On Thu, Mar 02, 2017 at 10:55:41AM +0100, Martin Pieuchot wrote: > Sleeping here is completely ok. The NET_LOCK() in this function is only > taken to make sure no other thread will try to do a route lookup in > ip_output() while we're messing with the routing table.
Then I think your change is ok. The kernel lock in route_input() should protect us. But please resend the diff after merging with krw@'s rtm proposal commit. bluhm
