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>
36 #include "vport-generic.h"
37 #include "vport-internal_dev.h"
39 #ifdef NEED_CACHE_TIMEOUT
41 * On kernels where we can't quickly detect changes in the rest of the system
42 * we use an expiration time to invalidate the cache. A shorter expiration
43 * reduces the length of time that we may potentially blackhole packets while
44 * a longer time increases performance by reducing the frequency that the
45 * cache needs to be rebuilt. A variety of factors may cause the cache to be
46 * invalidated before the expiration time but this is the maximum. The time
47 * is expressed in jiffies.
49 #define MAX_CACHE_EXP HZ
53 * Interval to check for and remove caches that are no longer valid. Caches
54 * are checked for validity before they are used for packet encapsulation and
55 * old caches are removed at that time. However, if no packets are sent through
56 * the tunnel then the cache will never be destroyed. Since it holds
57 * references to a number of system objects, the cache will continue to use
58 * system resources by not allowing those objects to be destroyed. The cache
59 * cleaner is periodically run to free invalid caches. It does not
60 * significantly affect system performance. A lower interval will release
61 * resources faster but will itself consume resources by requiring more frequent
62 * checks. A longer interval may result in messages being printed to the kernel
63 * message buffer about unreleased resources. The interval is expressed in
66 #define CACHE_CLEANER_INTERVAL (5 * HZ)
68 #define CACHE_DATA_ALIGN 16
70 /* Protected by RCU. */
71 static struct tbl *port_table __read_mostly;
73 static void cache_cleaner(struct work_struct *work);
74 DECLARE_DELAYED_WORK(cache_cleaner_wq, cache_cleaner);
77 * These are just used as an optimization: they don't require any kind of
78 * synchronization because we could have just as easily read the value before
79 * the port change happened.
81 static unsigned int key_local_remote_ports __read_mostly;
82 static unsigned int key_remote_ports __read_mostly;
83 static unsigned int local_remote_ports __read_mostly;
84 static unsigned int remote_ports __read_mostly;
86 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,36)
87 #define rt_dst(rt) (rt->dst)
89 #define rt_dst(rt) (rt->u.dst)
92 static inline struct vport *tnl_vport_to_vport(const struct tnl_vport *tnl_vport)
94 return vport_from_priv(tnl_vport);
97 static inline struct tnl_vport *tnl_vport_table_cast(const struct tbl_node *node)
99 return container_of(node, struct tnl_vport, tbl_node);
102 static inline void schedule_cache_cleaner(void)
104 schedule_delayed_work(&cache_cleaner_wq, CACHE_CLEANER_INTERVAL);
107 static void free_cache(struct tnl_cache *cache)
112 flow_put(cache->flow);
113 ip_rt_put(cache->rt);
117 static void free_config_rcu(struct rcu_head *rcu)
119 struct tnl_mutable_config *c = container_of(rcu, struct tnl_mutable_config, rcu);
123 static void free_cache_rcu(struct rcu_head *rcu)
125 struct tnl_cache *c = container_of(rcu, struct tnl_cache, rcu);
129 static void assign_config_rcu(struct vport *vport,
130 struct tnl_mutable_config *new_config)
132 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
133 struct tnl_mutable_config *old_config;
135 old_config = tnl_vport->mutable;
136 rcu_assign_pointer(tnl_vport->mutable, new_config);
137 call_rcu(&old_config->rcu, free_config_rcu);
140 static void assign_cache_rcu(struct vport *vport, struct tnl_cache *new_cache)
142 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
143 struct tnl_cache *old_cache;
145 old_cache = tnl_vport->cache;
146 rcu_assign_pointer(tnl_vport->cache, new_cache);
149 call_rcu(&old_cache->rcu, free_cache_rcu);
152 static unsigned int *find_port_pool(const struct tnl_mutable_config *mutable)
154 if (mutable->port_config.flags & TNL_F_IN_KEY_MATCH) {
155 if (mutable->port_config.saddr)
156 return &local_remote_ports;
158 return &remote_ports;
160 if (mutable->port_config.saddr)
161 return &key_local_remote_ports;
163 return &key_remote_ports;
167 struct port_lookup_key {
172 const struct tnl_mutable_config *mutable;
176 * Modifies 'target' to store the rcu_dereferenced pointer that was used to do
179 static int port_cmp(const struct tbl_node *node, void *target)
181 const struct tnl_vport *tnl_vport = tnl_vport_table_cast(node);
182 struct port_lookup_key *lookup = target;
184 lookup->mutable = rcu_dereference(tnl_vport->mutable);
186 return (lookup->mutable->tunnel_type == lookup->tunnel_type &&
187 lookup->mutable->port_config.daddr == lookup->daddr &&
