Re: [PATCH net-next v2 5/6] tsnep: Add XDP socket zero-copy RX support

[Gerhard Engleder <gerhard@xxxxxxxxxxxxxxxxxxxxx>]

On 18.04.23 10:22, Paolo Abeni wrote:
> On Sat, 2023-04-15 at 16:42 +0200, Gerhard Engleder wrote:
> > @@ -892,6 +900,37 @@ static int tsnep_rx_desc_available(struct tsnep_rx *rx)
> >   		return rx->read - rx->write - 1;
> >   }
> >   
> > +static void tsnep_rx_free_page_buffer(struct tsnep_rx *rx)
> > +{
> > +	struct page **page;
> > +
> > +	page = rx->page_buffer;
> > +	while (*page) {
> > +		page_pool_put_full_page(rx->page_pool, *page, false);
> > +		*page = NULL;
> > +		page++;
> > +	}
> > +}
>
> [...]
>
> >   static void tsnep_rx_close(struct tsnep_rx *rx)
> >   {
> > +	if (rx->xsk_pool)
> > +		tsnep_rx_free_page_buffer(rx);
>
> It looks like the above could call tsnep_rx_free_page_buffer() with
> each page ptr in rx->page_buffer not zero. If so
> tsnep_rx_free_page_buffer() will do an out of bound access.

rx->page_buffer has space for up to TSNEP_RING_SIZE ptr's. The
descriptor ring is filled with at most TSNEP_RING_SIZE - 1
pages. Thus, the last ptr in rx->page_buffer is always zero.

> Also, why testing rx->xsk_pool instead of rx->page_buffer?

Testing for rx->xsk_pool is done for all code, which is only needed
if XSK zero-copy is enabled. For me this is more consistent to the
rest of the code.

Thanks!

Gerhard