[net] ipv6: skip __ipv6_select_ident when there is no skb_dst

2015-03-18, 10:25:43 -0400, Vlad Yasevich wrote:
> On 03/18/2015 10:15 AM, Sabrina Dubroca wrote:
> > 2015-03-18, 10:04:10 -0400, Vlad Yasevich wrote:
> >> On 03/18/2015 09:52 AM, Sabrina Dubroca wrote:
> >>> Matt Grant reported frequent crashes in ipv6_select_ident when
> >>> udp6_ufo_fragment is called from openvswitch on a skb that doesn't
> >>> have a dst_entry set.
> >>>
> >>> Skip __ipv6_select_ident in case of a NULL rt.
> >>>
> >>> Fixes: 0508c07f5e0c ("ipv6: Select fragment id during UFO segmentation if not set.")
> >>> Cc: Vladislav Yasevich <vyasevic@redhat.com>
> >>> Reported-by: Matt Grant <matt@mattgrant.net.nz>
> >>> Tested-by: Matt Grant <matt@mattgrant.net.nz>
> >>> Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
> >>> ---
> >> [...]
> >>
> >> Hi Sabrina
> >>
> >> This would result in us using id 0 which is not what we want to do.
> >>
> >> In this case, udp6_ufo_fragment() should be calling ipv6_proxy_select_ident() so that
> >> the fragment id is properly generated.
> > 
> > Hi Vlad,
> > 
> > So, instead, something like this?  Or do you want to use
> > ipv6_proxy_select_ident even when we have a skb_dst?
> > 
> 
> Yes, this is what I was thinking...
> 
> -vlad

Okay, so:

Hi!

This fixes hard lock on 3.19.2 as well. Without it, machine drops dead -
hard locks - pretty quickly after boot.

Will try 4.0-rc4 tomorrow.  4.0-rc3 had same hard crash issues.

Get it out there, this is a IPv6 network Packet of Death!  We need
3.19.3 ASAP I think.  

I'll try and see if I can get a network trace of the cause on Saturday.

Regards,

Matt

On Wed, 2015-03-18 at 15:36 +0100, Sabrina Dubroca wrote:
> 2015-03-18, 10:25:43 -0400, Vlad Yasevich wrote:
> > On 03/18/2015 10:15 AM, Sabrina Dubroca wrote:
> > > 2015-03-18, 10:04:10 -0400, Vlad Yasevich wrote:
> > >> On 03/18/2015 09:52 AM, Sabrina Dubroca wrote:
> > >>> Matt Grant reported frequent crashes in ipv6_select_ident when
> > >>> udp6_ufo_fragment is called from openvswitch on a skb that doesn't
> > >>> have a dst_entry set.
> > >>>
> > >>> Skip __ipv6_select_ident in case of a NULL rt.
> > >>>
> > >>> Fixes: 0508c07f5e0c ("ipv6: Select fragment id during UFO segmentation if not set.")
> > >>> Cc: Vladislav Yasevich <vyasevic@redhat.com>
> > >>> Reported-by: Matt Grant <matt@mattgrant.net.nz>
> > >>> Tested-by: Matt Grant <matt@mattgrant.net.nz>
> > >>> Signed-off-by: Sabrina Dubroca <sd@queasysnail.net>
> > >>> ---
> > >> [...]
> > >>
> > >> Hi Sabrina
> > >>
> > >> This would result in us using id 0 which is not what we want to do.
> > >>
> > >> In this case, udp6_ufo_fragment() should be calling ipv6_proxy_select_ident() so that
> > >> the fragment id is properly generated.
> > > 
> > > Hi Vlad,
> > > 
> > > So, instead, something like this?  Or do you want to use
> > > ipv6_proxy_select_ident even when we have a skb_dst?
> > > 
> > 
> > Yes, this is what I was thinking...
> > 
> > -vlad
> 
> Okay, so:
> 
> diff --git a/net/ipv6/udp_offload.c b/net/ipv6/udp_offload.c
> index ab889bb16b3c..be2c0ba82c85 100644
> --- a/net/ipv6/udp_offload.c
> +++ b/net/ipv6/udp_offload.c
> @@ -112,11 +112,9 @@ static struct sk_buff *udp6_ufo_fragment(struct sk_buff *skb,
>  		fptr = (struct frag_hdr *)(skb_network_header(skb) + unfrag_ip6hlen);
>  		fptr->nexthdr = nexthdr;
>  		fptr->reserved = 0;
> -		if (skb_shinfo(skb)->ip6_frag_id)
> -			fptr->identification = skb_shinfo(skb)->ip6_frag_id;
> -		else
> -			ipv6_select_ident(fptr,
> -					  (struct rt6_info *)skb_dst(skb));
> +		if (!skb_shinfo(skb)->ip6_frag_id)
> +			ipv6_proxy_select_ident(skb);
> +		fptr->identification = skb_shinfo(skb)->ip6_frag_id;
>  
>  		/* Fragment the skb. ipv6 header and the remaining fields of the
>  		 * fragment header are updated in ipv6_gso_segment()
> 
>

2015-03-19, 22:39:50 +1300, Matt Grant wrote:
> Hi!
> 
> This fixes hard lock on 3.19.2 as well. Without it, machine drops dead -
> hard locks - pretty quickly after boot.
> 
> Will try 4.0-rc4 tomorrow.  4.0-rc3 had same hard crash issues.
> 
> Get it out there, this is a IPv6 network Packet of Death!  We need
> 3.19.3 ASAP I think.  
> 
> I'll try and see if I can get a network trace of the cause on Saturday.
> 
> Regards,
> 
> Matt

Thanks for testing this, Matt.
I will submit the patch in a few minutes.