosts to do fragmentation
757 * if the packet is above the minimum MTU.
759 if (df_inherit && packet_length > IPV6_MIN_MTU)
760 frag_off = htons(IP_DF);
763 mtu = max(mtu, IPV6_MIN_MTU);
765 if (packet_length > mtu &&
766 tnl_frag_needed(vport, mutable, skb, mtu,
767 OVS_CB(skb)->tun_id))
773 *frag_offp = frag_off;
777 static void create_tunnel_header(const struct vport *vport,
778 const struct tnl_mutable_config *mutable,
779 const struct rtable *rt, void *header)
781 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
782 struct iphdr *iph = header;
785 iph->ihl = sizeof(struct iphdr) >> 2;
786 iph->frag_off = htons(IP_DF);
787 iph->protocol = tnl_vport->tnl_ops->ipproto;
788 iph->tos = mutable->tos;
789 iph->daddr = rt->rt_dst;
790 iph->saddr = rt->rt_src;
791 iph->ttl = mutable->ttl;
793 iph->ttl = ip4_dst_hoplimit(&rt_dst(rt));
795 tnl_vport->tnl_ops->build_header(vport, mutable, iph + 1);
798 static void *get_cached_header(const struct tnl_cache *cache)
800 return (void *)cache + ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN);
803 static bool check_cache_valid(const struct tnl_cache *cache,
804 const struct tnl_mutable_config *mutable)
811 hh = rt_hh(cache->rt);
813 #ifdef NEED_CACHE_TIMEOUT
814 time_before(jiffies, cache->expiration) &&
817 atomic_read(&init_net.ipv4.rt_genid) == cache->rt->rt_genid &&
820 hh->hh_lock.sequence == cache->hh_seq &&
822 mutable->seq == cache->mutable_seq &&
823 (!is_internal_dev(rt_dst(cache->rt).dev) ||
824 (cache->flow && !cache->flow->dead));
827 static void __cache_cleaner(struct tnl_vport *tnl_vport)
829 const struct tnl_mutable_config *mutable =
830 rcu_dereference(tnl_vport->mutable);
831 const struct tnl_cache *cache = rcu_dereference(tnl_vport->cache);
833 if (cache && !check_cache_valid(cache, mutable) &&
834 spin_trylock_bh(&tnl_vport->cache_lock)) {
835 assign_cache_rcu(tnl_vport_to_vport(tnl_vport), NULL);
836 spin_unlock_bh(&tnl_vport->cache_lock);
840 static void cache_cleaner(struct work_struct *work)
844 schedule_cache_cleaner();
847 for (i = 0; i < PORT_TABLE_SIZE; i++) {
848 struct hlist_node *n;
849 struct hlist_head *bucket;
850 struct tnl_vport *tnl_vport;
852 bucket = &port_table[i];
853 hlist_for_each_entry_rcu(tnl_vport, n, bucket, hash_node)
854 __cache_cleaner(tnl_vport);
859 static void create_eth_hdr(struct tnl_cache *cache, struct hh_cache *hh)
861 void *cache_data = get_cached_header(cache);
868 hh_seq = read_seqbegin(&hh->hh_lock);
869 hh_off = HH_DATA_ALIGN(hh->hh_len) - hh->hh_len;
870 memcpy(cache_data, (void *)hh->hh_data + hh_off, hh->hh_len);
871 cache->hh_len = hh->hh_len;
872 } while (read_seqretry(&hh->hh_lock, hh_seq));
874 cache->hh_seq = hh_seq;
876 read_lock(&hh->hh_lock);
877 hh_off = HH_DATA_ALIGN(hh->hh_len) - hh->hh_len;
878 memcpy(cache_data, (void *)hh->hh_data + hh_off, hh->hh_len);
879 cache->hh_len = hh->hh_len;
880 read_unlock(&hh->hh_lock);
884 static struct tnl_cache *build_cache(struct vport *vport,
885 const struct tnl_mutable_config *mutable,
888 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
889 struct tnl_cache *cache;
894 if (!(mutable->flags & TNL_F_HDR_CACHE))
898 * If there is no entry in the ARP cache or if this device does not
899 * support hard header caching just fall back to the IP stack.
