From 0188f76c8d48dd77fad54172b9459c2e21020c7b Mon Sep 17 00:00:00 2001 From: "Jason A. Donenfeld" Date: Mon, 29 Nov 2021 10:39:26 -0500 Subject: wireguard: receive: use ring buffer for incoming handshakes Apparently the spinlock on incoming_handshake's skb_queue is highly contended, and a torrent of handshake or cookie packets can bring the data plane to its knees, simply by virtue of enqueueing the handshake packets to be processed asynchronously. So, we try switching this to a ring buffer to hopefully have less lock contention. This alleviates the problem somewhat, though it still isn't perfect, so future patches will have to improve this further. However, it at least doesn't completely diminish the data plane. Reported-by: Streun Fabio Reported-by: Joel Wanner Fixes: a8f1bc7bdea3 ("net: WireGuard secure network tunnel") Signed-off-by: Jason A. Donenfeld Signed-off-by: Jakub Kicinski --- drivers/net/wireguard/receive.c | 27 ++++++++++++--------------- 1 file changed, 12 insertions(+), 15 deletions(-) (limited to 'drivers/net/wireguard/receive.c') diff --git a/drivers/net/wireguard/receive.c b/drivers/net/wireguard/receive.c index 7dc84bc..f4e537e 100644 --- a/drivers/net/wireguard/receive.c +++ b/drivers/net/wireguard/receive.c @@ -116,8 +116,8 @@ static void wg_receive_handshake_packet(struct wg_device *wg, return; } - under_load = skb_queue_len(&wg->incoming_handshakes) >= - MAX_QUEUED_INCOMING_HANDSHAKES / 8; + under_load = atomic_read(&wg->handshake_queue_len) >= + MAX_QUEUED_INCOMING_HANDSHAKES / 8; if (under_load) { last_under_load = ktime_get_coarse_boottime_ns(); } else if (last_under_load) { @@ -212,13 +212,14 @@ static void wg_receive_handshake_packet(struct wg_device *wg, void wg_packet_handshake_receive_worker(struct work_struct *work) { - struct wg_device *wg = container_of(work, struct multicore_worker, - work)->ptr; + struct crypt_queue *queue = container_of(work, struct multicore_worker, work)->ptr; + struct wg_device *wg = container_of(queue, struct wg_device, handshake_queue); struct sk_buff *skb; - while ((skb = skb_dequeue(&wg->incoming_handshakes)) != NULL) { + while ((skb = ptr_ring_consume_bh(&queue->ring)) != NULL) { wg_receive_handshake_packet(wg, skb); dev_kfree_skb(skb); + atomic_dec(&wg->handshake_queue_len); cond_resched(); } } @@ -554,21 +555,17 @@ void wg_packet_receive(struct wg_device *wg, struct sk_buff *skb) case cpu_to_le32(MESSAGE_HANDSHAKE_RESPONSE): case cpu_to_le32(MESSAGE_HANDSHAKE_COOKIE): { int cpu; - - if (skb_queue_len(&wg->incoming_handshakes) > - MAX_QUEUED_INCOMING_HANDSHAKES || - unlikely(!rng_is_initialized())) { + if (unlikely(!rng_is_initialized() || + ptr_ring_produce_bh(&wg->handshake_queue.ring, skb))) { net_dbg_skb_ratelimited("%s: Dropping handshake packet from %pISpfsc\n", wg->dev->name, skb); goto err; } - skb_queue_tail(&wg->incoming_handshakes, skb); - /* Queues up a call to packet_process_queued_handshake_ - * packets(skb): - */ - cpu = wg_cpumask_next_online(&wg->incoming_handshake_cpu); + atomic_inc(&wg->handshake_queue_len); + cpu = wg_cpumask_next_online(&wg->handshake_queue.last_cpu); + /* Queues up a call to packet_process_queued_handshake_packets(skb): */ queue_work_on(cpu, wg->handshake_receive_wq, - &per_cpu_ptr(wg->incoming_handshakes_worker, cpu)->work); + &per_cpu_ptr(wg->handshake_queue.worker, cpu)->work); break; } case cpu_to_le32(MESSAGE_DATA): -- cgit v1.2.3 From a0b44acb448d7a4805f062fe4f776142305f89aa Mon Sep 17 00:00:00 2001 From: "Jason A. Donenfeld" Date: Mon, 29 Nov 2021 10:39:27 -0500 Subject: wireguard: receive: drop handshakes if queue lock is contended If we're being delivered packets from multiple CPUs so quickly that the ring lock is contended for CPU tries, then it's safe to assume that the queue is near capacity anyway, so just drop the packet rather than spinning. This helps deal with multicore DoS that can interfere with data path performance. It _still_ does not completely fix the issue, but it again chips away at it. Reported-by: Streun Fabio Fixes: a8f1bc7bdea3 ("net: WireGuard secure network tunnel") Signed-off-by: Jason A. Donenfeld Signed-off-by: Jakub Kicinski --- drivers/net/wireguard/receive.c | 16 +++++++++++++--- 1 file changed, 13 insertions(+), 3 deletions(-) (limited to 'drivers/net/wireguard/receive.c') diff --git a/drivers/net/wireguard/receive.c b/drivers/net/wireguard/receive.c index f4e537e..7b8df40 100644 --- a/drivers/net/wireguard/receive.c +++ b/drivers/net/wireguard/receive.c @@ -554,9 +554,19 @@ void wg_packet_receive(struct wg_device *wg, struct sk_buff *skb) case cpu_to_le32(MESSAGE_HANDSHAKE_INITIATION): case cpu_to_le32(MESSAGE_HANDSHAKE_RESPONSE): case cpu_to_le32(MESSAGE_HANDSHAKE_COOKIE): { - int cpu; - if (unlikely(!rng_is_initialized() || - ptr_ring_produce_bh(&wg->handshake_queue.ring, skb))) { + int cpu, ret = -EBUSY; + + if (unlikely(!rng_is_initialized())) + goto drop; + if (atomic_read(&wg->handshake_queue_len) > MAX_QUEUED_INCOMING_HANDSHAKES / 2) { + if (spin_trylock_bh(&wg->handshake_queue.ring.producer_lock)) { + ret = __ptr_ring_produce(&wg->handshake_queue.ring, skb); + spin_unlock_bh(&wg->handshake_queue.ring.producer_lock); + } + } else + ret = ptr_ring_produce_bh(&wg->handshake_queue.ring, skb); + if (ret) { + drop: net_dbg_skb_ratelimited("%s: Dropping handshake packet from %pISpfsc\n", wg->dev->name, skb); goto err; -- cgit v1.2.3