Re: use-after-free in sock_wake_async

[Eric Dumazet]

On Wed, 2015-11-25 at 23:32 +0100, Hannes Frederic Sowa wrote:
> On Wed, Nov 25, 2015, at 23:09, Eric Dumazet wrote:
> > On Wed, 2015-11-25 at 20:57 +0000, Rainer Weikusat wrote:
> > 
> > > I do agree that keeping the ->sk_data_ready outside of the lock will
> > > very likely have performance advantages. That's just something I
> > > wouldn't have undertaken because I'd be reluctant to make a fairly
> > > complicated change to a lot of code.
> > 
> > All I am saying is that we can keep current performance.
> > 
> > We already have the core infrastructure, we only need to properly use
> > it.
> > 
> > I will split my changes in two parts.
> > 
> > One part doing a very boring change of
> > 
> > rename SOCK_ASYNC_NOSPACE and SOCK_ASYNC_WAITDATA
> > for X in SOCK_ASYNC_NOSPACE SOCK_ASYNC_WAITDATA 
> > 
> >  set_bit(X, &sk->sk_socket->flags) -> sk_set_bit(X, sk)
> >  clear_bit(X, &sk->sk_socket->flags) -> sk_clear_bit(X, sk)
> 
> sk_set_bit and sk_clear_bit will forward the set_bit and clear_bit into
> the socket_wq like you explained above?

In the first patch (no functional change), the helpers will look like

static void inline sk_set_bit(int nr, struct sock *sk)
{
	set_bit(nr, &sk->sk_socket->flags);
}


Then the second patch will change the helper to :

static void inline sk_set_bit(int nr, struct sock *sk)
{
	set_bit(nr, &sk->sk_wq_raw->flags);
}


> 
> > The rename will help backports to catch code that might have been
> > removed in recent kernels.
> > 
> > Then the second patch will do the actual changes, and they will look
> > very sensible for people wanting to review them, and or familiar with
> > the stack, do not worry ;)
> 
> Do you see a chance to inline socket_wq into struct socket and discard
> struct socket_alloc in one go by rcu in socket_destroy_inode?

???

I guess you missed the whole point to have socket_wq allocated outside
of the inode :(

inodes are not rcu protected (yet). I certainly don't want to mess with
VFS, we have enough problems in net/ directory already.




--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/