907 * If lock is contended fall back to directly building the header.
908 * We're not going to help performance by sitting here spinning.
910 if (!spin_trylock(&tnl_vport->cache_lock))
913 cache = cache_dereference(tnl_vport);
914 if (check_cache_valid(cache, mutable))
919 cache_len = LL_RESERVED_SPACE(rt_dst(rt).dev) + mutable->tunnel_hlen;
921 cache = kzalloc(ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN) +
922 cache_len, GFP_ATOMIC);
926 create_eth_hdr(cache, hh);
927 cache_data = get_cached_header(cache) + cache->hh_len;
928 cache->len = cache->hh_len + mutable->tunnel_hlen;
930 create_tunnel_header(vport, mutable, rt, cache_data);
932 cache->mutable_seq = mutable->seq;
934 #ifdef NEED_CACHE_TIMEOUT
935 cache->expiration = jiffies + tnl_vport->cache_exp_interval;
938 if (is_internal_dev(rt_dst(rt).dev)) {
939 struct sw_flow_key flow_key;
940 struct vport *dst_vport;
944 struct sw_flow *flow;
946 dst_vport = internal_dev_get_vport(rt_dst(rt).dev);
950 skb = alloc_skb(cache->len, GFP_ATOMIC);
954 __skb_put(skb, cache->len);
955 memcpy(skb->data, get_cached_header(cache), cache->len);
957 err = flow_extract(skb, dst_vport->port_no, &flow_key,
964 flow = flow_tbl_lookup(rcu_dereference(dst_vport->dp->table),
965 &flow_key, flow_key_len);
973 assign_cache_rcu(vport, cache);
976 spin_unlock(&tnl_vport->cache_lock);
981 static struct rtable *__find_route(const struct tnl_mutable_config *mutable,
984 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,39)
985 struct flowi fl = { .nl_u = { .ip4_u = {
986 .daddr = mutable->key.daddr,
987 .saddr = mutable->key.saddr,
992 if (unlikely(ip_route_output_key(&init_net, &rt, &fl)))
993 return ERR_PTR(-EADDRNOTAVAIL);
997 struct flowi4 fl = { .daddr = mutable->key.daddr,
998 .saddr = mutable->key.saddr,
1000 .flowi4_proto = ipproto };
1002 return ip_route_output_key(&init_net, &fl);
1006 static struct rtable *find_route(struct vport *vport,
1007 const struct tnl_mutable_config *mutable,
1008 u8 tos, struct tnl_cache **cache)
1010 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1011 struct tnl_cache *cur_cache = rcu_dereference(tnl_vport->cache);
1016 if (likely(tos == mutable->tos &&
1017 check_cache_valid(cur_cache, mutable))) {
1019 return cur_cache->rt;
1023 rt = __find_route(mutable, tnl_vport->tnl_ops->ipproto, tos);
1027 if (likely(tos == mutable->tos))
1028 *cache = build_cache(vport, mutable, rt);
1034 static bool need_linearize(const struct sk_buff *skb)
1038 if (unlikely(skb_shinfo(skb)->frag_list))
1042 * Generally speaking we should linearize if there are paged frags.
1043 * However, if all of the refcounts are 1 we know nobody else can
1044 * change them from underneath us and we can skip the linearization.
1046 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1047 if (unlikely(page_count(skb_frag_page(&skb_shinfo(skb)->frags[i])) > 1))
1053 static struct sk_buff *handle_offloads(struct sk_buff *skb,
1054 const struct tnl_mutable_config *mutable,
1055 const struct rtable *rt)
1060 min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
1061 + mutable->tunnel_hlen
1062 + (vlan_tx_tag_present(skb) ? VLAN_HLEN : 0);
1064 if (skb_headroom(skb) < min_headroom || skb_header_cloned(skb)) {
1065 int head_delta = SKB_DATA_ALIGN(min_headroom -
1068 err = pskb_expand_head(skb, max_t(int, head_delta, 0),
1074 forward_ip_summed(skb, true);
1076 if (skb_is_gso(skb)) {
1077 struct sk_buff *nskb;
1079 nskb = skb_gso_segment(skb, 0);
1082 err = PTR_ERR(nskb);
1088 } else if (get_ip_summed(skb) == OVS_CSUM_PARTIAL) {
1089 /* Pages aren't locked and could change at any time.
