Re: [PATCH v2] x86: kaslr to avoid setup_data regions

[Dave Young <dyoung@redhat.com>]

On 05/14/14 at 04:23pm, Dave Young wrote:
> On 05/14/14 at 10:13am, Mateusz Guzik wrote:
> > On Wed, May 14, 2014 at 04:02:01PM +0800, Dave Young wrote:
> > > X86 will pass setup_data region while necessary, these regions could be
> > > overwitten by kernel due to kaslr.
> > > 
> > > Thus iterate and add setup regions to mem_avoid[] in this patch.
> > > Set max mem avoid entries 32, hopefully it will be enough.
> > > 
> > > Signed-off-by: Dave Young <dyoung@redhat.com>
> > > ---
> > >  arch/x86/boot/compressed/aslr.c |   30 ++++++++++++++++++++++++++++--
> > >  1 file changed, 28 insertions(+), 2 deletions(-)
> > > 
> > > Index: linux-2.6/arch/x86/boot/compressed/aslr.c
> > > ===================================================================
> > > --- linux-2.6.orig/arch/x86/boot/compressed/aslr.c
> > > +++ linux-2.6/arch/x86/boot/compressed/aslr.c
> > > @@ -110,8 +110,9 @@ struct mem_vector {
> > >  	unsigned long size;
> > >  };
> > >  
> > > -#define MEM_AVOID_MAX 5
> > > +#define MEM_AVOID_MAX 32
> > >  static struct mem_vector mem_avoid[MEM_AVOID_MAX];
> > > +static int mem_avoid_nr;
> > >  
> > >  static bool mem_contains(struct mem_vector *region, struct mem_vector *item)
> > >  {
> > > @@ -135,6 +136,28 @@ static bool mem_overlaps(struct mem_vect
> > >  	return true;
> > >  }
> > >  
> > > +static void mem_avoid_setup_data(void)
> > > +{
> > > +	struct setup_data *data;
> > > +	u64 pa_data;
> > > +
> > > +	if (mem_avoid_nr >= MEM_AVOID_MAX) {
> > > +		debug_putstr("KASLR: too many setup_data ranges.\n");
> > > +		return;
> > > +	}
> > > +
> > > +	pa_data = real_mode->hdr.setup_data;
> > > +	while (pa_data) {
> > > +		data = (struct setup_data *)pa_data;
> > > +		if (pa_data < CONFIG_RANDOMIZE_BASE_MAX_OFFSET) {
> > > +			mem_avoid[mem_avoid_nr].start = pa_data;
> > > +			mem_avoid[mem_avoid_nr].size = sizeof(*data) + data->len;
> > > +			mem_avoid_nr++;
> > > +		}
> > > +		pa_data = data->next;
> > > +	}
> > > +}
> > > +
> > 
> > I have no idea if this real_mode->hdr.setup_data examination is correct,
> > so not commenting on that.

Probably should return early when pa_data > 4G because of 4G early mapping.
Not sure for 32bit though. I'm not familar with the detail of x86 early boot.
Help will be appreciate. 

> > 
> > What prevents the loop from overflowing the table?
> 
> Ooooops, the checking should be in the loop instead of our of it.
> Will repost. Thanks for catching.
> 
> Dave