kvm.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Jason Gunthorpe <jgg@nvidia.com>
To: Peter Xu <peterx@redhat.com>
Cc: Alex Williamson <alex.williamson@redhat.com>, <cohuck@redhat.com>,
	<kvm@vger.kernel.org>, <linux-kernel@vger.kernel.org>
Subject: Re: [RFC PATCH 10/10] vfio/type1: Register device notifier
Date: Thu, 25 Feb 2021 15:17:14 -0400	[thread overview]
Message-ID: <20210225191714.GU4247@nvidia.com> (raw)
In-Reply-To: <20210225190646.GE250483@xz-x1>

On Thu, Feb 25, 2021 at 02:06:46PM -0500, Peter Xu wrote:

> Agreed.  I saw discussions around on redefining the vm_pgoff namespace, I can't
> say I followed that closely either, but yes it definitely makes sense to always
> use an unified namespace.  Maybe we should even comment it somewhere on how
> vm_pgoff is encoded?

Yes, it should be described, it is subtle
 
> > Correct. VFIO can map into the IOMMU PFNs it can get a reference
> > to. pin_user_pages() works for the majority, special VFIO VMAs cover
> > the rest, and everthing else must be blocked for security.
> 
> If we all agree that the current follow_pfn() should only apply to vfio
> internal vmas, 

I want to remvoe follow_pfn(). Internal VMAs can deduce the PFN from
the vm_pgoff, they don't need to do follow.

> then it seems we can drop it indeed, as long as the crash reported
> in 5cbf3264b would fail gracefully at e.g. VFIO_IOMMU_MAP_DMA rather
> than triggering a kernel warning somehow.

Yes, this will just fail the ioctl because pin_user_pages() failed and
the VMA was not VFIO.

> However I'm still confused on why it's more secure - the current process to do
> VFIO_IOMMU_MAP_DMA should at least has proper permission for everything to be
> setup, including the special vma, right?  Say, if the process can write to
> those memories, then shouldn't we also allow it to grant this write permission
> to other devices too?

It is a use-after-free. Once the PFN is programmed into the IOMMU it
becomes completely divorced from the VMA. Remember there is no
pin_user_page here, so the PFN has no reference count.

If the owner of the VMA decided to zap it or otherwise then the IOMMU
access keeps going - but now the owner thinks the PFN is free'd and
nobody is referencing it. Goes bad.

Jason

  reply	other threads:[~2021-02-25 19:21 UTC|newest]

Thread overview: 36+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-02-22 16:50 [RFC PATCH 00/10] vfio: Device memory DMA mapping improvements Alex Williamson
2021-02-22 16:50 ` [RFC PATCH 01/10] vfio: Create vfio_fs_type with inode per device Alex Williamson
2021-02-26  5:38   ` Christoph Hellwig
2021-02-26 13:15     ` Jason Gunthorpe
2021-02-22 16:50 ` [RFC PATCH 02/10] vfio: Update vfio_add_group_dev() API Alex Williamson
2021-02-22 17:01   ` Jason Gunthorpe
2021-02-22 16:50 ` [RFC PATCH 03/10] vfio: Export unmap_mapping_range() wrapper Alex Williamson
2021-02-22 16:51 ` [RFC PATCH 04/10] vfio/pci: Use vfio_device_unmap_mapping_range() Alex Williamson
2021-02-22 17:22   ` Jason Gunthorpe
2021-02-24 21:55     ` Alex Williamson
2021-02-25  0:57       ` Jason Gunthorpe
2021-02-22 16:51 ` [RFC PATCH 05/10] vfio: Create a vfio_device from vma lookup Alex Williamson
2021-02-22 17:29   ` Jason Gunthorpe
2021-02-24 21:55     ` Alex Williamson
2021-02-25  0:06       ` Jason Gunthorpe
2021-02-25 22:21         ` Alex Williamson
2021-02-25 23:49           ` Jason Gunthorpe
2021-03-04 21:37             ` Alex Williamson
2021-03-04 23:16               ` Jason Gunthorpe
2021-03-05  0:07                 ` Alex Williamson
2021-03-05  0:36                   ` Jason Gunthorpe
2021-02-22 16:51 ` [RFC PATCH 06/10] vfio: Add a device notifier interface Alex Williamson
2021-02-22 16:51 ` [RFC PATCH 07/10] vfio/pci: Notify on device release Alex Williamson
2021-02-22 16:52 ` [RFC PATCH 08/10] vfio/type1: Refactor pfn_list clearing Alex Williamson
2021-02-22 16:52 ` [RFC PATCH 09/10] vfio/type1: Pass iommu and dma objects through to vaddr_get_pfn Alex Williamson
2021-02-22 16:52 ` [RFC PATCH 10/10] vfio/type1: Register device notifier Alex Williamson
2021-02-22 17:55   ` Jason Gunthorpe
2021-02-24 21:55     ` Alex Williamson
2021-02-25  0:22       ` Jason Gunthorpe
2021-02-25 17:54         ` Peter Xu
2021-02-25 18:19           ` Jason Gunthorpe
2021-02-25 19:06             ` Peter Xu
2021-02-25 19:17               ` Jason Gunthorpe [this message]
2021-02-25 19:54                 ` Peter Xu
2021-02-26  5:47     ` Christoph Hellwig
2021-02-22 18:00 ` [RFC PATCH 00/10] vfio: Device memory DMA mapping improvements Jason Gunthorpe

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20210225191714.GU4247@nvidia.com \
    --to=jgg@nvidia.com \
    --cc=alex.williamson@redhat.com \
    --cc=cohuck@redhat.com \
    --cc=kvm@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=peterx@redhat.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body. 
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).