Re: [PATCH 5/5] vfio/iommu_type1: implement the VFIO_DMA_MAP_FLAG_NOEXEC flag

[Antonios Motakis <a.motakis@virtualopensystems.com>]

On Tue, Oct 21, 2014 at 4:37 PM, Alex Williamson
<alex.williamson@redhat.com> wrote:
> On Tue, 2014-10-21 at 14:40 +0200, Antonios Motakis wrote:
>> On Mon, Oct 20, 2014 at 11:13 PM, Alex Williamson
>> <alex.williamson@redhat.com> wrote:
>> > On Mon, 2014-10-13 at 15:09 +0200, Antonios Motakis wrote:
>> >> Some IOMMU drivers, such as the ARM SMMU driver, make available the
>> >> IOMMU_NOEXEC flag, to set the page tables for a device as XN (execute never).
>> >> This affects devices such as the ARM PL330 DMA Controller, which respects
>> >> this flag and will refuse to fetch DMA instructions from memory where the
>> >> XN flag has been set.
>> >>
>> >> The flag can be used only if all IOMMU domains behind the container support
>> >> the IOMMU_NOEXEC flag. Also, if any mappings are created with the flag, any
>> >> new domains with devices will have to support it as well.
>> >>
>> >> Signed-off-by: Antonios Motakis <a.motakis@virtualopensystems.com>
>> >> ---
>> >>  drivers/vfio/vfio_iommu_type1.c | 25 ++++++++++++++++++++++++-
>> >>  1 file changed, 24 insertions(+), 1 deletion(-)
>> >>
>> >> diff --git a/drivers/vfio/vfio_iommu_type1.c b/drivers/vfio/vfio_iommu_type1.c
>> >> index 8b4202a..e225e8f 100644
>> >> --- a/drivers/vfio/vfio_iommu_type1.c
>> >> +++ b/drivers/vfio/vfio_iommu_type1.c
>> >> @@ -569,6 +569,12 @@ static int vfio_dma_do_map(struct vfio_iommu *iommu,
>> >>       if (map->flags & VFIO_DMA_MAP_FLAG_READ)
>> >>               prot |= IOMMU_READ;
>> >>
>> >> +     if (map->flags & VFIO_DMA_MAP_FLAG_NOEXEC) {
>> >> +             if (!vfio_domains_have_iommu_cap(iommu, IOMMU_CAP_NOEXEC))
>> >> +                     return -EINVAL;
>> >> +             prot |= IOMMU_NOEXEC;
>> >> +     }
>> >> +
>> >>       if (!prot || !size || (size | iova | vaddr) & mask)
>> >>               return -EINVAL;
>> >
>> > I think this test needs to move above adding the NOEXEC flag, otherwise
>> > we now allow mappings without read or write, which is an ABI change.
>> >
>>
>> Ack.
>>
>> >>
>> >> @@ -662,6 +668,14 @@ static int vfio_iommu_replay(struct vfio_iommu *iommu,
>> >>               dma = rb_entry(n, struct vfio_dma, node);
>> >>               iova = dma->iova;
>> >>
>> >> +             /*
>> >> +              * if any of the mappings to be replayed has the NOEXEC flag
>> >> +              * set, then the new iommu domain must support it
>> >> +              */
>> >> +             if ((dma->prot | IOMMU_NOEXEC) &&
>> >
>> > I think you mean
>> >
>> > & IOMMU_NOEXEC
>>
>> Ack.
>>
>> >
>> >> +                             !(domain->caps & IOMMU_CAP_NOEXEC))
>> >> +                     return -EINVAL;
>> >> +
>> >
>> > In patch 2/5 you stated:
>> >
>> >         The IOMMU_NOEXEC flag needs to be available for all the IOMMUs
>> >         of the container used.
>> >
>> > But here you'll create heterogeneous containers so long as there are no
>> > NOEXEC mappings.  Is that intentional or a side effect of the above
>> > masking bug?
>> >
>>
>> Yeah, my intention was to not stop the user of having heterogeneous
>> containers, as long as he doesn't care about using the NOEXEC flag. As
>> soon as the user tries to apply this flag however, then it should be
>> supported by all the IOMMUs behind the container - otherwise it is not
>> enforceable.
>>
>> Do you think we should change this behavior? I think most users will
>> not care about using this flag, and we should not stop them from
>> mixing containers.
>
> I think that's a reasonable way to go, but let's add a comment in uapi
> vfio.h describing that expectation.  Thanks,

Ok, will do.

>
> Alex
>
>> >>               while (iova < dma->iova + dma->size) {
>> >>                       phys_addr_t phys = iommu_iova_to_phys(d->domain, iova);
>> >>                       size_t size;
>> >> @@ -749,6 +763,9 @@ static int vfio_iommu_type1_attach_group(void *iommu_data,
>> >>       if (iommu_capable(bus, IOMMU_CAP_CACHE_COHERENCY))
>> >>               domain->caps |= IOMMU_CAP_CACHE_COHERENCY;
>> >>
>> >> +     if (iommu_capable(bus, IOMMU_CAP_NOEXEC))
>> >> +             domain->caps |= IOMMU_CAP_NOEXEC;
>> >> +
>> >>       /*
>> >>        * Try to match an existing compatible domain.  We don't want to
>> >>        * preclude an IOMMU driver supporting multiple bus_types and being
>> >> @@ -900,6 +917,11 @@ static long vfio_iommu_type1_ioctl(void *iommu_data,
>> >>                               return 0;
>> >>                       return vfio_domains_have_iommu_cap(iommu,
>> >>                                                 IOMMU_CAP_CACHE_COHERENCY);
>> >> +             case VFIO_DMA_NOEXEC_IOMMU:
>> >> +                     if (!iommu)
>> >> +                             return 0;
>> >> +                     return vfio_domains_have_iommu_cap(iommu,
>> >> +                                                        IOMMU_CAP_NOEXEC);
>> >>               default:
>> >>                       return 0;
>> >>               }
>> >> @@ -923,7 +945,8 @@ static long vfio_iommu_type1_ioctl(void *iommu_data,
>> >>       } else if (cmd == VFIO_IOMMU_MAP_DMA) {
>> >>               struct vfio_iommu_type1_dma_map map;
>> >>               uint32_t mask = VFIO_DMA_MAP_FLAG_READ |
>> >> -                             VFIO_DMA_MAP_FLAG_WRITE;
>> >> +                             VFIO_DMA_MAP_FLAG_WRITE |
>> >> +                             VFIO_DMA_MAP_FLAG_NOEXEC;
>> >>
>> >>               minsz = offsetofend(struct vfio_iommu_type1_dma_map, size);
>> >>
>> >
>> >
>> >
>>
>>
>>
>
>
>



-- 
Antonios Motakis
Virtual Open Systems