Bugzilla – Bug 1142428
VUL-0: CVE-2019-17351: kernel-source: XSA-300 v2 - Linux: No grant table and foreign mapping limits
Last modified: 2020-07-10 14:07:28 UTC
Xen Security Advisory XSA-300
Linux: No grant table and foreign mapping limits
UPDATES IN VERSION 2
Drop inapplicable "Deployment during embargo" section.
Rewrite for clarity, and to remove most references to dom0. The issue
is equally applicable to domU's providing backend services.
Add information about the arbitrary limit for userspace backends.
Virtual device backends and device models running in domain 0, or
other backend driver domains, need to be able to map guest memory
(either via grant mappings, or via the foreign mapping interface).
Inside Xen, mapped grants are tracked by the maptrack structure. The
size of this structure is chosen during domain creation, and has a
fixed upper bound for the lifetime of the domain.
For Linux to keep track of these mappings, it needs to have a page
structure for each one. In practice the number of page structures is
usually limited. In PV guests, a range of pfns are typically set
aside at boot ("pre-ballooned") for this purpose. For HVM/PVH and Arm
guests, no memory is set aside to begin with. In either case, when
more of this "foreign / grant map pfn space" is needed, Linux will
balloon out extra pages to use for this purpose.
Unfortunately, in Linux, there are no limits, either on the total
amount of memory which the domain will attempt to balloon out, nor on
the amount of "foreign / grant map" memory which any individual guest
For Linux userspace backends (e.g. QEMU) which use /dev/xen/gnttab or
/proc/xen/gnttab, there is an arbitrary mapping limit which, if hit,
will prevent further mappings from being established.
As a result, a malicious guest may be able to, with crafted requests,
cause a backend Linux domain to either:
1) Fill the maptrack table in Xen and/or hit the userspace limit.
This will starve I/O from other guests served by the same backend.
2) Balloon out sufficient RAM to cause it to swap excessively, or run
completely out of memory. This may starve all operations from the
domain, including I/O from other guests, or may cause a crash of
Guest may be able to crash backend Linux domains, or starve operations
inside the domain, including the processing of guest I/O requests
If the backend is domain 0, which is the most common configuration,
then host-wide operations may be starved, or the host may crash (Host
All versions of Linux are vulnerable. Only Linux guests acting as
backend domains for other guests may be exploited.
All Arm domains are vulnerable, as are x86 PVH/HVM guests. The
vulnerability of x86 PV guests depends on how they were configured at
PV guests can be constructed with "pre-ballooned" memory, by building
it with maxmem > memory. See `man 5 xl.cfg` for full details of these
For PV dom0, these are controlled by Xen's "dom0_mem=$X,max:$Y"
command line parameter.
The larger the difference between memory and maxmem, the more space
Linux has to fill with grant/foreign mappings before it will start
ballooning out real memory to satisfy further mapping requests. This
makes the attack more difficult to accomplish.
This issue was discovered by Julien Grall of ARM.
Applying the appropriate attached patch resolves the backend memory
NOTE: This does NOT fix the guest starvation issue. Fixing fixing
this issue is more complex, and it was determined that it was better
to work on a robust fix for the issue in public. This advisory will
be updated when fixes are available.
xsa300-linux-5.2.patch Linux 4.4 ... 5.2
$ sha256sum xsa300*
Created attachment 811242 [details]
Perhaps 1142450 wants to be a duplicate of the one here.
*** Bug 1142450 has been marked as a duplicate of this bug. ***