1090 * If this happens after we compute the checksum, the
1091 * checksum will be wrong. We linearize now to avoid
1094 if (unlikely(need_linearize(skb))) {
1095 err = __skb_linearize(skb);
1100 err = skb_checksum_help(skb);
1105 set_ip_summed(skb, OVS_CSUM_NONE);
1112 return ERR_PTR(err);
1115 static int send_frags(struct sk_buff *skb,
1116 const struct tnl_mutable_config *mutable)
1122 struct sk_buff *next = skb->next;
1123 int frag_len = skb->len - mutable->tunnel_hlen;
1127 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
1129 err = ip_local_out(skb);
1131 if (unlikely(net_xmit_eval(err)))
1133 sent_len += frag_len;
1140 * There's no point in continuing to send fragments once one has been
1141 * dropped so just free the rest. This may help improve the congestion
1142 * that caused the first packet to be dropped.
1144 tnl_free_linked_skbs(skb);
1148 int tnl_send(struct vport *vport, struct sk_buff *skb)
1150 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1151 const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
1153 enum vport_err_type err = VPORT_E_TX_ERROR;
1155 struct dst_entry *unattached_dst = NULL;
1156 struct tnl_cache *cache;
1158 __be16 frag_off = 0;
1163 /* Validate the protocol headers before we try to use them. */
1164 if (skb->protocol == htons(ETH_P_8021Q) &&
1165 !vlan_tx_tag_present(skb)) {
1166 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
1169 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
1170 skb_set_network_header(skb, VLAN_ETH_HLEN);
1173 if (skb->protocol == htons(ETH_P_IP)) {
1174 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1175 + sizeof(struct iphdr))))
1178 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1179 else if (skb->protocol == htons(ETH_P_IPV6)) {
1180 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1181 + sizeof(struct ipv6hdr))))
1187 if (skb->protocol == htons(ETH_P_IP))
1188 inner_tos = ip_hdr(skb)->tos;
1189 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1190 else if (skb->protocol == htons(ETH_P_IPV6))
1191 inner_tos = ipv6_get_dsfield(ipv6_hdr(skb));
1196 if (mutable->flags & TNL_F_TOS_INHERIT)
1201 tos = INET_ECN_encapsulate(tos, inner_tos);
1204 rt = find_route(vport, mutable, tos, &cache);
1207 if (unlikely(!cache))
1208 unattached_dst = &rt_dst(rt);
1214 skb_clear_rxhash(skb);
1217 skb = handle_offloads(skb, mutable, rt);
1222 if (unlikely(!check_mtu(skb, vport, mutable, rt, &frag_off))) {
1223 err = VPORT_E_TX_DROPPED;
1228 * If we are over the MTU, allow the IP stack to handle fragmentation.
1229 * Fragmentation is a slow path anyways.
