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54 * Interval to check for and remove caches that are no longer valid.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 static struct tbl __rcu
*port_table __read_mostly
;
73 static void cache_cleaner(struct work_struct
*work
);
74 static 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
= rtnl_dereference(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
{
168 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_rtnl(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 u32 x
= jhash_3words((__force u32
)k
->saddr
, (__force u32
)k
->daddr
,
196 return jhash_2words((__force u64
)k
->key
>> 32, (__force u32
)k
->key
, x
);
199 static u32
mutable_hash(const struct tnl_mutable_config
*mutable)
201 struct port_lookup_key lookup
;
203 lookup
.saddr
= mutable->port_config
.saddr
;
204 lookup
.daddr
= mutable->port_config
.daddr
;
205 lookup
.key
= mutable->port_config
.in_key
;
206 lookup
.tunnel_type
= mutable->tunnel_type
;
208 return port_hash(&lookup
);
211 static void check_table_empty(void)
213 struct tbl
*old_table
= rtnl_dereference(port_table
);
215 if (tbl_count(old_table
) == 0) {
216 cancel_delayed_work_sync(&cache_cleaner_wq
);
217 rcu_assign_pointer(port_table
, NULL
);
218 tbl_deferred_destroy(old_table
, NULL
);
222 static int add_port(struct vport
*vport
)
224 struct tbl
*cur_table
= rtnl_dereference(port_table
);
225 struct tnl_vport
*tnl_vport
= tnl_vport_priv(vport
);
229 struct tbl
*new_table
;
231 new_table
= tbl_create(0);
235 rcu_assign_pointer(port_table
, new_table
);
236 schedule_cache_cleaner();
238 } else if (tbl_count(cur_table
) > tbl_n_buckets(cur_table
)) {
239 struct tbl
*new_table
;
241 new_table
= tbl_expand(cur_table
);
242 if (IS_ERR(new_table
))
243 return PTR_ERR(new_table
);
245 rcu_assign_pointer(port_table
, new_table
);
246 tbl_deferred_destroy(cur_table
, NULL
);
249 err
= tbl_insert(rtnl_dereference(port_table
), &tnl_vport
->tbl_node
,
250 mutable_hash(rtnl_dereference(tnl_vport
->mutable)));
256 (*find_port_pool(rtnl_dereference(tnl_vport
->mutable)))++;
261 static int move_port(struct vport
*vport
, struct tnl_mutable_config
*new_mutable
)
264 struct tbl
*cur_table
= rtnl_dereference(port_table
);
265 struct tnl_vport
*tnl_vport
= tnl_vport_priv(vport
);
268 hash
= mutable_hash(new_mutable
);
269 if (hash
== tnl_vport
->tbl_node
.hash
)
273 * Ideally we should make this move atomic to avoid having gaps in
274 * finding tunnels or the possibility of failure. However, if we do
275 * find a tunnel it will always be consistent.
277 err
= tbl_remove(cur_table
, &tnl_vport
->tbl_node
);
281 err
= tbl_insert(cur_table
, &tnl_vport
->tbl_node
, hash
);
283 (*find_port_pool(rtnl_dereference(tnl_vport
->mutable)))--;
289 (*find_port_pool(rtnl_dereference(tnl_vport
->mutable)))--;
290 assign_config_rcu(vport
, new_mutable
);
291 (*find_port_pool(rtnl_dereference(tnl_vport
->mutable)))++;
296 static int del_port(struct vport
*vport
)
298 struct tnl_vport
*tnl_vport
= tnl_vport_priv(vport
);
301 err
= tbl_remove(rtnl_dereference(port_table
), &tnl_vport
->tbl_node
);
306 (*find_port_pool(rtnl_dereference(tnl_vport
->mutable)))--;
311 struct vport
*tnl_find_port(__be32 saddr
, __be32 daddr
, __be64 key
,
313 const struct tnl_mutable_config
**mutable)
315 struct port_lookup_key lookup
;
316 struct tbl
*table
= rcu_dereference_rtnl(port_table
);
317 struct tbl_node
*tbl_node
;
319 if (unlikely(!table
))
322 lookup
.saddr
= saddr
;
323 lookup
.daddr
= daddr
;
325 if (tunnel_type
& TNL_T_KEY_EXACT
) {
327 lookup
.tunnel_type
= tunnel_type
& ~TNL_T_KEY_MATCH
;
329 if (key_local_remote_ports
) {
330 tbl_node
= tbl_lookup(table
, &lookup
, port_hash(&lookup
), port_cmp
);
335 if (key_remote_ports
) {
338 tbl_node
= tbl_lookup(table
, &lookup
, port_hash(&lookup
), port_cmp
);
342 lookup
.saddr
= saddr
;
346 if (tunnel_type
& TNL_T_KEY_MATCH
) {
348 lookup
.tunnel_type
= tunnel_type
& ~TNL_T_KEY_EXACT
;
350 if (local_remote_ports
) {
351 tbl_node
= tbl_lookup(table
, &lookup
, port_hash(&lookup
), port_cmp
);
359 tbl_node
= tbl_lookup(table
, &lookup
, port_hash(&lookup
), port_cmp
);
368 *mutable = lookup
.mutable;
369 return tnl_vport_to_vport(tnl_vport_table_cast(tbl_node
));
372 static inline void ecn_decapsulate(struct sk_buff
*skb
)
374 /* This is accessing the outer IP header of the tunnel, which we've
375 * already validated to be OK. skb->data is currently set to the start
376 * of the inner Ethernet header, and we've validated ETH_HLEN.
