Right now when a device node shows up in a nullfs layer, it gets marked as
an alias for the node it's on top of and put at the head of the hash chain
for that device.

This aliasing is one of the few things we do different from FreeBSD, whose
nullfs is reported to be closer to working than ours. I'm wondering if
it's right.

So what problems can people see with removing the V_ALIASED checking in
null_node_alloc? Also, nullfs vnodes wouldn't get inserted into the
spechash'es so checkalias would never find them.

null_revoke would of course just pass the revoke op onto the lower layer,
cognizant of the fact that the node'll get cleaned as a result of the
operation.

The one change I see that would need to be done is that sys_revoke would
need to loose the "if (vp->v_usecount > 1 || (vp->v_flag & VALIASED))"
test it has above VOP_REVOKE. I think that's fine as the test just seems
to be for efficiency. As best I can tell, the test dates to 4.4-lite.

Take care,

Bill