Re: Data corruption issue with splice() on 2.6.27.10

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

On Wed, Jan 07, 2009 at 03:57:56PM +0300, Evgeniy Polyakov wrote:
> On Wed, Jan 07, 2009 at 11:53:56PM +1100, Herbert Xu (herbert@gondor.apana.org.au) wrote:
> > On Wed, Jan 07, 2009 at 01:52:01PM +0100, Willy Tarreau wrote:
> > > 
> > > Evgeniy, I'd like to understand something related to our apparent lack of
> > > knowledge of when the data is effectively transmitted. If we're focusing
> > > on the send part, I can't understand why I never reproduce the corruption
> > > when the data source is a file or loopback, but I only see it when the source
> > > is an ethernet interface. How is it possible that a problem affecting only
> > 
> > It doesn't happen with a file because in that case you don't
> > start with an skb so there is no skb->head.  It probably doesn't
> > happen with loopback because loopback does GSO so again skb->head
> > does not exist (so to speak).
> 
> Yup, basically splice's transmit pipe buffer contains page references,
> where the first one is actually not a real page but skb, while in the
> case of sendfile() and/or splice() from the file first page is a real
> page of the appropriate file.

OK thanks guys for the clarifications.

Evgeniy, my printk() in tcp_sendpage() fired several times indicating we were
going through do_tcp_sendpage. During the same test, I observed a lot of
corruption.

Also, I have a good news. As you suggested, disabling both SG and GSO indeed
fixes the issue. do_tcp_sendpage() is not called anymore from tcp_sendpage()
in this case (according to dmesg).

Cheers,
Willy