On Fri, Aug 7, 2020 at 12:19 AM Christoph Hellwig <h...@lst.de> wrote:
>
> On Thu, Aug 06, 2020 at 11:23:34PM -0700, John Stultz wrote:
> > So I've finally rebase-bisected it down to:
> > a31edb2059ed ("net: improve the user pointer check in init_user_sockptr")
> > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=a31edb2059ed4e498f9aa8230c734b59d0ad797a
> >
> > And reverting that from linus/HEAD (at least from this morning) seems
> > to avoid it.
> >
> > Seems like it is just adding extra checks on the data passed, so maybe
> > existing trouble from a different driver is the issue here, but it's
> > not really clear from the crash what might be wrong.
> >
> > Suggestions would be greatly appreciated!
>
> I think the sockpt optimization is just a little to clever for its
> own sake, as also chown by the other issue pointed out by Eric.
>
> Can you try this revert that just goes back to the "boring" normal
> version for everyone?
Yes! This seems to avoid the crash and networking looks ok.
Tested-by: John Stultz <john.stu...@linaro.org>
thanks
-john
