2 * Copyright (c) 2010 Nicira Networks.
3 * Distributed under the terms of the GNU GPL version 2.
5 * Significant portions of this file may be copied from parts of the Linux
6 * kernel, by Linus Torvalds and others.
9 #include <linux/if_arp.h>
10 #include <linux/if_ether.h>
12 #include <linux/if_vlan.h>
14 #include <linux/in_route.h>
15 #include <linux/jhash.h>
16 #include <linux/kernel.h>
17 #include <linux/version.h>
18 #include <linux/workqueue.h>
20 #include <net/dsfield.h>
23 #include <net/inet_ecn.h>
25 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
28 #include <net/route.h>
37 #include "vport-generic.h"
38 #include "vport-internal_dev.h"
40 #ifdef NEED_CACHE_TIMEOUT
42 * On kernels where we can't quickly detect changes in the rest of the system
43 * we use an expiration time to invalidate the cache. A shorter expiration
44 * reduces the length of time that we may potentially blackhole packets while
45 * a longer time increases performance by reducing the frequency that the
46 * cache needs to be rebuilt. A variety of factors may cause the cache to be
47 * invalidated before the expiration time but this is the maximum. The time
48 * is expressed in jiffies.
50 #define MAX_CACHE_EXP HZ
54 * Interval to check for and remove caches that are no longer valid. Caches
55 * are checked for validity before they are used for packet encapsulation and
56 * old caches are removed at that time. However, if no packets are sent through
57 * the tunnel then the cache will never be destroyed. Since it holds
58 * references to a number of system objects, the cache will continue to use
59 * system resources by not allowing those objects to be destroyed. The cache
60 * cleaner is periodically run to free invalid caches. It does not
61 * significantly affect system performance. A lower interval will release
62 * resources faster but will itself consume resources by requiring more frequent
63 * checks. A longer interval may result in messages being printed to the kernel
64 * message buffer about unreleased resources. The interval is expressed in
67 #define CACHE_CLEANER_INTERVAL (5 * HZ)
69 #define CACHE_DATA_ALIGN 16
71 /* Protected by RCU. */
72 static struct tbl *port_table __read_mostly;
74 static void cache_cleaner(struct work_struct *work);
75 DECLARE_DELAYED_WORK(cache_cleaner_wq, cache_cleaner);
78 * These are just used as an optimization: they don't require any kind of
79 * synchronization because we could have just as easily read the value before
80 * the port change happened.
82 static unsigned int key_local_remote_ports __read_mostly;
83 static unsigned int key_remote_ports __read_mostly;
84 static unsigned int local_remote_ports __read_mostly;
85 static unsigned int remote_ports __read_mostly;
87 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
88 #define rt_dst(rt) (rt->dst)
90 #define rt_dst(rt) (rt->u.dst)
93 static inline struct vport *tnl_vport_to_vport(const struct tnl_vport *tnl_vport)
95 return vport_from_priv(tnl_vport);
98 static inline struct tnl_vport *tnl_vport_table_cast(const struct tbl_node *node)
100 return container_of(node, struct tnl_vport, tbl_node);
103 static inline void schedule_cache_cleaner(void)
105 schedule_delayed_work(&cache_cleaner_wq, CACHE_CLEANER_INTERVAL);
108 static void free_cache(struct tnl_cache *cache)
113 flow_put(cache->flow);
114 ip_rt_put(cache->rt);
118 static void free_config_rcu(struct rcu_head *rcu)
120 struct tnl_mutable_config *c = container_of(rcu, struct tnl_mutable_config, rcu);
124 static void free_cache_rcu(struct rcu_head *rcu)
126 struct tnl_cache *c = container_of(rcu, struct tnl_cache, rcu);
130 static void assign_config_rcu(struct vport *vport,
131 struct tnl_mutable_config *new_config)
133 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
134 struct tnl_mutable_config *old_config;
136 old_config = tnl_vport->mutable;
137 rcu_assign_pointer(tnl_vport->mutable, new_config);
138 call_rcu(&old_config->rcu, free_config_rcu);
141 static void assign_cache_rcu(struct vport *vport, struct tnl_cache *new_cache)
