On Tue, Jul 07, 2020 at 05:10:46AM +0100, Matthew Wilcox wrote:
> On Tue, Jul 07, 2020 at 07:01:51AM +0300, Jarkko Sakkinen wrote:
> > On Tue, Jul 07, 2020 at 04:24:08AM +0100, Matthew Wilcox wrote:
> > > On Mon, Jul 06, 2020 at 08:22:54PM -0700, Sean Christopherson wrote:
> > > > On Tue, Jul 07, 2020 at 04:14:24AM +0100, Matthew Wilcox wrote:
> > > > > > + if (vma->vm_ops && vma->vm_ops->mprotect) {
> > > > > > + error = vma->vm_ops->mprotect(vma, nstart, tmp, prot);
> > > > > > + if (error)
> > > > > > + goto out;
> > > > > > + }
> > > >
> > > > Based on "... and then the vma owner can do whatever it needs to before
> > > > calling mprotect_fixup(), which is already not static", my interpretation
> > > > is that Matthew's intent was to do:
> > > >
> > > > if (vma->vm_ops && vma->vm_ops->mprotect)
> > > > error = = vma->vm_ops->mprotect(vma, nstart, tmp, prot);
> > > > else
> > > > error = mprotect_fixup(vma, &prev, nstart, tmp, newflags);
> > > > if (error)
> > > > goto out;
> > > >
> > > > i.e. make .mprotect() a full replacement as opposed to a prereq hook.
> > >
> > > Yes, it was. I was just looking at the next patch to be sure this was
> > > how I'd been misunderstood.
> >
> > I'm don't get this part. If mprotect_fixup is called in the tail of the
> > callback, why it has to be called inside the callback and not be called
> > after the callback?
>
> Because that's how every other VM operation works. Look at your
> implementation of get_unmapped_area() for example.
I get the point but I don't think that your proposal could work given
that mprotect-callback takes neither 'prev' nor 'newflags' as its
parameters. The current callback has no means to call mprotect_fixup()
properly.
It would have to be extended
int (*mprotect)(struct vm_area_struct *vma,
struct vm_area_struct **pprev, unsigned long start,
unsigned long end, unsigned long prot,
unsigned long newflags);
Is this what you want?
/Jarkko
