f_ncm: Don't use netdev_start_xmit().

Unfortunately, the USB gadget layer has this weird things where NULL
skbs are passed into ops->ndo_start_xmit() in order to trigger the
dev->wrap() calls to build packets.

This is completely outside of the allowable range of sane arguments
for the ndo_start_xmit method.  All invocations of ndo_start_xmit()
should be with non-NULL SKB arguments.

Put back the direct call, but with a comment explaining how this
is not acceptable in the long term.

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/drivers/usb/gadget/function/f_ncm.c b/drivers/usb/gadget/function/f_ncm.c
index cb5d646db6a7c21d53acd30dd05d55f39bf84528..146f48cc65d7fb9c000565448121337f914c694c 100644 (file)
--- a/drivers/usb/gadget/function/f_ncm.c
+++ b/drivers/usb/gadget/function/f_ncm.c
@@ -1101,7 +1101,15 @@ static void ncm_tx_tasklet(unsigned long data)
        /* Only send if data is available. */
        if (ncm->skb_tx_data) {
                ncm->timer_force_tx = true;
-               netdev_start_xmit(NULL, ncm->netdev);
+
+               /* XXX This allowance of a NULL skb argument to ndo_start_xmit
+                * XXX is not sane.  The gadget layer should be redesigned so
+                * XXX that the dev->wrap() invocations to build SKBs is transparent
+                * XXX and performed in some way outside of the ndo_start_xmit
+                * XXX interface.
+                */
+               ncm->netdev->netdev_ops->ndo_start_xmit(NULL, ncm->netdev);
+
                ncm->timer_force_tx = false;
        }
 }