143 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
144 struct tnl_cache *old_cache;
146 old_cache = tnl_vport->cache;
147 rcu_assign_pointer(tnl_vport->cache, new_cache);
150 call_rcu(&old_cache->rcu, free_cache_rcu);
153 static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable)
155 if (mutable->port_config.flags & TNL_F_IN_KEY_MATCH) {
156 if (mutable->port_config.saddr)
157 return &local_remote_ports;
159 return &remote_ports;
161 if (mutable->port_config.saddr)
162 return &key_local_remote_ports;
164 return &key_remote_ports;
168 struct port_lookup_key {
173 const struct tnl_mutable_config *mutable;
177 * Modifies 'target' to store the rcu_dereferenced pointer that was used to do
180 static int port_cmp(const struct tbl_node *node, void *target)
182 const struct tnl_vport *tnl_vport = tnl_vport_table_cast(node);
183 struct port_lookup_key *lookup = target;
185 lookup->mutable = rcu_dereference(tnl_vport->mutable);
187 return (lookup->mutable->tunnel_type == lookup->tunnel_type &&
188 lookup->mutable->port_config.daddr == lookup->daddr &&
189 lookup->mutable->port_config.in_key == lookup->key &&
190 lookup->mutable->port_config.saddr == lookup->saddr);
193 static u32 port_hash(struct port_lookup_key *k)
195 return jhash_3words(k->key, k->saddr, k->daddr, k->tunnel_type);
198 static u32 mutable_hash(const struct tnl_mutable_config *mutable)
200 struct port_lookup_key lookup;
202 lookup.saddr = mutable->port_config.saddr;
203 lookup.daddr = mutable->port_config.daddr;
204 lookup.key = mutable->port_config.in_key;
205 lookup.tunnel_type = mutable->tunnel_type;
207 return port_hash(&lookup);
210 static void check_table_empty(void)
212 if (tbl_count(port_table) == 0) {
213 struct tbl *old_table = port_table;
215 cancel_delayed_work_sync(&cache_cleaner_wq);
216 rcu_assign_pointer(port_table, NULL);
217 tbl_deferred_destroy(old_table, NULL);
221 static int add_port(struct vport *vport)
223 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
227 struct tbl *new_table;
229 new_table = tbl_create(0);
233 rcu_assign_pointer(port_table, new_table);
234 schedule_cache_cleaner();
236 } else if (tbl_count(port_table) > tbl_n_buckets(port_table)) {
237 struct tbl *old_table = port_table;
238 struct tbl *new_table;
240 new_table = tbl_expand(old_table);
241 if (IS_ERR(new_table))
242 return PTR_ERR(new_table);
244 rcu_assign_pointer(port_table, new_table);
245 tbl_deferred_destroy(old_table, NULL);
248 err = tbl_insert(port_table, &tnl_vport->tbl_node, mutable_hash(tnl_vport->mutable));
254 (*find_port_pool(tnl_vport->mutable))++;
259 static int move_port(struct vport *vport, struct tnl_mutable_config *new_mutable)
262 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
265 hash = mutable_hash(new_mutable);
266 if (hash == tnl_vport->tbl_node.hash)
270 * Ideally we should make this move atomic to avoid having gaps in
271 * finding tunnels or the possibility of failure. However, if we do
272 * find a tunnel it will always be consistent.
274 err = tbl_remove(port_table, &tnl_vport->tbl_node);
278 err = tbl_insert(port_table, &tnl_vport->tbl_node, hash);
285 assign_config_rcu(vport, new_mutable);
290 static int del_port(struct vport *vport)
292 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
295 err = tbl_remove(port_table, &tnl_vport->tbl_node);
300 (*find_port_pool(tnl_vport->mutable))--;
305 struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be32 key,
307 const struct tnl_mutable_config **mutable)
309 struct port_lookup_key lookup;
310 struct tbl *table = rcu_dereference(port_table);
311 struct tbl_node *tbl_node;
313 if (unlikely(!table))
316 lookup.saddr = saddr;
317 lookup.daddr = daddr;
319 if (tunnel_type & TNL_T_KEY_EXACT) {
321 lookup.tunnel_type = tunnel_type & ~TNL_T_KEY_MATCH;
323 if (key_local_remote_ports) {
324 tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
329 if (key_remote_ports) {
332 tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
336 lookup.saddr = saddr;
340 if (tunnel_type & TNL_T_KEY_MATCH) {
342 lookup.tunnel_type = tunnel_type & ~TNL_T_KEY_EXACT;
