Re: [PATCH net] qmi_wwan: Clone the skb when in pass-through mode

On Mon, 14 Jun 2021 16:45:55 +0200 Bjørn Mork wrote:

Kristian Evensen [off-list ref] writes:

quoted

The skb that we pass to the rmnet driver is owned by usbnet and is freed
soon after the rx_fixup() callback is called (in usbnet_bh()).  There is
no guarantee that rmnet is done handling the skb before it is freed. We
should clone the skb before we call netif_rx() to prevent use-after-free
and misc. kernel oops.

Fixes: 59e139cf0b32 ("net: qmi_wwan: Add pass through mode")

Signed-off-by: Kristian Evensen <redacted>
---
 drivers/net/usb/qmi_wwan.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/drivers/net/usb/qmi_wwan.c b/drivers/net/usb/qmi_wwan.c
index db8d3a4f2678..5ac307eb0bfd 100644
--- a/drivers/net/usb/qmi_wwan.c
+++ b/drivers/net/usb/qmi_wwan.c

@@ -620,6 +620,10 @@ static int qmi_wwan_rx_fixup(struct usbnet *dev, struct sk_buff *skb)
 		return qmimux_rx_fixup(dev, skb);
 
 	if (info->flags & QMI_WWAN_FLAG_PASS_THROUGH) {
+		skb = skb_clone(skb, GFP_ATOMIC);
+		if (!skb)
+			return 0;
+
 		skb->protocol = htons(ETH_P_MAP);
 		return (netif_rx(skb) == NET_RX_SUCCESS);
 	}  

Thanks for pointing this out.  But it still looks strange to me.  Why do
we call netif_rx(skb) here instead of just returning 1 and leave that
for usbnet_skb_return()?  With cloning we end up doing eth_type_trans()
on the duplicate - is that wise?

Agreed on the cloning being a strange solution. Kristian, were you able
to reproduce the problem on upstream kernels?

It does look pretty strange that qmimux_rx_fixup() copies out all
packets and receives them, and then let's usbnet to process the
multi-frame skb without even fulling off the qmimux_hdr. I'm probably
missing something.. otherwise sth like FLAG_MULTI_PACKET may be in
order?