[LU-8321] kernel update [RHEL7.2 3.10.0-327.22.2.el7]

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.9.0
• Severity: 3
• Assignee: Bob Glossman (Inactive)
• Reporter: Bob Glossman (Inactive)

Description

Security Fixes:

• A flaw was found in the way certain interfaces of the Linux kernel's
  Infiniband subsystem used write() as bi-directional ioctl() replacement,
  which could lead to insufficient memory security checks when being invoked
  using the splice() system call. A local unprivileged user on a system
  with either Infiniband hardware present or RDMA Userspace Connection
  Manager Access module explicitly loaded, could use this flaw to escalate
  their privileges on the system. (CVE-2016-4565, Important)

• A race condition flaw was found in the way the Linux kernel's SCTP
  implementation handled sctp_accept() during the processing of heartbeat
  timeout events. A remote attacker could use this flaw to prevent further
  connections to be accepted by the SCTP server running on the system,
  resulting in a denial of service. (CVE-2015-8767, Moderate)

Bug Fixes:

• When Small Computer System Interface (SCSI) devices were removed or
  deleted, a system crash could occur due to a race condition between listing
  all SCSI devices and SCSI device removal. The provided patch ensures that
  the starting node for the klist_iter_init_node() function is actually a
  member of the list before using it. As a result, a system crash no longer
  occurs in the described scenario. (BZ#1333403)

• This update offers a reworked series of patches for the resizable hash
  table (rhashtable) including a number of backported bug fixes and
  enhancements from upstream. (BZ#1328801)

• Previously, the same value of the mperf Model-Specific Register (MSR)
  read twice in a row could lead to a kernel panic due to the divide-by-zero
  error. The provided patch fixes this bug, and the kernel now handles two
  identical values of mperf gracefully. (BZ#1334438)

• When a transparent proxy application was running and the number of
  established connections on the computer exceeded one million, unrelated
  processes, such as curl or ssh, were unable to bind to a local IP on the
  box to initiate a connection. The provided patch fixes the cooperation of
  the REUSEADDR/NOREUSEADDR socket option, and thus prevents the local port
  from being exhausted. As a result, the aforementioned bug no longer occurs
  in the described scenario. (BZ#1323960)

• Previously, the kernel support for non-local bind for the IPv6 protocol
  was incomplete. As a consequence, an attempt to bind a socket to an IPv6
  address that is not assigned to the host could fail. The provided patch
  includes changes in the ip_nonlocal_bind variable, which is now set to
  allow binding to an IPv6 address that is not assigned to the host. As a
  result, Linux servers are now able to bind to non-local IPv6 addresses as
  expected. (BZ#1324502)

• On some servers with a faster CPU, USB initialization could previously
  lead to a kernel hang during boot. If this inconvenience occurred when
  booting the second kernel during the kdump operation, the kdump service
  failed and the vmcore was lost. The provided upstream patch fixes this bug,
  and the kernel no longer hangs after USB initialization. (BZ#1327581)

• Previously, when running iperf servers using the mlx4_en module, a kernel
  panic occurred. The underlying source code has been fixed, and the kernel
  panic no longer occurs in the described scenario. (BZ#1327583)

Bugs fixed (https://bugzilla.redhat.com/):

1297389 - CVE-2015-8767 kernel: SCTP denial of service during timeout
1310570 - CVE-2016-4565 kernel: infiniband: Unprivileged process can overwrite kernel memory using rdma_ucm.ko