Re: [PATCH v1] trace: Fix race in trace_open and buffer resize call

[Greg KH <gregkh@linuxfoundation.org>]

On Thu, Jan 21, 2021 at 03:37:32PM -0500, Steven Rostedt wrote:
> On Thu, 21 Jan 2021 23:15:22 +0300
> Denis Efremov <efremov@linux.com> wrote:
> 
> > On 1/21/21 10:09 PM, Steven Rostedt wrote:
> > > On Thu, 21 Jan 2021 17:30:40 +0300
> > > Denis Efremov <efremov@linux.com> wrote:
> > >   
> > >> Hi,
> > >>
> > >> This patch (CVE-2020-27825) was tagged with
> > >> Fixes: b23d7a5f4a07a ("ring-buffer: speed up buffer resets by avoiding synchronize_rcu for each CPU")
> > >>
> > >> I'm not an expert here but it seems like b23d7a5f4a07a only refactored
> > >> ring_buffer_reset_cpu() by introducing reset_disabled_cpu_buffer() without
> > >> significant changes. Hence, mutex_lock(&buffer->mutex)/mutex_unlock(&buffer->mutex)
> > >> can be backported further than b23d7a5f4a07a~ and to all LTS kernels. Is
> > >> b23d7a5f4a07a the actual cause of the bug?
> > >>  
> > > 
> > > Ug, that looks to be a mistake. Looking back at the thread about this:
> > > 
> > >   https://lore.kernel.org/linux-arm-msm/20200915141304.41fa7c30@gandalf.local.home/  
> > 
> > I see from the link that it was planned to backport the patch to LTS kernels:
> > 
> > > Actually we are seeing issue in older kernel like 4.19/4.14/5.4 and there below patch was not 
> > > present in stable branches:
> > > Commit b23d7a5f4a07 ("ring-buffer: speed up buffer resets by avoiding synchronize_rcu for each CPU")  
> > 
> > The point is that it's not backported yet. Maybe because of Fixes tag. I've discovered
> > this while trying to formalize CVE-2020-27825 bug in cvehound
> > https://github.com/evdenis/cvehound/blob/master/cvehound/cve/CVE-2020-27825.cocci
> > 
> > I think that the backport to the 4.4+ should be something like:
> > 
> > diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
> > index 547a3a5ac57b..2171b377bbc1 100644
> > --- a/kernel/trace/ring_buffer.c
> > +++ b/kernel/trace/ring_buffer.c
> > @@ -4295,6 +4295,8 @@ void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu)
> >  	if (!cpumask_test_cpu(cpu, buffer->cpumask))
> >  		return;
> >  
> > +	mutex_lock(&buffer->mutex);
> > +
> >  	atomic_inc(&buffer->resize_disabled);
> >  	atomic_inc(&cpu_buffer->record_disabled);
> >  
> > @@ -4317,6 +4319,8 @@ void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu)
> >  
> >  	atomic_dec(&cpu_buffer->record_disabled);
> >  	atomic_dec(&buffer->resize_disabled);
> > +
> > +	mutex_unlock(&buffer->mutex);
> >  }
> >  EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu);
> >  
> 
> That could possibly work.

Ok, so what can I do here?  Can someone resend this as a backport to the
other stable kernels in this way so that I can queue it up?

thanks,

greg k-h