1231 if (unlikely(skb->len + mutable->tunnel_hlen > dst_mtu(&rt_dst(rt)) &&
1233 unattached_dst = &rt_dst(rt);
1234 dst_hold(unattached_dst);
1241 ttl = ip4_dst_hoplimit(&rt_dst(rt));
1243 if (mutable->flags & TNL_F_TTL_INHERIT) {
1244 if (skb->protocol == htons(ETH_P_IP))
1245 ttl = ip_hdr(skb)->ttl;
1246 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1247 else if (skb->protocol == htons(ETH_P_IPV6))
1248 ttl = ipv6_hdr(skb)->hop_limit;
1254 struct sk_buff *next_skb = skb->next;
1257 if (unlikely(vlan_deaccel_tag(skb)))
1260 if (likely(cache)) {
1261 skb_push(skb, cache->len);
1262 memcpy(skb->data, get_cached_header(cache), cache->len);
1263 skb_reset_mac_header(skb);
1264 skb_set_network_header(skb, cache->hh_len);
1267 skb_push(skb, mutable->tunnel_hlen);
1268 create_tunnel_header(vport, mutable, rt, skb->data);
1269 skb_reset_network_header(skb);
1272 skb_dst_set(skb, dst_clone(unattached_dst));
1274 skb_dst_set(skb, unattached_dst);
1275 unattached_dst = NULL;
1278 skb_set_transport_header(skb, skb_network_offset(skb) + sizeof(struct iphdr));
1283 iph->frag_off = frag_off;
1284 ip_select_ident(iph, &rt_dst(rt), NULL);
1286 skb = tnl_vport->tnl_ops->update_header(vport, mutable,
1291 if (likely(cache)) {
1292 int orig_len = skb->len - cache->len;
1293 struct vport *cache_vport;
1295 cache_vport = internal_dev_get_vport(rt_dst(rt).dev);
1296 skb->protocol = htons(ETH_P_IP);
1298 iph->tot_len = htons(skb->len - skb_network_offset(skb));
1302 if (unlikely(compute_ip_summed(skb, true))) {
1307 OVS_CB(skb)->flow = cache->flow;
1308 vport_receive(cache_vport, skb);
1309 sent_len += orig_len;
1313 skb->dev = rt_dst(rt).dev;
1314 xmit_err = dev_queue_xmit(skb);
1316 if (likely(net_xmit_eval(xmit_err) == 0))
1317 sent_len += orig_len;
1320 sent_len += send_frags(skb, mutable);
1326 if (unlikely(sent_len == 0))
1327 vport_record_error(vport, VPORT_E_TX_DROPPED);
1332 tnl_free_linked_skbs(skb);
1334 vport_record_error(vport, err);
1336 dst_release(unattached_dst);
1340 static const struct nla_policy tnl_policy[OVS_TUNNEL_ATTR_MAX + 1] = {
1341 [OVS_TUNNEL_ATTR_FLAGS] = { .type = NLA_U32 },
1342 [OVS_TUNNEL_ATTR_DST_IPV4] = { .type = NLA_U32 },
1343 [OVS_TUNNEL_ATTR_SRC_IPV4] = { .type = NLA_U32 },
1344 [OVS_TUNNEL_ATTR_OUT_KEY] = { .type = NLA_U64 },
1345 [OVS_TUNNEL_ATTR_IN_KEY] = { .type = NLA_U64 },
1346 [OVS_TUNNEL_ATTR_TOS] = { .type = NLA_U8 },
1347 [OVS_TUNNEL_ATTR_TTL] = { .type = NLA_U8 },
1350 /* Sets OVS_TUNNEL_ATTR_* fields in 'mutable', which must initially be
1352 static int tnl_set_config(struct nlattr *options, const struct tnl_ops *tnl_ops,
1353 const struct vport *cur_vport,
1354 struct tnl_mutable_config *mutable)
1356 const struct vport *old_vport;
1357 const struct tnl_mutable_config *old_mutable;
1358 struct nlattr *a[OVS_TUNNEL_ATTR_MAX + 1];
1364 err = nla_parse_nested(a, OVS_TUNNEL_ATTR_MAX, options, tnl_policy);
1368 if (!a[OVS_TUNNEL_ATTR_FLAGS] || !a[OVS_TUNNEL_ATTR_DST_IPV4])
1371 mutable->flags = nla_get_u32(a[OVS_TUNNEL_ATTR_FLAGS]) & TNL_F_PUBLIC;
1373 mutable->key.daddr = nla_get_be32(a[OVS_TUNNEL_ATTR_DST_IPV4]);