188 lookup->mutable->port_config.in_key == lookup->key &&
189 lookup->mutable->port_config.saddr == lookup->saddr);
192 static u32 port_hash(struct port_lookup_key *k)
194 return jhash_3words(k->key, k->saddr, k->daddr, k->tunnel_type);
197 static u32 mutable_hash(const struct tnl_mutable_config *mutable)
199 struct port_lookup_key lookup;
201 lookup.saddr = mutable->port_config.saddr;
202 lookup.daddr = mutable->port_config.daddr;
203 lookup.key = mutable->port_config.in_key;
204 lookup.tunnel_type = mutable->tunnel_type;
206 return port_hash(&lookup);
209 static void check_table_empty(void)
211 if (tbl_count(port_table) == 0) {
212 struct tbl *old_table = port_table;
214 cancel_delayed_work_sync(&cache_cleaner_wq);
215 rcu_assign_pointer(port_table, NULL);
216 tbl_deferred_destroy(old_table, NULL);
220 static int add_port(struct vport *vport)
222 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
226 struct tbl *new_table;
228 new_table = tbl_create(0);
232 rcu_assign_pointer(port_table, new_table);
233 schedule_cache_cleaner();
235 } else if (tbl_count(port_table) > tbl_n_buckets(port_table)) {
236 struct tbl *old_table = port_table;
237 struct tbl *new_table;
239 new_table = tbl_expand(old_table);
240 if (IS_ERR(new_table))
241 return PTR_ERR(new_table);
243 rcu_assign_pointer(port_table, new_table);
244 tbl_deferred_destroy(old_table, NULL);
247 err = tbl_insert(port_table, &tnl_vport->tbl_node, mutable_hash(tnl_vport->mutable));
253 (*find_port_pool(tnl_vport->mutable))++;
258 static int move_port(struct vport *vport, struct tnl_mutable_config *new_mutable)
261 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
264 hash = mutable_hash(new_mutable);
265 if (hash == tnl_vport->tbl_node.hash)
269 * Ideally we should make this move atomic to avoid having gaps in
270 * finding tunnels or the possibility of failure. However, if we do
271 * find a tunnel it will always be consistent.
273 err = tbl_remove(port_table, &tnl_vport->tbl_node);
277 err = tbl_insert(port_table, &tnl_vport->tbl_node, hash);
284 assign_config_rcu(vport, new_mutable);
289 static int del_port(struct vport *vport)
291 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
294 err = tbl_remove(port_table, &tnl_vport->tbl_node);
299 (*find_port_pool(tnl_vport->mutable))--;
304 struct vport *tnl_find_port(__be32 saddr, __be32 daddr, __be32 key,
306 const struct tnl_mutable_config **mutable)
308 struct port_lookup_key lookup;
309 struct tbl *table = rcu_dereference(port_table);
310 struct tbl_node *tbl_node;
312 if (unlikely(!table))
315 lookup.saddr = saddr;
316 lookup.daddr = daddr;
318 if (tunnel_type & TNL_T_KEY_EXACT) {
320 lookup.tunnel_type = tunnel_type & ~TNL_T_KEY_MATCH;
322 if (key_local_remote_ports) {
323 tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
328 if (key_remote_ports) {
331 tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
335 lookup.saddr = saddr;
339 if (tunnel_type & TNL_T_KEY_MATCH) {
341 lookup.tunnel_type = tunnel_type & ~TNL_T_KEY_EXACT;
343 if (local_remote_ports) {
344 tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
352 tbl_node = tbl_lookup(table, &lookup, port_hash(&lookup), port_cmp);
361 *mutable = lookup.mutable;
362 return tnl_vport_to_vport(tnl_vport_table_cast(tbl_node));
365 static inline void ecn_decapsulate(struct sk_buff *skb)
367 u8 tos = ip_hdr(skb)->tos;
369 if (INET_ECN_is_ce(tos)) {
370 __be16 protocol = skb->protocol;
371 unsigned int nw_header = skb_network_offset(skb);
373 if (skb->protocol == htons(ETH_P_8021Q)) {
374 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
377 protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
378 nw_header += VLAN_HLEN;
381 if (protocol == htons(ETH_P_IP)) {
382 if (unlikely(!pskb_may_pull(skb, nw_header
383 + sizeof(struct iphdr))))
386 IP_ECN_set_ce((struct iphdr *)(skb->data + nw_header));
388 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
389 else if (protocol == htons(ETH_P_IPV6)) {
390 if (unlikely(!pskb_may_pull(skb, nw_header
391 + sizeof(struct ipv6hdr))))
394 IP6_ECN_set_ce((struct ipv6hdr *)(skb->data + nw_header));
400 /* Called with rcu_read_lock. */
401 void tnl_rcv(struct vport *vport, struct sk_buff *skb)
403 skb->pkt_type = PACKET_HOST;
404 skb->protocol = eth_type_trans(skb, skb->dev);
409 skb_reset_network_header(skb);
411 ecn_decapsulate(skb);
413 skb_push(skb, ETH_HLEN);
414 compute_ip_summed(skb, false);
416 vport_receive(vport, skb);
419 static bool check_ipv4_address(__be32 addr)
421 if (ipv4_is_multicast(addr) || ipv4_is_lbcast(addr)
422 || ipv4_is_loopback(addr) || ipv4_is_zeronet(addr))
428 static bool ipv4_should_icmp(struct sk_buff *skb)
