Scientific Linux Security Update : kernel on SL5.x i386/x86_64
High Nessus Plugin ID 60477
SynopsisThe remote Scientific Linux host is missing one or more security updates.
DescriptionSecurity fixes :
- a missing capability check was found in the Linux kernel do_change_type routine. This could allow a local unprivileged user to gain privileged access or cause a denial of service. (CVE-2008-2931, Important)
- a flaw was found in the Linux kernel Direct-IO implementation. This could allow a local unprivileged user to cause a denial of service. (CVE-2007-6716, Important)
- Tobias Klein reported a missing check in the Linux kernel Open Sound System (OSS) implementation. This deficiency could lead to a possible information leak.
- a deficiency was found in the Linux kernel virtual filesystem (VFS) implementation. This could allow a local unprivileged user to attempt file creation within deleted directories, possibly causing a denial of service. (CVE-2008-3275, Moderate)
- a flaw was found in the Linux kernel tmpfs implementation. This could allow a local unprivileged user to read sensitive information from the kernel.
Bug fixes :
- when copying a small IPoIB packet from the original skb it was received in to a new, smaller skb, all fields in the new skb were not initialized. This may have caused a kernel oops.
- previously, data may have been written beyond the end of an array, causing memory corruption on certain systems, resulting in hypervisor crashes during context switching.
- a kernel crash may have occurred on heavily-used Samba servers after 24 to 48 hours of use.
- under heavy memory pressure, pages may have been swapped out from under the SGI Altix XPMEM driver, causing silent data corruption in the kernel.
- the ixgbe driver is untested, but support was advertised for the Intel 82598 network card. If this card was present when the ixgbe driver was loaded, a NULL pointer dereference and a panic occurred.
- on certain systems, if multiple InfiniBand queue pairs simultaneously fell into an error state, an overrun may have occurred, stopping traffic.
- with bridging, when forward delay was set to zero, setting an interface to the forwarding state was delayed by one or possibly two timers, depending on whether STP was enabled. This may have caused long delays in moving an interface to the forwarding state. This issue caused packet loss when migrating virtual machines, preventing them from being migrated without interrupting applications.
- on certain multinode systems, IPMI device nodes were created in reverse order of where they physically resided.
- process hangs may have occurred while accessing application data files via asynchronous direct I/O system calls.
- on systems with heavy lock traffic, a possible deadlock may have caused anything requiring locks over NFS to stop, or be very slow. Errors such as 'lockd: server [IP] not responding, timed out' were logged on client systems.
- unexpected removals of USB devices may have caused a NULL pointer dereference in kobject_get_path.
- on Itanium-based systems, repeatedly creating and destroying Windows guests may have caused Dom0 to crash, due to the 'XENMEM_add_to_physmap' hypercall, used by para-virtualized drivers on HVM, being SMP-unsafe.
- when using an MD software RAID, crashes may have occurred when devices were removed or changed while being iterated through. Correct locking is now used.
- break requests had no effect when using 'Serial Over Lan' with the Intel 82571 network card. This issue may have caused log in problems.
- on Itanium-based systems, module_free() referred the first parameter before checking it was valid. This may have caused a kernel panic when exiting SystemTap.
SolutionUpdate the affected packages.