[PATCH 8/9] [RFC] Example multi-bindable subsystem: a per-cgroup notes field

Paul Menage menage at google.com
Mon Jul 13 17:49:56 PDT 2009

On Fri, Jul 3, 2009 at 1:58 AM, Li Zefan<lizf at cn.fujitsu.com> wrote:
> If it's not per cgroup, it can be a global value.

No it can't, since the notes subsystem can be mounted on multiple
hierarchies simultaneously.

> But why not make it per cgroup?

Setting notes.size to a large value allows userspace to create large
kmalloc'd objects; so if it's per-cgroup, then it's not safe to
delegate control of part of a cgroups hierarchy to untrusted users.
(E.g. we want to be able to give an untrusted user process the power
to create sub-cgroups in the CPU scheduler hierarchy, so that it can
give different CPU guarantees to each of its threadpools).

I guess that an alternative would be to have a per-cgroup size field,
and use the min of the cgroup and all its ancestors when doing length

>> +     if (nbytes > cg_info(cgrp->top_cgroup)->max_len)
>> +             return -E2BIG;
>> +     if (nbytes) {
>> +             notes = kmalloc(nbytes, GFP_USER);
>> +             if (!notes)
>> +                     return -ENOMEM;
>> +             if (copy_from_user(notes, userbuf, nbytes))
> missing kfree(notes)

Good catch, thanks. Fixed (and all the newlines added).


More information about the Containers mailing list