Skip to content

Commit 7d442fa

Browse files
Tom Herbertdavem330
Tom Herbert
authored and
davem330
committed
ipv4: Cache dst in tunnels
Avoid doing a route lookup on every packet being tunneled. In ip_tunnel.c cache the route returned from ip_route_output if the tunnel is "connected" so that all the rouitng parameters are taken from tunnel parms for a packet. Specifically, not NBMA tunnel and tos is from tunnel parms (not inner packet). Signed-off-by: Tom Herbert <therbert@google.com> Signed-off-by: David S. Miller <davem@davemloft.net>
1 parent f916ec9 commit 7d442fa

File tree

2 files changed

+92
-24
lines changed

2 files changed

+92
-24
lines changed

‎include/net/ip_tunnels.h

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -54,6 +54,9 @@ struct ip_tunnel {
5454
int hlen; /* Precalculated header length */
5555
int mlink;
5656

57+
struct dst_entry __rcu *dst_cache;
58+
spinlock_t dst_lock;
59+
5760
struct ip_tunnel_parm parms;
5861

5962
/* for SIT */

‎net/ipv4/ip_tunnel.c

Lines changed: 89 additions & 24 deletions
Original file line numberDiff line numberDiff line change
@@ -68,6 +68,54 @@ static unsigned int ip_tunnel_hash(struct ip_tunnel_net *itn,
6868
IP_TNL_HASH_BITS);
6969
}
7070

71+
static inline void __tunnel_dst_set(struct ip_tunnel *t, struct dst_entry *dst)
72+
{
73+
struct dst_entry *old_dst;
74+
75+
if (dst && (dst->flags & DST_NOCACHE))
76+
dst = NULL;
77+
78+
spin_lock_bh(&t->dst_lock);
79+
old_dst = rcu_dereference_raw(t->dst_cache);
80+
rcu_assign_pointer(t->dst_cache, dst);
81+
dst_release(old_dst);
82+
spin_unlock_bh(&t->dst_lock);
83+
}
84+
85+
static inline void tunnel_dst_set(struct ip_tunnel *t, struct dst_entry *dst)
86+
{
87+
__tunnel_dst_set(t, dst);
88+
}
89+
90+
static inline void tunnel_dst_reset(struct ip_tunnel *t)
91+
{
92+
tunnel_dst_set(t, NULL);
93+
}
94+
95+
static inline struct dst_entry *tunnel_dst_get(struct ip_tunnel *t)
96+
{
97+
struct dst_entry *dst;
98+
99+
rcu_read_lock();
100+
dst = rcu_dereference(t->dst_cache);
101+
if (dst)
102+
dst_hold(dst);
103+
rcu_read_unlock();
104+
return dst;
105+
}
106+
107+
struct dst_entry *tunnel_dst_check(struct ip_tunnel *t, u32 cookie)
108+
{
109+
struct dst_entry *dst = tunnel_dst_get(t);
110+
111+
if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
112+
tunnel_dst_reset(t);
113+
return NULL;
114+
}
115+
116+
return dst;
117+
}
118+
71119
/* Often modified stats are per cpu, other are shared (netdev->stats) */
72120
struct rtnl_link_stats64 *ip_tunnel_get_stats64(struct net_device *dev,
73121
struct rtnl_link_stats64 *tot)
@@ -318,11 +366,10 @@ static struct net_device *__ip_tunnel_create(struct net *net,
318366
return ERR_PTR(err);
319367
}
320368

321-
static inline struct rtable *ip_route_output_tunnel(struct net *net,
322-
struct flowi4 *fl4,
323-
int proto,
324-
__be32 daddr, __be32 saddr,
325-
__be32 key, __u8 tos, int oif)
369+
static inline void init_tunnel_flow(struct flowi4 *fl4,
370+
int proto,
371+
__be32 daddr, __be32 saddr,
372+
__be32 key, __u8 tos, int oif)
326373
{
327374
memset(fl4, 0, sizeof(*fl4));
328375
fl4->flowi4_oif = oif;
@@ -331,7 +378,6 @@ static inline struct rtable *ip_route_output_tunnel(struct net *net,
331378
fl4->flowi4_tos = tos;
332379
fl4->flowi4_proto = proto;
333380
fl4->fl4_gre_key = key;
334-
return ip_route_output_key(net, fl4);
335381
}
336382

337383
static int ip_tunnel_bind_dev(struct net_device *dev)
@@ -350,14 +396,14 @@ static int ip_tunnel_bind_dev(struct net_device *dev)
350396
struct flowi4 fl4;
351397
struct rtable *rt;
352398

