Re: [lwip-users] TCP causing out of mem pool [RAW]

[Chris_S]

TCP dies after that.  This is a real problem.

I tried changing TCP_MSL from 60000 to 6000,
and it sure had an effect.  I saw some TCB blocks being reallocated,
but again after more refreshes it finally crashed the CPU and went to
restart.

I had KEEPALIVE turned ON, and tried it OFF, still the same.

I noticed an old bug 24830 that seems like the same thing.
Not fixed yet, says v1.4

I'm just mystified how people get this RAW working with this kind of a
problem.
Are there some special settings in lwipopts.h to deal with this?
Is this something I can bandaid in my HTTPd?    Is that what people do?

Chris.




> On Tue, 2009-07-28 at 04:03 -0700, Chris_S wrote:
> > > > Obviously the TCP PCB is never being freed.
> > > > So what routine is suppose to do this?
> > >
> > > They're probably all in the TIME_WAIT state.  This is correct.
They'll
> > > get freed once they leave TIME_WAIT and go to CLOSED.
> > >
> > > Kieran
> >
> > So when do they go to CLOSED?
> > I waited minutes and the TCP pile keeps building.
> > I closed my browser window, opened another,
> > and the TCP still keeps piling up.
> >
> > Not happening.  When do they get freed?
>
> 2 * TCP_MSL is the default timeout I think.
>
> However, I think we try and recycle PCBs from TIME_WAIT when we try to
> alloc and there isn't one available.  I wonder if the message you're
> seeing is just the "I failed to find a free one" and then the following
> line of code gets one from TIME_WAIT and everything is OK.  Do you
> actually see anything go wrong, or is it just the message that you don't
> like?