[PATCH net] gre: Fix regression in gretap TSO support

[PATCH net] gre: Fix regression in gretap TSO support

[alexander.duyck]

From: Alexander Duyck <alexander.h.duyck@redhat.com>

On recent kernels I found that TSO on gretap interfaces didn't work.  After
bisecting it I found that commit b884b1a4 had introduced a regression in
which the Ethernet header was being included in the GRE header length.

This change corrects that by basing the GRE header length on the inner mac
header in the case of GRE tunnels using transparent Ethernet bridging, and
uses the network header for all other GRE tunnel types.

Fixes: b884b1a4 ("gre_offload: simplify GRE header length calculation in gre_gso_segment()")
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: H.K. Jerry Chu <hkchu@google.com>
Cc: Pravin B Shelar <pshelar@nicira.com>
Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>
---
 net/ipv4/gre_offload.c |    5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/net/ipv4/gre_offload.c b/net/ipv4/gre_offload.c
index f6e345c..67a1f66 100644
--- a/net/ipv4/gre_offload.c
+++ b/net/ipv4/gre_offload.c
@@ -47,7 +47,10 @@ static struct sk_buff *gre_gso_segment(struct sk_buff *skb,
 
 	greh = (struct gre_base_hdr *)skb_transport_header(skb);
 
-	ghl = skb_inner_network_header(skb) - skb_transport_header(skb);
+	if (greh->protocol == htons(ETH_P_TEB))
+		ghl = skb_inner_mac_header(skb) - skb_transport_header(skb);
+	else
+		ghl = skb_inner_network_header(skb) - skb_transport_header(skb);
 	if (unlikely(ghl < sizeof(*greh)))
 		goto out;
 

Re: [PATCH net] gre: Fix regression in gretap TSO support

[Pravin Shelar]

On Wed, Oct 29, 2014 at 8:26 PM,  <alexander.duyck@gmail.com> wrote:
> From: Alexander Duyck <alexander.h.duyck@redhat.com>
>
> On recent kernels I found that TSO on gretap interfaces didn't work.  After
> bisecting it I found that commit b884b1a4 had introduced a regression in
> which the Ethernet header was being included in the GRE header length.
>
> This change corrects that by basing the GRE header length on the inner mac
> header in the case of GRE tunnels using transparent Ethernet bridging, and
> uses the network header for all other GRE tunnel types.
>
> Fixes: b884b1a4 ("gre_offload: simplify GRE header length calculation in gre_gso_segment()")
> Cc: Neal Cardwell <ncardwell@google.com>
> Cc: Eric Dumazet <edumazet@google.com>
> Cc: H.K. Jerry Chu <hkchu@google.com>
> Cc: Pravin B Shelar <pshelar@nicira.com>
> Signed-off-by: Alexander Duyck <alexander.h.duyck@redhat.com>

Patch also fixed problem with ovs-gre.

Acked-by: Pravin B Shelar <pshelar@nicira.com>

Thanks.

> ---
>  net/ipv4/gre_offload.c |    5 ++++-
>  1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/net/ipv4/gre_offload.c b/net/ipv4/gre_offload.c
> index f6e345c..67a1f66 100644
> --- a/net/ipv4/gre_offload.c
> +++ b/net/ipv4/gre_offload.c
> @@ -47,7 +47,10 @@ static struct sk_buff *gre_gso_segment(struct sk_buff *skb,
>
>         greh = (struct gre_base_hdr *)skb_transport_header(skb);
>
> -       ghl = skb_inner_network_header(skb) - skb_transport_header(skb);
> +       if (greh->protocol == htons(ETH_P_TEB))
> +               ghl = skb_inner_mac_header(skb) - skb_transport_header(skb);
> +       else
> +               ghl = skb_inner_network_header(skb) - skb_transport_header(skb);
>         if (unlikely(ghl < sizeof(*greh)))
>                 goto out;
>
>

Re: [PATCH net] gre: Fix regression in gretap TSO support

[Neal Cardwell]

On Thu, Oct 30, 2014 at 1:14 AM, Pravin Shelar <pshelar@nicira.com> wrote:
> On Wed, Oct 29, 2014 at 8:26 PM,  <alexander.duyck@gmail.com> wrote:
>> From: Alexander Duyck <alexander.h.duyck@redhat.com>
>>
>> On recent kernels I found that TSO on gretap interfaces didn't work.  After
>> bisecting it I found that commit b884b1a4 had introduced a regression in
>> which the Ethernet header was being included in the GRE header length.
>>
>> This change corrects that by basing the GRE header length on the inner mac
>> header in the case of GRE tunnels using transparent Ethernet bridging, and
>> uses the network header for all other GRE tunnel types.
>>
>> Fixes: b884b1a4 ("gre_offload: simplify GRE header length calculation in gre_gso_segment()")

Hmm. There may be other protocols, either now or in the future, where
we want to be able to have a mac header inside the GRE header, rather
than a network header. AFAICT it would be safer to revert b884b1a4,
and go back to the previous code (from c50cd357), where we parse the
GRE header to figure out its length.

neal

Re: [PATCH net] gre: Fix regression in gretap TSO support

[Alexander Duyck]

