Revision v1 of 2 in this series.
4 messages, 3 authors, 2016-08-04 · open the first message on its own page
From: <hidden>
Date: 2016-08-01 17:19:17
From: Gerard Garcia <redacted> This patch applies over VSOCK RFC v6. Reset connection and close rx socket when the sender is ignoring our announced available buffer. This avoids a DoS. Gerard Garcia (1): Fix unbound rx buffer net/vmw_vsock/virtio_transport_common.c | 22 ++++++++++++++++------ 1 file changed, 16 insertions(+), 6 deletions(-) -- 2.9.1
From: <hidden>
Date: 2016-08-01 18:21:43
From: Gerard Garcia <redacted> Reset connection and close rx socket when the sender is ignoring our announced available buffer. Signed-off-by: Gerard Garcia <redacted> --- net/vmw_vsock/virtio_transport_common.c | 22 ++++++++++++++++------ 1 file changed, 16 insertions(+), 6 deletions(-)
diff --git a/net/vmw_vsock/virtio_transport_common.c b/net/vmw_vsock/virtio_transport_common.c
index a53b3a1..3c38d43 100644
--- a/net/vmw_vsock/virtio_transport_common.c
+++ b/net/vmw_vsock/virtio_transport_common.c@@ -773,13 +773,23 @@ virtio_transport_recv_connected(struct sock *sk, pkt->len = le32_to_cpu(pkt->hdr.len); pkt->off = 0; - spin_lock_bh(&vvs->rx_lock); - virtio_transport_inc_rx_pkt(vvs, pkt); - list_add_tail(&pkt->list, &vvs->rx_queue); - spin_unlock_bh(&vvs->rx_lock); + if (vvs->rx_bytes + pkt->len <= vvs->buf_size) { + spin_lock_bh(&vvs->rx_lock); + virtio_transport_inc_rx_pkt(vvs, pkt); + list_add_tail(&pkt->list, &vvs->rx_queue); + spin_unlock_bh(&vvs->rx_lock); + + sk->sk_data_ready(sk); + return err; + } else { + /* Sender is ignoring our buf_alloc */ + spin_unlock_bh(&vvs->rx_lock); + + virtio_transport_reset(vsk, pkt); + virtio_transport_do_close(vsk, true); + break; + } - sk->sk_data_ready(sk); - return err; case VIRTIO_VSOCK_OP_CREDIT_UPDATE: sk->sk_write_space(sk); break;
--
2.9.1
From: David Miller <davem@davemloft.net>
Date: 2016-08-03 18:43:48
From: ggarcia@abra.uab.cat Date: Mon, 01 Aug 2016 19:18:21 +0200
From: Gerard Garcia <redacted> Reset connection and close rx socket when the sender is ignoring our announced available buffer. Signed-off-by: Gerard Garcia <redacted> --- net/vmw_vsock/virtio_transport_common.c | 22 ++++++++++++++++------ 1 file changed, 16 insertions(+), 6 deletions(-)diff --git a/net/vmw_vsock/virtio_transport_common.c b/net/vmw_vsock/virtio_transport_common.c index a53b3a1..3c38d43 100644 --- a/net/vmw_vsock/virtio_transport_common.c +++ b/net/vmw_vsock/virtio_transport_common.c@@ -773,13 +773,23 @@ virtio_transport_recv_connected(struct sock *sk, pkt->len = le32_to_cpu(pkt->hdr.len); pkt->off = 0; - spin_lock_bh(&vvs->rx_lock); - virtio_transport_inc_rx_pkt(vvs, pkt); - list_add_tail(&pkt->list, &vvs->rx_queue); - spin_unlock_bh(&vvs->rx_lock); + if (vvs->rx_bytes + pkt->len <= vvs->buf_size) { + spin_lock_bh(&vvs->rx_lock); + virtio_transport_inc_rx_pkt(vvs, pkt); + list_add_tail(&pkt->list, &vvs->rx_queue); + spin_unlock_bh(&vvs->rx_lock); + + sk->sk_data_ready(sk); + return err; + } else { + /* Sender is ignoring our buf_alloc */ + spin_unlock_bh(&vvs->rx_lock);
You don't hold vvs->rx_lock here, so it's a bug to try and release it. I'd also expect the maintainers of this driver to give some feedback before I apply this.
From: Stefan Hajnoczi <stefanha@redhat.com>
Date: 2016-08-04 13:40:17
On Mon, Aug 01, 2016 at 07:18:21PM +0200, ggarcia@abra.uab.cat wrote: Besides the other locking issue that was already mentioned:
diff --git a/net/vmw_vsock/virtio_transport_common.c b/net/vmw_vsock/virtio_transport_common.c index a53b3a1..3c38d43 100644 --- a/net/vmw_vsock/virtio_transport_common.c +++ b/net/vmw_vsock/virtio_transport_common.c@@ -773,13 +773,23 @@ virtio_transport_recv_connected(struct sock *sk, pkt->len = le32_to_cpu(pkt->hdr.len); pkt->off = 0; - spin_lock_bh(&vvs->rx_lock); - virtio_transport_inc_rx_pkt(vvs, pkt); - list_add_tail(&pkt->list, &vvs->rx_queue); - spin_unlock_bh(&vvs->rx_lock); + if (vvs->rx_bytes + pkt->len <= vvs->buf_size) {
vvs->rx_bytes must be accessed under vvs->rx_lock.