378 if (unlikely(INET_ECN_is_ce(ip_hdr(skb
)->tos
))) {
379 __be16 protocol
= skb
->protocol
;
381 skb_set_network_header(skb
, ETH_HLEN
);
383 if (skb
->protocol
== htons(ETH_P_8021Q
)) {
384 if (unlikely(!pskb_may_pull(skb
, VLAN_ETH_HLEN
)))
387 protocol
= vlan_eth_hdr(skb
)->h_vlan_encapsulated_proto
;
388 skb_set_network_header(skb
, VLAN_ETH_HLEN
);
391 if (protocol
== htons(ETH_P_IP
)) {
392 if (unlikely(!pskb_may_pull(skb
, skb_network_offset(skb
)
393 + sizeof(struct iphdr
))))
396 IP_ECN_set_ce(ip_hdr(skb
));
398 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
399 else if (protocol
== htons(ETH_P_IPV6
)) {
400 if (unlikely(!pskb_may_pull(skb
, skb_network_offset(skb
)
401 + sizeof(struct ipv6hdr
))))
404 IP6_ECN_set_ce(ipv6_hdr(skb
));
410 /* Called with rcu_read_lock. */
411 void tnl_rcv(struct vport
*vport
, struct sk_buff
*skb
)
413 /* Packets received by this function are in the following state:
414 * - skb->data points to the inner Ethernet header.
415 * - The inner Ethernet header is in the linear data area.
416 * - skb->csum does not include the inner Ethernet header.
417 * - The layer pointers point at the outer headers.
420 struct ethhdr
*eh
= (struct ethhdr
*)skb
->data
;
422 if (likely(ntohs(eh
->h_proto
) >= 1536))
423 skb
->protocol
= eh
->h_proto
;
425 skb
->protocol
= htons(ETH_P_802_2
);
431 ecn_decapsulate(skb
);
432 compute_ip_summed(skb
, false);
434 vport_receive(vport
, skb
);
437 static bool check_ipv4_address(__be32 addr
)
439 if (ipv4_is_multicast(addr
) || ipv4_is_lbcast(addr
)
440 || ipv4_is_loopback(addr
) || ipv4_is_zeronet(addr
))
446 static bool ipv4_should_icmp(struct sk_buff
*skb
)
448 struct iphdr
*old_iph
= ip_hdr(skb
);
450 /* Don't respond to L2 broadcast. */
451 if (is_multicast_ether_addr(eth_hdr(skb
)->h_dest
))
454 /* Don't respond to L3 broadcast or invalid addresses. */
455 if (!check_ipv4_address(old_iph
->daddr
) ||
456 !check_ipv4_address(old_iph
->saddr
))
459 /* Only respond to the first fragment. */
460 if (old_iph
->frag_off
& htons(IP_OFFSET
))
463 /* Don't respond to ICMP error messages. */
464 if (old_iph
->protocol
== IPPROTO_ICMP
) {
465 u8 icmp_type
, *icmp_typep
;
467 icmp_typep
= skb_header_pointer(skb
, (u8
*)old_iph
+
468 (old_iph
->ihl
<< 2) +
469 offsetof(struct icmphdr
, type
) -
470 skb
->data
, sizeof(icmp_type
),
476 if (*icmp_typep
> NR_ICMP_TYPES
477 || (*icmp_typep
<= ICMP_PARAMETERPROB
478 && *icmp_typep
!= ICMP_ECHOREPLY
479 && *icmp_typep
!= ICMP_ECHO
))
486 static void ipv4_build_icmp(struct sk_buff
*skb
, struct sk_buff
*nskb
,
487 unsigned int mtu
, unsigned int payload_length
)
489 struct iphdr
*iph
, *old_iph
= ip_hdr(skb
);
490 struct icmphdr
*icmph
;
493 iph
= (struct iphdr
*)skb_put(nskb
, sizeof(struct iphdr
));
494 icmph
= (struct icmphdr
*)skb_put(nskb
, sizeof(struct icmphdr
));
495 payload
= skb_put(nskb
, payload_length
);
499 iph
->ihl
= sizeof(struct iphdr
) >> 2;
500 iph
->tos
= (old_iph
->tos
& IPTOS_TOS_MASK
) |
501 IPTOS_PREC_INTERNETCONTROL
;
502 iph
->tot_len
= htons(sizeof(struct iphdr
)
503 + sizeof(struct icmphdr
)
505 get_random_bytes(&iph
->id
, sizeof(iph
->id
));
508 iph
->protocol
= IPPROTO_ICMP
;
509 iph
->daddr
= old_iph
->saddr
;