1374 if (a[OVS_TUNNEL_ATTR_SRC_IPV4]) {
1375 if (ipv4_is_multicast(mutable->key.daddr))
1377 mutable->key.saddr = nla_get_be32(a[OVS_TUNNEL_ATTR_SRC_IPV4]);
1380 if (a[OVS_TUNNEL_ATTR_TOS]) {
1381 mutable->tos = nla_get_u8(a[OVS_TUNNEL_ATTR_TOS]);
1382 if (mutable->tos != RT_TOS(mutable->tos))
1386 if (a[OVS_TUNNEL_ATTR_TTL])
1387 mutable->ttl = nla_get_u8(a[OVS_TUNNEL_ATTR_TTL]);
1389 mutable->key.tunnel_type = tnl_ops->tunnel_type;
1390 if (!a[OVS_TUNNEL_ATTR_IN_KEY]) {
1391 mutable->key.tunnel_type |= TNL_T_KEY_MATCH;
1392 mutable->flags |= TNL_F_IN_KEY_MATCH;
1394 mutable->key.tunnel_type |= TNL_T_KEY_EXACT;
1395 mutable->key.in_key = nla_get_be64(a[OVS_TUNNEL_ATTR_IN_KEY]);
1398 if (!a[OVS_TUNNEL_ATTR_OUT_KEY])
1399 mutable->flags |= TNL_F_OUT_KEY_ACTION;
1401 mutable->out_key = nla_get_be64(a[OVS_TUNNEL_ATTR_OUT_KEY]);
1403 mutable->tunnel_hlen = tnl_ops->hdr_len(mutable);
1404 if (mutable->tunnel_hlen < 0)
1405 return mutable->tunnel_hlen;
1407 mutable->tunnel_hlen += sizeof(struct iphdr);
1409 old_vport = port_table_lookup(&mutable->key, &old_mutable);
1410 if (old_vport && old_vport != cur_vport)
1414 if (ipv4_is_multicast(mutable->key.daddr)) {
1415 struct net_device *dev;
1418 rt = __find_route(mutable, tnl_ops->ipproto, mutable->tos);
1420 return -EADDRNOTAVAIL;
1421 dev = rt_dst(rt).dev;
1423 if (__in_dev_get_rtnl(dev) == NULL)
1424 return -EADDRNOTAVAIL;
1425 mutable->mlink = dev->ifindex;
1426 ip_mc_inc_group(__in_dev_get_rtnl(dev), mutable->key.daddr);
1432 struct vport *tnl_create(const struct vport_parms *parms,
1433 const struct vport_ops *vport_ops,
1434 const struct tnl_ops *tnl_ops)
1436 struct vport *vport;
1437 struct tnl_vport *tnl_vport;
1438 struct tnl_mutable_config *mutable;
1439 int initial_frag_id;
1442 vport = vport_alloc(sizeof(struct tnl_vport), vport_ops, parms);
1443 if (IS_ERR(vport)) {
1444 err = PTR_ERR(vport);
1448 tnl_vport = tnl_vport_priv(vport);
1450 strcpy(tnl_vport->name, parms->name);
1451 tnl_vport->tnl_ops = tnl_ops;
1453 mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1456 goto error_free_vport;
1459 random_ether_addr(mutable->eth_addr);
1461 get_random_bytes(&initial_frag_id, sizeof(int));
1462 atomic_set(&tnl_vport->frag_id, initial_frag_id);
1464 err = tnl_set_config(parms->options, tnl_ops, NULL, mutable);
1466 goto error_free_mutable;
1468 spin_lock_init(&tnl_vport->cache_lock);
1470 #ifdef NEED_CACHE_TIMEOUT
1471 tnl_vport->cache_exp_interval = MAX_CACHE_EXP -
1472 (net_random() % (MAX_CACHE_EXP / 2));
1475 rcu_assign_pointer(tnl_vport->mutable, mutable);
1477 port_table_add_port(vport);
1481 free_mutable_rtnl(mutable);
1486 return ERR_PTR(err);
1489 int tnl_set_options(struct vport *vport, struct nlattr *options)
1491 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1492 const struct tnl_mutable_config *old_mutable;
1493 struct tnl_mutable_config *mutable;
1496 mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1502 /* Copy fields whose values should be retained. */
1503 old_mutable = rtnl_dereference(tnl_vport->mutable);
1504 mutable->seq = old_mutable->seq + 1;
