On Tue, Jan 31, 2023 at 03:40:32PM +0800, Heng Qi wrote:

On Tue, Jan 31, 2023 at 02:20:49AM -0500, Michael S. Tsirkin wrote:

quoted

On Tue, Jan 31, 2023 at 11:43:37AM +0800, Heng Qi wrote:

quoted

When the single-buffer xdp is loaded and after xdp_linearize_page()
is called, *num_buf becomes 0 and (*num_buf - 1) may overflow into
a large integer in virtnet_build_xdp_buff_mrg(), resulting in
unexpected packet dropping.

Fixes: ef75cb51f139 ("virtio-net: build xdp_buff with multi buffers")
Signed-off-by: Heng Qi <redacted>
Reviewed-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>

Given the confusion, just make num_buf an int?

In the structure virtio_net_hdr_mrg_rxbuf, \field{num_buffers} is unsigned int,
which matches each other.

Because hardware buffers are all unsigned. Does not mean we need
to do all math unsigned now.

And num_buf is used in many different places, it seems
to be a lot of work to modify it to int.

Are you sure?

Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 18b3de854aeb..97f2e9bfc6f9 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c

@@ -722,7 +722,7 @@ static unsigned int virtnet_get_headroom(struct virtnet_info *vi)
  * have enough headroom.
  */
 static struct page *xdp_linearize_page(struct receive_queue *rq,
-				       u16 *num_buf,
+				       int *num_buf,
 				       struct page *p,
 				       int offset,
 				       int page_off,

@@ -822,7 +822,7 @@ static struct sk_buff *receive_small(struct net_device *dev,
 		if (unlikely(xdp_headroom < virtnet_get_headroom(vi))) {
 			int offset = buf - page_address(page) + header_offset;
 			unsigned int tlen = len + vi->hdr_len;
-			u16 num_buf = 1;
+			int num_buf = 1;
 
 			xdp_headroom = virtnet_get_headroom(vi);
 			header_offset = VIRTNET_RX_PAD + xdp_headroom;

@@ -948,7 +948,7 @@ static struct sk_buff *receive_mergeable(struct net_device *dev,
 					 struct virtnet_rq_stats *stats)
 {
 	struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
-	u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
+	int num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
 	struct page *page = virt_to_head_page(buf);
 	int offset = buf - page_address(page);
 	struct sk_buff *head_skb, *curr_skb;

Feel free to reuse.

quoted

quoted

---
 drivers/net/virtio_net.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index aaa6fe9b214a..a8e9462903fa 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c

@@ -1007,6 +1007,9 @@ static int virtnet_build_xdp_buff_mrg(struct net_device *dev,
 	xdp_prepare_buff(xdp, buf - VIRTIO_XDP_HEADROOM,
 			 VIRTIO_XDP_HEADROOM + vi->hdr_len, len - vi->hdr_len, true);
 
+	if (!*num_buf)
+		return 0;
+
 	if (*num_buf > 1) {
 		/* If we want to build multi-buffer xdp, we need
 		 * to specify that the flags of xdp_buff have the

This means truesize won't be set.

Do you mean xdp_frags_truesize please? If yes, the answer is yes, this fix
is only for single-buffer xdp, which doesn't need xdp_frags_truesize, and
already set it to 0 in its wrapper receive_mergeable().

It seems cleaner to always set the value not rely on caller to do so.

quoted

quoted

@@ -1020,10 +1023,10 @@ static int virtnet_build_xdp_buff_mrg(struct net_device *dev,
 		shinfo->xdp_frags_size = 0;
 	}
 
-	if ((*num_buf - 1) > MAX_SKB_FRAGS)
+	if (*num_buf > MAX_SKB_FRAGS + 1)
 		return -EINVAL;

Admittedly this is cleaner.

quoted

 
-	while ((--*num_buf) >= 1) {
+	while (--*num_buf) {

A bit more fragile, > 0 would be better.

Sure.

Thanks.

quoted

quoted

 		buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx);
 		if (unlikely(!buf)) {
 			pr_debug("%s: rx error: %d buffers out of %d missing\n",
-- 
2.19.1.6.gb485710b