Contents
ChangesEdit
Since 028stab064.8:
- Rebase to 2.6.18-164.2.1.el5 kernel, number of CVEs fixed (see below), ext4, driver updates, etc.
- vzquota: ext4 support
- Bug fixes for the following issues:
- a Hardware Node may hang while printing the calltraces for all processes (caused by Alt+SysRQ+p magic key interrupt) due to incorrect NMI handling
- the default size of "tmpfs" filesystem was set to half amount of Hardware Node's memory
- two different UDP sockets may be bound to the same port inside a Container which in particular results in "aim7" test suite fail
- a Container based on Ubuntu 8.04 template may fail to migrate (checkpoint) with the following message: "Error: d_path is invisible /dev" (OpenVZ Bug #1198)
- the control script for the Oracle CSS daemon ('init.cssd', a part of Oracle 11g Release 1 RAC software) may fail to work and fill logs with the following messages: "Oprocd detected system hang while in nonfatal mode. Respawning."
- a kernel crash may occur after Packet Generator tool kernel module (pktgen.ko) is loaded (OpenVZ Bug #1325)
- the sched_getcpu() function may return incorrect number if current CPU in case a process was migrated to another CPU via user request (OpenVZ Bug #1149)
- DRBD update up to v8.3.4
- Improvements:
- memory reclaiming algorithm has been modified, which resulted in significant NUMA nodes performance improvement
- "magic" Alt+SysRQ keys management has been enhanced: now disabling SysRQ keys (via
echo 0 > /proc/sys/sysrq) also disables the ability to execute SysRQ key handlers via the /proc/sysrq-trigger file - a kmemsize debugging mechanism has been added, which simplifies kmemsize leaks investigation
- veth driver statistic has been enhanced to report correct "tx_dropped" packets value
- UDP hash table size has been significantly increased to speed up UDP traffic handling, which is important for example for Containers running Asterisk servers especially in the Full Proxy mode. Alternatively, UDP hash table size can be configured via
uhash_entries=kernel boot parameter - softirq statistics can be now monitored via the /proc/softirqs file
- root on a Hardware Node may now run some new processes even if the overall number of processes/threads on the Node exceeds the value of
kernel.threads-maxsysctl. In particular, this feature allows root to login to an overloaded Hardware Node
CompatibilityEdit
- No new issues
ConfigsEdit
- No OpenVZ-specific changes