430 struct iphdr *old_iph = ip_hdr(skb);
432 /* Don't respond to L2 broadcast. */
433 if (is_multicast_ether_addr(eth_hdr(skb)->h_dest))
436 /* Don't respond to L3 broadcast or invalid addresses. */
437 if (!check_ipv4_address(old_iph->daddr) ||
438 !check_ipv4_address(old_iph->saddr))
441 /* Only respond to the first fragment. */
442 if (old_iph->frag_off & htons(IP_OFFSET))
445 /* Don't respond to ICMP error messages. */
446 if (old_iph->protocol == IPPROTO_ICMP) {
447 u8 icmp_type, *icmp_typep;
449 icmp_typep = skb_header_pointer(skb, (u8 *)old_iph +
450 (old_iph->ihl << 2) +
451 offsetof(struct icmphdr, type) -
452 skb->data, sizeof(icmp_type),
458 if (*icmp_typep > NR_ICMP_TYPES
459 || (*icmp_typep <= ICMP_PARAMETERPROB
460 && *icmp_typep != ICMP_ECHOREPLY
461 && *icmp_typep != ICMP_ECHO))
468 static void ipv4_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
469 unsigned int mtu, unsigned int payload_length)
471 struct iphdr *iph, *old_iph = ip_hdr(skb);
472 struct icmphdr *icmph;
475 iph = (struct iphdr *)skb_put(nskb, sizeof(struct iphdr));
476 icmph = (struct icmphdr *)skb_put(nskb, sizeof(struct icmphdr));
477 payload = skb_put(nskb, payload_length);
481 iph->ihl = sizeof(struct iphdr) >> 2;
482 iph->tos = (old_iph->tos & IPTOS_TOS_MASK) |
483 IPTOS_PREC_INTERNETCONTROL;
484 iph->tot_len = htons(sizeof(struct iphdr)
485 + sizeof(struct icmphdr)
487 get_random_bytes(&iph->id, sizeof(iph->id));
490 iph->protocol = IPPROTO_ICMP;
491 iph->daddr = old_iph->saddr;
492 iph->saddr = old_iph->daddr;
497 icmph->type = ICMP_DEST_UNREACH;
498 icmph->code = ICMP_FRAG_NEEDED;
499 icmph->un.gateway = htonl(mtu);
502 nskb->csum = csum_partial((u8 *)icmph, sizeof(struct icmphdr), 0);
503 nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_iph - skb->data,
504 payload, payload_length,
506 icmph->checksum = csum_fold(nskb->csum);
509 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
510 static bool ipv6_should_icmp(struct sk_buff *skb)
512 struct ipv6hdr *old_ipv6h = ipv6_hdr(skb);
514 int payload_off = (u8 *)(old_ipv6h + 1) - skb->data;
515 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
517 /* Check source address is valid. */
518 addr_type = ipv6_addr_type(&old_ipv6h->saddr);
519 if (addr_type & IPV6_ADDR_MULTICAST || addr_type == IPV6_ADDR_ANY)
522 /* Don't reply to unspecified addresses. */
523 if (ipv6_addr_type(&old_ipv6h->daddr) == IPV6_ADDR_ANY)
526 /* Don't respond to ICMP error messages. */
527 payload_off = ipv6_skip_exthdr(skb, payload_off, &nexthdr);
531 if (nexthdr == NEXTHDR_ICMP) {
532 u8 icmp_type, *icmp_typep;
534 icmp_typep = skb_header_pointer(skb, payload_off +
535 offsetof(struct icmp6hdr,
537 sizeof(icmp_type), &icmp_type);
539 if (!icmp_typep || !(*icmp_typep & ICMPV6_INFOMSG_MASK))
546 static void ipv6_build_icmp(struct sk_buff *skb, struct sk_buff *nskb,
547 unsigned int mtu, unsigned int payload_length)
549 struct ipv6hdr *ipv6h, *old_ipv6h = ipv6_hdr(skb);
550 struct icmp6hdr *icmp6h;
553 ipv6h = (struct ipv6hdr *)skb_put(nskb, sizeof(struct ipv6hdr));
554 icmp6h = (struct icmp6hdr *)skb_put(nskb, sizeof(struct icmp6hdr));
555 payload = skb_put(nskb, payload_length);
560 memset(&ipv6h->flow_lbl, 0, sizeof(ipv6h->flow_lbl));
561 ipv6h->payload_len = htons(sizeof(struct icmp6hdr)
563 ipv6h->nexthdr = NEXTHDR_ICMP;
564 ipv6h->hop_limit = IPV6_DEFAULT_HOPLIMIT;
565 ipv6_addr_copy(&ipv6h->daddr, &old_ipv6h->saddr);
566 ipv6_addr_copy(&ipv6h->saddr, &old_ipv6h->daddr);
569 icmp6h->icmp6_type = ICMPV6_PKT_TOOBIG;
570 icmp6h->icmp6_code = 0;
571 icmp6h->icmp6_cksum = 0;
572 icmp6h->icmp6_mtu = htonl(mtu);
574 nskb->csum = csum_partial((u8 *)icmp6h, sizeof(struct icmp6hdr), 0);
575 nskb->csum = skb_copy_and_csum_bits(skb, (u8 *)old_ipv6h - skb->data,
576 payload, payload_length,
578 icmp6h->icmp6_cksum = csum_ipv6_magic(&ipv6h->saddr, &ipv6h->daddr,
579 sizeof(struct icmp6hdr)
581 ipv6h->nexthdr, nskb->csum);
585 bool tnl_frag_needed(struct vport *vport, const struct tnl_mutable_config *mutable,
586 struct sk_buff *skb, unsigned int mtu, __be32 flow_key)
588 unsigned int eth_hdr_len = ETH_HLEN;
589 unsigned int total_length = 0, header_length = 0, payload_length;
590 struct ethhdr *eh, *old_eh = eth_hdr(skb);
591 struct sk_buff *nskb;
594 if (skb->protocol == htons(ETH_P_IP)) {
595 if (mtu < IP_MIN_MTU)
598 if (!ipv4_should_icmp(skb))
601 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
602 else if (skb->protocol == htons(ETH_P_IPV6)) {
603 if (mtu < IPV6_MIN_MTU)
607 * In theory we should do PMTUD on IPv6 multicast messages but
608 * we don't have an address to send from so just fragment.
