ipv4: do not use per netns icmp sockets

Back in linux-2.6.25 (commit 4a6ad7a141cb "[NETNS]: Make icmp_sk per namespace."),
we added private per-cpu/per-netns ipv4 icmp sockets.

This adds memory and cpu costs, which do not seem needed.
Now typical servers have 256 or more cores, this adds considerable
tax to netns users.

icmp sockets are used from BH context, are not receiving packets,
and do not store any persistent state but the 'struct net' pointer.

icmp_xmit_lock() already makes sure to lock the chosen per-cpu
socket.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 31 insertions, 60 deletions

diff --git a/net/ipv4/icmp.c b/net/ipv4/icmp.c
index b7e277d8a84d..72a375c7f417 100644
--- a/net/ipv4/icmp.c
+++ b/net/ipv4/icmp.c
@@ -192,24 +192,14 @@ struct icmp_control {
 
 static const struct icmp_control icmp_pointers[NR_ICMP_TYPES+1];
 
-/*
- *	The ICMP socket(s). This is the most convenient way to flow control
- *	our ICMP output as well as maintain a clean interface throughout
- *	all layers. All Socketless IP sends will soon be gone.
- *
- *	On SMP we have one ICMP socket per-cpu.
- */
-static struct sock *icmp_sk(struct net *net)
-{
-	return this_cpu_read(*net->ipv4.icmp_sk);
-}
+static DEFINE_PER_CPU(struct sock *, ipv4_icmp_sk);
 
 /* Called with BH disabled */
 static inline struct sock *icmp_xmit_lock(struct net *net)
 {
 	struct sock *sk;
 
-	sk = icmp_sk(net);
+	sk = this_cpu_read(ipv4_icmp_sk);
 
 	if (unlikely(!spin_trylock(&sk->sk_lock.slock))) {
 		/* This can happen if the output path signals a
@@ -217,11 +207,13 @@ static inline struct sock *icmp_xmit_lock(struct net *net)
 		 */
 		return NULL;
 	}
+	sock_net_set(sk, net);
 	return sk;
 }
 
 static inline void icmp_xmit_unlock(struct sock *sk)
 {
+	sock_net_set(sk, &init_net);
 	spin_unlock(&sk->sk_lock.slock);
 }
 
@@ -363,14 +355,13 @@ static int icmp_glue_bits(void *from, char *to, int offset, int len, int odd,
 	return 0;
 }
 
-static void icmp_push_reply(struct icmp_bxm *icmp_param,
+static void icmp_push_reply(struct sock *sk,
+			    struct icmp_bxm *icmp_param,
 			    struct flowi4 *fl4,
 			    struct ipcm_cookie *ipc, struct rtable **rt)
 {
-	struct sock *sk;
 	struct sk_buff *skb;
 
-	sk = icmp_sk(dev_net((*rt)->dst.dev));
 	if (ip_append_data(sk, fl4, icmp_glue_bits, icmp_param,
 			   icmp_param->data_len+icmp_param->head_len,
 			   icmp_param->head_len,
@@ -452,7 +443,7 @@ static void icmp_reply(struct icmp_bxm *icmp_param, struct sk_buff *skb)
 	if (IS_ERR(rt))
 		goto out_unlock;
 	if (icmpv4_xrlim_allow(net, rt, &fl4, type, code))
-		icmp_push_reply(icmp_param, &fl4, &ipc, &rt);
+		icmp_push_reply(sk, icmp_param, &fl4, &ipc, &rt);
 	ip_rt_put(rt);
 out_unlock:
 	icmp_xmit_unlock(sk);
@@ -766,7 +757,7 @@ void __icmp_send(struct sk_buff *skb_in, int type, int code, __be32 info,
 	if (!fl4.saddr)
 		fl4.saddr = htonl(INADDR_DUMMY);
 
-	icmp_push_reply(&icmp_param, &fl4, &ipc, &rt);
+	icmp_push_reply(sk, &icmp_param, &fl4, &ipc, &rt);
 ende:
 	ip_rt_put(rt);
 out_unlock:
@@ -1434,46 +1425,8 @@ static const struct icmp_control icmp_pointers[NR_ICMP_TYPES + 1] = {
 	},
 };
 
-static void __net_exit icmp_sk_exit(struct net *net)
-{
-	int i;
-
-	for_each_possible_cpu(i)
-		inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.icmp_sk, i));
-	free_percpu(net->ipv4.icmp_sk);
-	net->ipv4.icmp_sk = NULL;
-}
-
 static int __net_init icmp_sk_init(struct net *net)
 {
-	int i, err;
-
-	net->ipv4.icmp_sk = alloc_percpu(struct sock *);
-	if (!net->ipv4.icmp_sk)
-		return -ENOMEM;
-
-	for_each_possible_cpu(i) {
-		struct sock *sk;
-
-		err = inet_ctl_sock_create(&sk, PF_INET,
-					   SOCK_RAW, IPPROTO_ICMP, net);
-		if (err < 0)
-			goto fail;
-
-		*per_cpu_ptr(net->ipv4.icmp_sk, i) = sk;
-
-		/* Enough space for 2 64K ICMP packets, including
-		 * sk_buff/skb_shared_info struct overhead.
-		 */
-		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
-
-		/*
-		 * Speedup sock_wfree()
-		 */
-		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
-		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
-	}
-
 	/* Control parameters for ECHO replies. */
 	net->ipv4.sysctl_icmp_echo_ignore_all = 0;
 	net->ipv4.sysctl_icmp_echo_enable_probe = 0;
@@ -1499,18 +1452,36 @@ static int __net_init icmp_sk_init(struct net *net)
 	net->ipv4.sysctl_icmp_errors_use_inbound_ifaddr = 0;
 
 	return 0;
-
-fail:
-	icmp_sk_exit(net);
-	return err;
 }
 
 static struct pernet_operations __net_initdata icmp_sk_ops = {
        .init = icmp_sk_init,
-       .exit = icmp_sk_exit,
 };
 
 int __init icmp_init(void)
 {
+	int err, i;
+
+	for_each_possible_cpu(i) {
+		struct sock *sk;
+
+		err = inet_ctl_sock_create(&sk, PF_INET,
+					   SOCK_RAW, IPPROTO_ICMP, &init_net);
+		if (err < 0)
+			return err;
+
+		per_cpu(ipv4_icmp_sk, i) = sk;
+
+		/* Enough space for 2 64K ICMP packets, including
+		 * sk_buff/skb_shared_info struct overhead.
+		 */
+		sk->sk_sndbuf =	2 * SKB_TRUESIZE(64 * 1024);
+
+		/*
+		 * Speedup sock_wfree()
+		 */
+		sock_set_flag(sk, SOCK_USE_WRITE_QUEUE);
+		inet_sk(sk)->pmtudisc = IP_PMTUDISC_DONT;
+	}
 	return register_pernet_subsys(&icmp_sk_ops);
 }