[CRIU] [PATCH] cr-super: Initial commit

Pavel Emelyanov xemul at parallels.com
Thu Sep 17 06:14:42 PDT 2015


On 09/17/2015 04:04 PM, Florian Weimer wrote:
> On 09/16/2015 04:24 PM, Pavel Emelyanov wrote:
> 
>>> It may have its reasons for that.  If it does not, then the kernel should
>>> be fixed and simply provide access (if the process is dumpable, a check
>>> which is much safer to implement inside the kernel).
>>
>> Of course, but we still have to work on older kernels without the fix.
> 
> What's the reason for that?  

Well, backward compatibility? We claim that criu works on kernels starting from
3.11 and there are people using it in this way. Once we have this fixed in, say,
4.3, what do we do with the older kernel users?

> libffi closures support?  Those memory
> segments are so small they could copy them with ptrace (more tricky is
> that you have to restore aliasing).

It's not only about copying the memory, it's also (and mainly) about getting
the mnt_id for a mapped file.

> Or is checkpointing libraries which use outdated and removed binaries
> functionality that is critical to you?
> 



More information about the CRIU mailing list