6,534
edits
Changes
created
== Changes ==
Since {{kernel link|rhel6-testing|042stab065.4}}:
<!-- 065.4 to 070.1 -->
* [ext4] fdatasync optimizations are disabled by default (PSBM-16727)
* [fs] a lot of patches backported from mainstream in order to fix possible deadlocks on fs freezing (PSBM-16485)
* [perf] perf stats are functional again (PSBM-16825)
<!-- 070.2 -->
* [cgroups] rcu puts accounting has been fixed (PSBM-16539)
* [mm] fix irq handling while processing Alt+SysRQ+M
* [ubc] various possible crashes fixed caused by extra BC put ({{B|2446}})
* [nfs] prevent nfs module unload in case a CT with enabled nfs feature is running, otherwise this leads to a crash ({{B|2427}})
* [nfs] extra debug has been added to the kernel in order to proceed with PSBM-13851. This is not a fix, this a debug only
* [mm] clear page-checkpointed flag after reusing swapcache, otherwise checkpointing can corrupt user data if this happens in the middle of iterative migration
* [cpt] open files after children processes clone during the restore process. This is needed to migrate online processes with opened children-owned files like /proc/<pid>/... ({{B|2404}})
* [cpt] drop dentries DCACHE_NFSFS_RENAMED flag on CT kill cause otherwise we will be unable to rollback failed migration (files will be already unlinked) if private in on NFS and SimFS is used as container's file system ({{B|2242}})
* [cpt] iterative migration swap pages installation fixed (possible panic avoided)
* [cpt] nf_conntracks expectation class online migration support added (PCLIN-31321)
* [fuse] i_size setting races with fuse_file_fallocate() have been fixed
* [fuse] cleanup errors handling in __fuse_direct_io()
* [scheduler] make cpulimit work like nr cpus limit. So now if you set CPULIMIT=X, scheduler will try to run CT processes on no more than X/100 + 1 physical cpus (PSBM-16856)
* [scheduler] clear boost flag on fork, otherwise a newly created task will get boost, which is unfair
<!-- 070.3 -->
* [cpt] active conntrack expectation online migration support (PCLIN-31321)
* [fs] memory corruption has been fixed which was caused the backported set intended to fix fs freeze (PSBM-16485, PSBM-16987)
<!-- 070.4 is just a rebuild -->
<!-- 070.5 -->
* [fuse] don't update i_size if FALLOC_FL_KEEP_SIZE (PSBM-17114)
* [fs] i_mutex leak on an error path has been fixed (PSBM-17019)
<!-- 070.6 -->
* [cpt] kernel panic has been fixed due to DEAD task restoration attempt (PCLIN-30483, PSBM-17107)
* [net] initialize skb properly in build_skb() in order to prevent a kernel panic ({{B|2401}})
* [ploop] support for several ploop images per CT, including hidden mount options (pfcache,balloon,...) (PSBM-16891, PSBM-16843, PSBM-16839)
* [ploop] ploop minor number has been randomized in order to prevent Container owner to guess the number of Containers running on the Hardware Node. The ploop number as a random from 10000 - 65535 range if possible. (PSBM-16818)
<!-- 070.7 only vzfs -->
<!-- 070.8 -->
* [fs] a hang during filesystem freezing (while preparing a snapshot for example) has been fixed (PSBM-17218)
* [devpts] handle dentry allocation failure properly in devpts_pty_new() ({{B|1758}})
<!-- 070.9 -->
* [x86] an independent vector MONITOR_IPI_VECTOR (internal Monitor IPI vector) has been added for Parallels hypervisor for performance improvement (PSBM-17133)
* [scheduler] some enhancements in order to run processes on no more than assigned CT "vcpus" pcpus has been fixed. In particular, "vcpus" are made more sticky to pcpus, they are not migrated faster than "kernel.sched_vcpu_hotslice" sysctl (5ms by default) (PSBM-16913)
* [scheduler] scheduler has been enhanced to run a lot of lightweight containers (PSBM-17289) (Iterate only through busy groups while balancing load; Reduce rq->lock hold times for large cgroup hierarchies)
* [fuse] unitialized data on writeback should be zeroed
* [fuse] fuse_wait_on_writeback has been disabled for CUSE (a kernel module implementing character device in FUSE way)
* [cpt] debug has been added to investigate checkpointing failures due to timeouts or unexpected signal received (related to PSBM-17015)
<!-- 070.10 -->
* [fs] online migration hang in call_rwsem_down_write_failed() has been fixed: sb->s_umount should be locked before calling freeze_super() (PSBM-17273)
* [net] possible deadlock on fsfreeze in unix_bind() has been fixed
* [ploop] kernel BUG in kaio_fsync_thread() has been fixed (PSBM-17273) (kaio_resubmit() is now ready to handle PLOOP_E_DATA_WBI state)
<!-- 070.11 -->
* [cgroups] kernel-cgroup-add-cgroup_root_mutex.patch from Red Hat kernel-2.6.32-279.14.1.el6 has been applied (PSBM-16529)