510 iph
->saddr
= old_iph
->daddr
;
515 icmph
->type
= ICMP_DEST_UNREACH
;
516 icmph
->code
= ICMP_FRAG_NEEDED
;
517 icmph
->un
.gateway
= htonl(mtu
);
520 nskb
->csum
= csum_partial((u8
*)icmph
, sizeof(struct icmphdr
), 0);
521 nskb
->csum
= skb_copy_and_csum_bits(skb
, (u8
*)old_iph
- skb
->data
,
522 payload
, payload_length
,
524 icmph
->checksum
= csum_fold(nskb
->csum
);
527 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
528 static bool ipv6_should_icmp(struct sk_buff
*skb
)
530 struct ipv6hdr
*old_ipv6h
= ipv6_hdr(skb
);
532 int payload_off
= (u8
*)(old_ipv6h
+ 1) - skb
->data
;
533 u8 nexthdr
= ipv6_hdr(skb
)->nexthdr
;
535 /* Check source address is valid. */
536 addr_type
= ipv6_addr_type(&old_ipv6h
->saddr
);
537 if (addr_type
& IPV6_ADDR_MULTICAST
|| addr_type
== IPV6_ADDR_ANY
)
540 /* Don't reply to unspecified addresses. */
541 if (ipv6_addr_type(&old_ipv6h
->daddr
) == IPV6_ADDR_ANY
)
544 /* Don't respond to ICMP error messages. */
545 payload_off
= ipv6_skip_exthdr(skb
, payload_off
, &nexthdr
);
549 if (nexthdr
== NEXTHDR_ICMP
) {
550 u8 icmp_type
, *icmp_typep
;
552 icmp_typep
= skb_header_pointer(skb
, payload_off
+
553 offsetof(struct icmp6hdr
,
555 sizeof(icmp_type
), &icmp_type
);
557 if (!icmp_typep
|| !(*icmp_typep
& ICMPV6_INFOMSG_MASK
))
564 static void ipv6_build_icmp(struct sk_buff
*skb
, struct sk_buff
*nskb
,
565 unsigned int mtu
, unsigned int payload_length
)
567 struct ipv6hdr
*ipv6h
, *old_ipv6h
= ipv6_hdr(skb
);
568 struct icmp6hdr
*icmp6h
;
571 ipv6h
= (struct ipv6hdr
*)skb_put(nskb
, sizeof(struct ipv6hdr
));
572 icmp6h
= (struct icmp6hdr
*)skb_put(nskb
, sizeof(struct icmp6hdr
));
573 payload
= skb_put(nskb
, payload_length
);
578 memset(&ipv6h
->flow_lbl
, 0, sizeof(ipv6h
->flow_lbl
));
579 ipv6h
->payload_len
= htons(sizeof(struct icmp6hdr
)
581 ipv6h
->nexthdr
= NEXTHDR_ICMP
;
582 ipv6h
->hop_limit
= IPV6_DEFAULT_HOPLIMIT
;
583 ipv6_addr_copy(&ipv6h
->daddr
, &old_ipv6h
->saddr
);
584 ipv6_addr_copy(&ipv6h
->saddr
, &old_ipv6h
->daddr
);
587 icmp6h
->icmp6_type
= ICMPV6_PKT_TOOBIG
;
588 icmp6h
->icmp6_code
= 0;
589 icmp6h
->icmp6_cksum
= 0;
590 icmp6h
->icmp6_mtu
= htonl(mtu
);
592 nskb
->csum
= csum_partial((u8
*)icmp6h
, sizeof(struct icmp6hdr
), 0);
593 nskb
->csum
= skb_copy_and_csum_bits(skb
, (u8
*)old_ipv6h
- skb
->data
,
594 payload
, payload_length
,
596 icmp6h
->icmp6_cksum
= csum_ipv6_magic(&ipv6h
->saddr
, &ipv6h
->daddr
,
597 sizeof(struct icmp6hdr
)
599 ipv6h
->nexthdr
, nskb
->csum
);
603 bool tnl_frag_needed(struct vport
*vport
, const struct tnl_mutable_config
*mutable,
604 struct sk_buff
*skb
, unsigned int mtu
, __be64 flow_key
)
606 unsigned int eth_hdr_len
= ETH_HLEN
;
607 unsigned int total_length
= 0, header_length
= 0, payload_length
;
608 struct ethhdr
*eh
, *old_eh
= eth_hdr(skb
);
609 struct sk_buff
*nskb
;
612 if (skb
->protocol
== htons(ETH_P_IP
)) {
613 if (mtu
< IP_MIN_MTU
)
616 if (!ipv4_should_icmp(skb
))
619 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
620 else if (skb
->protocol
== htons(ETH_P_IPV6
)) {
621 if (mtu
< IPV6_MIN_MTU
)
625 * In theory we should do PMTUD on IPv6 multicast messages but
626 * we don't have an address to send from so just fragment.
