Open main menu

OpenVZ Virtuozzo Containers Wiki β

Changes

Download/kernel/rhel6/042stab057.1/changes

8,617 bytes added, 09:14, 2 July 2012
created
== Changes ==
Since {{kernel link|rhel6|042stab055.16}}:
* Rebase to RHEL6.2 kernel-2.6.32-220.23.1.el6 kernel (security and bug fixes, see below)
* Note that CONFIG_OPENVSWITCH is not set in this kernel
* [cpt] cleanup: unneeded check for <code>PF_EXIT_RESTART</code> removed ({{B|2266}})
* [fuse] <code>fuse_get_context()->pid</code> has been virtualized ({{B|2245}})
* [fuse] [major] kaio support has been added
* [ext4] [major, performance] possibility to submit big BIOs has been added (PCLIN-30841)
** enabled by default, disable by setting <code>/sys/fs/ext4/features/batched_writeback</code> to 0.
* [ext4] fdatasync optimization feature switch has been added (PSBM-12094)
** disable by setting <code>/sys/fs/ext4/features/optimize_fsync</code> to 0
* [sysrq] add sysrq-a, move sched-debug from sysrq-t to it
* [sysctl] Add kernel.hung_task_verbosity sysctl (see /[[proc/sys/kernel/hung_task_verbosity]])
* [ipc] allow <code>mqueue</code> fs to be mounted in container
* [bridge] put brmark check code under <code>CONFIG_BRIDGE</code>
* [cpt] check/fix negative hrtimer expires on nanosleep futex_wait rst
* [fuse] show active requests, open files, connection info in <code>fusectl</code> files
* [fuse] fix handling IO errors
* [fuse] don't leave a page locked in case of write_being error
* [sched] make <code>CONFIG_DEBUG_SPINLOCK_SLEEP</code> work with <code>CONFIG_PREEMPT</code> off
* [ploop] fix handling preallocated space on truncate (PSBM-13082, PSBM-13083)
* [ploop] avoid overflow of 'long' in ploop_bd_full ({{B|2249}})
* [ploop] update modify time if ploop image is mounted rw (PSBM-12847)
* [netdev] simplify dev_hard_start_xmit (PSBM-11782)
* [uevent] send events in correct order according to seqnum (PSBM-11728)
* [ext4] memory leak of biovecs has been fixed (PSBM-13208)
* [ubc] UB maxvalue is shown as "inf" now instead of "9223372036854775807" (on Alt+SysRQ+m)
* [ubc] a lot of internal enhancements/rework in ubc and memory reclaimer code
* [ubc/debug] total count of pruned dentries is shown in /proc/bc/id/debug:dcache_pruned
* [ubc] empty/non-regular shmem inodes are accounted into kmemsize as well (debug for PCLIN-30869)
* [ploop] free blocks discard has been optimized to remove large extents only in case it's enough for freeing asked amount of disk
* [ext4] additional sanity checks has been added to prevent crash on a corrupted fs (PSBM-12241)
* [vzwdog] vzwdog prints now max sched and alloc latencies additionally for the whole sleep period
* [cpt] correct traced processes handling behavior during the checkpoint (PSBM-13065)
* [cpt] correct vzctl hang in case "init" process has died during the restore process (PCLIN-30886)
* [fuse] pending requests queue is flushed at the very end now to make sure it will stay empty
* [fuse] support for file holes in case of writeback enabled has been added (PSBM-13216)
* [fuse] API for userspace reconnecting has been added
* [fairsched] make processes be able to set realtime priority in CT0, this makes pltp testcase happy (PCLIN-29945)
* [ploop] prohibit ploop mounting over ext3, only support ext4 ({{B|2277}}, PSBM-13225)
* [perf/debug] perf tool is now able to collect D-state statistics (PCLIN-30022)
* [kmemleak/debug] make kmemleak be able to work with our kernel (kmemleak is available in debug kernels only)
* [cfq] kernel BUG at block/cfq-iosched.c:386 has been fixed (PCLIN-30890)
* [fuse] added possibility to disable fuse O_DIRECT with help of fs.fuse_ext_caps sysctl (enabled by default) (debug for PSBM-13075)
* [fuse] flush writeback data on file close. In particular this fixes checkpointing (PSBM-13226)
* [fuse] several other bugfixes
* [quota] remounting of FS with quota on ploop could crash host (PSBM-13231)
* [fs/writeback] writeback writer throttler algorithm has been enhanced
* [mount namespaces] mount namespaces support for has been added. This is required for Fedora 17 (PSBM-13004). Note: online migration support has not been added yet
* [cpt] kernel BUG at kernel/cpt/rst_socket.c:407 (on a self-connected socket restore) fixed (PSBM-13367)
* [cpt] sleep-in-atomic at dumping cgroup has been fixes (PSBM-13297)
* [cpt] timerfs irq-safe locking has been fixed
* [cpt] false positive error messages "_rst_get_object: wrong object type" has been fixed (PSBM-13303)
