Re: [PATCH] exit: Retain nsproxy for exit_task_work() work entries

[Christian Brauner]

On Thu, Dec 09, 2021 at 02:44:19PM +0100, Michal Koutný wrote:
> On Wed, Dec 08, 2021 at 01:07:54PM -1000, Tejun Heo <tj@xxxxxxxxxx> wrote:
> 
> > +	saved_cred = override_creds(of->file->f_cred);
> >  	ret = cgroup_attach_permissions(src_cgrp, dst_cgrp,
> > -					of->file->f_path.dentry->d_sb, threadgroup);
> > +					of->file->f_path.dentry->d_sb,
> > +					threadgroup, ctx->ns);
> > +	revert_creds(saved_cred);
> 
> I wonder now whether such a wrap shouldn't also be around cgroup_kill()
> too (+ replacement of send_sig() with group_send_sig_info() [1])?

send_sig() isn't used that was changed in response to a review. I'm
confused. 

> 
> This shouldn't break the use case of passing cgroup kill fd to a less
> privileged task for (auto)destruction purposes but on the other hand it
> would prevent subverting the fd to a more privileged confused task to
> kill otherwise disallowed processes.

Kill and freeze only do time permission checking at open. Why would you
introduce another write time check?