628 if (ipv6_addr_type(&ipv6_hdr(skb
)->daddr
) & IPV6_ADDR_MULTICAST
)
631 if (!ipv6_should_icmp(skb
))
639 if (old_eh
->h_proto
== htons(ETH_P_8021Q
))
640 eth_hdr_len
= VLAN_ETH_HLEN
;
642 payload_length
= skb
->len
- eth_hdr_len
;
643 if (skb
->protocol
== htons(ETH_P_IP
)) {
644 header_length
= sizeof(struct iphdr
) + sizeof(struct icmphdr
);
645 total_length
= min_t(unsigned int, header_length
+
646 payload_length
, 576);
648 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
650 header_length
= sizeof(struct ipv6hdr
) +
651 sizeof(struct icmp6hdr
);
652 total_length
= min_t(unsigned int, header_length
+
653 payload_length
, IPV6_MIN_MTU
);
657 total_length
= min(total_length
, mutable->mtu
);
658 payload_length
= total_length
- header_length
;
660 nskb
= dev_alloc_skb(NET_IP_ALIGN
+ eth_hdr_len
+ header_length
+
665 skb_reserve(nskb
, NET_IP_ALIGN
);
667 /* Ethernet / VLAN */
668 eh
= (struct ethhdr
*)skb_put(nskb
, eth_hdr_len
);
669 memcpy(eh
->h_dest
, old_eh
->h_source
, ETH_ALEN
);
670 memcpy(eh
->h_source
, mutable->eth_addr
, ETH_ALEN
);
671 nskb
->protocol
= eh
->h_proto
= old_eh
->h_proto
;
672 if (old_eh
->h_proto
== htons(ETH_P_8021Q
)) {
673 struct vlan_ethhdr
*vh
= (struct vlan_ethhdr
*)eh
;
675 vh
->h_vlan_TCI
= vlan_eth_hdr(skb
)->h_vlan_TCI
;
676 vh
->h_vlan_encapsulated_proto
= skb
->protocol
;
678 skb_reset_mac_header(nskb
);
681 if (skb
->protocol
== htons(ETH_P_IP
))
682 ipv4_build_icmp(skb
, nskb
, mtu
, payload_length
);
683 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
685 ipv6_build_icmp(skb
, nskb
, mtu
, payload_length
);
689 * Assume that flow based keys are symmetric with respect to input
690 * and output and use the key that we were going to put on the
691 * outgoing packet for the fake received packet. If the keys are
692 * not symmetric then PMTUD needs to be disabled since we won't have
693 * any way of synthesizing packets.
695 if ((mutable->port_config
.flags
& (TNL_F_IN_KEY_MATCH
| TNL_F_OUT_KEY_ACTION
)) ==
696 (TNL_F_IN_KEY_MATCH
| TNL_F_OUT_KEY_ACTION
))
697 OVS_CB(nskb
)->tun_id
= flow_key
;
699 compute_ip_summed(nskb
, false);
700 vport_receive(vport
, nskb
);
705 static bool check_mtu(struct sk_buff
*skb
,
707 const struct tnl_mutable_config
*mutable,
708 const struct rtable
*rt
, __be16
*frag_offp
)
713 frag_off
= (mutable->port_config
.flags
& TNL_F_PMTUD
) ? htons(IP_DF
) : 0;
715 mtu
= dst_mtu(&rt_dst(rt
))
717 - mutable->tunnel_hlen
718 - (eth_hdr(skb
)->h_proto
== htons(ETH_P_8021Q
) ? VLAN_HLEN
: 0);
722 if (skb
->protocol
== htons(ETH_P_IP
)) {
723 struct iphdr
*old_iph
= ip_hdr(skb
);
725 frag_off
|= old_iph
->frag_off
& htons(IP_DF
);
726 mtu
= max(mtu
, IP_MIN_MTU
);
728 if ((old_iph
->frag_off
& htons(IP_DF
)) &&
729 mtu
< ntohs(old_iph
->tot_len
)) {
730 if (tnl_frag_needed(vport
, mutable, skb
, mtu
, OVS_CB(skb
)->tun_id
))
734 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
735 else if (skb
->protocol
== htons(ETH_P_IPV6
)) {
736 unsigned int packet_length
= skb
->len
- ETH_HLEN
737 - (eth_hdr(skb
)->h_proto
== htons(ETH_P_8021Q
) ? VLAN_HLEN
: 0);
739 mtu
= max(mtu
, IPV6_MIN_MTU
);
741 /* IPv6 requires PMTUD if the packet is above the minimum MTU. */
742 if (packet_length
> IPV6_MIN_MTU
)
743 frag_off
= htons(IP_DF
);
745 if (mtu
< packet_length
) {
746 if (tnl_frag_needed(vport
, mutable, skb
, mtu
, OVS_CB(skb
)->tun_id
))
752 *frag_offp
= frag_off
;
760 static void create_tunnel_header(const struct vport
*vport
,
761 const struct tnl_mutable_config
*mutable,
762 const struct rtable
*rt
, void *header
)
764 struct tnl_vport
*tnl_vport
= tnl_vport_priv(vport
);
765 struct iphdr
*iph
= header
;
768 iph
->ihl
= sizeof(struct iphdr
) >> 2;
769 iph
->frag_off
= htons(IP_DF
);
770 iph
->protocol
= tnl_vport
->tnl_ops
->ipproto
;
771 iph
->tos
= mutable->port_config
.tos
;
772 iph
->daddr
= rt
->rt_dst
;
773 iph
->saddr
= rt
->rt_src
;
774 iph
->ttl
= mutable->port_config
.ttl
;
776 iph
->ttl
= dst_metric(&rt_dst(rt
), RTAX_HOPLIMIT
);
778 tnl_vport
->tnl_ops
->build_header(vport
, mutable, iph
+ 1);
781 static inline void *get_cached_header(const struct tnl_cache
*cache
)
783 return (void *)cache
+ ALIGN(sizeof(struct tnl_cache
), CACHE_DATA_ALIGN
);
786 static inline bool check_cache_valid(const struct tnl_cache
*cache
,
787 const struct tnl_mutable_config
*mutable)
790 #ifdef NEED_CACHE_TIMEOUT
791 time_before(jiffies
, cache
->expiration
) &&
794 atomic_read(&init_net
.ipv4
.rt_genid
) == cache
->rt
->rt_genid
&&
797 rt_dst(cache
->rt
).hh
->hh_lock
.sequence
== cache