1505 memcpy(mutable->eth_addr, old_mutable->eth_addr, ETH_ALEN);
1507 /* Parse the others configured by userspace. */
1508 err = tnl_set_config(options, tnl_vport->tnl_ops, vport, mutable);
1512 if (port_hash(&mutable->key) != port_hash(&old_mutable->key))
1513 port_table_move_port(vport, mutable);
1515 assign_config_rcu(vport, mutable);
1520 free_mutable_rtnl(mutable);
1526 int tnl_get_options(const struct vport *vport, struct sk_buff *skb)
1528 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1529 const struct tnl_mutable_config *mutable = rcu_dereference_rtnl(tnl_vport->mutable);
1531 NLA_PUT_U32(skb, OVS_TUNNEL_ATTR_FLAGS, mutable->flags & TNL_F_PUBLIC);
1532 NLA_PUT_BE32(skb, OVS_TUNNEL_ATTR_DST_IPV4, mutable->key.daddr);
1534 if (!(mutable->flags & TNL_F_IN_KEY_MATCH))
1535 NLA_PUT_BE64(skb, OVS_TUNNEL_ATTR_IN_KEY, mutable->key.in_key);
1536 if (!(mutable->flags & TNL_F_OUT_KEY_ACTION))
1537 NLA_PUT_BE64(skb, OVS_TUNNEL_ATTR_OUT_KEY, mutable->out_key);
1538 if (mutable->key.saddr)
1539 NLA_PUT_BE32(skb, OVS_TUNNEL_ATTR_SRC_IPV4, mutable->key.saddr);
1541 NLA_PUT_U8(skb, OVS_TUNNEL_ATTR_TOS, mutable->tos);
1543 NLA_PUT_U8(skb, OVS_TUNNEL_ATTR_TTL, mutable->ttl);
1551 static void free_port_rcu(struct rcu_head *rcu)
1553 struct tnl_vport *tnl_vport = container_of(rcu,
1554 struct tnl_vport, rcu);
1556 free_cache((struct tnl_cache __force *)tnl_vport->cache);
1557 kfree((struct tnl_mutable __force *)tnl_vport->mutable);
1558 vport_free(tnl_vport_to_vport(tnl_vport));
1561 void tnl_destroy(struct vport *vport)
1563 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1564 struct tnl_mutable_config *mutable;
1566 mutable = rtnl_dereference(tnl_vport->mutable);
1567 port_table_remove_port(vport);
1568 free_mutable_rtnl(mutable);
1569 call_rcu(&tnl_vport->rcu, free_port_rcu);
1572 int tnl_set_addr(struct vport *vport, const unsigned char *addr)
1574 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1575 struct tnl_mutable_config *old_mutable, *mutable;
1577 old_mutable = rtnl_dereference(tnl_vport->mutable);
1578 mutable = kmemdup(old_mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL);
1582 old_mutable->mlink = 0;
1584 memcpy(mutable->eth_addr, addr, ETH_ALEN);
1585 assign_config_rcu(vport, mutable);
1590 const char *tnl_get_name(const struct vport *vport)
1592 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1593 return tnl_vport->name;
1596 const unsigned char *tnl_get_addr(const struct vport *vport)
1598 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1599 return rcu_dereference_rtnl(tnl_vport->mutable)->eth_addr;
1602 void tnl_free_linked_skbs(struct sk_buff *skb)
1605 struct sk_buff *next = skb->next;
1615 port_table = kmalloc(PORT_TABLE_SIZE * sizeof(struct hlist_head *),
1620 for (i = 0; i < PORT_TABLE_SIZE; i++)
1621 INIT_HLIST_HEAD(&port_table[i]);
1630 for (i = 0; i < PORT_TABLE_SIZE; i++) {
1631 struct tnl_vport *tnl_vport;
1632 struct hlist_head *hash_head;
1633 struct hlist_node *n;
1635 hash_head = &port_table[i];
1636 hlist_for_each_entry(tnl_vport, n, hash_head, hash_node) {