Re: [PATCH V3 1/5] RDMA/core: Transport-independent access flags

[Jason Gunthorpe <jgunthorpe@obsidianresearch.com>]

On Sat, Jul 11, 2015 at 03:17:36AM -0700, 'Christoph Hellwig' wrote:
> On Fri, Jul 10, 2015 at 01:54:20PM -0600, Jason Gunthorpe wrote:
> > diff --git a/drivers/infiniband/core/verbs.c b/drivers/infiniband/core/verbs.c
> > index bac3fb406a74..6ed7e0f6c162 100644
> > +++ b/drivers/infiniband/core/verbs.c
> > @@ -1126,6 +1126,12 @@ struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags)
> >         struct ib_mr *mr;
> >         int err;
> >  
> > +       /* Granting remote access to the physical MR is a security hole, don't
> > +          do it. */
> > +       WARN_ON_ONCE(mr_access_flags &
> > +               (IB_ACCESS_REMOTE_WRITE | IB_ACCESS_REMOTE_READ |
> > +                IB_ACCESS_REMOTE_ATOMIC));
> > +
> 
> How about providing a system-wide IB_ACCESS_LOCAL_READ |
> IB_ACCESS_LOCAL_WRITE MR that all drivers can use and get rid of
> ib_get_dma_mr in the long run?  That would help to nicely simplify
> drivers?

That is more or less what I was talking about when I suggested
removing the lkey from the API ULPs use.

Ultimately ib_get_dma_mr is just one call, and switching to
(eg) pd->local_memory_lkey is tidier, but not much simpler..

> Currently various drivers are using ib_get_dma_mr with remote flags
> unfortunately, e.g. the SRP initiator driver uses it to optimize away
> memory registrtions for single SGL entry requests.

Unconditionally? Ugh. Maybe we do need the warn_on :(

> That looks fixable realtively easily, but I don't understand the
> other consumers as good.

The ones I looked at used it as a fallback if FMR or FRMR are not
available.

Jason