344 if (local_remote_ports) {
345 tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
353 tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
362 *mutable = lookup.mutable;
363 return tnl_vport_to_vport(tnl_vport_table_cast(tbl_node));
366 static inline void ecn_decapsulate(struct sk_buff *skb)
368 u8 tos = ip_hdr(skb)->tos;
370 if (INET_ECN_is_ce(tos)) {
371 __be16 protocol = skb->protocol;
372 unsigned int nw_header = skb_network_offset(skb);
374 if (skb->protocol == htons(ETH_P_8021Q)) {
375 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
378 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
379 nw_header += VLAN_HLEN;
382 if (protocol == htons(ETH_P_IP)) {
383 if (unlikely(!pskb_may_pull(skb, nw_header
384 + sizeof(struct iphdr))))
387 IP_ECN_set_ce((struct iphdr *)(skb->data + nw_header));
389 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
390 else if (protocol == htons(ETH_P_IPV6)) {
391 if (unlikely(!pskb_may_pull(skb, nw_header
392 + sizeof(struct ipv6hdr))))
395 IP6_ECN_set_ce((struct ipv6hdr *)(skb->data + nw_header));
401 /* Called with rcu_read_lock. */
402 void tnl_rcv(struct vport *vport, struct sk_buff *skb)
404 skb->pkt_type = PACKET_HOST;
405 skb->protocol = eth_type_trans(skb, skb->dev);
410 skb_reset_network_header(skb);
412 ecn_decapsulate(skb);
414 skb_push(skb, ETH_HLEN);
415 compute_ip_summed(skb, false);
417 vport_receive(vport, skb);
420 static bool check_ipv4_address(__be32 addr)
422 if (ipv4_is_multicast(addr) || ipv4_is_lbcast(addr)
423 || ipv4_is_loopback(addr) || ipv4_is_zeronet(addr))
429 static bool ipv4_should_icmp(struct sk_buff *skb)
431 struct iphdr *old_iph = ip_hdr(skb);
433 /* Don't respond to L2 broadcast. */
434 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
437 /* Don't respond to L3 broadcast or invalid addresses. */
438 if (!check_ipv4_address(old_iph->daddr) ||
439 !check_ipv4_address(old_iph->saddr))
442 /* Only respond to the first fragment. */
443 if (old_iph->frag_off & htons(IP_OFFSET))
446 /* Don't respond to ICMP error messages. */
447 if (old_iph->protocol == IPPROTO_ICMP) {
448 u8 icmp_type, *icmp_typep;
450 icmp_typep = skb_header_pointer(skb, (u8 *)old_iph +
451 (old_iph->ihl << 2) +
452 offsetof(struct icmphdr, type) -
453 skb->data, sizeof(icmp_type),
459 if (*icmp_typep > NR_ICMP_TYPES
460 || (*icmp_typep <= ICMP_PARAMETERPROB
461 && *icmp_typep != ICMP_ECHOREPLY
462 && *icmp_typep != ICMP_ECHO))
469 static void ipv4_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
470 unsigned int mtu, unsigned int payload_length)
472 struct iphdr *iph, *old_iph = ip_hdr(skb);
473 struct icmphdr *icmph;
476 iph = (struct iphdr *)skb_put(nskb, sizeof(struct iphdr));
477 icmph = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr));
478 payload = skb_put(nskb, payload_length);
482 iph->ihl = sizeof(struct iphdr) >> 2;
483 iph->tos = (old_iph->tos & IPTOS_TOS_MASK) |
484 IPTOS_PREC_INTERNETCONTROL;
485 iph->tot_len = htons(sizeof(struct iphdr)
486 + sizeof(struct icmphdr)
488 get_random_bytes(&iph->id, sizeof(iph->id));
491 iph->protocol = IPPROTO_ICMP;
492 iph->daddr = old_iph->saddr;
493 iph->saddr = old_iph->daddr;
498 icmph->type = ICMP_DEST_UNREACH;
499 icmph->code = ICMP_FRAG_NEEDED;
500 icmph->un.gateway = htonl(mtu);
503 nskb->csum = csum_partial((u8 *)icmph, sizeof(struct icmphdr), 0);
504 nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_iph - skb->data,
505 payload, payload_length,
507 icmph->checksum = csum_fold(nskb->csum);
510 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
511 static bool ipv6_should_icmp(struct sk_buff *skb)
513 struct ipv6hdr *old_ipv6h = ipv6_hdr(skb);
515 int payload_off = (u8 *)(old_ipv6h + 1) - skb->data;
516 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
518 /* Check source address is valid. */
519 addr_type = ipv6_addr_type(&old_ipv6h->saddr);
520 if (addr_type & IPV6_ADDR_MULTICAST || addr_type == IPV6_ADDR_ANY)
523 /* Don't reply to unspecified addresses. */
524 if (ipv6_addr_type(&old_ipv6h->daddr) == IPV6_ADDR_ANY)