610 if (ipv6_addr_type(&ipv6_hdr(skb)->daddr) & IPV6_ADDR_MULTICAST)
613 if (!ipv6_should_icmp(skb))
621 if (old_eh->h_proto == htons(ETH_P_8021Q))
622 eth_hdr_len = VLAN_ETH_HLEN;
624 payload_length = skb->len - eth_hdr_len;
625 if (skb->protocol == htons(ETH_P_IP)) {
626 header_length = sizeof(struct iphdr) + sizeof(struct icmphdr);
627 total_length = min_t(unsigned int, header_length +
628 payload_length, 576);
630 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
632 header_length = sizeof(struct ipv6hdr) +
633 sizeof(struct icmp6hdr);
634 total_length = min_t(unsigned int, header_length +
635 payload_length, IPV6_MIN_MTU);
639 total_length = min(total_length, mutable->mtu);
640 payload_length = total_length - header_length;
642 nskb = dev_alloc_skb(NET_IP_ALIGN + eth_hdr_len + header_length +
647 skb_reserve(nskb, NET_IP_ALIGN);
649 /* Ethernet / VLAN */
650 eh = (struct ethhdr *)skb_put(nskb, eth_hdr_len);
651 memcpy(eh->h_dest, old_eh->h_source, ETH_ALEN);
652 memcpy(eh->h_source, mutable->eth_addr, ETH_ALEN);
653 nskb->protocol = eh->h_proto = old_eh->h_proto;
654 if (old_eh->h_proto == htons(ETH_P_8021Q)) {
655 struct vlan_ethhdr *vh = (struct vlan_ethhdr *)eh;
657 vh->h_vlan_TCI = vlan_eth_hdr(skb)->h_vlan_TCI;
658 vh->h_vlan_encapsulated_proto = skb->protocol;
660 skb_reset_mac_header(nskb);
663 if (skb->protocol == htons(ETH_P_IP))
664 ipv4_build_icmp(skb, nskb, mtu, payload_length);
665 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
667 ipv6_build_icmp(skb, nskb, mtu, payload_length);
671 * Assume that flow based keys are symmetric with respect to input
672 * and output and use the key that we were going to put on the
673 * outgoing packet for the fake received packet. If the keys are
674 * not symmetric then PMTUD needs to be disabled since we won't have
675 * any way of synthesizing packets.
677 if ((mutable->port_config.flags & (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION)) ==
678 (TNL_F_IN_KEY_MATCH | TNL_F_OUT_KEY_ACTION))
679 OVS_CB(nskb)->tun_id = flow_key;
681 compute_ip_summed(nskb, false);
682 vport_receive(vport, nskb);
687 static bool check_mtu(struct sk_buff *skb,
689 const struct tnl_mutable_config *mutable,
690 const struct rtable *rt, __be16 *frag_offp)
695 frag_off = (mutable->port_config.flags & TNL_F_PMTUD) ? htons(IP_DF) : 0;
697 mtu = dst_mtu(&rt_dst(rt))
699 - mutable->tunnel_hlen
700 - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0);
704 if (skb->protocol == htons(ETH_P_IP)) {
705 struct iphdr *old_iph = ip_hdr(skb);
707 frag_off |= old_iph->frag_off & htons(IP_DF);
708 mtu = max(mtu, IP_MIN_MTU);
710 if ((old_iph->frag_off & htons(IP_DF)) &&
711 mtu < ntohs(old_iph->tot_len)) {
712 if (tnl_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id))
716 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
717 else if (skb->protocol == htons(ETH_P_IPV6)) {
718 unsigned int packet_length = skb->len - ETH_HLEN
719 - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ? VLAN_HLEN : 0);
721 mtu = max(mtu, IPV6_MIN_MTU);
723 /* IPv6 requires PMTUD if the packet is above the minimum MTU. */
724 if (packet_length > IPV6_MIN_MTU)
