iSCSI in 7.0

[James Mansion]

Scott Long wrote:
> inspected it closely for this.  Unfortunately, the GEOM block layer is
> not free of this deadlock, so this whole discussion is mostly academic
> (the GEOM problem is small and very hard to trigger since it uses the
> zone allocator, but problems have been observed in real-world
> situations, which is a shame).
Sorry if this is blindingly obvious to everyone else, but does this mean
'FreeBSD is prone to swap deadlock when swapping against local
partition and file swap areas'?

I hadn't considered that you need to be non-pageable AND that you
have to have pre-allocated the zones that you allocate from for avoid
a deadlock like this.  (ie, code paths doing IO driven by swap must
use a different memory management strategy - and that mkes life
interesting when receiving network packats that might be in response
to such an IO request, which is where we came in)

Is there any work underway to fix it?

James