353-
rt = ip_route_output_tunnel(tunnel->net, &fl4,
354-
tunnel->parms.iph.protocol,
355-
iph->daddr, iph->saddr,
356-
tunnel->parms.o_key,
357-
RT_TOS(iph->tos),
358-
tunnel->parms.link);
399+
init_tunnel_flow(&fl4, iph->protocol, iph->daddr,
400+
iph->saddr, tunnel->parms.o_key,
401+
RT_TOS(iph->tos), tunnel->parms.link);
402+
rt = ip_route_output_key(tunnel->net, &fl4);
403+
359404
if (!IS_ERR(rt)) {
360405
tdev = rt->dst.dev;
406+
tunnel_dst_set(tunnel, dst_clone(&rt->dst));
361407
ip_rt_put(rt);
362408
}
363409
if (dev->type != ARPHRD_ETHER)
@@ -528,10 +574,11 @@ void ip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev,
528574
struct flowi4 fl4;
529575
u8 tos, ttl;
530576
__be16 df;
531-
struct rtable *rt; /* Route to the other host */
577+
struct rtable *rt = NULL; /* Route to the other host */
532578
unsigned int max_headroom; /* The extra header space needed */
533579
__be32 dst;
534580
int err;
581+
bool connected = true;
535582

536583
inner_iph = (const struct iphdr *)skb_inner_network_header(skb);
537584

@@ -581,27 +628,39 @@ void ip_tunnel_xmit(struct sk_buff *skb, struct net_device *dev,
581628
#endif
582629
else
583630
goto tx_error;
631+
632+
connected = false;
584633
}
585634

586635
tos = tnl_params->tos;
587636
if (tos & 0x1) {
588637
tos &= ~0x1;
589-
if (skb->protocol == htons(ETH_P_IP))
638+
if (skb->protocol == htons(ETH_P_IP)) {
590639
tos = inner_iph->tos;
591-
else if (skb->protocol == htons(ETH_P_IPV6))
640+
connected = false;
641+
} else if (skb->protocol == htons(ETH_P_IPV6)) {
592642
tos = ipv6_get_dsfield((const struct ipv6hdr *)inner_iph);
643+
connected = false;
644+
}
593645
}
594646

595-
rt = ip_route_output_tunnel(tunnel->net, &fl4,
596-
protocol,
597-
dst, tnl_params->saddr,
598-
tunnel->parms.o_key,
599-
RT_TOS(tos),
600-
tunnel->parms.link);
601-
if (IS_ERR(rt)) {
602-
dev->stats.tx_carrier_errors++;
603-
goto tx_error;
647+
init_tunnel_flow(&fl4, protocol, dst, tnl_params->saddr,
648+
tunnel->parms.o_key, RT_TOS(tos), tunnel->parms.link);
649+
650+
if (connected)
651+
rt = (struct rtable *)tunnel_dst_check(tunnel, 0);
652+
653+
if (!rt) {
654+
rt = ip_route_output_key(tunnel->net, &fl4);
655+
656+
if (IS_ERR(rt)) {
657+
dev->stats.tx_carrier_errors++;
658+
goto tx_error;
659+
}
660+
if (connected)
661+
tunnel_dst_set(tunnel, dst_clone(&rt->dst));
604662
}
663+
605664
if (rt->dst.dev == dev) {
606665
ip_rt_put(rt);
607666
dev->stats.collisions++;
@@ -696,6 +755,7 @@ static void ip_tunnel_update(struct ip_tunnel_net *itn,
696755
if (set_mtu)
697756
dev->mtu = mtu;
698757
}
758+
tunnel_dst_reset(t);
699759
netdev_state_change(dev);
700760
}
701761

@@ -1001,6 +1061,9 @@ int ip_tunnel_init(struct net_device *dev)
10011061
iph->version = 4;
10021062
iph->ihl = 5;
10031063

1064+
tunnel->dst_cache = NULL;
1065+
spin_lock_init(&tunnel->dst_lock);
1066+
10041067
return 0;
10051068
}
10061069
EXPORT_SYMBOL_GPL(ip_tunnel_init);
@@ -1015,6 +1078,8 @@ void ip_tunnel_uninit(struct net_device *dev)
10151078
/* fb_tunnel_dev will be unregisted in net-exit call. */
10161079
if (itn->fb_tunnel_dev != dev)
10171080
ip_tunnel_del(netdev_priv(dev));
1081+
1082+
tunnel_dst_reset(tunnel);
10181083
}
10191084
EXPORT_SYMBOL_GPL(ip_tunnel_uninit);
10201085

0 commit comments

Comments
 (0)