725 frag_off = htons(IP_DF);
727 if (mtu < packet_length) {
728 if (tnl_frag_needed(vport, mutable, skb, mtu, OVS_CB(skb)->tun_id))
734 *frag_offp = frag_off;
742 static void create_tunnel_header(const struct vport *vport,
743 const struct tnl_mutable_config *mutable,
744 const struct rtable *rt, void *header)
746 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
747 struct iphdr *iph = header;
750 iph->ihl = sizeof(struct iphdr) >> 2;
751 iph->frag_off = htons(IP_DF);
752 iph->protocol = tnl_vport->tnl_ops->ipproto;
753 iph->tos = mutable->port_config.tos;
754 iph->daddr = rt->rt_dst;
755 iph->saddr = rt->rt_src;
756 iph->ttl = mutable->port_config.ttl;
758 iph->ttl = dst_metric(&rt_dst(rt), RTAX_HOPLIMIT);
760 tnl_vport->tnl_ops->build_header(vport, mutable, iph + 1);
763 static inline void *get_cached_header(const struct tnl_cache *cache)
765 return (void *)cache + ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN);
768 static inline bool check_cache_valid(const struct tnl_cache *cache,
769 const struct tnl_mutable_config *mutable)
772 #ifdef NEED_CACHE_TIMEOUT
773 time_before(jiffies, cache->expiration) &&
776 atomic_read(&init_net.ipv4.rt_genid) == cache->rt->rt_genid &&
779 rt_dst(cache->rt).hh->hh_lock.sequence == cache->hh_seq &&
781 mutable->seq == cache->mutable_seq &&
782 (!is_internal_dev(rt_dst(cache->rt).dev) ||
783 (cache->flow && !cache->flow->dead));
786 static int cache_cleaner_cb(struct tbl_node *tbl_node, void *aux)
788 struct tnl_vport *tnl_vport = tnl_vport_table_cast(tbl_node);
789 const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
790 const struct tnl_cache *cache = rcu_dereference(tnl_vport->cache);
792 if (cache && !check_cache_valid(cache, mutable) &&
793 spin_trylock_bh(&tnl_vport->cache_lock)) {
794 assign_cache_rcu(tnl_vport_to_vport(tnl_vport), NULL);
795 spin_unlock_bh(&tnl_vport->cache_lock);
801 static void cache_cleaner(struct work_struct *work)
803 schedule_cache_cleaner();
806 tbl_foreach(port_table, cache_cleaner_cb, NULL);
810 static inline void create_eth_hdr(struct tnl_cache *cache,
811 const struct rtable *rt)
813 void *cache_data = get_cached_header(cache);
814 int hh_len = rt_dst(rt).hh->hh_len;
815 int hh_off = HH_DATA_ALIGN(rt_dst(rt).hh->hh_len) - hh_len;
821 hh_seq = read_seqbegin(&rt_dst(rt).hh->hh_lock);
822 memcpy(cache_data, (void *)rt_dst(rt).hh->hh_data + hh_off, hh_len);
823 } while (read_seqretry(&rt_dst(rt).hh->hh_lock, hh_seq));
825 cache->hh_seq = hh_seq;
827 read_lock_bh(&rt_dst(rt).hh->hh_lock);
828 memcpy(cache_data, (void *)rt_dst(rt).hh->hh_data + hh_off, hh_len);
829 read_unlock_bh(&rt_dst(rt).hh->hh_lock);
833 static struct tnl_cache *build_cache(struct vport *vport,
834 const struct tnl_mutable_config *mutable,
837 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
838 struct tnl_cache *cache;
842 if (!(mutable->port_config.flags & TNL_F_HDR_CACHE))
846 * If there is no entry in the ARP cache or if this device does not
847 * support hard header caching just fall back to the IP stack.
853 * If lock is contended fall back to directly building the header.
854 * We're not going to help performance by sitting here spinning.