527 /* Don't respond to ICMP error messages. */
528 payload_off = ipv6_skip_exthdr(skb, payload_off, &nexthdr);
532 if (nexthdr == NEXTHDR_ICMP) {
533 u8 icmp_type, *icmp_typep;
535 icmp_typep = skb_header_pointer(skb, payload_off +
536 offsetof(struct icmp6hdr,
538 sizeof(icmp_type), &icmp_type);
540 if (!icmp_typep || !(*icmp_typep & ICMPV6_INFOMSG_MASK))
547 static void ipv6_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
548 unsigned int mtu, unsigned int payload_length)
550 struct ipv6hdr *ipv6h, *old_ipv6h = ipv6_hdr(skb);
551 struct icmp6hdr *icmp6h;
554 ipv6h = (struct ipv6hdr *)skb_put(nskb, sizeof(struct ipv6hdr));
555 icmp6h = (struct icmp6hdr *)skb_put(nskb, sizeof(struct icmp6hdr));
556 payload = skb_put(nskb, payload_length);
561 memset(&ipv6h->flow_lbl, 0, sizeof(ipv6h->flow_lbl));
562 ipv6h->payload_len = htons(sizeof(struct icmp6hdr)
564 ipv6h->nexthdr = NEXTHDR_ICMP;
565 ipv6h->hop_limit = IPV6_DEFAULT_HOPLIMIT;
566 ipv6_addr_copy(&ipv6h->daddr, &old_ipv6h->saddr);
567 ipv6_addr_copy(&ipv6h->saddr, &old_ipv6h->daddr);
570 icmp6h->icmp6_type = ICMPV6_PKT_TOOBIG;
571 icmp6h->icmp6_code = 0;
572 icmp6h->icmp6_cksum = 0;
573 icmp6h->icmp6_mtu = htonl(mtu);
575 nskb->csum = csum_partial((u8 *)icmp6h, sizeof(struct icmp6hdr), 0);
576 nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_ipv6h - skb->data,
577 payload, payload_length,
579 icmp6h->icmp6_cksum = csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
580 sizeof(struct icmp6hdr)
582 ipv6h->nexthdr, nskb->csum);
586 bool tnl_frag_needed(struct vport *vport, const struct tnl_mutable_config *mutable,
587 struct sk_buff *skb, unsigned int mtu, __be32 flow_key)
589 unsigned int eth_hdr_len = ETH_HLEN;
590 unsigned int total_length = 0, header_length = 0, payload_length;
591 struct ethhdr *eh, *old_eh = eth_hdr(skb);
592 struct sk_buff *nskb;
595 if (skb->protocol == htons(ETH_P_IP)) {
596 if (mtu < IP_MIN_MTU)
599 if (!ipv4_should_icmp(skb))
602 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
603 else if (skb->protocol == htons(ETH_P_IPV6)) {
604 if (mtu < IPV6_MIN_MTU)
608 * In theory we should do PMTUD on IPv6 multicast messages but
609 * we don't have an address to send from so just fragment.
611 if (ipv6_addr_type(&ipv6_hdr(skb)->daddr) & IPV6_ADDR_MULTICAST)
614 if (!ipv6_should_icmp(skb))
622 if (old_eh->h_proto == htons(ETH_P_8021Q))
623 eth_hdr_len = VLAN_ETH_HLEN;
625 payload_length = skb->len - eth_hdr_len;
626 if (skb->protocol == htons(ETH_P_IP)) {
627 header_length = sizeof(struct iphdr) + sizeof(struct icmphdr);
628 total_length = min_t(unsigned int, header_length +
629 payload_length, 576);
631 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
633 header_length = sizeof(struct ipv6hdr) +
634 sizeof(struct icmp6hdr);
635 total_length = min_t(unsigned int, header_length +
636 payload_length, IPV6_MIN_MTU);
640 total_length = min(total_length, mutable->mtu);
641 payload_length = total_length - header_length;
643 nskb = dev_alloc_skb(NET_IP_ALIGN + eth_hdr_len + header_length +
648 skb_reserve(nskb, NET_IP_ALIGN);
650 /* Ethernet / VLAN */
651 eh = (struct ethhdr *)skb_put(nskb, eth_hdr_len);
652 memcpy(eh->h_dest, old_eh->h_source, ETH_ALEN);
653 memcpy(eh->h_source, mutable->eth_addr, ETH_ALEN);
654 nskb->protocol = eh->h_proto = old_eh->h_proto;
655 if (old_eh->h_proto == htons(ETH_P_8021Q)) {
656 struct vlan_ethhdr *vh = (struct vlan_ethhdr *)eh;
658 vh->h_vlan_TCI = vlan_eth_hdr(skb)->h_vlan_TCI;
659 vh->h_vlan_encapsulated_proto = skb->protocol;
661 skb_reset_mac_header(nskb);
664 if (skb->protocol == htons(ETH_P_IP))
665 ipv4_build_icmp(skb, nskb, mtu, payload_length);
666 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
668 ipv6_build_icmp(skb, nskb, mtu, payload_length);
672 * Assume that flow based keys are symmetric with respect to input
673 * and output and use the key that we were going to put on the
674 * outgoing packet for the fake received packet. If the keys are
675 * not symmetric then PMTUD needs to be disabled since we won't have
676 * any way of synthesizing packets.
