Re: [PATCH] scatterlist: Comment on pages for sg_set_page()

[Alan Mikhak]

On Wed, Oct 2, 2019 at 7:27 AM Jason Gunthorpe <jgg@xxxxxxxx> wrote:
>
> On Tue, Oct 01, 2019 at 11:32:50AM -0700, Alan Mikhak wrote:
> > From: Alan Mikhak <alan.mikhak@xxxxxxxxxx>
> >
> > Update the description of sg_set_page() to communicate current
> > requirements for the page pointer parameter.
> >
> > Signed-off-by: Alan Mikhak <alan.mikhak@xxxxxxxxxx>
> >  include/linux/scatterlist.h | 6 ++++++
> >  1 file changed, 6 insertions(+)
> >
> > diff --git a/include/linux/scatterlist.h b/include/linux/scatterlist.h
> > index 6eec50fb36c8..6dda865893aa 100644
> > +++ b/include/linux/scatterlist.h
> > @@ -112,6 +112,12 @@ static inline void sg_assign_page(struct scatterlist *sg, struct page *page)
> >   *   of the page pointer. See sg_page() for looking up the page belonging
> >   *   to an sg entry.
> >   *
> > + *   Scatterlist currently expects the page parameter to be a pointer to
> > + *   a page that is backed by a page struct.
> > + *
> > + *   Page pointers derived from addresses obtained from ioremap() are
> > + *   currently not supported since they require use of iomem safe memcpy.
> > + *
> >   **/
> >  static inline void sg_set_page(struct scatterlist *sg, struct page *page,
> >                              unsigned int len, unsigned int offset)
>
> It seems a bit weird to have a comment explaining that 'struct page
> *page' must actually be a valid pointer. Of course it must.
>
> Computing a 'struct page *' to something that doesn't actually have a
> struct page is simply a bug in whoever did that.
>
> Code should never be interchanging ioremap results with the struct
> page* world.
>
> Jason

Agreed. Should scatterlist API sg_set_page() check the page pointer to
prevent such usage that could crash the system? Is it still reasonable
to at least tell the programmer what is not allowed by adding a
comment in the description?

Alan