On 11/16/21 04:27, Jakub Kicinski wrote:
I'd ignore that path, it's just special casing that's supposed to keep the driver-visible API sane. Nobody should be touching netdev past free_netdev(). Actually if you can it'd be interesting to add checks for using whatever netdev_priv(ndev) returned past free_netdev(ndev).
Most UAFs that come to mind from the past were people doing something like:
struct my_priv *mine = netdev_priv(ndev);
netdev_unregister(ndev); free_netdev(ndev);
free(mine->bla); /* UAF, free_netdev() frees the priv */
I've implemented this checker couple of months ago. The latest smatch (v1.72) should warn about this type of bugs. All reported bugs are fixed already :)
My checker warns about using priv pointer after free_netdev() and free_candev() calls. There are a few more wrappers like free_sja1000dev(), so it worth to add them to check list too. Will add them today later
Important thing, that there are complex situations like
struct priv *priv = get_priv_from_smth(smth);
free_netdev(priv->netdev); clean_up_priv(priv);
and for now I have no idea how to handle it (ex: ems_usb_disconnect).
With regards, Pavel Skripkin