The kernel packages contain the Linux kernel, the core of any Linux operating system.
It was found that the x86 ISA (Instruction Set Architecture) is prone to a denial of service attack inside a virtualized environment in the form of an infinite loop in the microcode due to the way (sequential) delivering of benign exceptions such as #DB (debug exception) is handled. A privileged user inside a guest could use this flaw to create denial of service conditions on the host kernel. (CVE-2015-8104, Important)
A use-after-free flaw was found in the way the Linux kernel's key management subsystem handled keyring object reference counting in certain error path of the join_session_keyring() function. A local, unprivileged user could use this flaw to escalate their privileges on the system. (CVE-2016-0728, Important)
It was found that the fix for CVE-2015-1805 incorrectly kept buffer offset and buffer length in sync on a failed atomic read, potentially resulting in a pipe buffer state corruption. A local, unprivileged user could use this flaw to crash the system or leak kernel memory to user space. (CVE-2016-0774, Moderate)
Red Hat would like to thank the Perception Point research team for reporting the CVE-2016-0728 issue. The security impact of the CVE-2016-0774 issue was discovered by Red Hat.
NMI watchdog of guests using legacy LVT0-based NMI delivery did not work with APICv. Now, NMI works with LVT0 regardless of APICv. (BZ#1244726)
Parallel file-extending direct I/O writes could previously race to update the size of the file. If they executed out-of-order, the file size could move backwards and push a previously completed write beyond the end of the file, causing it to be lost. (BZ#1258942)
The GHES NMI handler had a global spin lock that significantly increased the latency of each perf sample collection. This update simplifies locking inside the handler. (BZ#1280200)
Sometimes, iptables rules are updated along with ip rules, and routes are reloaded. Previously, skb->sk was mistakenly attached to some IPv6 forwarding traffic packets, which could cause kernel panic. Now, such packets are checked and not processed. (BZ#1281700)
The NUMA node was not reported for PCI adapters, which affected every POWER system deployed with Red Hat Enterprise Linux 7 and caused significant decrease in the system performance. (BZ#1283525)
Processing packets with a lot of different IPv6 source addresses caused the kernel to return warnings concerning soft-lockups due to high lock contention and latency increase. (BZ#1285369)
Running edge triggered interrupts with an ack notifier when simultaneously reconfiguring the Intel I/O IOAPIC did not work correctly, so EOI in the interrupt did not cause a VM to exit if APICv was enabled. Consequently, the VM sometimes became unresponsive. (BZ#1287001)
Block device readahead was artificially limited, so the read performance was poor, especially on RAID devices. Now, per-device readahead limits are used for each device, which has improved read performance. (BZ#1287548)
Identical expectations could not be tracked simultaneously even if they resided in different connection tracking zones. Now, an expectation insert attempt is rejected only if the zone is also identical. (BZ#1290093)
The storvsc kernel driver for Microsoft Hyper-V storage was setting incorrect SRB flags, and Red Hat Enterprise Linux 7 guests running on Microsoft Hyper-V were experiencing slow I/O as well as I/O failures when they were connected to a virtual SAN. Now, SRB flags are set correctly. (BZ#1290095)
When a NUMA system with no memory in node 0 was used, the system terminated unexpectedly during boot or when using OpenVSwitch. Now, the kernel tries to allocate memory from other nodes when node 0 is not present. (BZ#1300950)