Synopsis: Important: kernel security and bug fix update
Advisory ID: SLSA-2016:2766-1
Issue Date: 2016-11-15
CVE Numbers: CVE-2016-2143
* It was found that stacking a file system over procfs in the Linux kernel
could lead to a kernel stack overflow due to deep nesting, as demonstrated
by mounting ecryptfs over procfs and creating a recursion by mapping
/proc/environ. An unprivileged, local user could potentially use this flaw
to escalate their privileges on the system. (CVE-2016-1583, Important)
* It was reported that on s390x, the fork of a process with four page
table levels will cause memory corruption with a variety of symptoms. All
processes are created with three level page table and a limit of 4TB for
the address space. If the parent process has four page table levels with a
limit of 8PB, the function that duplicates the address space will try to
copy memory areas outside of the address space limit for the child
process. (CVE-2016-2143, Moderate)
* Use of a multi-threaded workload with high memory mappings sometiems
caused a kernel panic, due to a race condition between the context switch
and the pagetable upgrade. This update fixes the switch_mm() by using the
complete asce parameter instead of the asce_bits parameter. As a result,
the kernel no longer panics in the described scenario.
* When iptables created the Transmission Control Protocol (TCP) reset
packet, a kernel crash could occur due to uninitialized pointer to the TCP
header within the Socket Buffer (SKB). This update fixes the transport
header pointer in TCP reset for both IPv4 and IPv6, and the kernel no
longer crashes in the described situation.
* Previously, when the Enhanced Error Handling (EEH) mechanism did not
block the PCI configuration space access and an error was detected, a
kernel panic occurred. This update fixes EEH to fix this problem. As a
result, the kernel no longer panics in the described scenario.
* When the lockd service failed to start up completely, the notifier
blocks were in some cases registered on a notification chain multiple
times, which caused the occurrence of a circular list on the notification
chain. Consequently, a soft lock-up or a kernel oops occurred. With this
update, the notifier blocks are unregistered if lockd fails to start up
completely, and the soft lock-ups or the kernel oopses no longer occur
under the described circumstances.
* When the Fibre Channel over Ethernet (FCoE) was configured, the FCoE
MaxFrameSize parameter was incorrectly restricted to 1452. With this
update, the NETIF_F_ALL_FCOE symbol is no longer ignored, which fixes this
bug. MaxFrameSize is now restricted to 2112, which is the correct value.
* When the fnic driver was installed on Cisco UCS Blade Server, the discs
were under certain circumstances put into the offline state with the
following error message: “Medium access timeout failure. Offlining disk!”.
This update fixes fnic to set the Small Computer System Interface (SCSI)
status as DID_ABORT after a successful abort operation. As a result, the
discs are no longer put into the offlined state in the described
– Scientific Linux Development Team