(RHSA-2018:2387) Important: kernel security and bug fix update

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

Security Fix(es):

• Modern operating systems implement virtualization of physical memory to efficiently use available system resources and provide inter-domain protection through access control and isolation. The L1TF issue was found in the way the x86 microprocessor designs have implemented speculative execution of instructions (a commonly used performance optimisation) in combination with handling of page-faults caused by terminated virtual to physical address resolving process. As a result, an unprivileged attacker could use this flaw to read privileged memory of the kernel or other processes and/or cross guest/host boundaries to read host memory by conducting targeted cache side-channel attacks. (CVE-2018-3620, CVE-2018-3646)

• An industry-wide issue was found in the way many modern microprocessor designs have implemented speculative execution of Load & Store instructions (a commonly used performance optimization). It relies on the presence of a precisely-defined instruction sequence in the privileged code as well as the fact that memory read from address to which a recent memory write has occurred may see an older value and subsequently cause an update into the microprocessor’s data cache even for speculatively executed instructions that never actually commit (retire). As a result, an unprivileged attacker could use this flaw to read privileged memory by conducting targeted cache side-channel attacks. (CVE-2018-3639)

Red Hat would like to thank Intel OSSIRT (Intel.com) for reporting CVE-2018-3620 and CVE-2018-3646 and Ken Johnson (Microsoft Security Response Center) and Jann Horn (Google Project Zero) for reporting CVE-2018-3639.

Bug Fix(es):

• Previously, configurations with the little-endian variant of IBM Power Systems CPU architectures and Hard Disk Drives (HDD) designed according to Nonvolatile Memory Express (NVMe) open standards, experienced crashes during shutdown or reboot due to race conditions of CPUs. As a consequence, the sysfs pseudo file system threw a stack trace report about an attempt to create a duplicate entry in sysfs. This update modifies the source code so that the irq_dispose_mapping() function is called first and the msi_bitmap_free_hwirqs() function is called afterwards. As a result, the race condition no longer appears in the described scenario. (BZ#1570510)

• When switching from the indirect branch speculation (IBRS) feature to the retpolines feature, the IBRS state of some CPUs was sometimes not handled correctly. Consequently, some CPUs were left with the IBRS Model-Specific Register (MSR) bit set to 1, which could lead to performance issues. With this update, the underlying source code has been fixed to clear the IBRS MSR bits correctly, thus fixing the bug. (BZ#1586147)

• During a balloon reset, page pointers were not correctly initialized after unmapping the memory. Consequently, on the VMware ESXi hypervisor with “Fault Tolerance” and “ballooning” enabled, the following messages repeatedly occurred in the kernel log:

[3014611.640148] WARNING: at mm/vmalloc.c:1491 __vunmap+0xd3/0x100()
[3014611.640269] Trying to vfree() nonexistent vm area (ffffc90000697000)

With this update, the underlying source code has been fixed to initialize page pointers properly. As a result, the mm/vmalloc.c warnings no longer occur under the described circumstances. (BZ#1595600)