Changes
Since 042stab072.10:
- [security] Rebase to 2.6.32-279.22.1.el6 RHEL6.3 kernel (security and bug fixes, see link below)
- [security/utrace] utrace: ensure arch_ptrace/ptrace_request can never race with SIGKILL, ported from 2.6.32-358.0.1.el6 RHEL6.4 kernel (CVE-2013-0871) (RHSA-2013-0567) (PSBM-18575)
- [net/tg3] kernel panic when netconsole is using (PCLIN-31420)
- [mm/vswap] kernel panic in handle_pte_fault() due to do_vswap_page() race with reclaimer (OVZ 2491)
- [vzquota/nfs] nfs client quota access to inode reserved space is protected by inode lock, preventing possible races (found during PSBM-17619 investigation (not a fix))
- [sched] do not throttle on cond_resched(), preventing one more case of possible priority inversion (PSBM-15441)
- [cgroup] calling release agent for already dead khelper caused a warning. This could happened if Fedora 18 CT is stopped with "--fast". (PSBM-17915)
- [sched] running tasks could be throttled and never unthrottled thus causing random node hangs. (PSBM-17658)
- [cgroup] added possibility to define the CT0 minimal cpu units in %% of all units assigned by all CTs on the Node (min_shares_pct)
- [fence-watchdog] kernel fencing watchgod is implemented.
- [fence-watchdog] The ability to choose, which action to perform in case of timeout. The configuration is done via sysfs files: /sys/kernel/watchdog_available_actions and /sys/kernel/watchdog_action (PSBM-13631)
- [proc] several issues on error paths have been fixed in proc_fill_super() (PSBM-17869)
- [fuse] background requests accounting has been fixed (PSBM-18114)
- [sched] possible node lockups have been fixed by correcting throttling algorithm (PCLIN-31507)
- [sched] the number of context switches should be reported correctly inside a CT in /proc/stat (PSBM-18113)
- [sched] the number of processes should be reported correctly inside a CT in /proc/stat (PSBM-18113)
- [sched] fixed another case of cpu priority inversion (PCLIN-31507)
- [sched] corrected handling of cpu shares settings (PSBM-18193)
- [fs] added missing mnt_drop_write(), its absence triggered a warning (PSBM-17612)
- [netlink] netlink should advertise userspace about incomplete dumps (needed to fix PSBM-17633)
- [mm/debug] added some useful vmstat counters into sysrq-m output.
- [mm/vmscan] fixed possible races in slab shrinker batching
- [cpt] preserve old failcounters during online migration in case the appropriate beancounter already exists on the destination node (PSBM-17465)
- [ve/net] fixed kernel panic in case venetdev module is unloaded (#2487)
- [ext4] removed ext4 "optimize_fsync" attribute, leaving the optimization code disabled, we do not need this optimization anymore and it's broken (PSBM-16762)
- [block/stats] %WA statistic in 'top' should account the time spent in fdatasync() call (PSBM-11692)
- [cpt] debug prints corrected: unfrozen tasks should be printed before CT wakeups (debug for PSBM-17361)
- [fuse] code has been reviewed and made working after rebase from 2.6.32-279.19.1.el6 to 2.6.32-279.22.1.el6 (PSBM-18215)
- [fs] handle "hidden" mount options properly, required for secondary ploop inside a CT feature (required for PSBM-15338)
- [timers] corrected TFD_TIMER_ABSTIME timer handling, the issue led to high cpu usage inside a Fedora 18 CT by 'init' process (PSBM-18284)
- [fuse] one more kernel panic due to incorrect fuse rebase on new RHEL kernel update fixed (PSBM-18347)
- [fs] deadlock in unix_bind() has been fixed (PSBM-18292)
- [fuse] fuse patches cleanups were performed (PSBM-18211)
- [fs] get_active_super/umount race has been fixed (backport from mainstream) (PSBM-18397)
- [ploop] possible ploop image corruption has been fixed (PSBM-18293)
- [ploop] possible deadlock on ploop snapshot has been fixed (PSBM-18376)
- [fs] CAP_FS_FREEZE capability should be used only when fs is really frozen => prevent possible data corruption
- [ext4] fixed ext4_end_aio_dio_nolock() error handling. Before this we could get an unkillable process waiting in ..._aios() on exit.
- [ext4] fixed potential deadlock in ext4_nonda_switch()
- [fs] fixed unbalanced write counter in __dentry_open(). Before this we could get a process stuck in mntput_no_expire() (PSBM-18369)
- [fs] CAP_FS_FREEZE capability should be used only when fs is really frozen => prevent possible data corruption
- [vedev] ioctl(VZCTL_SETDEVPERMS) has been enhanced to pass a device to a CT in a more consistent way: register device in CT' devtmpfs, make hardlink in CT's /sys/device/... pointing to the host' /sys/device/..., make symlink in CT's /sys/class/{block|char} pointing to device, set device permissions by adding device to CT' devcgroup whitelist (as it used to be before this patchset). This fixes quota issues on ploop for Fedora 18 CT (PSBM-18337)
- [cgroups] cgroups should be automatically destroyed on CT stop (related to PSBM-17934)
- [fence-watchdog] a message is printed if the node is fenced by watchdog (PSBM-18312)
- [fuse] severe kernel memory leak while writing to pStorage (PSBM-18516)