856 if (!spin_trylock_bh(&tnl_vport->cache_lock))
859 cache = tnl_vport->cache;
860 if (check_cache_valid(cache, mutable))
865 cache_len = rt_dst(rt).hh->hh_len + mutable->tunnel_hlen;
867 cache = kzalloc(ALIGN(sizeof(struct tnl_cache), CACHE_DATA_ALIGN) +
868 cache_len, GFP_ATOMIC);
872 cache->len = cache_len;
874 create_eth_hdr(cache, rt);
875 cache_data = get_cached_header(cache) + rt_dst(rt).hh->hh_len;
877 create_tunnel_header(vport, mutable, rt, cache_data);
879 cache->mutable_seq = mutable->seq;
881 #ifdef NEED_CACHE_TIMEOUT
882 cache->expiration = jiffies + tnl_vport->cache_exp_interval;
885 if (is_internal_dev(rt_dst(rt).dev)) {
888 struct dp_port *dp_port;
891 struct odp_flow_key flow_key;
892 struct tbl_node *flow_node;
894 vport = internal_dev_get_vport(rt_dst(rt).dev);
898 dp_port = vport_get_dp_port(vport);
902 skb = alloc_skb(cache->len, GFP_ATOMIC);
906 __skb_put(skb, cache->len);
907 memcpy(skb->data, get_cached_header(cache), cache->len);
909 err = flow_extract(skb, dp_port->port_no, &flow_key, &is_frag);
915 flow_node = tbl_lookup(rcu_dereference(dp_port->dp->table),
916 &flow_key, flow_hash(&flow_key),
919 struct sw_flow *flow = flow_cast(flow_node);
927 assign_cache_rcu(vport, cache);
930 spin_unlock_bh(&tnl_vport->cache_lock);
935 static struct rtable *find_route(struct vport *vport,
936 const struct tnl_mutable_config *mutable,
937 u8 tos, struct tnl_cache **cache)
939 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
940 struct tnl_cache *cur_cache = rcu_dereference(tnl_vport->cache);
945 if (likely(tos == mutable->port_config.tos &&
946 check_cache_valid(cur_cache, mutable))) {
948 return cur_cache->rt;
951 struct flowi fl = { .nl_u = { .ip4_u =
952 { .daddr = mutable->port_config.daddr,
953 .saddr = mutable->port_config.saddr,
955 .proto = tnl_vport->tnl_ops->ipproto };
957 if (unlikely(ip_route_output_key(&init_net, &rt, &fl)))
960 if (likely(tos == mutable->port_config.tos))
961 *cache = build_cache(vport, mutable, rt);
967 static struct sk_buff *check_headroom(struct sk_buff *skb, int headroom)
969 if (skb_headroom(skb) < headroom || skb_header_cloned(skb)) {
970 struct sk_buff *nskb = skb_realloc_headroom(skb, headroom + 16);
971 if (unlikely(!nskb)) {
973 return ERR_PTR(-ENOMEM);
976 set_skb_csum_bits(skb, nskb);
979 skb_set_owner_w(nskb, skb->sk);
988 static inline bool need_linearize(const struct sk_buff *skb)
992 if (unlikely(skb_shinfo(skb)->frag_list))
996 * Generally speaking we should linearize if there are paged frags.
997 * However, if all of the refcounts are 1 we know nobody else can
998 * change them from underneath us and we can skip the linearization.
1000 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1001 if (unlikely(page_count(skb_shinfo(skb)->frags[0].page) > 1))
1007 static struct sk_buff *handle_offloads(struct sk_buff *skb,
1008 const struct tnl_mutable_config *mutable,
1009 const struct rtable *rt)
1014 forward_ip_summed(skb);
1016 err = vswitch_skb_checksum_setup(skb);
1020 min_headroom = LL_RESERVED_SPACE(rt_dst(rt).dev) + rt_dst(rt).header_len
1021 + mutable->tunnel_hlen;
1023 if (skb_is_gso(skb)) {
1024 struct sk_buff *nskb;
1027 * If we are doing GSO on a pskb it is better to make sure that
1028 * the headroom is correct now. We will only have to copy the
1029 * portion in the linear data area and GSO will preserve
1030 * headroom when it creates the segments. This is particularly
1031 * beneficial on Xen where we get a lot of GSO pskbs.
1032 * Conversely, we avoid copying if it is just to get our own
1033 * writable clone because GSO will do the copy for us.
1035 if (skb_headroom(skb) < min_headroom) {
1036 skb = check_headroom(skb, min_headroom);
1037 if (unlikely(IS_ERR(skb))) {
1043 nskb = skb_gso_segment(skb, 0);
1045 if (unlikely(IS_ERR(nskb))) {
1046 err = PTR_ERR(nskb);
1052 skb = check_headroom(skb, min_headroom);
1053 if (unlikely(IS_ERR(skb))) {
1058 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1060 * Pages aren't locked and could change at any time.
1061 * If this happens after we compute the checksum, the
1062 * checksum will be wrong. We linearize now to avoid
1065 if (unlikely(need_linearize(skb))) {
1066 err = __skb_linearize(skb);
1071 err = skb_checksum_help(skb);
1074 } else if (skb->ip_summed == CHECKSUM_COMPLETE)
1075 skb->ip_summed = CHECKSUM_NONE;
1083 return ERR_PTR(err);
1086 static int send_frags(struct sk_buff *skb,
1087 const struct tnl_mutable_config *mutable)
1094 struct sk_buff *next = skb->next;
1095 int frag_len = skb->len - mutable->tunnel_hlen;
1098 memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
1100 err = ip_local_out(skb);
1101 if (likely(net_xmit_eval(err) == 0))
1102 sent_len += frag_len;
1115 * There's no point in continuing to send fragments once one has been
1116 * dropped so just free the rest. This may help improve the congestion
1117 * that caused the first packet to be dropped.
