Re: [RFC] knfsd: per-client fair scheduling to prevent single-client starvation

On Sat, 06 Jun 2026, Chuck Lever wrote:
> On Fri, Jun 5, 2026, at 5:29 AM, Benjamin Coddington wrote:
> > Also, because knfsd doesn't have different resource pools for each version
> > we're going to want to continue to balance the pool for all versions
> > exported.
> 
> At a higher altitude, I am treating this issue as a partial DoS surface,
> so it does indeed have to be addressed for all NFS versions to eliminate
> that surface.
> 

I can't see that DoS is a useful way to look at this.  For that we would
need a threat model which I don't think we have.

The problem situations that have been mentioned involve actors that would
like to co-operate but have no obvious mechanism to enable that
co-operation.
For v4, I think session flow control should be the "obvious" mechanism.
For v3, nothing is obvious so we have to look at heuristics, classifying
by recent behaviour or some sort or identity.  We don't need a system
that is always perfect, just a system that clients are able to
co-operate with.

Thanks,
NeilBrown