678 if ((mutable->port_config.flags & (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION)) ==
679 (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION))
680 OVS_CB(nskb)->tun_id = flow_key;
682 compute_ip_summed(nskb, false);
683 vport_receive(vport, nskb);
688 static bool check_mtu(struct sk_buff *skb,
690 const struct tnl_mutable_config *mutable,
691 const struct rtable *rt, __be16 *frag_offp)
696 frag_off = (mutable->port_config.flags & TNL_F_PMTUD) ? htons(IP_DF) : 0;
698 mtu = dst_mtu(&rt_dst(rt))
700 - mutable->tunnel_hlen
701 - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0);
705 if (skb->protocol == htons(ETH_P_IP)) {
706 struct iphdr *old_iph = ip_hdr(skb);
708 frag_off |= old_iph->frag_off & htons(IP_DF);
709 mtu = max(mtu, IP_MIN_MTU);
711 if ((old_iph->frag_off & htons(IP_DF)) &&
712 mtu < ntohs(old_iph->tot_len)) {
713 if (tnl_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id))
717 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
718 else if (skb->protocol == htons(ETH_P_IPV6)) {
719 unsigned int packet_length = skb->len - ETH_HLEN
720 - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0);
722 mtu = max(mtu, IPV6_MIN_MTU);
724 /* IPv6 requires PMTUD if the packet is above the minimum MTU. */
725 if (packet_length > IPV6_MIN_MTU)
726 frag_off = htons(IP_DF);
728 if (mtu < packet_length) {
729 if (tnl_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id))
735 *frag_offp = frag_off;
743 static void create_tunnel_header(const struct vport *vport,
744 const struct tnl_mutable_config *mutable,
745 const struct rtable *rt, void *header)
747 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
748 struct iphdr *iph = header;
751 iph->ihl = sizeof(struct iphdr) >> 2;
752 iph->frag_off = htons(IP_DF);
753 iph->protocol = tnl_vport->tnl_ops->ipproto;
754 iph->tos = mutable->port_config.tos;
755 iph->daddr = rt->rt_dst;
756 iph->saddr = rt->rt_src;
757 iph->ttl = mutable->port_config.ttl;
759 iph->ttl = dst_metric(&rt_dst(rt), RTAX_HOPLIMIT);
761 tnl_vport->tnl_ops->build_header(vport, mutable, iph + 1);
764 static inline void *get_cached_header(const struct tnl_cache *cache)
766 return (void *)cache + ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN);
769 static inline bool check_cache_valid(const struct tnl_cache *cache,
770 const struct tnl_mutable_config *mutable)
773 #ifdef NEED_CACHE_TIMEOUT
774 time_before(jiffies, cache->expiration) &&
777 atomic_read(&init_net.ipv4.rt_genid) == cache->rt->rt_genid &&
780 rt_dst(cache->rt).hh->hh_lock.sequence == cache->hh_seq &&
782 mutable->seq == cache->mutable_seq &&
783 (!is_internal_dev(rt_dst(cache->rt).dev) ||
784 (cache->flow && !cache->flow->dead));
787 static int cache_cleaner_cb(struct tbl_node *tbl_node, void *aux)
789 struct tnl_vport *tnl_vport = tnl_vport_table_cast(tbl_node);
790 const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
791 const struct tnl_cache *cache = rcu_dereference(tnl_vport->cache);
793 if (cache && !check_cache_valid(cache, mutable) &&
794 spin_trylock_bh(&tnl_vport->cache_lock)) {
795 assign_cache_rcu(tnl_vport_to_vport(tnl_vport), NULL);
796 spin_unlock_bh(&tnl_vport->cache_lock);
802 static void cache_cleaner(struct work_struct *work)
804 schedule_cache_cleaner();
807 tbl_foreach(port_table, cache_cleaner_cb, NULL);
811 static inline void create_eth_hdr(struct tnl_cache *cache,
812 const struct rtable *rt)
814 void *cache_data = get_cached_header(cache);
815 int hh_len = rt_dst(rt).hh->hh_len;
816 int hh_off = HH_DATA_ALIGN(rt_dst(rt).hh->hh_len) - hh_len;
822 hh_seq = read_seqbegin(&rt_dst(rt).hh->hh_lock);
823 memcpy(cache_data, (void *)rt_dst(rt).hh->hh_data + hh_off, hh_len);
824 } while (read_seqretry(&rt_dst(rt).hh->hh_lock, hh_seq));
826 cache->hh_seq = hh_seq;
828 read_lock_bh(&rt_dst(rt).hh->hh_lock);
829 memcpy(cache_data, (void *)rt_dst(rt).hh->hh_data + hh_off, hh_len);
830 read_unlock_bh(&rt_dst(rt).hh->hh_lock);
834 static struct tnl_cache *build_cache(struct vport *vport,
835 const struct tnl_mutable_config *mutable,
838 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
839 struct tnl_cache *cache;
843 if (!(mutable->port_config.flags & TNL_F_HDR_CACHE))
847 * If there is no entry in the ARP cache or if this device does not
848 * support hard header caching just fall back to the IP stack.
854 * If lock is contended fall back to directly building the header.
855 * We're not going to help performance by sitting here spinning.
