On Tue, Nov 28, 2017 at 07:50:58PM +0200, Michael S. Tsirkin wrote:

On Tue, Nov 28, 2017 at 12:17:16PM -0500, wexu@redhat.com wrote:

quoted

From: Wei Xu <redacted>

Matthew found a roughly 40% tcp throughput regression with commit
c67df11f(vhost_net: try batch dequing from skb array) as discussed
in the following thread:
https://www.mail-archive.com/netdev@vger.kernel.org/msg187936.html

Eventually we figured out that it was a skb leak in handle_rx()
when sending packets to the VM. This usually happens when a guest
can not drain out vq as fast as vhost fills in, afterwards it sets
off the traffic jam and leaks skb(s) which occurs as no headcount
to send on the vq from vhost side.

This can be avoided by making sure we have got enough headcount
before actually consuming a skb from the batched rx array while
transmitting, which is simply done by deferring it a moment later
in this patch.

Signed-off-by: Wei Xu <redacted>

Looks like a good way to fix it, but it will still leak if recvmsg
returns a wrong length (I'm not sure this can happen in practice, but
best to keep it simple).

Right, it is better to defend this case, will include it in v2.

Also, we need to add this before each recvmsg, including overrun,
and discard on error.

OK.

Wei

quoted

---
 drivers/vhost/net.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
index 8d626d7..e76535e 100644
--- a/drivers/vhost/net.c
+++ b/drivers/vhost/net.c

@@ -778,8 +778,6 @@ static void handle_rx(struct vhost_net *net)
 		/* On error, stop handling until the next kick. */
 		if (unlikely(headcount < 0))
 			goto out;
-		if (nvq->rx_array)
-			msg.msg_control = vhost_net_buf_consume(&nvq->rxq);
 		/* On overrun, truncate and discard */
 		if (unlikely(headcount > UIO_MAXIOV)) {
 			iov_iter_init(&msg.msg_iter, READ, vq->iov, 1, 1);

@@ -809,6 +807,8 @@ static void handle_rx(struct vhost_net *net)
 			 */
 			iov_iter_advance(&msg.msg_iter, vhost_hlen);
 		}
+		if (nvq->rx_array)
+			msg.msg_control = vhost_net_buf_consume(&nvq->rxq);
 		err = sock->ops->recvmsg(sock, &msg,
 					 sock_len, MSG_DONTWAIT | MSG_TRUNC);
 		/* Userspace might have consumed the packet meanwhile:

-- 
1.8.3.1