Since {{kernel link|rhel6-testing|042stab065.4}}:
<!-- 065.4 to 070.1 -->
* [ext4] fdatasync optimizations are disabled by default (PSBM-16727)
* [fs] a lot of patches backported from mainstream in order to fix possible deadlocks on fs freezing (PSBM-16485)
* [perf] perf stats are functional again (PSBM-16825)
<!-- 070.2 -->
* [cgroups] rcu puts accounting has been fixed (PSBM-16539)
* [mm] fix irq handling while processing Alt+SysRQ+M
* [ubc] various possible crashes fixed caused by extra BC put ({{B|2446}})
* [nfs] prevent nfs module unload in case a CT with enabled nfs feature is running, otherwise this leads to a crash ({{B|2427}})
* [nfs] extra debug has been added to the kernel in order to proceed with PSBM-13851. This is not a fix, this a debug only
* [mm] clear page-checkpointed flag after reusing swapcache, otherwise checkpointing can corrupt user data if this happens in the middle of iterative migration
* [cpt] open files after children processes clone during the restore process. This is needed to migrate online processes with opened children-owned files like /proc/<pid>/... ({{B|2404}})
* [cpt] drop dentries DCACHE_NFSFS_RENAMED flag on CT kill cause otherwise we will be unable to rollback failed migration (files will be already unlinked) if private in on NFS and SimFS is used as container's file system ({{B|2242}})
* [cpt] iterative migration swap pages installation fixed (possible panic avoided)
* [cpt] nf_conntracks expectation class online migration support added (PCLIN-31321)
* [fuse] i_size setting races with fuse_file_fallocate() have been fixed
* [fuse] cleanup errors handling in __fuse_direct_io()
* [scheduler] make cpulimit work like nr cpus limit. So now if you set CPULIMIT=X, scheduler will try to run CT processes on no more than X/100 + 1 physical cpus (PSBM-16856)
* [scheduler] clear boost flag on fork, otherwise a newly created task will get boost, which is unfair
<!-- 070.3 -->
* [cpt] active conntrack expectation online migration support (PCLIN-31321)
* [fs] memory corruption has been fixed which was caused the backported set intended to fix fs freeze (PSBM-16485, PSBM-16987)
<!-- 070.4 is just a rebuild -->
<!-- 070.5 -->
* [fuse] don't update i_size if FALLOC_FL_KEEP_SIZE (PSBM-17114)
* [fs] i_mutex leak on an error path has been fixed (PSBM-17019)
<!-- 070.6 -->
* [cpt] kernel panic has been fixed due to DEAD task restoration attempt (PCLIN-30483, PSBM-17107)
* [net] initialize skb properly in build_skb() in order to prevent a kernel panic ({{B|2401}})
* [ploop] support for several ploop images per CT, including hidden mount options (pfcache,balloon,...) (PSBM-16891, PSBM-16843, PSBM-16839)
* [ploop] ploop minor number has been randomized in order to prevent Container owner to guess the number of Containers running on the Hardware Node. The ploop number as a random from 10000 - 65535 range if possible. (PSBM-16818)
<!-- 070.7 only vzfs -->
<!-- 070.8 -->
* [fs] a hang during filesystem freezing (while preparing a snapshot for example) has been fixed (PSBM-17218)
* [devpts] handle dentry allocation failure properly in devpts_pty_new() ({{B|1758}})
<!-- 070.9 -->
* [x86] an independent vector MONITOR_IPI_VECTOR (internal Monitor IPI vector) has been added for Parallels hypervisor for performance improvement (PSBM-17133)
* [scheduler] some enhancements in order to run processes on no more than assigned CT "vcpus" pcpus has been fixed. In particular, "vcpus" are made more sticky to pcpus, they are not migrated faster than "kernel.sched_vcpu_hotslice" sysctl (5ms by default) (PSBM-16913)
* [scheduler] scheduler has been enhanced to run a lot of lightweight containers (PSBM-17289) (Iterate only through busy groups while balancing load; Reduce rq->lock hold times for large cgroup hierarchies)
* [fuse] unitialized data on writeback should be zeroed
* [fuse] fuse_wait_on_writeback has been disabled for CUSE (a kernel module implementing character device in FUSE way)
* [cpt] debug has been added to investigate checkpointing failures due to timeouts or unexpected signal received (related to PSBM-17015)
<!-- 070.10 -->
* [fs] online migration hang in call_rwsem_down_write_failed() has been fixed: sb->s_umount should be locked before calling freeze_super() (PSBM-17273)
* [net] possible deadlock on fsfreeze in unix_bind() has been fixed
* [ploop] kernel BUG in kaio_fsync_thread() has been fixed (PSBM-17273) (kaio_resubmit() is now ready to handle PLOOP_E_DATA_WBI state)
<!-- 070.11 -->
* [cgroups] kernel-cgroup-add-cgroup_root_mutex.patch from Red Hat kernel-2.6.32-279.14.1.el6 has been applied (PSBM-16529)
