Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Sender: Konrad Rzeszutek Wilk <konrad.r.wilk@gmail.com>
Date:   Mon, 18 Jan 2021 10:14:28 -0500
From:   Konrad Rzeszutek Wilk <konrad@darnok.org>
To:     Martin Radev <martin.b.radev@gmail.com>
Cc:     Christoph Hellwig <hch@lst.de>, thomas.lendacky@amd.com,
        file@sect.tu-berlin.de, robert.buhren@sect.tu-berlin.de,
        kvm@vger.kernel.org, konrad.wilk@oracle.com,
        mathias.morbitzer@aisec.fraunhofer.de,
        linux-kernel@vger.kernel.org,
        virtualization@lists.linux-foundation.org,
        iommu@lists.linux-foundation.org, robin.murphy@arm.com,
        kirill.shutemov@linux.intel.com
Subject: Re: [PATCH] swiotlb: Validate bounce size in the sync/unmap path
Message-ID: <20210118151428.GA72213@fedora>
References: <X/27MSbfDGCY9WZu@martin>
 <20210113113017.GA28106@lst.de>
 <YAV0uhfkimXn1izW@martin>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <YAV0uhfkimXn1izW@martin>
Precedence: bulk

On Mon, Jan 18, 2021 at 12:44:58PM +0100, Martin Radev wrote:
> On Wed, Jan 13, 2021 at 12:30:17PM +0100, Christoph Hellwig wrote:
> > On Tue, Jan 12, 2021 at 04:07:29PM +0100, Martin Radev wrote:
> > > The size of the buffer being bounced is not checked if it happens
> > > to be larger than the size of the mapped buffer. Because the size
> > > can be controlled by a device, as it's the case with virtio devices,
> > > this can lead to memory corruption.
> > > 
> > 
> > I'm really worried about all these hodge podge hacks for not trusted
> > hypervisors in the I/O stack.  Instead of trying to harden protocols
> > that are fundamentally not designed for this, how about instead coming
> > up with a new paravirtualized I/O interface that is specifically
> > designed for use with an untrusted hypervisor from the start?
> 
> Your comment makes sense but then that would require the cooperation
> of these vendors and the cloud providers to agree on something meaningful.
> I am also not sure whether the end result would be better than hardening
> this interface to catch corruption. There is already some validation in
> unmap path anyway.
> 
> Another possibility is to move this hardening to the common virtio code,
> but I think the code may become more complicated there since it would
> require tracking both the dma_addr and length for each descriptor.

Christoph,

I've been wrestling with the same thing - this is specific to busted
drivers. And in reality you could do the same thing with a hardware
virtio device (see example in http://thunderclap.io/) - where the
mitigation is 'enable the IOMMU to do its job.'.

AMD SEV documents speak about utilizing IOMMU to do this (AMD SEV-SNP)..
and while that is great in the future, SEV without IOMMU is now here.

Doing a full circle here, this issue can be exploited with virtio
but you could say do that with real hardware too if you hacked the
firmware, so if you say used Intel SR-IOV NIC that was compromised
on an AMD SEV machine, and plumbed in the guest - the IOMMU inside
of the guest would be SWIOTLB code. Last line of defense against
bad firmware to say.

As such I am leaning towards taking this code, but I am worried
about the performance hit .. but perhaps I shouldn't as if you
are using SWIOTLB=force already you are kind of taking a
performance hit?