Re: Hang in 2.6.23-rc5

[Patrick Mau <mau@oscar.ping.de>]

On Mon, Sep 03, 2007 at 04:15:01AM +0530, Satyam Sharma wrote:
> On Mon, 3 Sep 2007, Alexey Dobriyan wrote:
> > 
> > Try this from net-2.6 tree:
> > 
> > --- a/net/ipv4/tcp_input.c
> > +++ b/net/ipv4/tcp_input.c
> > @@ -560,7 +560,7 @@ static u32 tcp_rto_min(struct sock *sk)
> >  	struct dst_entry *dst = __sk_dst_get(sk);
> >  	u32 rto_min = TCP_RTO_MIN;
> >  
> > -	if (dst_metric_locked(dst, RTAX_RTO_MIN))
> > +	if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
> >  		rto_min = dst->metrics[RTAX_RTO_MIN-1];
> >  	return rto_min;
> >  }
> 
> That's my impression as well. That's way too core/busy a codepath to have
> a bug in. As I said earlier, almost anybody testing -rc5 is sure to hit
> this within a few hours (probably less) -- sad, it greatly erodes from the
> usefulness of -rc5 as a release candidate.

The above patch also fixed my problems using NFS mounted home directories.

I already applied it yesterday without giving any feedback, so I just
thought I should comment.

In addition I totally agree with Satyam's comment above: either
anybody is testing rc's these days, or people simply stopped reporting.

My concern is that we have to live with a git tree that has this
problem, because core developers are unavailable and nobody has the
authority to commit this fix. (That was already posted before).

I am not voting for public commit rights and personally I don't want
them anyway. It's just my observation that nothing really happens
whenever a point release is only a few days away.

Maybe most people believe that they have to wait for the 2.6.24
cycly to start before they join the discussion again ...

Patrick