857 if (!spin_trylock_bh(&tnl_vport->cache_lock))
860 cache = tnl_vport->cache;
861 if (check_cache_valid(cache, mutable))
866 cache_len = rt_dst(rt).hh->hh_len + mutable->tunnel_hlen;
868 cache = kzalloc(ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN) +
869 cache_len, GFP_ATOMIC);
873 cache->len = cache_len;
875 create_eth_hdr(cache, rt);
876 cache_data = get_cached_header(cache) + rt_dst(rt).hh->hh_len;
878 create_tunnel_header(vport, mutable, rt, cache_data);
880 cache->mutable_seq = mutable->seq;
882 #ifdef NEED_CACHE_TIMEOUT
883 cache->expiration = jiffies + tnl_vport->cache_exp_interval;
886 if (is_internal_dev(rt_dst(rt).dev)) {
887 struct odp_flow_key flow_key;
888 struct tbl_node *flow_node;
894 vport = internal_dev_get_vport(rt_dst(rt).dev);
898 skb = alloc_skb(cache->len, GFP_ATOMIC);
902 __skb_put(skb, cache->len);
903 memcpy(skb->data, get_cached_header(cache), cache->len);
905 err = flow_extract(skb, vport->port_no, &flow_key, &is_frag);
911 flow_node = tbl_lookup(rcu_dereference(vport->dp->table),
912 &flow_key, flow_hash(&flow_key),
915 struct sw_flow *flow = flow_cast(flow_node);
923 assign_cache_rcu(vport, cache);
926 spin_unlock_bh(&tnl_vport->cache_lock);
931 static struct rtable *find_route(struct vport *vport,
932 const struct tnl_mutable_config *mutable,
933 u8 tos, struct tnl_cache **cache)
935 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
936 struct tnl_cache *cur_cache = rcu_dereference(tnl_vport->cache);
941 if (likely(tos == mutable->port_config.tos &&
942 check_cache_valid(cur_cache, mutable))) {
944 return cur_cache->rt;
947 struct flowi fl = { .nl_u = { .ip4_u =
948 { .daddr = mutable->port_config.daddr,
949 .saddr = mutable->port_config.saddr,
951 .proto = tnl_vport->tnl_ops->ipproto };
953 if (unlikely(ip_route_output_key(&init_net, &rt, &fl)))
956 if (likely(tos == mutable->port_config.tos))
957 *cache = build_cache(vport, mutable, rt);
963 static struct sk_buff *check_headroom(struct sk_buff *skb, int headroom)
965 if (skb_headroom(skb) < headroom || skb_header_cloned(skb)) {
966 struct sk_buff *nskb = skb_realloc_headroom(skb, headroom + 16);
967 if (unlikely(!nskb)) {
969 return ERR_PTR(-ENOMEM);
972 set_skb_csum_bits(skb, nskb);
975 skb_set_owner_w(nskb, skb->sk);
984 static inline bool need_linearize(const struct sk_buff *skb)
988 if (unlikely(skb_shinfo(skb)->frag_list))
992 * Generally speaking we should linearize if there are paged frags.
993 * However, if all of the refcounts are 1 we know nobody else can
994 * change them from underneath us and we can skip the linearization.
996 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
997 if (unlikely(page_count(skb_shinfo(skb)->frags[0].page) > 1))
1003 static struct sk_buff *handle_offloads(struct sk_buff *skb,
1004 const struct tnl_mutable_config *mutable,
1005 const struct rtable *rt)
1010 forward_ip_summed(skb);
1012 err = vswitch_skb_checksum_setup(skb);
1016 min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
1017 + mutable->tunnel_hlen;
1019 if (skb_is_gso(skb)) {
1020 struct sk_buff *nskb;
1023 * If we are doing GSO on a pskb it is better to make sure that
1024 * the headroom is correct now. We will only have to copy the
1025 * portion in the linear data area and GSO will preserve
1026 * headroom when it creates the segments. This is particularly
1027 * beneficial on Xen where we get a lot of GSO pskbs.
1028 * Conversely, we avoid copying if it is just to get our own
1029 * writable clone because GSO will do the copy for us.
1031 if (skb_headroom(skb) < min_headroom) {
1032 skb = check_headroom(skb, min_headroom);
1033 if (unlikely(IS_ERR(skb))) {
1039 nskb = skb_gso_segment(skb, 0);
1041 if (unlikely(IS_ERR(nskb))) {
1042 err = PTR_ERR(nskb);
1048 skb = check_headroom(skb, min_headroom);
1049 if (unlikely(IS_ERR(skb))) {
1054 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1056 * Pages aren't locked and could change at any time.
1057 * If this happens after we compute the checksum, the
1058 * checksum will be wrong. We linearize now to avoid
1061 if (unlikely(need_linearize(skb))) {
1062 err = __skb_linearize(skb);
1067 err = skb_checksum_help(skb);
1070 } else if (skb->ip_summed == CHECKSUM_COMPLETE)
1071 skb->ip_summed = CHECKSUM_NONE;
1079 return ERR_PTR(err);
1082 static int send_frags(struct sk_buff *skb,
1083 const struct tnl_mutable_config *mutable)
1090 struct sk_buff *next = skb->next;
1091 int frag_len = skb->len - mutable->tunnel_hlen;
1094 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
1096 err = ip_local_out(skb);
1097 if (likely(net_xmit_eval(err) == 0))
1098 sent_len += frag_len;
1111 * There's no point in continuing to send fragments once one has been
1112 * dropped so just free the rest. This may help improve the congestion
1113 * that caused the first packet to be dropped.
