|| ||Troy Dawson <firstname.lastname@example.org> |
|| ||"email@example.com" <firstname.lastname@example.org> |
|| ||Security ERRATA Moderate: libvirt on SL5.x i386/x86_64 |
|| ||Tue, 30 Aug 2011 11:45:08 -0500|
|| ||Article, Thread
Synopsis: Moderate: libvirt security, bug fix, and enhancement update
Issue date: 2011-07-21
CVE Names: CVE-2011-2511
The libvirt library is a C API for managing and interacting with the
virtualization capabilities of Linux and other operating systems.
An integer overflow flaw was found in libvirtd's RPC call handling. An
attacker able to establish read-only connections to libvirtd could
trigger this flaw by calling virDomainGetVcpus() with specially-crafted
parameters, causing libvirtd to crash. (CVE-2011-2511)
This update fixes the following bugs:
* libvirt was rebased from version 0.6.3 to version 0.8.2 in
Scientific Linux 5.6. A code audit found a minor API change that
effected error messages seen by libvirt 0.8.2 clients talking to libvirt
0.7.1 - 0.7.7 (0.7.x) servers. A libvirt 0.7.x server could send
VIR_ERR_BUILD_FIREWALL errors where a libvirt 0.8.2 client expected
VIR_ERR_CONFIG_UNSUPPORTED errors. In other circumstances, a libvirt
0.8.2 client saw a "Timed out during operation" message where it should
see an "Invalid network filter" error. This update adds a backported
patch that allows libvirt 0.8.2 clients to interoperate with the API as
used by libvirt 0.7.x servers, ensuring correct error messages are sent.
* libvirt could crash if the maximum number of open file descriptors
(_SC_OPEN_MAX) grew larger than the FD_SETSIZE value because it accessed
file descriptors outside the bounds of the set. With this update the
maximum number of open file descriptors can no longer grow larger than
the FD_SETSIZE value.
* A libvirt race condition was found. An array in the libvirt event
handlers was accessed with a lock temporarily released. In rare cases,
if one thread attempted to access this array but a second thread
reallocated the array before the first thread reacquired a lock, it
could lead to the first thread attempting to access freed memory,
potentially causing libvirt to crash. With this update libvirt no longer
refers to the old array and, consequently, behaves as expected.
* Guests connected to a passthrough NIC would kernel panic if a
system_reset signal was sent through the QEMU monitor. With this update
you can reset such guests as expected.
* When using the Xen kernel, the rpmbuild command failed on the
xencapstest test. With this update you can run rpmbuild successfully
when using the Xen kernel.
* When a disk was hot unplugged, "ret >= 0" was passed to the
qemuAuditDisk calls in disk hotunplug operations before ret was, in
fact, set to 0. As well, the error path jumped to the "cleanup" label
prematurely. As a consequence, hotunplug failures were not audited and
hotunplug successes were audited as failures. This was corrected and hot
unplugging checks now behave as expected.
* A conflict existed between filter update locking sequences and virtual
machine startup locking sequences. When a filter update occurred on one
or more virtual machines, a deadlock could consequently occur if a
virtual machine referencing a filter was started. This update changes
and makes more flexible several qemu locking sequences ensuring this
deadlock no longer occurs.
* qemudDomainSaveImageStartVM closed some incoming file descriptor (fd)
arguments without informing the caller. The consequent double-closes
could cause Domain restoration failure. This update alters the
qemudDomainSaveImageStartVM signature to prevent the double-closes.
This update also adds the following enhancements:
* The libvirt Xen driver now supports more than one serial port.
* Enabling and disabling the High Precision Event Timer (HPET) in Xen
domains is now possible.
All libvirt users should install this update which addresses this
vulnerability, fixes these bugs and adds these enhancements. After
installing the updated packages, libvirtd must be restarted ("service
libvirtd restart") for this update to take effect.
- Scientific Linux Development Team
to post comments)