Re: Data corruption issue with splice() on 2.6.27.10

[Willy Tarreau <w@1wt.eu>]

Hi Jarek,

On Tue, Jan 06, 2009 at 08:54:42AM +0000, Jarek Poplawski wrote:
> On 24-12-2008 16:28, Willy Tarreau wrote:
> > Hi Jens,
> > 
> > I'm facing a data corruption problem with splice() between two
> > non-blocking TCP sockets on 2.6.27.10. I could finally write a
> > simpler proof of concept, and capture a snapshot of the issue
> > with the associated strace result.
> ...
> > I found an analysis [1] for a potential corruption problem between two
> > sockets, but I noticed there were no responses and I did not fully
> > understand the report anyway.
> > 
> > What can I do to help debug the problem ? I'm really willing to help
> > getting this fixed, and I also have at least one user who definitely
> > wants splice() to work because the recv/send model currently limits
> > haproxy to 3 Gbps on his machines, while I have no problem reaching
> > 10 Gbps with splice().
> ...
> > ----
> > [1] http://lkml.org/lkml/2008/2/26/210
> 
> Great story! Alas I don't understand this fully either, but it seems
> Changli Gao was concerned with sendpage sending this "as pages", so
> when NETIF_F_SG flag is available. Did you try this without SG btw?

No I did not. I can try, it's not too hard. It would in part defeat the
purpose of the mechanism (especially at 10 Gbps) but at least it will
help narrow the problem down.

Thanks for the tip, I'll keep you informed !
Willy