->hh_seq
&&
799 mutable->seq
== cache
->mutable_seq
&&
800 (!is_internal_dev(rt_dst(cache
->rt
).dev
) ||
801 (cache
->flow
&& !cache
->flow
->dead
));
804 static int cache_cleaner_cb(struct tbl_node
*tbl_node
, void *aux
)
806 struct tnl_vport
*tnl_vport
= tnl_vport_table_cast(tbl_node
);
807 const struct tnl_mutable_config
*mutable = rcu_dereference(tnl_vport
->mutable);
808 const struct tnl_cache
*cache
= rcu_dereference(tnl_vport
->cache
);
810 if (cache
&& !check_cache_valid(cache
, mutable) &&
811 spin_trylock_bh(&tnl_vport
->cache_lock
)) {
812 assign_cache_rcu(tnl_vport_to_vport(tnl_vport
), NULL
);
813 spin_unlock_bh(&tnl_vport
->cache_lock
);
819 static void cache_cleaner(struct work_struct
*work
)
821 schedule_cache_cleaner();
824 tbl_foreach(rcu_dereference(port_table
), cache_cleaner_cb
, NULL
);
828 static inline void create_eth_hdr(struct tnl_cache
*cache
,
829 const struct rtable
*rt
)
831 void *cache_data
= get_cached_header(cache
);
832 int hh_len
= rt_dst(rt
).hh
->hh_len
;
833 int hh_off
= HH_DATA_ALIGN(rt_dst(rt
).hh
->hh_len
) - hh_len
;
839 hh_seq
= read_seqbegin(&rt_dst(rt
).hh
->hh_lock
);
840 memcpy(cache_data
, (void *)rt_dst(rt
).hh
->hh_data
+ hh_off
, hh_len
);
841 } while (read_seqretry(&rt_dst(rt
).hh
->hh_lock
, hh_seq
));
843 cache
->hh_seq
= hh_seq
;
845 read_lock_bh(&rt_dst(rt
).hh
->hh_lock
);
846 memcpy(cache_data
, (void *)rt_dst(rt
).hh
->hh_data
+ hh_off
, hh_len
);
847 read_unlock_bh(&rt_dst(rt
).hh
->hh_lock
);
851 static struct tnl_cache
*build_cache(struct vport
*vport
,
852 const struct tnl_mutable_config
*mutable,
855 struct tnl_vport
*tnl_vport
= tnl_vport_priv(vport
);
856 struct tnl_cache
*cache
;
860 if (!(mutable->port_config
.flags
& TNL_F_HDR_CACHE
))
864 * If there is no entry in the ARP cache or if this device does not
865 * support hard header caching just fall back to the IP stack.
871 * If lock is contended fall back to directly building the header.
872 * We're not going to help performance by sitting here spinning.
874 if (!spin_trylock_bh(&tnl_vport
->cache_lock
))
877 cache
= tnl_vport
->cache
;
878 if (check_cache_valid(cache
, mutable))
883 cache_len
= rt_dst(rt
).hh
->hh_len
+ mutable->tunnel_hlen
;
885 cache
= kzalloc(ALIGN(sizeof(struct tnl_cache
), CACHE_DATA_ALIGN
) +
886 cache_len
, GFP_ATOMIC
);
890 cache
->len
= cache_len
;
892 create_eth_hdr(cache
, rt
);
893 cache_data
= get_cached_header(cache
) + rt_dst(rt
).hh
->hh_len
;
895 create_tunnel_header(vport
, mutable, rt
, cache_data
);
897 cache
->mutable_seq
= mutable->seq
;
899 #ifdef NEED_CACHE_TIMEOUT
900 cache
->expiration
= jiffies
+ tnl_vport
->cache_exp_interval
;
903 if (is_internal_dev(rt_dst(rt
).dev
)) {
904 struct odp_flow_key flow_key
;
905 struct tbl_node
*flow_node
;
911 vport
= internal_dev_get_vport(rt_dst(rt
).dev
);
915 skb
= alloc_skb(cache
->len
, GFP_ATOMIC
);
919 __skb_put(skb
, cache
->len
);
920 memcpy(skb
->data
, get_cached_header(cache
), cache
->len
);
922 err
= flow_extract(skb
, vport
->port_no
, &flow_key
, &is_frag
);
928 flow_node
= tbl_lookup(rcu_dereference(vport
->dp
->table
),
929 &flow_key
, flow_hash(&flow_key
),
932 struct sw_flow
*flow
= flow_cast(flow_node
);
940 assign_cache_rcu(vport
, cache
);
943 spin_unlock_bh(&tnl_vport
->cache_lock
);
948 static struct rtable
*find_route(struct vport
*vport
,
949 const struct tnl_mutable_config
*mutable,
950 u8 tos
, struct tnl_cache
**cache
)
952 struct tnl_vport
*tnl_vport
= tnl_vport_priv(vport
);
953 struct tnl_cache
*cur_cache
= rcu_dereference(tnl_vport
->cache
);
958 if (likely(tos
== mutable->port_config
.tos
&&
959 check_cache_valid(cur_cache
, mutable))) {
961 return cur_cache
->rt
;
964 struct flowi fl
= { .nl_u
= { .ip4_u
=
965 { .daddr
= mutable->port_config
.daddr
,
966 .saddr
= mutable->port_config
.saddr
,
968 .proto
= tnl_vport
->tnl_ops
->ipproto
};
970 if (unlikely(ip_route_output_key(&init_net
, &rt
, &fl
)))
973 if (likely(tos
== mutable->port_config
.tos
))
974 *cache
= build_cache(vport
, mutable, rt
);
980 static struct sk_buff
*check_headroom(struct sk_buff
*skb
, int headroom
)
982 if (skb_headroom(skb
) < headroom
|| skb_header_cloned(skb
)) {
983 struct sk_buff
*nskb
= skb_realloc_headroom(skb
, headroom
+ 16);
984 if (unlikely(!nskb
)) {
986 return ERR_PTR(-ENOMEM
);
989 set_skb_csum_bits(skb
, nskb
);
992 skb_set_owner_w(nskb
, skb
->sk
);
1001 static inline bool need_linearize(const struct sk_buff
*skb
)
1005 if (unlikely(skb_shinfo(skb
)->frag_list
))
1009 * Generally speaking we should linearize if there are paged frags.
