Re: [PATCHv4 2/2] mm/page_poisoning.c: Allow for zero poisoning

[Laura Abbott]

On 03/04/2016 04:07 PM, Andrew Morton wrote:
> On Fri,  4 Mar 2016 15:50:48 -0800 Laura Abbott <labbott@xxxxxxxxxxxxxxxxx> wrote:
>
> > By default, page poisoning uses a poison value (0xaa) on free. If this
> > is changed to 0, the page is not only sanitized but zeroing on alloc
> > with __GFP_ZERO can be skipped as well. The tradeoff is that detecting
> > corruption from the poisoning is harder to detect. This feature also
> > cannot be used with hibernation since pages are not guaranteed to be
> > zeroed after hibernation.
> >
> > Credit to Grsecurity/PaX team for inspiring this work
> >
> > --- a/kernel/power/hibernate.c
> > +++ b/kernel/power/hibernate.c
> > @@ -1158,6 +1158,22 @@ static int __init kaslr_nohibernate_setup(char *str)
> >   	return nohibernate_setup(str);
> >   }
> >
> > +static int __init page_poison_nohibernate_setup(char *str)
> > +{
> > +#ifdef CONFIG_PAGE_POISONING_ZERO
> > +	/*
> > +	 * The zeroing option for page poison skips the checks on alloc.
> > +	 * since hibernation doesn't save free pages there's no way to
> > +	 * guarantee the pages will still be zeroed.
> > +	 */
> > +	if (!strcmp(str, "on")) {
> > +		pr_info("Disabling hibernation due to page poisoning\n");
> > +		return nohibernate_setup(str);
> > +	}
> > +#endif
> > +	return 1;
> > +}
>
> It seems a bit unfriendly to silently accept the boot option but not
> actually do anything with it.  Perhaps a `#else pr_info("sorry")' is
> needed.
>
> But I bet we made the same mistake in 1000 other places.
>
> What happens if page_poison_nohibernate_setup() simply doesn't exist
> when CONFIG_PAGE_POISONING_ZERO=n?  It looks like
> kernel/params.c:parse_args() says "Unknown parameter".

I didn't see that behavior when I tested, even with nonsense parameters.
It looks like it might fall back to some other behavior before giving
-ENOENT?

It's also worth noting the page_poison= option is also parsed in
mm/page_poison.c to do other on/off of the poisoning feature. The
option code supported it and it seemed to match better with what the
existing hibernate code was doing with turning off options.

Thanks,
Laura