[Propose] Isolate core_pattern in mnt namespace.

Eric W. Biederman ebiederm at xmission.com
Tue Jan 5 07:58:34 UTC 2016


Dongsheng Yang <yangds.fnst at cn.fujitsu.com> writes:

> On 12/24/2015 12:36 AM, Eric W. Biederman wrote:
>> Dongsheng Yang <yangds.fnst at cn.fujitsu.com> writes:
> [...]
>
> Hi Eric,
> 	Happy new year and sorry for the late reply.
>>
>> Given the other constraints on an implementation the pid namespace looks
>> by far the one best suited to host such a sysctl if it is possible to
>> implement safely.
>
> So you think it's better to isolate the core_pattern in pid_namespace,
> am I right?

Roughly.

> But, core_file_path and user_mode_helper_path in core_pattern are much
> more related with mnt_namespace IMO.
>
> Could you help to explain it more?

You need a full complement of namespaces, to execute a user mode helper.

Really roughly you need a namespaced equivalent of kthreadd, with a full
complement of namespaces and cgroups setup in the container.

Further it is necessary to have a clear rule that says which processes
that dump core are affected.    For a hierarchical pid namespace this is
straight forward.  For a mount namespace I don't know how that could be
implemented.

And yes the whole kthreadd thing that user mode helper does to launch a
task is necessary to have a clean and predicatable environment.

Of course the default rule of dropping a file named core in the current
directory of the process that died works for everyone, with no kernel
modifications needed.

Eric


More information about the Containers mailing list