Since 042stab059.7:
- Rebase to 2.6.32-279.2.1.el6 kernel (bug fix update, see below)
- [timers] apps that use sys_clock_nanosleep() to suspend their execution could hang after CT migration.
- [sysfs] /sys/devices/system/cpu has been "virtualized": the "virtualization" consists in creating empty cpu# dir for each possible cpu (+1 on UP systems). This makes glibc think that _SC_NPROCESSORS_CONF always > 1 so that various userspace apps (JAVA) won't be tempted to "optimize" when running inside a CT with --cpus=1. (PSBM-14103, #2206)
- [OOM/debug] added line "oom_score_adj: manual|auto" into /proc/bc/$ID/debug (PSBM-12215)
- [devtmpfs] account devtmpfs inodes in tmpfs in proper context. This fixes the accounting memory leak found by QA vzt-pltp test. (PCLIN-30869)
- [cpt] support for restart blocks used for CLOCK_PROCESS_CPUTIME_ID timers (per-process clock that measures CPU time consumed by all threads in the process) has been added
- [cpt] a race has been fixed which can result in tasks left in frozen state (PSBM-14184)
- [fuse] sync case of direct_IO should send a set of async requests (splitted incoming request) and wait for their completion. (PSBM-14010)
- [ploop] a warning calltrace shown when a ploop goes to "ABORTED" state has been removed as considered to be too noisy (PSBM-14148)
- [net/tg3] ~20% network performance decrease on Broadcom cards has been fixed (PSBM-14240)
- [cgroups] notify_on_release and release_agent features of cgroups have been fixed (#2261)
- [sysctl] sysctl code has been cleaned up in order for sysctl checker to not complain. In particular, several vz specific sysctls were converted to "unnumbered" => sysctl() syscall won't work anymore for them.
- [iptables] if ip conntracks and NAT were disabled on the node, reloading "nf_conntrack" module with parameter ip_conntrack_disable_ve0=0 was enabling them again (PCLIN-31123)
- [nfsd] stopping a CT with NFS server inside caused a kernel BUG (PCLIN-31107)
- [sunrpc] kernel panic on mounting nfs in 2 CTs has been fixed (PCLIN-31080)
- [fuse] O_DIRECT operations vs cached writeback misorder has been fixed (PSBM-14110)
- [ploop] kernel BUG in kaio_fsync_thread() has been fixed (PSBM-14249)
- [net] netns batched cleanup support for veth, venet, and nf_conntrack. This decreases time needed to kill 100 ves by ~7 s (from ~13 to ~6 s) (PSBM-11724)
- [jbd] fix for BUG at fs/jbd/commit.c:325 has been updated. Patches were updated in order to synchronize them with mainstream variants, only warning parts have been modified (PCLIN-30972)
- [scheduler] "WARNING: at kernel/sched_rt.c:13" has been fixed (PCLIN-31148)
- [cpuinfo] if some cpu features are masked using msr-tools (and thus cpuid() does not report them anymore), /proc/cpuinfo inside CTs will also mask these features. Hardware Node /proc/cpuinfo will still report all cpu features possible available on the node (PSBM-14131)
- [cpt] checkpointing code now checks cpu features available for CT using cpuid() => it's possible to mask out some cpu features using msr-tools (PSBM-14131)
- [fuse] async processing of sync direct I/O has been corrected, now we do not use async processing of sync direct I/O in case we are writing beyond the end of the file (PSBM-14282)
x86 (i686, IA32)Edit
x86_64 (AMD64, EM64T)Edit
Source RPMEditKernel patchEdit
This patch is applicable to vanilla Linux 2.6.32 kernel (not to 2.6.32.y),
available from kernel.org.
Official configs of this OpenVZ kernel used to build binary RPMS.