ipv4: tcp: get rid of ugly unicast_sock

[ Upstream commit bdbbb852 ]

In commit be9f4a44 ("ipv4: tcp: remove per net tcp_sock")
I tried to address contention on a socket lock, but the solution
I chose was horrible :

commit 3a7c384f ("ipv4: tcp: unicast_sock should not land outside
of TCP stack") addressed a selinux regression.

commit 0980e56e ("ipv4: tcp: set unicast_sock uc_ttl to -1")
took care of another regression.

commit b5ec8eea ("ipv4: fix ip_send_skb()") fixed another regression.

commit 811230cd ("tcp: ipv4: initialize unicast_sock sk_pacing_rate")
was another shot in the dark.

Really, just use a proper socket per cpu, and remove the skb_orphan()
call, to re-enable flow control.

This solves a serious problem with FQ packet scheduler when used in
hostile environments, as we do not want to allocate a flow structure
for every RST packet sent in response to a spoofed packet.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

include/net/ip.h

@@ -163,7 +163,7 @@ static inline __u8 ip_reply_arg_flowi_flags(const struct ip_reply_arg *arg)
 	return (arg->flags & IP_REPLY_ARG_NOSRCCHECK) ? FLOWI_FLAG_ANYSRC : 0;
 }
 
-void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,
+void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb, __be32 daddr,
 			   __be32 saddr, const struct ip_reply_arg *arg,
 			   unsigned int len);
 

include/net/netns/ipv4.h

@@ -43,6 +43,7 @@ struct netns_ipv4 {
 	struct inet_peer_base	*peers;
 	struct tcpm_hash_bucket	*tcp_metrics_hash;
 	unsigned int		tcp_metrics_hash_log;
+	struct sock  * __percpu	*tcp_sk;
 	struct netns_frags	frags;
 #ifdef CONFIG_NETFILTER
 	struct xt_table		*iptable_filter;

net/ipv4/ip_output.c

@@ -1454,24 +1454,8 @@ static int ip_reply_glue_bits(void *dptr, char *to, int offset,
 /*
  *	Generic function to send a packet as reply to another packet.
  *	Used to send some TCP resets/acks so far.
- *
- *	Use a fake percpu inet socket to avoid false sharing and contention.
  */
-static DEFINE_PER_CPU(struct inet_sock, unicast_sock) = {
-	.sk = {
-		.__sk_common = {
-			.skc_refcnt = ATOMIC_INIT(1),
-		},
-		.sk_wmem_alloc	= ATOMIC_INIT(1),
-		.sk_allocation	= GFP_ATOMIC,
-		.sk_flags	= (1UL << SOCK_USE_WRITE_QUEUE),
-		.sk_pacing_rate = ~0U,
-	},
-	.pmtudisc	= IP_PMTUDISC_WANT,
-	.uc_ttl		= -1,
-};
-
-void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,
+void ip_send_unicast_reply(struct sock *sk, struct sk_buff *skb, __be32 daddr,
 			   __be32 saddr, const struct ip_reply_arg *arg,
 			   unsigned int len)
 {
@@ -1479,9 +1463,8 @@ void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,
 	struct ipcm_cookie ipc;
 	struct flowi4 fl4;
 	struct rtable *rt = skb_rtable(skb);
+	struct net *net = sock_net(sk);
 	struct sk_buff *nskb;
-	struct sock *sk;
-	struct inet_sock *inet;
 	int err;
 
 	if (ip_options_echo(&replyopts.opt.opt, skb))
@@ -1509,15 +1492,11 @@ void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,
 	if (IS_ERR(rt))
 		return;
 
-	inet = &get_cpu_var(unicast_sock);
+	inet_sk(sk)->tos = arg->tos;
 
-	inet->tos = arg->tos;
-	sk = &inet->sk;
 	sk->sk_priority = skb->priority;
 	sk->sk_protocol = ip_hdr(skb)->protocol;
 	sk->sk_bound_dev_if = arg->bound_dev_if;
-	sock_net_set(sk, net);
-	__skb_queue_head_init(&sk->sk_write_queue);
 	sk->sk_sndbuf = sysctl_wmem_default;
 	err = ip_append_data(sk, &fl4, ip_reply_glue_bits, arg->iov->iov_base,
 			     len, 0, &ipc, &rt, MSG_DONTWAIT);
@@ -1533,13 +1512,10 @@ void ip_send_unicast_reply(struct net *net, struct sk_buff *skb, __be32 daddr,
 			  arg->csumoffset) = csum_fold(csum_add(nskb->csum,
 								arg->csum));
 		nskb->ip_summed = CHECKSUM_NONE;
-		skb_orphan(nskb);
 		skb_set_queue_mapping(nskb, skb_get_queue_mapping(skb));
 		ip_push_pending_frames(sk, &fl4);
 	}
 out:
-	put_cpu_var(unicast_sock);
-
 	ip_rt_put(rt);
 }
 

net/ipv4/tcp_ipv4.c

@@ -707,7 +707,8 @@ static void tcp_v4_send_reset(struct sock *sk, struct sk_buff *skb)
 
 	net = dev_net(skb_dst(skb)->dev);
 	arg.tos = ip_hdr(skb)->tos;
-	ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr,
+	ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
+			      skb, ip_hdr(skb)->saddr,
 			      ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len);
 
 	TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
@@ -790,7 +791,8 @@ static void tcp_v4_send_ack(struct sk_buff *skb, u32 seq, u32 ack,
 	if (oif)
 		arg.bound_dev_if = oif;
 	arg.tos = tos;
-	ip_send_unicast_reply(net, skb, ip_hdr(skb)->saddr,
+	ip_send_unicast_reply(*this_cpu_ptr(net->ipv4.tcp_sk),
+			      skb, ip_hdr(skb)->saddr,
 			      ip_hdr(skb)->daddr, &arg, arg.iov[0].iov_len);
 
 	TCP_INC_STATS_BH(net, TCP_MIB_OUTSEGS);
@@ -2898,14 +2900,39 @@ struct proto tcp_prot = {
 };
 EXPORT_SYMBOL(tcp_prot);
 
+static void __net_exit tcp_sk_exit(struct net *net)
+{
+	int cpu;
+
+	for_each_possible_cpu(cpu)
+		inet_ctl_sock_destroy(*per_cpu_ptr(net->ipv4.tcp_sk, cpu));
+	free_percpu(net->ipv4.tcp_sk);
+}
+
 static int __net_init tcp_sk_init(struct net *net)
 {
+	int res, cpu;
+
+	net->ipv4.tcp_sk = alloc_percpu(struct sock *);
+	if (!net->ipv4.tcp_sk)
+		return -ENOMEM;
+
+	for_each_possible_cpu(cpu) {
+		struct sock *sk;
+
+		res = inet_ctl_sock_create(&sk, PF_INET, SOCK_RAW,
+					   IPPROTO_TCP, net);
+		if (res)
+			goto fail;
+		*per_cpu_ptr(net->ipv4.tcp_sk, cpu) = sk;
+	}
 	net->ipv4.sysctl_tcp_ecn = 2;
 	return 0;
-}
 
-static void __net_exit tcp_sk_exit(struct net *net)
-{
+fail:
+	tcp_sk_exit(net);
+
+	return res;
 }
 
 static void __net_exit tcp_sk_exit_batch(struct list_head *net_exit_list)