[Devel] Re: [PATCH 4/4] The control group itself

Paul Jackson pj at sgi.com
Tue Feb 12 18:32:15 PST 2008


Serge wrote:
> Paul (actually both Menage and Jackson :) do you have an opinion on
> this?  Are there sites which eg do 'chown -R some_user_id /cgroup/cpusets/'
> to have some non-root user be able to dole out cpusets?  Is there any
> way it would be ok to have cgroup_file_write() check for CAP_SYS_ADMIN?

I don't know what my users actually do here ... I'm a couple layers
removed from that reality.  But certainly I've recommended that they
sometimes do things like having the batch scheduler chown the files
of each jobs cpuset to the uid of the user running that job, so that
the job can manipulate its own cpuset allocate resources in finer
detail.

One of the more elaborate ways of doing this nests a pair of cpusets,
with the parent owned by the batch scheduler confining the child
owned by the individual job.  The job can actually do things like
write its own cpus and mems files, but is confined by the parent
cpuset to only specify cpus and mems assigned to that job.

As to how this affects your question ... I'm not sure.  However I
suspect that an added requirement for CAP_SYS_ADMIN would cause
breakage and not be a good idea.

-- 
                  I won't rest till it's the best ...
                  Programmer, Linux Scalability
                  Paul Jackson <pj at sgi.com> 1.940.382.4214
_______________________________________________
Containers mailing list
Containers at lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers




More information about the Devel mailing list