On Fri, Feb 12, 2021 at 12:22 PM Jens Axboe <axboe@xxxxxxxxx> wrote:
>
> My other idea was
> to add a check in path resolution that catches it, for future safe
> guards outside of send/recvmsg. That's obviously a separate change
> from the comment, but would be nice to have.
I wonder how painful it would be to just make sure that kernel threads
have a NULL ->fs and ->files by default.
But maybe the oops in a kernel thread ends up being too painful and
harder to debug and deal with, and a special check is preferred just
because a WARN_ON_ONCE() wouldn't cause any other downstream issues.
Yes, some kernel threads do need to do pathname lookups (and io_uring
isn't the only such thing), but I think they are generally fairly
special cases (ie things like usermode helper, firmware loaders, etc).
Linus
