Re: mpd client timeouts (bisected) 2.6.35-rc3

[Eric Dumazet <eric.dumazet@gmail.com>]

Le dimanche 13 juin 2010 à 13:36 -0700, John Fastabend a écrit :
> John Fastabend wrote:
> > David Miller wrote:
> >> From: Markus Trippelsdorf <markus@trippelsdorf.de>
> >> Date: Sat, 12 Jun 2010 12:28:02 +0200
> >>
> >>> Commit 597a264b1a9c7e36d1728f677c66c5c1f7e3b837:
> >>> »net: deliver skbs on inactive slaves to exact matches«
> >>>
> >>> causes large timeouts when mpd clients try to connect to a locally
> >>> running mpd (music player demon) on my machine. This makes it 
> >>> impossible to control mpd.
> >>>
> >>> I bisected this down to the commit mentioned above.
> >>> Reverting the commit from 2.6.35-rc3 also solves the problem.
> >> John, find an easy and fast way to fix this or else I am
> >> going to revert.
> >>
> >> Thanks.
> > 
> > Looks like skbs are hitting loopback_xmit() with deliver_no_wcard set.  Then in 
> > the receive path these skbs are only delivered to exact matches.  Not sure why 
> > this bit is set here, I'll track this down first thing tomorrow.
> > 
> > Thanks,
> > John.
> > --
> 
> Needed to set the wcard bit in copy_skb_header otherwise it will not be cleared 
> when called from skb_clone.  Which then hits the loopback device gets pushed 
> into the rx path and is eventually dropped. The following patch fixes this. 
> Hopefully, this is easy and fast enough for you Dave.
> 
> 
> [PATCH] net: fix deliver_no_wcard regression on loopback device
> 
> deliver_no_wcard is not being set in skb_copy_header.
> In the skb_cloned case it is not being cleared and
> may cause the skb to be dropped when the loopback device
> pushes it back up the stack.
> 
> Signed-off-by: John Fastabend <john.r.fastabend@intel.com>
> ---
> 
>   net/core/skbuff.c |    1 +
>   1 files changed, 1 insertions(+), 0 deletions(-)
> 
> diff --git a/net/core/skbuff.c b/net/core/skbuff.c
> index 9f07e74..bcf2fa3 100644
> --- a/net/core/skbuff.c
> +++ b/net/core/skbuff.c
> @@ -532,6 +532,7 @@ static void __copy_skb_header(struct sk_buff *new, const 
> struct sk_buff *old)
>   	new->ip_summed		= old->ip_summed;
>   	skb_copy_queue_mapping(new, old);
>   	new->priority		= old->priority;
> +	new->deliver_no_wcard	= old->deliver_no_wcard;
>   #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
>   	new->ipvs_property	= old->ipvs_property;
>   #endif
> --

Acked-by: Eric Dumazet <eric.dumazet@gmail.com>

BTW, David, it seems there is a double rxhash copy...

[PATCH] net: rxhash already set in __copy_skb_header

No need to copy rxhash again in __skb_clone()

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
---
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 9f07e74..a58e63b 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -569,7 +569,6 @@ static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
 	C(len);
 	C(data_len);
 	C(mac_len);
-	C(rxhash);
 	n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
 	n->cloned = 1;
 	n->nohdr = 0;