1119 tnl_free_linked_skbs(skb);
1123 int tnl_send(struct vport *vport, struct sk_buff *skb)
1125 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1126 const struct tnl_mutable_config *mutable = rcu_dereference(tnl_vport->mutable);
1128 enum vport_err_type err = VPORT_E_TX_ERROR;
1130 struct dst_entry *unattached_dst = NULL;
1131 struct tnl_cache *cache;
1138 /* Validate the protocol headers before we try to use them. */
1139 if (skb->protocol == htons(ETH_P_8021Q)) {
1140 if (unlikely(!pskb_may_pull(skb, VLAN_ETH_HLEN)))
1143 skb->protocol = vlan_eth_hdr(skb)->h_vlan_encapsulated_proto;
1144 skb_set_network_header(skb, VLAN_ETH_HLEN);
1147 if (skb->protocol == htons(ETH_P_IP)) {
1148 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1149 + sizeof(struct iphdr))))
1152 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1153 else if (skb->protocol == htons(ETH_P_IPV6)) {
1154 if (unlikely(!pskb_may_pull(skb, skb_network_offset(skb)
1155 + sizeof(struct ipv6hdr))))
1161 if (skb->protocol == htons(ETH_P_IP))
1162 inner_tos = ip_hdr(skb)->tos;
1163 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1164 else if (skb->protocol == htons(ETH_P_IPV6))
1165 inner_tos = ipv6_get_dsfield(ipv6_hdr(skb));
1170 if (mutable->port_config.flags & TNL_F_TOS_INHERIT)
1173 tos = mutable->port_config.tos;
1175 tos = INET_ECN_encapsulate(tos, inner_tos);
1178 rt = find_route(vport, mutable, tos, &cache);
1181 if (unlikely(!cache))
1182 unattached_dst = &rt_dst(rt);
1190 skb = handle_offloads(skb, mutable, rt);
1191 if (unlikely(IS_ERR(skb)))
1195 if (unlikely(!check_mtu(skb, vport, mutable, rt, &frag_off))) {
1196 err = VPORT_E_TX_DROPPED;
1201 * If we are over the MTU, allow the IP stack to handle fragmentation.
1202 * Fragmentation is a slow path anyways.
1204 if (unlikely(skb->len + mutable->tunnel_hlen > dst_mtu(&rt_dst(rt)) &&
1206 unattached_dst = &rt_dst(rt);
1207 dst_hold(unattached_dst);
1212 ttl = mutable->port_config.ttl;
1214 ttl = dst_metric(&rt_dst(rt), RTAX_HOPLIMIT);
1216 if (mutable->port_config.flags & TNL_F_TTL_INHERIT) {
1217 if (skb->protocol == htons(ETH_P_IP))
1218 ttl = ip_hdr(skb)->ttl;
1219 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1220 else if (skb->protocol == htons(ETH_P_IPV6))
1221 ttl = ipv6_hdr(skb)->hop_limit;
1227 struct sk_buff *next_skb = skb->next;
1230 if (likely(cache)) {
1231 skb_push(skb, cache->len);
1232 memcpy(skb->data, get_cached_header(cache), cache->len);
1233 skb_reset_mac_header(skb);
1234 skb_set_network_header(skb, rt_dst(rt).hh->hh_len);
1237 skb_push(skb, mutable->tunnel_hlen);
1238 create_tunnel_header(vport, mutable, rt, skb->data);
1239 skb_reset_network_header(skb);
1242 skb_dst_set(skb, dst_clone(unattached_dst));
1244 skb_dst_set(skb, unattached_dst);
1245 unattached_dst = NULL;
1248 skb_set_transport_header(skb, skb_network_offset(skb) + sizeof(struct iphdr));
1253 iph->frag_off = frag_off;
1254 ip_select_ident(iph, &rt_dst(rt), NULL);
1256 skb = tnl_vport->tnl_ops->update_header(vport, mutable, &rt_dst(rt), skb);
1260 if (likely(cache)) {
1261 int orig_len = skb->len - cache->len;
1262 struct vport *cache_vport = internal_dev_get_vport(rt_dst(rt).dev);
1264 skb->protocol = htons(ETH_P_IP);
1265 iph->tot_len = htons(skb->len - skb_network_offset(skb));
1269 OVS_CB(skb)->flow = cache->flow;
1270 compute_ip_summed(skb, true);
1271 vport_receive(cache_vport, skb);
1272 sent_len += orig_len;
1276 skb->dev = rt_dst(rt).dev;
1277 err = dev_queue_xmit(skb);
1279 if (likely(net_xmit_eval(err) == 0))
1280 sent_len += orig_len;
1283 sent_len += send_frags(skb, mutable);
1289 if (unlikely(sent_len == 0))
1290 vport_record_error(vport, VPORT_E_TX_DROPPED);
1295 tnl_free_linked_skbs(skb);
1297 dst_release(unattached_dst);
1298 vport_record_error(vport, err);
1303 static int set_config(const void __user *uconfig, const struct tnl_ops *tnl_ops,
1304 const struct vport *cur_vport,
1305 struct tnl_mutable_config *mutable)
1307 const struct vport *old_vport;
1308 const struct tnl_mutable_config *old_mutable;
1310 if (copy_from_user(&mutable->port_config, uconfig, sizeof(struct tnl_port_config)))
1313 if (mutable->port_config.daddr == 0)
1316 if (mutable->port_config.tos != RT_TOS(mutable->port_config.tos))
