Re: [PATCH net] virtio_net: Fix resize of the RX ring
From: Vadim Fedorenko <vadim.fedorenko@linux.dev>
Date: 2026-08-10 13:07:26
Also in:
lkml, netdev
On 10/08/2026 13:07, Anton Protopopov wrote:
quoted hunk ↗ jump to hunk
When a AF_XDP socket is attached, the virtnet_rx_resize should resize the rq->xsk_buffs XSK buffer array. Otherwise, when the size grows, the virtnet_rx_resume() causes a write past the end of the array. This is easily reproducable with ethtool -G ens3 rx 32 ./xdpsock -i eth0 -q 0 -r -z & ethtool -G eth0 rx 256 Fixes: e9f3962441c0 ("virtio_net: xsk: rx: support fill with xsk buffer") Signed-off-by: Anton Protopopov <redacted> --- drivers/net/virtio_net.c | 14 ++++++++++++++ 1 file changed, 14 insertions(+)diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c index 3e2a5876c6c8..e34c52d059d3 100644 --- a/drivers/net/virtio_net.c +++ b/drivers/net/virtio_net.c@@ -3444,17 +3444,31 @@ static void virtnet_rx_resume_all(struct virtnet_info *vi) static int virtnet_rx_resize(struct virtnet_info *vi, struct receive_queue *rq, u32 ring_num) { + unsigned int old_ring_num = virtqueue_get_vring_size(rq->vq); + struct xdp_buff **tmp_xsk_buffs = NULL; int err, qindex; qindex = rq - vi->rq; + if (rq->xsk_pool && ring_num > old_ring_num) {
why do you cover only the case for growing buffer? Don't we expect to free some memory in case of shrinking? Should be fine given you are swapping buffers completely...
+ tmp_xsk_buffs = kvzalloc_objs(*tmp_xsk_buffs, ring_num); + if (!tmp_xsk_buffs) + return -ENOMEM; + } + virtnet_rx_pause(vi, rq); err = virtqueue_resize(rq->vq, ring_num, virtnet_rq_unmap_free_buf, NULL); + + /* virtqueue_resize may have changed the size even if err != 0 */ + if (tmp_xsk_buffs && virtqueue_get_vring_size(rq->vq) > old_ring_num) + swap(rq->xsk_buffs, tmp_xsk_buffs); + if (err) netdev_err(vi->dev, "resize rx fail: rx queue index: %d err: %d\n", qindex, err); virtnet_rx_resume(vi, rq, true); + kvfree(tmp_xsk_buffs); return err; }