Re: [PATCH 4/6] xen/netback: Always pull through PKT_PROT_LEN bytes into the linear part of an skb.

[Ian Campbell <Ian.Campbell@citrix.com>]

On Wed, 2010-02-24 at 08:28 +0000, Jan Beulich wrote: 
> Could you point out what problem this addresses?

It ensures that at least the TCP/IP headers will be pulled into the
linear part of the SKB. At least skb_checksum_setup relies on this and I
think it is a more generic assumption in at least some parts of the
network stack as well. The next patch increases PKT_PROT_LEN to include
the TCP options as well since we have observed cases where Windows
guests with PV drivers can generate a frame with a split at the point.

In the common case the first fragment should already contain
PKT_PROT_LEN bytes so I don't think it will trigger often.

Ian.

> 
> Thanks, Jan
> 
> >>> Ian Campbell <ian.campbell@citrix.com> 23.02.10 17:47 >>>
> Previously PKT_PROT_LEN would only have an effect on the first fragment.
> 
> Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
> ---
>  drivers/xen/netback/netback.c |   10 ++++++++++
>  1 files changed, 10 insertions(+), 0 deletions(-)
> 
> diff --git a/drivers/xen/netback/netback.c b/drivers/xen/netback/netback.c
> index 7e1dfd1..e668704 100644
> --- a/drivers/xen/netback/netback.c
> +++ b/drivers/xen/netback/netback.c
> @@ -1346,6 +1346,16 @@ static void net_tx_submit(void)
>  
>  		netbk_fill_frags(skb);
>  
> +		/*
> +		 * If the initial fragment was < PKT_PROT_LEN then
> +		 * pull through some bytes from the other fragments to
> +		 * increase the linear region to PKT_PROT_LEN bytes.
> +		 */
> +		if (skb_headlen(skb) < PKT_PROT_LEN && skb_is_nonlinear(skb)) {
> +			int target = min_t(int, skb->len, PKT_PROT_LEN);
> +			__pskb_pull_tail(skb, target - skb_headlen(skb));
> +		}
> +
>  		skb->dev      = netif->dev;
>  		skb->protocol = eth_type_trans(skb, skb->dev);
>