1010 * However, if all of the refcounts are 1 we know nobody else can
1011 * change them from underneath us and we can skip the linearization.
1013 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++)
1014 if (unlikely(page_count(skb_shinfo(skb
)->frags
[0].page
) > 1))
1020 static struct sk_buff
*handle_offloads(struct sk_buff
*skb
,
1021 const struct tnl_mutable_config
*mutable,
1022 const struct rtable
*rt
)
1027 forward_ip_summed(skb
);
1029 err
= vswitch_skb_checksum_setup(skb
);
1033 min_headroom
= LL_RESERVED_SPACE(rt_dst(rt
).dev
) + rt_dst(rt
).header_len
1034 + mutable->tunnel_hlen
;
1036 if (skb_is_gso(skb
)) {
1037 struct sk_buff
*nskb
;
1040 * If we are doing GSO on a pskb it is better to make sure that
1041 * the headroom is correct now. We will only have to copy the
1042 * portion in the linear data area and GSO will preserve
1043 * headroom when it creates the segments. This is particularly
1044 * beneficial on Xen where we get a lot of GSO pskbs.
1045 * Conversely, we avoid copying if it is just to get our own
1046 * writable clone because GSO will do the copy for us.
1048 if (skb_headroom(skb
) < min_headroom
) {
1049 skb
= check_headroom(skb
, min_headroom
);
1056 nskb
= skb_gso_segment(skb
, 0);
1059 err
= PTR_ERR(nskb
);
1065 skb
= check_headroom(skb
, min_headroom
);
1071 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
1073 * Pages aren't locked and could change at any time.
1074 * If this happens after we compute the checksum, the
1075 * checksum will be wrong. We linearize now to avoid
1078 if (unlikely(need_linearize(skb
))) {
1079 err
= __skb_linearize(skb
);
1084 err
= skb_checksum_help(skb
);
1087 } else if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1088 skb
->ip_summed
= CHECKSUM_NONE
;
1096 return ERR_PTR(err
);
1099 static int send_frags(struct sk_buff
*skb
,
1100 const struct tnl_mutable_config
*mutable)
1107 struct sk_buff
*next
= skb
->next
;
1108 int frag_len
= skb
->len
- mutable->tunnel_hlen
;
1111 memset(IPCB(skb
), 0, sizeof(*IPCB(skb
)));
1113 err
= ip_local_out(skb
);
1114 if (likely(net_xmit_eval(err
) == 0))
1115 sent_len
+= frag_len
;
1128 * There's no point in continuing to send fragments once one has been
1129 * dropped so just free the rest. This may help improve the congestion
1130 * that caused the first packet to be dropped.
1132 tnl_free_linked_skbs(skb
);
1136 int tnl_send(struct vport
*vport
, struct sk_buff
*skb
)
1138 struct tnl_vport
*tnl_vport
= tnl_vport_priv(vport
);
1139 const struct tnl_mutable_config
*mutable = rcu_dereference(tnl_vport
->mutable);
1141 enum vport_err_type err
= VPORT_E_TX_ERROR
;
1143 struct dst_entry
*unattached_dst
= NULL
;
1144 struct tnl_cache
*cache
;
1151 /* Validate the protocol headers before we try to use them. */
1152 if (skb
->protocol
== htons(ETH_P_8021Q
)) {
1153 if (unlikely(!pskb_may_pull(skb
, VLAN_ETH_HLEN
)))
1156 skb
->protocol
= vlan_eth_hdr(skb
)->h_vlan_encapsulated_proto
;
1157 skb_set_network_header(skb
, VLAN_ETH_HLEN
);
1160 if (skb
->protocol
== htons(ETH_P_IP
)) {
1161 if (unlikely(!pskb_may_pull(skb
, skb_network_offset(skb
)
1162 + sizeof(struct iphdr
))))
1165 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1166 else if (skb
->protocol
== htons(ETH_P_IPV6
)) {
1167 if (unlikely(!pskb_may_pull(skb
, skb_network_offset(skb
)
1168 + sizeof(struct ipv6hdr
))))
1174 if (skb
->protocol
== htons(ETH_P_IP
))
1175 inner_tos
= ip_hdr(skb
)->tos
;
1176 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1177 else if (skb
->protocol
== htons(ETH_P_IPV6
))
1178 inner_tos
= ipv6_get_dsfield(ipv6_hdr(skb
));
1183 if (mutable->port_config
.flags
& TNL_F_TOS_INHERIT
)
1186 tos
= mutable->port_config
.tos
;
1188 tos
= INET_ECN_encapsulate(tos
, inner_tos
);
1191 rt
= find_route(vport
, mutable, tos
, &cache
);
1194 if (unlikely(!cache
))
1195 unattached_dst
= &rt_dst(rt
);
1203 skb
= handle_offloads(skb
, mutable, rt
);
1208 if (unlikely(!check_mtu(skb
, vport
, mutable, rt
, &frag_off
))) {
1209 err
= VPORT_E_TX_DROPPED
;
1214 * If we are over the MTU, allow the IP stack to handle fragmentation.