* [fuse] "fs.fuse_ext_caps" sysctl has been removed. Now we need to provide fuse mount options to enable features:
** "writeback_enable" mount option to enable writeback for fuse mount
** "direct_enable" mount option to enable direct IO for fuse mount
* [fuse] show writeback and dio options in proc
* [fuse] file release should make sure writeback is finished
* [iov_iter] direct_IO_page interface has been added. Required for efficient fuse io_kaio (PSBM-12830)
* [oom] oom reports now the current fuse writeback pages counter
* [time] CT uptime is calculated incorrectly in cases like host suspend (PCLIN-30011)
* [fairsched] accuracy of idle/iowait time stats reported through /proc/stat has been increased (PCLIN-30724)
* [fairsched] context switch latency on an almost idle node has been significantly decreased (PSBM-11477)
* [ploop] support for 'expanded' format to io_kaio, support for snapshot, merge and pcopy (migrate) operations has been added (PSBM-12830)
* [ploop] we support only ploop over NFS v3 => ploop mount over NFSv4 has been prohibited (PSBM-13376)
* [ploop] ploop balloon discard could ignore errors and hang (PSBM-13290)
* [ovs/ipv6] kernel ABI has been restored, hypervisor modules should compile fine now (PSBM-13383)
* [ploop] disable ENOSPC checks for preallocated and raw images
* [blkcg] allocate per-cpu statistics from worker thread
* [ipv6] change neighbour table hash for ipv6
* [ext4/xfs] warnings were added to detect a situation when direct reclaim tries to write writeback pages
* [block] sleep while atomic has been fixed (bug introduced in 56.8 kernel)
* [mm] several enhancements on direct reclaim and writeback joint work
* [cpt] CT restore could hang in case of any cpt error handling (PSBM-13440)
* [cpt] deleted /dev/zero file handling has been enhanced during cpt, fixing migration failures on recent kernels. (PSBM-13440)
* [cpt] mount namespaces (for Fedora 17) checkpoint/restore support has been added (PSBM-13004)
* [scheduler] idle time stat reported by /proc/stat has been corrected (PCLIN-30773)
* [scheduler] scheduler bug report has been corrected to print proper calltrace
* [ipv6] deadlock on network interface shutdown fixed (PSBM-13449)
* [watchdog] prohibit access to hardware watchdog devices from inside CTs ({{B|2286}})
* [mm] BUG_ON on freeing a writeback page has been fixed (PSBM-13075)
* [fuse] some fairness in handling writeback has been added (PSBM-13465)
* [mm] dirty pages reclaim process now can be interrupted by SIGKILL to appropriate process. This will help in case the underlying fs is stuck. (PSBM-13566)
* [cgroups] locking order has been corrected => deadlock has been fixed (PSBM-13297, {{B|2278}})
* [cpt/mqueue/Fedora17] mqueue fs mount inside a CT has been enabled, this fs is needed for Fedora 17. Online migration of a Fedora 17 CT with _empty_ mqueue fs mounted is supported (at the moment we do not know who uses mqueue fs in real life). If mqueue mount is not empty - the online migration fails (PSBM-13213)
* [cpt] a kernel panic has been fixed
* [fuse] fuse_flush() must wait till writeback is completed (PSBM-13697)
* [fuse] synchronous close has been reimplemented
* [fuse] file size handling has been corrected (PSBM-13216)
* [ubc/net] potential infinite loop fixed in case network code
* [pid namespaces] missing BSD_PROCESS_ACCT ifdef added. Required to compile kernel without CONFIG_BSD_PROCESS_ACCT kernel option ({{B|2273}})
* [fairsched] vmstat floating point operation fixed: idle scaling has been enhanced once again (PCLIN-30773)
* [scheduler] idle time should not be accounted on "possible" but offline cpus
* [mm] fix a corner case in memory reclaimer - guarantee to invoke OOM in case of no progress (PCLIN-30976)
* [cpt] Fedora 17 CTs should not migrate to older kernels without mount namespaces support, other CTs should migrate fine (PSBM-13534)
* [cpt] allow migration of tasks with SCHED_IDLE or SCHED_BATCH scheduler policy set ({{B|2283}})
* [fuse] PCS special files i_size handling has been fixed in case writeback is enabled (PSBM-13636)
* [pfcache] lockdep false-positive report fixed. The report is false-positive only in case we do not enable pfcache for pfcache cache.
* [scheduler] compilation without CONFIG_SCHEDSTATS config option fixed ({{B|2299}})

=== See also ===
* {{RHSA|2012:0571}}
* {{RHSA|2012:0743}}