1115 tnl_free_linked_skbs(skb);
1119 int tnl_send(struct vport *vport, struct sk_buff *skb)
1121 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1122 const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
1124 enum vport_err_type err = VPORT_E_TX_ERROR;
1126 struct dst_entry *unattached_dst = NULL;
1127 struct tnl_cache *cache;
1134 /* Validate the protocol headers before we try to use them. */
1135 if (skb->protocol == htons(ETH_P_8021Q)) {
1136 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
1139 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
1140 skb_set_network_header(skb, VLAN_ETH_HLEN);
1143 if (skb->protocol == htons(ETH_P_IP)) {
1144 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1145 + sizeof(struct iphdr))))
1148 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1149 else if (skb->protocol == htons(ETH_P_IPV6)) {
1150 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1151 + sizeof(struct ipv6hdr))))
1157 if (skb->protocol == htons(ETH_P_IP))
1158 inner_tos = ip_hdr(skb)->tos;
1159 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1160 else if (skb->protocol == htons(ETH_P_IPV6))
1161 inner_tos = ipv6_get_dsfield(ipv6_hdr(skb));
1166 if (mutable->port_config.flags & TNL_F_TOS_INHERIT)
1169 tos = mutable->port_config.tos;
1171 tos = INET_ECN_encapsulate(tos, inner_tos);
1174 rt = find_route(vport, mutable, tos, &cache);
1177 if (unlikely(!cache))
1178 unattached_dst = &rt_dst(rt);
1186 skb = handle_offloads(skb, mutable, rt);
1187 if (unlikely(IS_ERR(skb)))
1191 if (unlikely(!check_mtu(skb, vport, mutable, rt, &frag_off))) {
1192 err = VPORT_E_TX_DROPPED;
1197 * If we are over the MTU, allow the IP stack to handle fragmentation.
1198 * Fragmentation is a slow path anyways.
1200 if (unlikely(skb->len + mutable->tunnel_hlen > dst_mtu(&rt_dst(rt)) &&
1202 unattached_dst = &rt_dst(rt);
1203 dst_hold(unattached_dst);
1208 ttl = mutable->port_config.ttl;
1210 ttl = dst_metric(&rt_dst(rt), RTAX_HOPLIMIT);
1212 if (mutable->port_config.flags & TNL_F_TTL_INHERIT) {
1213 if (skb->protocol == htons(ETH_P_IP))
1214 ttl = ip_hdr(skb)->ttl;
1215 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1216 else if (skb->protocol == htons(ETH_P_IPV6))
1217 ttl = ipv6_hdr(skb)->hop_limit;
1223 struct sk_buff *next_skb = skb->next;
1226 if (likely(cache)) {
1227 skb_push(skb, cache->len);
1228 memcpy(skb->data, get_cached_header(cache), cache->len);
1229 skb_reset_mac_header(skb);
1230 skb_set_network_header(skb, rt_dst(rt).hh->hh_len);
1233 skb_push(skb, mutable->tunnel_hlen);
1234 create_tunnel_header(vport, mutable, rt, skb->data);
1235 skb_reset_network_header(skb);
1238 skb_dst_set(skb, dst_clone(unattached_dst));
1240 skb_dst_set(skb, unattached_dst);
1241 unattached_dst = NULL;
1244 skb_set_transport_header(skb, skb_network_offset(skb) + sizeof(struct iphdr));
1249 iph->frag_off = frag_off;
1250 ip_select_ident(iph, &rt_dst(rt), NULL);
1252 skb = tnl_vport->tnl_ops->update_header(vport, mutable, &rt_dst(rt), skb);
1256 if (likely(cache)) {
1257 int orig_len = skb->len - cache->len;
1258 struct vport *cache_vport = internal_dev_get_vport(rt_dst(rt).dev);
1260 skb->protocol = htons(ETH_P_IP);
1261 iph->tot_len = htons(skb->len - skb_network_offset(skb));
1265 OVS_CB(skb)->flow = cache->flow;
1266 compute_ip_summed(skb, true);
1267 vport_receive(cache_vport, skb);
1268 sent_len += orig_len;
1272 skb->dev = rt_dst(rt).dev;
1273 err = dev_queue_xmit(skb);
1275 if (likely(net_xmit_eval(err) == 0))
1276 sent_len += orig_len;
1279 sent_len += send_frags(skb, mutable);
1285 if (unlikely(sent_len == 0))
1286 vport_record_error(vport, VPORT_E_TX_DROPPED);
1291 tnl_free_linked_skbs(skb);
1293 dst_release(unattached_dst);
1294 vport_record_error(vport, err);
1299 static int set_config(const void *config, const struct tnl_ops *tnl_ops,
1300 const struct vport *cur_vport,
1301 struct tnl_mutable_config *mutable)
1303 const struct vport *old_vport;
1304 const struct tnl_mutable_config *old_mutable;
1306 mutable->port_config = *(struct tnl_port_config *)config;
1308 if (mutable->port_config.daddr == 0)
1311 if (mutable->port_config.tos != RT_TOS(mutable->port_config.tos))
1314 mutable->tunnel_hlen = tnl_ops->hdr_len(&mutable->port_config);