1215 * Fragmentation is a slow path anyways.
1217 if (unlikely(skb
->len
+ mutable->tunnel_hlen
> dst_mtu(&rt_dst(rt
)) &&
1219 unattached_dst
= &rt_dst(rt
);
1220 dst_hold(unattached_dst
);
1225 ttl
= mutable->port_config
.ttl
;
1227 ttl
= dst_metric(&rt_dst(rt
), RTAX_HOPLIMIT
);
1229 if (mutable->port_config
.flags
& TNL_F_TTL_INHERIT
) {
1230 if (skb
->protocol
== htons(ETH_P_IP
))
1231 ttl
= ip_hdr(skb
)->ttl
;
1232 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1233 else if (skb
->protocol
== htons(ETH_P_IPV6
))
1234 ttl
= ipv6_hdr(skb
)->hop_limit
;
1240 struct sk_buff
*next_skb
= skb
->next
;
1243 if (likely(cache
)) {
1244 skb_push(skb
, cache
->len
);
1245 memcpy(skb
->data
, get_cached_header(cache
), cache
->len
);
1246 skb_reset_mac_header(skb
);
1247 skb_set_network_header(skb
, rt_dst(rt
).hh
->hh_len
);
1250 skb_push(skb
, mutable->tunnel_hlen
);
1251 create_tunnel_header(vport
, mutable, rt
, skb
->data
);
1252 skb_reset_network_header(skb
);
1255 skb_dst_set(skb
, dst_clone(unattached_dst
));
1257 skb_dst_set(skb
, unattached_dst
);
1258 unattached_dst
= NULL
;
1261 skb_set_transport_header(skb
, skb_network_offset(skb
) + sizeof(struct iphdr
));
1266 iph
->frag_off
= frag_off
;
1267 ip_select_ident(iph
, &rt_dst(rt
), NULL
);
1269 skb
= tnl_vport
->tnl_ops
->update_header(vport
, mutable, &rt_dst(rt
), skb
);
1273 if (likely(cache
)) {
1274 int orig_len
= skb
->len
- cache
->len
;
1275 struct vport
*cache_vport
= internal_dev_get_vport(rt_dst(rt
).dev
);
1277 skb
->protocol
= htons(ETH_P_IP
);
1278 iph
->tot_len
= htons(skb
->len
- skb_network_offset(skb
));
1282 OVS_CB(skb
)->flow
= cache
->flow
;
1283 compute_ip_summed(skb
, true);
1284 vport_receive(cache_vport
, skb
);
1285 sent_len
+= orig_len
;
1289 skb
->dev
= rt_dst(rt
).dev
;
1290 err
= dev_queue_xmit(skb
);
1292 if (likely(net_xmit_eval(err
) == 0))
1293 sent_len
+= orig_len
;
1296 sent_len
+= send_frags(skb
, mutable);
1302 if (unlikely(sent_len
== 0))
1303 vport_record_error(vport
, VPORT_E_TX_DROPPED
);
1308 tnl_free_linked_skbs(skb
);
1310 dst_release(unattached_dst
);
1311 vport_record_error(vport
, err
);
1316 static int set_config(const void *config
, const struct tnl_ops
*tnl_ops
,
1317 const struct vport
*cur_vport
,
1318 struct tnl_mutable_config
*mutable)
1320 const struct vport
*old_vport
;
1321 const struct tnl_mutable_config
*old_mutable
;
1323 mutable->port_config
= *(struct tnl_port_config
*)config
;
1325 if (mutable->port_config
.daddr
== 0)
1328 if (mutable->port_config
.tos
!= RT_TOS(mutable->port_config
.tos
))
1331 mutable->tunnel_hlen
= tnl_ops
->hdr_len(&mutable->port_config
);
1332 if (mutable->tunnel_hlen
< 0)
1333 return mutable->tunnel_hlen
;
1335 mutable->tunnel_hlen
+= sizeof(struct iphdr
);
1337 mutable->tunnel_type
= tnl_ops
->tunnel_type
;
1338 if (mutable->port_config
.flags
& TNL_F_IN_KEY_MATCH
) {
1339 mutable->tunnel_type
|= TNL_T_KEY_MATCH
;
1340 mutable->port_config
.in_key
= 0;
1342 mutable->tunnel_type
|= TNL_T_KEY_EXACT
;
1344 old_vport
= tnl_find_port(mutable->port_config
.saddr
,
1345 mutable->port_config
.daddr
,
1346 mutable->port_config
.in_key
,
1347 mutable->tunnel_type
,
1350 if (old_vport
&& old_vport
!= cur_vport
)
1353 if (mutable->port_config
.flags
& TNL_F_OUT_KEY_ACTION
)
1354 mutable->port_config
.out_key
= 0;
1359 struct vport
*tnl_create(const struct vport_parms
*parms
,
1360 const struct vport_ops
*vport_ops
,
1361 const struct tnl_ops
*tnl_ops
)
1363 struct vport
*vport
;
