net: sk_add_backlog() take rmem_alloc into account

Current socket backlog limit is not enough to really stop DDOS attacks,
because user thread spend many time to process a full backlog each
round, and user might crazy spin on socket lock.

We should add backlog size and receive_queue size (aka rmem_alloc) to
pace writers, and let user run without being slow down too much.

Introduce a sk_rcvqueues_full() helper, to avoid taking socket lock in
stress situations.

Under huge stress from a multiqueue/RPS enabled NIC, a single flow udp
receiver can now process ~200.000 pps (instead of ~100 pps before the
patch) on a 8 core machine.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

5 files changed, 27 insertions, 6 deletions

diff --git a/include/net/sock.h b/include/net/sock.h
index 0782228..cf12b1e 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -256,7 +256,6 @@ struct sock {
 		struct sk_buff *head;
 		struct sk_buff *tail;
 		int len;
-		int limit;
 	} sk_backlog;
 	wait_queue_head_t	*sk_sleep;
 	struct dst_entry	*sk_dst_cache;
@@ -608,10 +607,20 @@ static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
 	skb->next = NULL;
 }
 
+/*
+ * Take into account size of receive queue and backlog queue
+ */
+static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb)
+{
+	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
+
+	return qsize + skb->truesize > sk->sk_rcvbuf;
+}
+
 /* The per-socket spinlock must be held here. */
 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb)
 {
-	if (sk->sk_backlog.len >= max(sk->sk_backlog.limit, sk->sk_rcvbuf << 1))
+	if (sk_rcvqueues_full(sk, skb))
 		return -ENOBUFS;
 
 	__sk_add_backlog(sk, skb);
diff --git a/net/core/sock.c b/net/core/sock.c
index 58ebd14..5104175 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -327,6 +327,10 @@ int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
 
 	skb->dev = NULL;
 
+	if (sk_rcvqueues_full(sk, skb)) {
+		atomic_inc(&sk->sk_drops);
+		goto discard_and_relse;
+	}
 	if (nested)
 		bh_lock_sock_nested(sk);
 	else
@@ -1885,7 +1889,6 @@ void sock_init_data(struct socket *sock, struct sock *sk)
 	sk->sk_allocation	=	GFP_KERNEL;
 	sk->sk_rcvbuf		=	sysctl_rmem_default;
 	sk->sk_sndbuf		=	sysctl_wmem_default;
-	sk->sk_backlog.limit	=	sk->sk_rcvbuf << 1;
 	sk->sk_state		=	TCP_CLOSE;
 	sk_set_socket(sk, sock);
 
diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index fa3d287..63eb56b 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -1372,6 +1372,10 @@ int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 			goto drop;
 	}
 
+
+	if (sk_rcvqueues_full(sk, skb))
+		goto drop;
+
 	rc = 0;
 
 	bh_lock_sock(sk);
diff --git a/net/ipv6/udp.c b/net/ipv6/udp.c
index 2850e35..3ead20a 100644
--- a/net/ipv6/udp.c
+++ b/net/ipv6/udp.c
@@ -584,6 +584,10 @@ static void flush_stack(struct sock **stack, unsigned int count,
 
 		sk = stack[i];
 		if (skb1) {
+			if (sk_rcvqueues_full(sk, skb)) {
+				kfree_skb(skb1);
+				goto drop;
+			}
 			bh_lock_sock(sk);
 			if (!sock_owned_by_user(sk))
 				udpv6_queue_rcv_skb(sk, skb1);
@@ -759,6 +763,10 @@ int __udp6_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
 
 	/* deliver */
 
+	if (sk_rcvqueues_full(sk, skb)) {
+		sock_put(sk);
+		goto discard;
+	}
 	bh_lock_sock(sk);
 	if (!sock_owned_by_user(sk))
 		udpv6_queue_rcv_skb(sk, skb);
diff --git a/net/sctp/socket.c b/net/sctp/socket.c
index f34adcc..13d8229 100644
--- a/net/sctp/socket.c
+++ b/net/sctp/socket.c
@@ -3721,9 +3721,6 @@ SCTP_STATIC int sctp_init_sock(struct sock *sk)
 	SCTP_DBG_OBJCNT_INC(sock);
 	percpu_counter_inc(&sctp_sockets_allocated);
 
-	/* Set socket backlog limit. */
-	sk->sk_backlog.limit = sysctl_sctp_rmem[1];
-
 	local_bh_disable();
 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 	local_bh_enable();