Re: Race between neigh_timer_handler and neigh_event_send

["David S. Miller" <davem@xxxxxxxxxxxxx>]

Good catch Greg.

All of this trouble is because this is the one spot where we
make reference to the arp_queue packets without neigh->lock
held.

I wish we could avoiding dropping the lock instead, but that
is not the case.  All of these ->solicit() methods can end
up calling back down into the neighbour layer and try to grab
the lock again.

I'll apply your patch, thanks.