Hi,

	I think this addresses the problems reported by David Laight and
others, where errors saved on a per socket area could be delivered to a
different thread, so I just followed David Laight's suggestion and
stopped saving it, we'll return it only if it happens for the first
datagram, else we return less entries than asked for.

	Steven, IIRC you was the one that suggested using this
mechanism, no? Do you have anything against this move?

- Arnaldo