RHBA-2016:0554 - Bug Fix Advisory

Topic

Updated kernel-rt packages that fix several bugs are now available for Red Hat Enterprise Linux 7.

Description

The kernel-rt packages contain the Linux kernel, the core of any Linux operating system.

The kernel-rt packages have been upgraded to version 3.10.0-327.13.1, which provides a number of bug fixes including:

• [md] dm-thin-metadata: fix bugs in dm_thin_remove_range()
• [md] dm-thin: fix regression in advertised discard limits
• [scsi] hpsa: Update driver revision to RH2 and fix multilun device problems
• [net] tcp: honour SO_BINDTODEVICE for TW_RST case too
• [net] tcp: send_reset: test for non-NULL sk first
• [net] add inet_sk_transparent() and sk_fullsock() helpers
• [net] ipv6: Fix regression in udp_v6_mcast_next()
• [netdrv] mlx4_en: Remove BUG_ON assert when checking if ring is full
• [netdrv] cxgb4: changes for new firmware 1.14.4.0
• [netdrv] iwlwifi: modify PCI IDs for the 3165, 7265, 8000, and 8260 series
• [mm] memcg: oom_notify use-after-free fix

(BZ#1311476)

This update also fixes the following bugs:

• The realtime kernel was carrying four lines of stale code in the block devices core code and was missing an upstream fix to this same area of code. Consequently, users of the NVM Express device driver (nvme module), and possibly other block devices enabling multi-queue support, experienced a kernel crash at boot, which made the system unusable. These two bugs have been fixed, and the kernel now boots and functions correctly. (BZ#1304862)
• While merging new Fibre Channel over Ethernet (FCoE) code, the get_cpu() and put_cpu() function calls were converted to the get_cpu_light() and put_cpu_light() calls on the realtime kernel. One spot was missed during the code update that caused an imbalance in the get_cpu_light() and put_cpu_light() call pairing, which left preemption disabled. Consequently, this resulted in the "BUG: scheduling while atomic" error. This update ensures that all the remaining function calls are converted correctly, and the FCoE code now works as expected. (BZ#1306958)
• A ksoftirqd kernel thread can serve many requests including network-related requests. Previously, ksoftirqd always ran at a real time priority (FIFO:1) and the timer softirqs (timer and hrtimer) were a part of ksoftirqd. As a consequence, a system under a heavy network load could keep ksoftirqd busy for a long period of time, and prevent important tasks running at lower priorities from getting CPU time. If the read-copy update (RCU) threads were neglected, RCU stalls, out of memory (OOM) situations, or kernel panic could occur. With this update, the code has been modified to match the upstream realtime kernel where the timer softirqs are separated from ksoftirqd, ksoftirqd can run at a lower priority, and the new ktimersoftd thread can also be prioritized according to the specific system load. As a result, the system now works as expected under a heavy network load. (BZ#1306962)

All kernel-rt users are advised to upgrade to these updated packages, which fix these bugs. The system must be rebooted for this update to take effect.

Solution

Before applying this update, make sure all previously released errata
relevant to your system have been applied.

For details on how to apply this update, refer to:

https://access.redhat.com/articles/11258

Affected Products

• Red Hat Enterprise Linux for Real Time 7 x86_64
• Red Hat Enterprise Linux for Real Time for x86_64 - Extended Life Cycle Support 7 x86_64

Fixes

• BZ - 1306962 - backport of: "softirq: split timer softirqs out of ksoftirqd"
• BZ - 1311476 - kernel-rt: update to the RHEL7.2.z batch#3 source tree

CVEs

(none)

References

• http://www.redhat.com/security/updates/classification/#normal