Synopsis: Important: kernel security and bug fix update
Advisory ID: SLSA-2018:0512-1
Issue Date: 2018-03-13
CVE Numbers: CVE-2017-5753
* hw: cpu: speculative execution branch target injection (s390-only)
* hw: cpu: speculative execution bounds-check bypass (s390 and powerpc)
* hw: cpu: speculative execution permission faults handling (powerpc-only)
* If a fibre channel (FC) switch was powered down and then powered on
again, the SCSI device driver stopped permanently the SCSI device's
request queue. Consequently, the FC port login failed, leaving the port
state as "Bypassed" instead of "Online", and users had to reboot the
operating system. This update fixes the driver to avoid the permanent stop
of the request queue. As a result, SCSI device now continues working as
expected after power cycling the FC switch.
* Previously, on final close or unlink of a file, the find_get_pages()
function in the memory management sometimes found no pages even if there
were some pages left to save. Consequently, a kernel crash occurred when
attempting to enter the unlink() function. This update fixes the
find_get_pages() function in the memory management code to not return 0
too early. As a result, the kernel no longer crashes due to this behavior.
* Using IPsec connections under a heavy load could previously lead to a
network performance degradation, especially when using the aesni-intel
module. This update fixes the issue by making the cryptd queue length
configurable so that it can be increased to prevent an overflow and packet
drop. As a result, using IPsec under a heavy load no longer reduces
* Previously, a deadlock in the bnx2fc driver caused all adapters to block
and the SCSI error handler to become unresponsive. As a result, data
transferring through the adapter was sometimes blocked. This update fixes
bnx2fc, and data transferring through the adapter is no longer blocked due
to this behavior.
* If an NFSv3 client mounted a subdirectory of an exported file system, a
directory entry to the mount hosting the export was incorrectly held even
after clearing the cache. Consequently, attempts to unmount the
subdirectory with the umount command failed with the EBUSY error. With
this update, the underlying source code has been fixed, and the unmount
operation now succeeds as expected in the described situation.
The system must be rebooted for this update to take effect.
- Scientific Linux Development Team