Hello!
On Tue, 2008-06-10 at 15:32 -0700, David Miller wrote:
I took a close look at this, it seems this patch adds
an ABBA deadlock. But I might be wrong.
Yes, this is not a real deadlock. But also it is not even a lock.
bh_lock_sock on listening socket is not enough to lock it.
Seems, the solution would be forcing bh_lock_sock() to protect accept_queue,
which is a mess. Or adding a separate spinlock, like it was done with
syn_wait_lock. Also not good.
Alexey