Dear Qubes community,

We have just published Qubes Security Bulletin (QSB) #29: Critical Xen bug in PV memory virtualization code (XSA-212).

The current text of this QSB is reproduced below. The latest version, including any future corrections, will always be available in the Qubes Security Pack (qubes-secpack).

View QSB #29 in the qubes-secpack:

https://github.com/QubesOS/qubes-secpack/blob/master/QSBs/qsb-029-2017.txt

Learn about the qubes-secpack, including how to obtain, verify, and read it:

https://www.qubes-os.org/doc/security-pack/

View all past QSBs:

https://www.qubes-os.org/doc/security-bulletins/

View XSA-212 in the XSA Tracker:

https://www.qubes-os.org/security/xsa/#212

             ---===[ Qubes Security Bulletin #29 ]===---

                          April 4, 2017


      Critical Xen bug in PV memory virtualization code (XSA-212)


Quick Summary
==============

The Xen Security Team has discovered a serious security bug (XSA-212)
in the hypervisor code handling memory virtualization for
paravirtualized (PV) VMs [1]:

| The XSA-29 fix introduced an insufficient check on XENMEM_exchange
| input, allowing the caller to drive hypervisor memory accesses outside
| of the guest provided input/output arrays.
| 
| A malicious or buggy 64-bit PV guest may be able to access all of
| system memory, allowing for all of privilege escalation, host crashes,
| and information leaks.

An attacker who exploits this bug can break Qubes-provided isolation.
This means that if an attacker has already exploited another
vulnerability, e.g. in a Web browser or networking or USB stack, then
the attacker would be able to compromise a whole Qubes system.

Description of the bug
=======================

Arguments of XENMEM_exchange hypercall contains array of page numbers
to be exchanged and array for exchanged page numbers. Validation of
those arrays (supposedly living in guest memory) assumes a certain
memory layout -- that Xen memory is just above non-canonical addresses.
It also assumes that array accesses are sequential starting from the
beginning of the array. With this assumption, validation code checks
only the address of the first array element and assumes (thanks to
sequential access) that further code will hit a non-canonical address,
resulting in page fault, before reaching Xen memory.

Relevant part of xen/include/asm-x86/x86_64/uaccess.h:

    /*
     * Valid if in +ve half of 48-bit address space, or above Xen-reserved area.
     * This is also valid for range checks (addr, addr+size). As long as the
     * start address is outside the Xen-reserved area, sequential accesses
     * (starting at addr) will hit a non-canonical address (and thus fault)
     * before ever reaching VIRT_START.
     */
    #define __addr_ok(addr) \
        (((unsigned long)(addr) < (1UL<<47)) || \
         ((unsigned long)(addr) >= HYPERVISOR_VIRT_END))

    #define access_ok(addr, size) \
        (__addr_ok(addr) || is_compat_arg_xlat_range(addr, size))

    #define array_access_ok(addr, count, size) \
        (access_ok(addr, (count)*(size)))

Unfortunately, this assumption is false in XENMEM_exchange. The caller
of this hypercall has full control over the place at which array
processing starts (using the exch.nr_exchanged argument). This means
that the caller can trick the Xen hypervisor into reading input page
numbers from the Xen memory area or write exchanged page numbers into
the Xen memory area. The attacker does not directly control values
written there, but using, for example, balloon driver, can influence
what pages will be available in the system and later chosen for
allocation by XENMEM_exchange.

This means that a sufficiently determined attacker can use this bug to
write arbitrary data into arbitrary places in Xen memory and thereby
take full control of the system.

Discussion
===========

This is another bug resulting from the overly-complex memory
virtualization required for PV in Xen. As we announced last year [5],
the upcoming Qubes OS 4.0 will no longer use PV. Instead, we will be
switching to HVM-based virtualization:

| One of the most important security improvements that we plan to
| introduce with the release of Qubes 4 is to ditch paravirtualization
| (PV) technology and replace it with hardware-enforced memory
| virtualization, which recent processors have made possible thanks to
| so-called Second Level Address Translation (SLAT), also known as EPT
| in Intel parlance. SLAT (EPT) is an extension to Intel VT-x
| virtualization, which originally was capable of only CPU
| virtualization but not memory virtualization and hence required a
| complex Shadow Page Tables approach (which  we believed back then  was
| actually less attractive than the PV approach). We hope that embracing
| SLAT-based memory virtualization will allow us to prevent disastrous
| security bugs, such as the infamous XSA 148, publicly disclosed in
| October of last year, which  unlike many other major Xen bugs
| regrettably did affect Qubes OS. Consequently, we will be requiring
| SLAT support of all certified hardware for Qubes OS 4 and later.

At the same time, we would like to point out that the security of Qubes
OS has so far been affected by less than 10% of publicly disclosed Xen
bugs, as tracked by the recently created Xen Security Advisory (XSA)
Tracker. [6]

Availability of patches
========================

Patched packages will be built and uploaded to the security-testing
repository shortly after this advisory is published. We have recently
implemented and published the details of a new, transparent build
infrastructure. [3] In this new infrastructure, the source code for
packages is pushed to a public repository, and logs from the build
process are also publicly published. However, the Xen security policy
does not permit us to make this data public until after the XSA-212
embargo has been lifted. [4] While we have already privately built
and tested these packages, we must wait until the embargo has been
lifted before transparently building the public packages using our
new infrastructure.

Thanks to the new infrastructure, it's also much easier for us handle
updates. Because of this, we've decided to release fixed packages
for Qubes OS 3.1, even though it recently reached EOL. Both Qubes OS
3.1 and Qubes OS 3.2 use the same Xen version, so this requires no
additional work.

Patching
=========

The specific packages that resolve the problem discussed in this
bulletin will be uploaded to the security-testing repository:

For Qubes 3.1 and 3.2:
* Xen packages, version 4.6.4-26

The packages are to be installed in Dom0 via the qubes-dom0-update
command or via the Qubes VM Manager.

A system restart will be required afterwards.

If you use Anti Evil Maid, you will need to reseal your secret
passphrase to new PCR values, as PCR18 will change due to the new
xen.gz binary.

These packages will migrate to the current (stable) repository over
the coming days after being tested by the community.

Credits
========

This issue was discovered by Jann Horn of Google Project Zero and
reported to the Xen Security Team.


References
===========

[1] https://xenbits.xen.org/xsa/advisory-212.html
[2] https://xenbits.xen.org/xsa/
[3] https://github.com/QubesOS/qubes-infrastructure/
[4] https://www.xenproject.org/security-policy.html
[5] https://www.qubes-os.org/news/2016/07/21/new-hw-certification-for-q4/
[6] https://www.qubes-os.org/security/xsa/

--
The Qubes Security Team
https://www.qubes-os.org/security/