Re: [PATCH 07/18] [media] uvcvideo: prevent bounds-check bypass via speculative execution

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

Hi Greg,

On Tuesday, 9 January 2018 12:04:10 EET Greg KH wrote:
> On Tue, Jan 09, 2018 at 10:40:21AM +0200, Laurent Pinchart wrote:
> > On Saturday, 6 January 2018 11:40:26 EET Greg KH wrote:
> >> On Sat, Jan 06, 2018 at 10:09:07AM +0100, Greg KH wrote:
> >> 
> >> While I'm all for fixing this type of thing, I feel like we need to do
> >> something "else" for this as playing whack-a-mole for this pattern is
> >> going to be a never-ending battle for all drivers for forever.
> > 
> > That's my concern too, as even if we managed to find and fix all the
> > occurrences of the problematic patterns (and we won't), new ones will keep
> > being merged all the time.
> 
> And what about the millions of lines of out-of-tree drivers that we all
> rely on every day in our devices?  What about the distro kernels that
> add random new drivers?

Of course, even though the out-of-tree drivers probably come with lots of 
security issues worse than this one.

> We need some sort of automated way to scan for this.

Is there any initiative to implement such a scan in an open-source tool ?

We also need to educate developers. An automatic scanner could help there, but 
in the end the information has to spread to all our brains. It won't be easy, 
and is likely not fully feasible, but it's no different than how developers 
have to be educated about race conditions and locking for instance. It's a 
mind set.

> Intel, any chance we can get your coverity rules?  Given that the date
> of this original patchset was from last August, has anyone looked at
> what is now in Linus's tree?  What about linux-next?  I just added 3
> brand-new driver subsystems to the kernel tree there, how do we know
> there isn't problems in them?
> 
> And what about all of the other ways user-data can be affected?  Again,
> as Peter pointed out, USB devices.  I want some chance to be able to at
> least audit the codebase we have to see if that path is an issue.
> Without any hint of how to do this in an automated manner, we are all
> in deep shit for forever.

Or at least until the hardware architecture evolves. Let's drop the x86 
instruction set, expose the µops, and have gcc handle the scheduling. Sure, it 
will mean recompiling everything for every x86 CPU model out there, but we 
have source-based distros to the rescue :-D

> >> Either we need some way to mark this data path to make it easy for tools
> >> like sparse to flag easily, or we need to catch the issue in the driver
> >> subsystems, which unfortunatly, would harm the drivers that don't have
> >> this type of issue (like here.)
> > 
> > But how would you do so ?
> 
> I do not know, it all depends on the access pattern, right?

Any data coming from userspace could trigger such accesses. If we want 
complete coverage the only way I can think of is starting from syscalls and 
tainting data down the call stacks (__user could help to some extend), but 
we'll likely be drowned in false positives. I don't see how we could mark 
paths manually.

> >> I'm guessing that other operating systems, which don't have the luxury
> >> of auditing all of their drivers are going for the "big hammer in the
> >> subsystem" type of fix, right?
> > 
> > Other operating systems that ship closed-source drivers authored by
> > hardware vendors and not reviewed by third parties will likely stay
> > vulnerable forever. That's a small concern though as I expect those
> > drivers to contain much large security holes anyway.
> 
> Well yes, but odds are those operating systems are doing something to
> mitigate this, right?  Slowing down all user/kernel data paths?
> Targeted code analysis tools?  Something else?  I doubt they just don't
> care at all about it.  At the least, I would think Coverity would be
> trying to sell licenses for this :(

Given their track record of security issues in drivers (and I won't even 
mention the ones that are present by design, such as root kits in game copy 
protection systems for instance) I doubt they will do much beside sprinkling a 
bit of PR dust, at least for the consumer market. On the server market it 
might be different as there's less hardware variation, and thus less drivers 
to handle.

-- 
Regards,

Laurent Pinchart