1319 mutable->tunnel_hlen = tnl_ops->hdr_len(&mutable->port_config);
1320 if (mutable->tunnel_hlen < 0)
1321 return mutable->tunnel_hlen;
1323 mutable->tunnel_hlen += sizeof(struct iphdr);
1325 mutable->tunnel_type = tnl_ops->tunnel_type;
1326 if (mutable->port_config.flags & TNL_F_IN_KEY_MATCH) {
1327 mutable->tunnel_type |= TNL_T_KEY_MATCH;
1328 mutable->port_config.in_key = 0;
1330 mutable->tunnel_type |= TNL_T_KEY_EXACT;
1332 old_vport = tnl_find_port(mutable->port_config.saddr,
1333 mutable->port_config.daddr,
1334 mutable->port_config.in_key,
1335 mutable->tunnel_type,
1338 if (old_vport && old_vport != cur_vport)
1341 if (mutable->port_config.flags & TNL_F_OUT_KEY_ACTION)
1342 mutable->port_config.out_key = 0;
1347 struct vport *tnl_create(const struct vport_parms *parms,
1348 const struct vport_ops *vport_ops,
1349 const struct tnl_ops *tnl_ops)
1351 struct vport *vport;
1352 struct tnl_vport *tnl_vport;
1353 int initial_frag_id;
1356 vport = vport_alloc(sizeof(struct tnl_vport), vport_ops);
1357 if (IS_ERR(vport)) {
1358 err = PTR_ERR(vport);
1362 tnl_vport = tnl_vport_priv(vport);
1364 strcpy(tnl_vport->name, parms->name);
1365 tnl_vport->tnl_ops = tnl_ops;
1367 tnl_vport->mutable = kzalloc(sizeof(struct tnl_mutable_config), GFP_KERNEL);
1368 if (!tnl_vport->mutable) {
1370 goto error_free_vport;
1373 vport_gen_rand_ether_addr(tnl_vport->mutable->eth_addr);
1374 tnl_vport->mutable->mtu = ETH_DATA_LEN;
1376 get_random_bytes(&initial_frag_id, sizeof(int));
1377 atomic_set(&tnl_vport->frag_id, initial_frag_id);
1379 err = set_config(parms->config, tnl_ops, NULL, tnl_vport->mutable);
1381 goto error_free_mutable;
1383 spin_lock_init(&tnl_vport->cache_lock);
1385 #ifdef NEED_CACHE_TIMEOUT
1386 tnl_vport->cache_exp_interval = MAX_CACHE_EXP -
1387 (net_random() % (MAX_CACHE_EXP / 2));
1390 err = add_port(vport);
1392 goto error_free_mutable;
1397 kfree(tnl_vport->mutable);
1401 return ERR_PTR(err);
1404 int tnl_modify(struct vport *vport, const void __user *config)
1406 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1407 struct tnl_mutable_config *mutable;
1410 mutable = kmemdup(tnl_vport->mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL);
1416 err = set_config(config, tnl_vport->tnl_ops, vport, mutable);
1422 err = move_port(vport, mutable);
1434 static void free_port_rcu(struct rcu_head *rcu)
1436 struct tnl_vport *tnl_vport = container_of(rcu, struct tnl_vport, rcu);
1438 spin_lock_bh(&tnl_vport->cache_lock);
1439 free_cache(tnl_vport->cache);
1440 spin_unlock_bh(&tnl_vport->cache_lock);
1442 kfree(tnl_vport->mutable);
1443 vport_free(tnl_vport_to_vport(tnl_vport));
1446 int tnl_destroy(struct vport *vport)
1448 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1449 const struct tnl_mutable_config *old_mutable;
1451 if (vport == tnl_find_port(tnl_vport->mutable->port_config.saddr,
1452 tnl_vport->mutable->port_config.daddr,
1453 tnl_vport->mutable->port_config.in_key,
1454 tnl_vport->mutable->tunnel_type,
1458 call_rcu(&tnl_vport->rcu, free_port_rcu);
1463 int tnl_set_mtu(struct vport *vport, int mtu)
1465 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1466 struct tnl_mutable_config *mutable;
1468 mutable = kmemdup(tnl_vport->mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL);
1473 assign_config_rcu(vport, mutable);
1478 int tnl_set_addr(struct vport *vport, const unsigned char *addr)
1480 struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1481 struct tnl_mutable_config *mutable;
1483 mutable = kmemdup(tnl_vport->mutable, sizeof(struct tnl_mutable_config), GFP_KERNEL);
1487 memcpy(mutable->eth_addr, addr, ETH_ALEN);
1488 assign_config_rcu(vport, mutable);
1493 const char *tnl_get_name(const struct vport *vport)
1495 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1496 return tnl_vport->name;
1499 const unsigned char *tnl_get_addr(const struct vport *vport)
1501 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1502 return rcu_dereference(tnl_vport->mutable)->eth_addr;
1505 int tnl_get_mtu(const struct vport *vport)
1507 const struct tnl_vport *tnl_vport = tnl_vport_priv(vport);
1508 return rcu_dereference(tnl_vport->mutable)->mtu;
1511 void tnl_free_linked_skbs(struct sk_buff *skb)
1517 struct sk_buff *next = skb->next;