[Xen-devel] DRAFT XSA 142 - libxl fails to honour readonly flag on disks with qemu-xen

[Xen.org security team <security@xxxxxxx>]

***** DRAFT DRAFT DRAFT *****

                    Xen Security Advisory XSA-142

        libxl fails to honour readonly flag on disks with qemu-xen

ISSUE DESCRIPTION
=================

Callers of libxl can specify that a disk should be read-only to the
guest.  However, there is no code in libxl to pass this information to
qemu-xen (the upstream-based qemu); and indeed there is no way in qemu
to make a disk read-only.

The vulnerability is exploitable only via devices emulated by the
device model, not the parallel PV devices for supporting PVHVM.
Normally the PVHVM device unplug protocol renders the emulated devices
inaccessible early in boot.

IMPACT
======

Malicious guest administrators or (in some situations) users may be
able to write to supposedly read-only disk images.

CDROM devices (that is, devices specified to be presented to the guest
as CDROMs, regardless of the nature of the backing storage on the
host) are not affected.

VULNERABLE SYSTEMS
==================

Only systems using qemu-xen (rather than qemu-xen-traditional) as the
device model version are vulnerable.

Only systems using libxl or libxl-based toolstacks are vulnerable.
(This includes xl, and libvirt with the libxl driver.)

All versions of libxl which support qemu-xen are vulnerable.  The
affected code was introduced in Xen 4.1.

If the host and guest together usually support PVHVM, the issue is
exploitable only if the malicious guest administrator has control of
the guest kernel or guest kernel command line.

MITIGATION
==========

Switching to qemu-xen-traditional will avoid this vulnerability.
This can be done with
   device_model_version="qemu-xen-traditional"
in the xl configuration file.

Using stub domain device models (which necessarily involves switching
to qemu-xen-traditional) will also avoid this vulnerability.
This can be done with
   device_model_stubdomain_override=true
in the xl configuration file.

Either of these mitigations is liable to have other guest-visible
effects or even regressions.

It may be possible, depending on the configuration, to make the
underlying storage object readonly, or to make it reject writes.

RESOLUTION
==========

There is no reasonable resolution because Qemu does not (at the time
of writing) support presenting a read-only block device to a guest as
a disk.

The attached patch corrects the weakness in the libxl code, by
rejecting the unsupported configurations, rather than allowing them to
run but with the device perhaps writeable by the guest.  Applying it
should increase confidence and avoid future configuration errors, but
will break configurations specifying read-only disk devices.

xsa142.patch                 Xen 4.3.x and later

$ sha256sum xsa142*.patch
de0d6d19becac199037dce5a6a49e35cb7de5c99b8e2950600ed71fdc2d5a752  xsa142.patch
$

NOTE REGARDING LACK OF EMBARGO
==============================

This issue was discussed in public in the Red Hat bugzilla:
  https://bugzilla.redhat.com/show_bug.cgi?id=1257893

CREDITS
=======

Thanks to Michael Young of Durham University for bring this problem to
our attention.

Attachment: xsa142.patch
Description: Binary data

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel