linux-kernel - Re: [PATCH v4 3/4] vfio/pci: Disable mmap() non-compliant BARs

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date: Thu, 27 Jun 2024 16:05:47 -0600
From: Alex Williamson <alex.williamson@...hat.com>
To: Niklas Schnelle <schnelle@...ux.ibm.com>
Cc: Bjorn Helgaas <helgaas@...nel.org>, Christoph Hellwig <hch@....de>,
 Gerald Schaefer <gerald.schaefer@...ux.ibm.com>, Heiko Carstens
 <hca@...ux.ibm.com>, Vasily Gorbik <gor@...ux.ibm.com>, Alexander Gordeev
 <agordeev@...ux.ibm.com>, Christian Borntraeger
 <borntraeger@...ux.ibm.com>, Sven Schnelle <svens@...ux.ibm.com>, Gerd
 Bayer <gbayer@...ux.ibm.com>, Matthew Rosato <mjrosato@...ux.ibm.com>,
 Jason Gunthorpe <jgg@...pe.ca>, linux-s390@...r.kernel.org,
 linux-kernel@...r.kernel.org, kvm@...r.kernel.org
Subject: Re: [PATCH v4 3/4] vfio/pci: Disable mmap() non-compliant BARs

On Wed, 26 Jun 2024 13:15:50 +0200
Niklas Schnelle <schnelle@...ux.ibm.com> wrote:

> When VFIO_PCI_MMAP is enabled for s390 in a future commit and the ISM
> device is passed-through to a KVM guest QEMU attempts to eagerly mmap()
> its BAR. This fails because the 256 TiB large BAR does not fit in the
> virtual map. Besides this issue mmap() of the ISM device's BAR is not
> useful either as even a partial mapping won't be usable from user-space
> without a vfio-pci variant driver. A previous commit ensures that pdev->
> non_compliant_bars is set for ISM so use this to disallow mmap() with
> the expecation that mmap() of non-compliant BARs is not advisable in the
> general case either.
> 
> Signed-off-by: Niklas Schnelle <schnelle@...ux.ibm.com>
> ---
>  drivers/vfio/pci/vfio_pci_core.c | 5 ++---
>  1 file changed, 2 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/vfio/pci/vfio_pci_core.c b/drivers/vfio/pci/vfio_pci_core.c
> index 987c7921affa..0e9d46575776 100644
> --- a/drivers/vfio/pci/vfio_pci_core.c
> +++ b/drivers/vfio/pci/vfio_pci_core.c
> @@ -128,10 +128,9 @@ static void vfio_pci_probe_mmaps(struct vfio_pci_core_device *vdev)
>  
>  		/*
>  		 * The PCI core shouldn't set up a resource with a
> -		 * type but zero size. But there may be bugs that
> -		 * cause us to do that.
> +		 * type but zero size or non-compliant BARs.
>  		 */
> -		if (!resource_size(res))
> +		if (!resource_size(res) || vdev->pdev->non_compliant_bars)
>  			goto no_mmap;
>  
>  		if (resource_size(res) >= PAGE_SIZE) {
> 

The non_compliant_bars flag causes pci_read_bases() to exit, shouldn't
that mean the resource is not setup and resource_size() is zero and
explicitly testing the non_compliant_bars flag is redundant?  Or does
s390 do this somewhere else?

The non_compliant_bars flag is defined as /* Broken BARs; ignore them */
so it'd be pretty strange if they had a resource size and we chose to
still expose them with read-write access... why wouldn't we just
deny-list the device from use with vfio-pci?

Also probably worth an explicit comment in the commit log why pci-sysfs
mmap support doesn't need to be bypassed on s390.  Thanks,

Alex