On Thu, Dec 14, 2017 at 12:44:58PM -0800, Dave Hansen wrote:
> On 12/14/2017 06:37 AM, Peter Zijlstra wrote:
> > I'm also looking at pte_access_permitted() in handle_pte_fault(); that
> > looks very dodgy to me. How does that not result in endlessly CoW'ing
> > the same page over and over when we have a PKEY disallowing write access
> > on that page?
>
> I'm not seeing the pte_access_permitted() in handle_pte_fault(). I
> assume that's something you added in this series.
No, Dan did in 5c9d2d5c269c4.
> But, one of the ways that we keep pkeys from causing these kinds of
> repeating loops when interacting with other things is this hunk in the
> page fault code:
>
> > static inline int
> > access_error(unsigned long error_code, struct vm_area_struct *vma)
> > {
> ...
> > /*
> > * Read or write was blocked by protection keys. This is
> > * always an unconditional error and can never result in
> > * a follow-up action to resolve the fault, like a COW.
> > */
> > if (error_code & PF_PK)
> > return 1;
>
> That short-circuits the page fault pretty quickly. So, basically, the
> rule is: if the hardware says you tripped over pkey permissions, you
> die. We don't try to do anything to the underlying page *before* saying
> that you die.
That only works when you trip the fault from hardware. Not if you do a
software fault using gup().
AFAIK __get_user_pages(FOLL_FORCE|FOLL_WRITE|FOLL_GET) will loop
indefinitely on the case I described.
