Download/kernel/rhel6/042stab063.2/changes
< Download | kernel | rhel6 | 042stab063.2
Changes[edit]
Since 042stab062.2:
- Rebase to 2.6.32-279.11.1.el6 RHEL6.3 kernel (security and bug fixes, see RHSA link below)
- [netfilter] iptables inability to handle > 2000 rules inside 32bit CT on 64bit node (PCLIN-31246)
- [cpuinfo] cpuflags are reported empty inside a CT on a node with single physical cpu. In particular this breaks yum (#2373)
- [cpt] fix raising kernel BUG() on cpt/rst of a ptraced task
- [cpt/delayfs] NFS remote locks handling has been fixed (PSBM-14815)
- [fairsched] a lot of scheduler enhancements mostly intended to improve performance on NUMA nodes, now tries to run CT processes on no more than vcpus configured for the CT (PSBM-14332)
- [framebuffer] Draw Parallels logo if node is booted in graphical mode and "psbm" kernel boot option is provided (PSBM-15257)
- [debug] added the ability to dump slab entries
- [mm debug] added an ability to catch double free in per-cpu page lists. Triggers BUG() if double free detected.
- [net] netlink messages handling code has been fixed (env context corrected). In particular this fixes network configuration using Network Manager (#2353, #2354)
- [fuse] trim 'count' of bytes to read to avoid reading beyond eof when racing with writeback (PSBM-15340)
- [fuse] one more race fixed in writeback (PSBM-15008)
- [cpt] ipv6 module detection error message has been improved
- [cpt] online migration failure between 2 nodes without ipv6 modules loaded/enabled has been fixed (PSBM-15431)
- [cpt] fixed tty struct leak (PSBM-15373)
- [mm] new sysctl has been introduced: vm.strict_mem_cpuset (=0 by default). If 0, mem cpuset will work as a hint, i.e. if allocation from allowed NUMA nodes fails, cpuset constraints will be ignored (PSBM-15613)
- [ploop] hang on fsync() on ro mount has been fixed (PSBM-15435)
- [fuse] a page-cache corruption has been eliminated
- [kstat] fixed a kernel panic due to cpu hard lockup caused by a deadlock in gathering statistics code (PCLIN-31259)
- [ubc/mm] a special hack for restricting backup tools and preventing cache thrashing. Container with CTID of /proc/sys/ubc/tools_bcid is marked so that CT reclaimer will not move its pages to HN inactive list once they were reclaimed in the CT. This should prevent backup from trashing HN cache (PSBM-15797)
- [tty] BUG_ON from n_tty_read() has been removed (mainstream patch backport). This is not a real fix, but thus we eliminate the panic (#2381)
- [sunrpc] destroy virtuzalized data moved to khelper, this eliminates possible fail to stop a CT with kernel workqueue (rpciod) in D state (PSBM-15630)
- [oom_score_adj] fix printing of oom_score_adj (PSBM-15718)
- [oom_score_adj] set a minimum adjustment for "init", it should be killed last in CT (PSBM-15718)
- [ubc/dcache] postpone oom if dcache still shrinkable (if dcache reclaimer reports real or possible future progress). Plus, show amount of currenly unused dcache entries in /proc/bc/$BCID/debug (PCLIN-31294)
- [cpt/delayfs] show hidden fs type instead of hardcoded "delayfs". New NFS utils become very clever on umount - they check whether mount point information in /proc/mounts suits some NFS-mount rules. In particular this fixes a failures on CT stop with NFS volume mounted inside (PCLIN-31282)
- [fuse] fixed one more page-cache corruption by fuse (If a page is under fuse-writeback now, fuse_launder_page() must wait on its completion)
- [fairsched] since now we boost throttled tasks on wakeups (in kernel) and throttle them only when switching to userspace. This should eliminate priority inversion issues (PSBM-15441)
- [ploop/pstorage] failed mount file system after snapshot creation for RAW ploop fixed (PSBM-15643)
- [drbd] driver updated up to version 8.3.13. Sources were taken from vendor git (#2254)
- [fairsched] do not account io-throttled tasks in loadavg
- [ploop] prevent using logic of reallocation for raw images (it's only suitable for expanded images) (PSBM-15793)
- [cpt/posix_timers] overrun counter migration support added (PSBM-15580, #2324)
