On Fri, Aug 31, 2018 at 05:26:03PM -0700, Tejun Heo wrote:
> Hello,
>
> On Thu, Aug 30, 2018 at 09:53:52PM -0400, Dennis Zhou wrote:
> > - css = cgroup_get_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys);
> >
> > - return __bio_associate_blkg_from_css(bio, css);
> > + rcu_read_lock();
> > +
> > + while (true) {
> > + css = cgroup_e_css(page->mem_cgroup->css.cgroup,
> > + &io_cgrp_subsys);
>
> So, while they seem similar cgroup_e_css() and cgroup_get_e_css()
> behave very differently in terms of locking. cgroup_e_css() can only
> be used under cgroup_mutex because it is used during migration and has
> to test cgroup_ss_mask(). The right thing to do here would be
> renaming cgroup_e_css() to something else and add a new implementation
> which operates in the same way as cgroup_get_e_css().
>
> BTW, this should have triggered lockdep warning. I'd strongly
> recommend testing with lockdep enabled.
>
> Other than that, looks good to me.
>
I see. I've renamed the original cgroup_e_css() to
cgroup_e_css_by_mask() and then did what cgroup_get_e_css() did without
the get part in the new cgroup_e_css().
Thanks,
Dennis
