[CRIU] Dealing with VDSO remap

Laurent Dufour ldufour at linux.vnet.ibm.com
Fri Mar 20 02:44:03 PDT 2015


On 18/03/2015 14:20, Pavel Emelyanov wrote:
>> No my concern was about the remapping of the vDSO and the non updating
>> of kernel's reference to its base.
>>
>> On my ppc64 system, the attached test case leads to a process's core
>> dump when it is returning from the signal handler once the vDSO has been
>> remapped.
> 
> Just for clarification -- is it specific of ARM and PPC? I mean, does it
> crash on x86 too?

There are multiple architectures affected by this issue. It has been
demonstrated on ppc64 and ARM64 but, reading the kernel code, I think
also powerpc (32bits), mips, tile, hexagon, s390 and x86 (32bits) are
also concerned by this sigreturn vDSO's trampoline issue.

X86/64 is not concerned since it always using SA_RESTORER.


>> I'd appreciate if you could give it a try on an ARM64 box/guest. I tried
>> to set up a ARM64 guest on my side, but I failed to make it run :(
>> My though is that the process should core dump on ARM64 since the
>> sigreturn stack frame will point to its old base address.
>>
>> I'm currently working on a kernel patch to handle the vDSO remapping,
>> which could solve part of this issue (there is still a window which
>> can't be addressed).
>>
>> Thanks,
>> Laurent.
>>
> 
> _______________________________________________
> CRIU mailing list
> CRIU at openvz.org
> https://lists.openvz.org/mailman/listinfo/criu
> 



More information about the CRIU mailing list