This chapter describes fixed issues for the Unbreakable Enterprise Kernel Release 4.

• aacraid driver issue that caused a kernel panic fixed

  An issue that caused a kernel panic when the Adaptec RAID controller was loaded and the system attempted to perform an asynchronous scan of the SCSI disks is fixed by an upstream patch to the aacraid driver. The fix initializes the SCSI shared tag map when the driver is loaded.

• i40e driver module fixed for DHCP requests when bridged with a virtio network interface

  An issue in the i40e driver module that caused DHCP to fail for a KVM guest when bridged with a virtio network interface has been fixed.

• lpfc driver module fixed for an issue connecting to targets that send PRLI under P2P mode

  An issue in the lpfc driver module has been patched to enable HBAs to connect to targets that send a Process Login (PRLI) in P2P mode. The issue prevented the resending of the PRLI and resulted in the connection being rejected.

• mgag200 driver issue that caused random system hangs fixed

  An issue in the Matrox GP200e4 driver that caused random system hangs is fixed by an upstream patch to the driver.

• xfs: Directory readahead completions fixed for unmount issue

  A bug that caused a system to hang if an XFS formatted file system is unmounted suddenly after mount has been fixed. A mechanism has been added to track all in-flight, asynchronous buffers using per-cpu counters and to explicitly exclude buffers from the I/O accounting.

• Memory management code for virtual memory area adjustment fixed for use-after-free bug

  A use-after-free issue in the Linux memory management code that resulted when memory allocation failed in vma_adjust() is fixed. This bug was replicated when running fdisk -l, which caused a kernel panic.

• RDS patched to fix memory allocation for scatter-gather based on message sizes

  Reliable Datagram Sockets (RDS) protocol support has been patched to fix a bug that was resulting in page allocation failures, by reducing the memory footprint in rds_sendmsg. Since InfiniBand supports scatter-gather in hardware, RDS fragment sizes don't need to match page size, which was causing unnecessary pressure on the memory allocation system. The fix only allocates the buffer according to RDS_MAX_FRAG_SIZE.

• RDS patched to fix a portstate transition issue that caused ports to get stuck in the INIT state

  RDS is patched for a rare race condition between RDS module initialization and links coming up late, that resulted in the state not getting initialized correctly. In most cases this problem would not occur but could trigger if a host and switch were rebooted at the same time. The issue appeared after reboot as one or more RDS InfiniBand ports never transitioned out of INIT state. If the initialization did not complete correctly, subsequent port failovers would fail and disrupt cluster failover and failback robustness.

• InfiniBand code patched to resolve a race condition that occurred when accessing FMR resources

  Code for InfiniBand was patched to resolve a race condition that resulted when accessing Fast Memory Registration (FMR) resources in the InfiniBand FMR pool. This resulted in a bad syncing issue that could cause a system crash or reboot due to deadlock on a hash list lookup.

• IPoIB patched to improve ioctl handling

  Code for InfiniBand was patched to improve ioctl handling. The code was unable to fully handle the IPoIB data structure and was triggering a warning message in /var/log.messages:

  ib0: ioctl fail to copy request data

  The patch ensures that only unsupported structures trigger the warning and only if debug is enabled.

• Hyper-V fcopy processes fixed to allow copying large files from host to guest

  Patches and driver updates applied to the Hyper-V driver modules have resolved an issue that caused trouble for Oracle Linux guests, running UEK R4 and hosted on Microsoft® Windows 2012 R2 Hyper-V systems, when copying large files from the host to the guest virtual machine.

• Hyper-V service start-up issue resolved

  Patches and driver updates applied to the Hyper-V driver modules have resolved an issue that prevented the hypervkvpd and hypervvssd services from starting after reboot. Note that in previous release notes for Oracle Linux 7, the recommended workaround for this issue was to downgrade packages and to modify /etc/yum.conf to exclude the Hyper-V packages from future updates. In this update release of UEK R4, the issue is resolved and you can safely upgrade Hyper-V packages to their most current version.

• Xen problem hot adding CPU for live Guest VM fixed

  A patch was applied to the Xen event handling code to fix the potential for soft lock-ups when adding CPUs to a guest VM.

• Xen networking fix for stalling receive during network stress

  A patch was applied to the Xen netfront driver to resolve an issue that caused virtual NICs to stall when memory usage was high and where available memory was limited. This high load and out-of-memory scenario caused the allocation of sk_buff in the xennet_alloc_rx_buffers() to fail, resulting in a stall.

• IPv4 code fixed for an issue that crashed dom0 running on Xen

  An upstream patch that was applied to the IPv4 code to use coalescing during defragmentation was reverted to fix an issue that caused dom0 to crash in the Xen hypervisor when handling some network traffic. The original patch helped to reduce overhead and simplify processing for the receiving socket, however it also destroyed the geometry of the original packet fragments. In some situations, it is possible that fastpath processing can result in fragment mismatches.