Re: UIO: missing resource mapping

[Hans J. Koch]

On Mon, Jul 16, 2012 at 08:16:12PM +0200, Dominic Eschweiler wrote:
> Am Freitag, den 13.07.2012, 21:19 +0300 schrieb Michael S. Tsirkin:
> > 
> > UIO has the same property, doesn't it? Multiple users can
> > access device memory through sysfs. 
> 
> 
> Indeed, that's a similar problem. I haven't tried it (yet), but this
> particular problem can maybe circumvented by using mmap with the
> MAP_PRIVATE flag. Doing so is the responsibility of the driver
> programmer (like Hans already said). Even if that mmap trick does not
> work, it is pretty much sure that a BAR is already used by another
> program, if a related kernel driver is loaded. In that case the kernel
> has a chance to avoid such BAR race conditions by not giving the
> possibility to map them to the userspace.

Don't make it more complicated than it is. I see no general problem in
mapping BARs in uio_pci_generic like in any other UIO PCI driver.

> 
> Nevertheless, I'm pretty sure that the possibility via sysfs to access
> BARs, which are already managed by a kernel driver, opens the door for
> denial of service attacks.

There are also a few other possible attack scenarios, depending on the
hardware.
That's a general problem of all userspace drivers (e.g. X graphics drivers).
You have to make sure access rights are correct. Making a UIO device node
available to all normal users is foolishly dangerous.

> 
> On the other hand, I'm quite a newbie on this topic and maybe I don't
> see the big picture here. Therefore it is up to you guys to make the
> right decision (if needed).

Try to hack up a patch to add generic BAR mapping to uio_pci_generic.c
and post it for review.

Thanks,
Hans
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/