1364 struct tnl_vport
*tnl_vport
;
1365 int initial_frag_id
;
1368 vport
= vport_alloc(sizeof(struct tnl_vport
), vport_ops
, parms
);
1369 if (IS_ERR(vport
)) {
1370 err
= PTR_ERR(vport
);
1374 tnl_vport
= tnl_vport_priv(vport
);
1376 strcpy(tnl_vport
->name
, parms
->name
);
1377 tnl_vport
->tnl_ops
= tnl_ops
;
1379 tnl_vport
->mutable = kzalloc(sizeof(struct tnl_mutable_config
), GFP_KERNEL
);
1380 if (!tnl_vport
->mutable) {
1382 goto error_free_vport
;
1385 vport_gen_rand_ether_addr(tnl_vport
->mutable->eth_addr
);
1386 tnl_vport
->mutable->mtu
= ETH_DATA_LEN
;
1388 get_random_bytes(&initial_frag_id
, sizeof(int));
1389 atomic_set(&tnl_vport
->frag_id
, initial_frag_id
);
1391 err
= set_config(parms
->config
, tnl_ops
, NULL
, tnl_vport
->mutable);
1393 goto error_free_mutable
;
1395 spin_lock_init(&tnl_vport
->cache_lock
);
1397 #ifdef NEED_CACHE_TIMEOUT
1398 tnl_vport
->cache_exp_interval
= MAX_CACHE_EXP
-
1399 (net_random() % (MAX_CACHE_EXP
/ 2));
1402 err
= add_port(vport
);
1404 goto error_free_mutable
;
1409 kfree(tnl_vport
->mutable);
1413 return ERR_PTR(err
);
1416 int tnl_modify(struct vport
*vport
, struct odp_port
*port
)
1418 struct tnl_vport
*tnl_vport
= tnl_vport_priv(vport
);
1419 struct tnl_mutable_config
*mutable;
1422 mutable = kmemdup(tnl_vport
->mutable, sizeof(struct tnl_mutable_config
), GFP_KERNEL
);
1428 err
= set_config(port
->config
, tnl_vport
->tnl_ops
, vport
, mutable);
1434 err
= move_port(vport
, mutable);
1446 static void free_port_rcu(struct rcu_head
*rcu
)
1448 struct tnl_vport
*tnl_vport
= container_of(rcu
, struct tnl_vport
, rcu
);
1450 spin_lock_bh(&tnl_vport
->cache_lock
);
1451 free_cache(tnl_vport
->cache
);
1452 spin_unlock_bh(&tnl_vport
->cache_lock
);
1454 kfree(tnl_vport
->mutable);
1455 vport_free(tnl_vport_to_vport(tnl_vport
));
1458 int tnl_destroy(struct vport
*vport
)
1460 struct tnl_vport
*tnl_vport
= tnl_vport_priv(vport
);
1461 const struct tnl_mutable_config
*old_mutable
;
1463 if (vport
== tnl_find_port(tnl_vport
->mutable->port_config
.saddr
,
1464 tnl_vport
->mutable->port_config
.daddr
,
1465 tnl_vport
->mutable->port_config
.in_key
,
1466 tnl_vport
->mutable->tunnel_type
,
1470 call_rcu(&tnl_vport
->rcu
, free_port_rcu
);
1475 int tnl_set_mtu(struct vport
*vport
, int mtu
)
1477 struct tnl_vport
*tnl_vport
= tnl_vport_priv(vport
);
1478 struct tnl_mutable_config
*mutable;
1480 mutable = kmemdup(tnl_vport
->mutable, sizeof(struct tnl_mutable_config
), GFP_KERNEL
);
1485 assign_config_rcu(vport
, mutable);
1490 int tnl_set_addr(struct vport
*vport
, const unsigned char *addr
)
1492 struct tnl_vport
*tnl_vport
= tnl_vport_priv(vport
);
1493 struct tnl_mutable_config
*mutable;
1495 mutable = kmemdup(tnl_vport
->mutable, sizeof(struct tnl_mutable_config
), GFP_KERNEL
);
1499 memcpy(mutable->eth_addr
, addr
, ETH_ALEN
);
1500 assign_config_rcu(vport
, mutable);
1505 const char *tnl_get_name(const struct vport
*vport
)
1507 const struct tnl_vport
*tnl_vport
= tnl_vport_priv(vport
);
1508 return tnl_vport
->name
;
1511 const unsigned char *tnl_get_addr(const struct vport
*vport
)
1513 const struct tnl_vport
*tnl_vport
= tnl_vport_priv(vport
);
1514 return rcu_dereference_rtnl(tnl_vport
->mutable)->eth_addr
;
1517 int tnl_get_mtu(const struct vport
*vport
)
1519 const struct tnl_vport
*tnl_vport
= tnl_vport_priv(vport
);
1520 return rcu_dereference_rtnl(tnl_vport
->mutable)->mtu
;
1523 void tnl_free_linked_skbs(struct sk_buff
*skb
)
1529 struct sk_buff
*next
= skb
->next
;