queue/net-packet-make-tp_drops-atomic.patch

From 8e8e2951e3095732d7e780c241f61ea130955a57 Mon Sep 17 00:00:00 2001
From: Eric Dumazet <edumazet@google.com>
Date: Wed, 12 Jun 2019 09:52:30 -0700
Subject: [PATCH] net/packet: make tp_drops atomic

commit 8e8e2951e3095732d7e780c241f61ea130955a57 upstream.

Under DDOS, we want to be able to increment tp_drops without
touching the spinlock. This will help readers to drain
the receive queue slightly faster :/

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

diff --git a/net/packet/af_packet.c b/net/packet/af_packet.c
index a0564855ed9d..2d499679811a 100644
--- a/net/packet/af_packet.c
+++ b/net/packet/af_packet.c
@@ -758,7 +758,7 @@ static void prb_close_block(struct tpacket_kbdq_core *pkc1,
 	struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
 	struct sock *sk = &po->sk;
 
-	if (po->stats.stats3.tp_drops)
+	if (atomic_read(&po->tp_drops))
 		status |= TP_STATUS_LOSING;
 
 	last_pkt = (struct tpacket3_hdr *)pkc1->prev;
@@ -2128,10 +2128,8 @@ static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
 
 drop_n_acct:
 	is_drop_n_account = true;
-	spin_lock(&sk->sk_receive_queue.lock);
-	po->stats.stats1.tp_drops++;
+	atomic_inc(&po->tp_drops);
 	atomic_inc(&sk->sk_drops);
-	spin_unlock(&sk->sk_receive_queue.lock);
 
 drop_n_restore:
 	if (skb_head != skb->data && skb_shared(skb)) {
@@ -2265,7 +2263,7 @@ static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 	 * Anyways, moving it for V1/V2 only as V3 doesn't need this
 	 * at packet level.
 	 */
-		if (po->stats.stats1.tp_drops)
+		if (atomic_read(&po->tp_drops))
 			status |= TP_STATUS_LOSING;
 	}
 
@@ -2381,9 +2379,9 @@ static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
 	return 0;
 
 drop_n_account:
-	is_drop_n_account = true;
-	po->stats.stats1.tp_drops++;
 	spin_unlock(&sk->sk_receive_queue.lock);
+	atomic_inc(&po->tp_drops);
+	is_drop_n_account = true;
 
 	sk->sk_data_ready(sk);
 	kfree_skb(copy_skb);
@@ -3879,6 +3877,7 @@ static int packet_getsockopt(struct socket *sock, int level, int optname,
 	void *data = &val;
 	union tpacket_stats_u st;
 	struct tpacket_rollover_stats rstats;
+	int drops;
 
 	if (level != SOL_PACKET)
 		return -ENOPROTOOPT;
@@ -3895,14 +3894,17 @@ static int packet_getsockopt(struct socket *sock, int level, int optname,
 		memcpy(&st, &po->stats, sizeof(st));
 		memset(&po->stats, 0, sizeof(po->stats));
 		spin_unlock_bh(&sk->sk_receive_queue.lock);
+		drops = atomic_xchg(&po->tp_drops, 0);
 
 		if (po->tp_version == TPACKET_V3) {
 			lv = sizeof(struct tpacket_stats_v3);
-			st.stats3.tp_packets += st.stats3.tp_drops;
+			st.stats3.tp_drops = drops;
+			st.stats3.tp_packets += drops;
 			data = &st.stats3;
 		} else {
 			lv = sizeof(struct tpacket_stats);
-			st.stats1.tp_packets += st.stats1.tp_drops;
+			st.stats1.tp_drops = drops;
+			st.stats1.tp_packets += drops;
 			data = &st.stats1;
 		}
 
diff --git a/net/packet/internal.h b/net/packet/internal.h
index 3bb7c5fb3bff..b5bcff2b7a43 100644
--- a/net/packet/internal.h
+++ b/net/packet/internal.h
@@ -131,6 +131,7 @@ struct packet_sock {
 	struct net_device __rcu	*cached_dev;
 	int			(*xmit)(struct sk_buff *skb);
 	struct packet_type	prot_hook ____cacheline_aligned_in_smp;
+	atomic_t		tp_drops ____cacheline_aligned_in_smp;
 };
 
 static struct packet_sock *pkt_sk(struct sock *sk)
-- 
1.9.1