1315 if (mutable->tunnel_hlen < 0)
1316 return mutable->tunnel_hlen;
1318 mutable->tunnel_hlen += sizeof(struct iphdr);
1320 mutable->tunnel_type = tnl_ops->tunnel_type;
1321 if (mutable->port_config.flags & TNL_F_IN_KEY_MATCH) {
1322 mutable->tunnel_type |= TNL_T_KEY_MATCH;
1323 mutable->port_config.in_key = 0;
1325 mutable->tunnel_type |= TNL_T_KEY_EXACT;
1327 old_vport = tnl_find_port(mutable->port_config.saddr,
1328 mutable->port_config.daddr,
1329 mutable->port_config.in_key,
1330 mutable->tunnel_type,
1333 if (old_vport && old_vport != cur_vport)
1336 if (mutable->port_config.flags & TNL_F_OUT_KEY_ACTION)
1337 mutable->port_config.out_key = 0;
1342 struct vport *tnl_create(const struct vport_parms *parms,
1343 const struct vport_ops *vport_ops,
1344 const struct tnl_ops *tnl_ops)
1346 struct vport *vport;
1347 struct tnl_vport *tnl_vport;
1348 int initial_frag_id;
1351 vport = vport_alloc(sizeof(struct tnl_vport), vport_ops, parms);
1352 if (IS_ERR(vport)) {
1353 err = PTR_ERR(vport);
1357 tnl_vport = tnl_vport_priv(vport);
1359 strcpy(tnl_vport->name, parms->name);
1360 tnl_vport->tnl_ops = tnl_ops;
1362 tnl_vport->mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1363 if (!tnl_vport->mutable) {
1365 goto error_free_vport;
1368 vport_gen_rand_ether_addr(tnl_vport->mutable->eth_addr);
1369 tnl_vport->mutable->mtu = ETH_DATA_LEN;
1371 get_random_bytes(&initial_frag_id, sizeof(int));
1372 atomic_set(&tnl_vport->frag_id, initial_frag_id);
1374 err = set_config(parms->config, tnl_ops, NULL, tnl_vport->mutable);
1376 goto error_free_mutable;
1378 spin_lock_init(&tnl_vport->cache_lock);
1380 #ifdef NEED_CACHE_TIMEOUT
1381 tnl_vport->cache_exp_interval = MAX_CACHE_EXP -
1382 (net_random() % (MAX_CACHE_EXP / 2));
1385 err = add_port(vport);
1387 goto error_free_mutable;
1392 kfree(tnl_vport->mutable);
1396 return ERR_PTR(err);
1399 int tnl_modify(struct vport *vport, struct odp_port *port)
1401 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1402 struct tnl_mutable_config *mutable;
1405 mutable = kmemdup(tnl_vport->mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL);
1411 err = set_config(port->config, tnl_vport->tnl_ops, vport, mutable);
1417 err = move_port(vport, mutable);
1429 static void free_port_rcu(struct rcu_head *rcu)
1431 struct tnl_vport *tnl_vport = container_of(rcu, struct tnl_vport, rcu);
1433 spin_lock_bh(&tnl_vport->cache_lock);
1434 free_cache(tnl_vport->cache);
1435 spin_unlock_bh(&tnl_vport->cache_lock);
1437 kfree(tnl_vport->mutable);
1438 vport_free(tnl_vport_to_vport(tnl_vport));
1441 int tnl_destroy(struct vport *vport)
1443 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1444 const struct tnl_mutable_config *old_mutable;
1446 if (vport == tnl_find_port(tnl_vport->mutable->port_config.saddr,
1447 tnl_vport->mutable->port_config.daddr,
1448 tnl_vport->mutable->port_config.in_key,
1449 tnl_vport->mutable->tunnel_type,
1453 call_rcu(&tnl_vport->rcu, free_port_rcu);
1458 int tnl_set_mtu(struct vport *vport, int mtu)
1460 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1461 struct tnl_mutable_config *mutable;
1463 mutable = kmemdup(tnl_vport->mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL);
1468 assign_config_rcu(vport, mutable);
1473 int tnl_set_addr(struct vport *vport, const unsigned char *addr)
1475 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1476 struct tnl_mutable_config *mutable;
1478 mutable = kmemdup(tnl_vport->mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL);
1482 memcpy(mutable->eth_addr, addr, ETH_ALEN);
1483 assign_config_rcu(vport, mutable);
1488 const char *tnl_get_name(const struct vport *vport)
1490 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1491 return tnl_vport->name;
1494 const unsigned char *tnl_get_addr(const struct vport *vport)
1496 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1497 return rcu_dereference(tnl_vport->mutable)->eth_addr;
1500 int tnl_get_mtu(const struct vport *vport)
1502 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1503 return rcu_dereference(tnl_vport->mutable)->mtu;
1506 void tnl_free_linked_skbs(struct sk_buff *skb)
1512 struct sk_buff *next = skb->next;