Re: [PATCH rdma-next 6/6] RDMA/srpt: Use private_data_len instead of hardcoded value

[Leon Romanovsky <leon@kernel.org>]

On Mon, Oct 28, 2019 at 10:43:51AM -0300, Jason Gunthorpe wrote:
> On Mon, Oct 28, 2019 at 03:19:17PM +0200, Leon Romanovsky wrote:
> > On Mon, Oct 28, 2019 at 10:13:19AM -0300, Jason Gunthorpe wrote:
> > > On Sun, Oct 20, 2019 at 10:15:59AM +0300, Leon Romanovsky wrote:
> > > > diff --git a/drivers/infiniband/ulp/srpt/ib_srpt.c b/drivers/infiniband/ulp/srpt/ib_srpt.c
> > > > index daf811abf40a..e66366de11e9 100644
> > > > +++ b/drivers/infiniband/ulp/srpt/ib_srpt.c
> > > > @@ -2609,7 +2609,7 @@ static int srpt_cm_handler(struct ib_cm_id *cm_id,
> > > >  	case IB_CM_REJ_RECEIVED:
> > > >  		srpt_cm_rej_recv(ch, event->param.rej_rcvd.reason,
> > > >  				 event->private_data,
> > > > -				 IB_CM_REJ_PRIVATE_DATA_SIZE);
> > > > +				 event->private_data_len);
> > >
> > > So, I took a look and found a heck of a lot more places assuming the
> > > size of private data that really should be checked if we are going to
> > > introduce a buffer length here.
> >
> > But we are not interested to make it dynamic, "private_data_len" has
> > constant size according to IBTA spec, I just don't want users to be
> > aware of this.
> >
> > Why should we add the below checks if it wasn't before?
>
> Why are we doing any of this then?

The final goal will be to present all CM messages as binary blobs with
access through specific GET/SET helpers, those including access to
private data. It is needed to completely eliminate be32 madness in
IB/core code.

Thanks

>
> Jason