Re: [PATCH V4 01/20] rv: Add Runtime Verification (RV) interface

From: Tao Zhou
Date: Sun Jul 10 2022 - 11:10:47 EST


On Fri, Jul 08, 2022 at 04:39:47PM +0200,
Daniel Bristot de Oliveira wrote:

> Hey Tao!
>
> On 7/6/22 19:49, Tao Zhou wrote:
> >> +static void *enabled_monitors_start(struct seq_file *m, loff_t *pos)
> >> +{
> >> + struct rv_monitor_def *m_def;
> >> + loff_t l;
> >> +
> >> + mutex_lock(&rv_interface_lock);
> >> + m_def = list_entry(&rv_monitors_list, struct rv_monitor_def, list);
> > I realized this m_def is not real but vain. Is it possible the loop is
> > skiped and just return m_def that is not valid.
>
> that is empty... not a problem.
>
> I am not seeing (the possible) problem here. Could you simulate/reproduce the problem?

The @*pos of enable_monitors_start() can not be -1 or other negative value.
And I checked that the *pos is 0(right?). That is safe. Sorry for not being
that ture and maybe this is a notice here. Because if it is a negative value,
the returned m_def is a point to a data place 16 bytes before &rv_monitors_list.
That is a not ture rv_monitors_list stucture data. But it is not possiable now.
Maybe "inspired" from your question. Look it more, I image this simulation.
If the monitor(and all is enabled) is more enough to let the *pos to increase
to -1. And the returned m_def is last monitor that returned from enable_monitors_start().
The enable_monitors_next() check from the last monitor and return NULL.
Only show the last monitor. This will not really happen I think.
But I am not focus enough to the seq file code or others now, so this may be
more possible to be not right. Late reply continued from me..

Thanks,
Tao