On 10/30/2014 06:51 AM, Neal Cardwell wrote:
> On Thu, Oct 30, 2014 at 1:14 AM, Pravin Shelar <pshelar@nicira.com> wrote:
>> On Wed, Oct 29, 2014 at 8:26 PM,  <alexander.duyck@gmail.com> wrote:
>>> From: Alexander Duyck <alexander.h.duyck@redhat.com>
>>>
>>> On recent kernels I found that TSO on gretap interfaces didn't work.  After
>>> bisecting it I found that commit b884b1a4 had introduced a regression in
>>> which the Ethernet header was being included in the GRE header length.
>>>
>>> This change corrects that by basing the GRE header length on the inner mac
>>> header in the case of GRE tunnels using transparent Ethernet bridging, and
>>> uses the network header for all other GRE tunnel types.
>>>
>>> Fixes: b884b1a4 ("gre_offload: simplify GRE header length calculation in gre_gso_segment()")
> Hmm. There may be other protocols, either now or in the future, where
> we want to be able to have a mac header inside the GRE header, rather
> than a network header. AFAICT it would be safer to revert b884b1a4,
> and go back to the previous code (from c50cd357), where we parse the
> GRE header to figure out its length.
>
> neal

The change is consistent with how we handle this in other spots 
throughout the kernel.  If nothing else you can just search for 
ETH_P_TEB and you will find multiple spots in the kernel where IP 
tunnels differentiate between transparent Ethernet bridging and regular 
IP in IP tunnels by checking for the protocol ETH_P_TEB.

Thanks,

Alex

Re: [PATCH net] gre: Fix regression in gretap TSO support

[Eric Dumazet]

On Thu, 2014-10-30 at 07:30 -0700, Alexander Duyck wrote:

> The change is consistent with how we handle this in other spots 
> throughout the kernel.  If nothing else you can just search for 
> ETH_P_TEB and you will find multiple spots in the kernel where IP 
> tunnels differentiate between transparent Ethernet bridging and regular 
> IP in IP tunnels by checking for the protocol ETH_P_TEB.

Agreed, I think that GUE might supersedes GRE usage anyway ;)

Acked-by: Eric Dumazet <edumazet@google.com>

Re: [PATCH net] gre: Fix regression in gretap TSO support

[Tom Herbert]

On Thu, Oct 30, 2014 at 7:30 AM, Alexander Duyck
<alexander.h.duyck@redhat.com> wrote:
>
> On 10/30/2014 06:51 AM, Neal Cardwell wrote:
>>
>> On Thu, Oct 30, 2014 at 1:14 AM, Pravin Shelar <pshelar@nicira.com> wrote:
>>>
>>> On Wed, Oct 29, 2014 at 8:26 PM,  <alexander.duyck@gmail.com> wrote:
>>>>
>>>> From: Alexander Duyck <alexander.h.duyck@redhat.com>
>>>>
>>>> On recent kernels I found that TSO on gretap interfaces didn't work.
>>>> After
>>>> bisecting it I found that commit b884b1a4 had introduced a regression in
>>>> which the Ethernet header was being included in the GRE header length.
>>>>
>>>> This change corrects that by basing the GRE header length on the inner
>>>> mac
>>>> header in the case of GRE tunnels using transparent Ethernet bridging,
>>>> and
>>>> uses the network header for all other GRE tunnel types.
>>>>
>>>> Fixes: b884b1a4 ("gre_offload: simplify GRE header length calculation in
>>>> gre_gso_segment()")
>>
>> Hmm. There may be other protocols, either now or in the future, where
>> we want to be able to have a mac header inside the GRE header, rather
>> than a network header. AFAICT it would be safer to revert b884b1a4,
>> and go back to the previous code (from c50cd357), where we parse the
>> GRE header to figure out its length.
>>
>> neal
>
>
> The change is consistent with how we handle this in other spots throughout
> the kernel.  If nothing else you can just search for ETH_P_TEB and you will
> find multiple spots in the kernel where IP tunnels differentiate between
> transparent Ethernet bridging and regular IP in IP tunnels by checking for
> the protocol ETH_P_TEB.
>
I'm not sure I understand this. We always use inner mac header in
__skb_udp_tunnel_segment for computing tunnel length and don't
distinguish between Ethernet or IP encapsulation. Presumably, in the
case of IP encapsulation inner mac header is equal to inner network
header. Why is this different for GRE?

Thanks,
Tom

> Thanks,
>
> Alex
>
> --
> To unsubscribe from this list: send the line "unsubscribe netdev" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH net] gre: Fix regression in gretap TSO support

[Tom Herbert]

> I'm not sure I understand this. We always use inner mac header in
> __skb_udp_tunnel_segment for computing tunnel length and don't
> distinguish between Ethernet or IP encapsulation. Presumably, in the
> case of IP encapsulation inner mac header is equal to inner network
> header. Why is this different for GRE?
>

Using skb_inner_mac_header seems to work okay for IP encapsulation.
I'll post the path momentarily.

Tom


> Thanks,
> Tom
>
>> Thanks,
>>
>> Alex
>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe netdev" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH net] gre: Fix regression in gretap TSO support

[Alexander Duyck]

On 10/30/2014 08:05 AM, Tom Herbert wrote:
> On Thu, Oct 30, 2014 at 7:30 AM, Alexander Duyck
> <alexander.h.duyck@redhat.com> wrote:
>> On 10/30/2014 06:51 AM, Neal Cardwell wrote:
>>> On Thu, Oct 30, 2014 at 1:14 AM, Pravin Shelar <pshelar@nicira.com> wrote:
>>>> On Wed, Oct 29, 2014 at 8:26 PM,  <alexander.duyck@gmail.com> wrote:
>>>>> From: Alexander Duyck <alexander.h.duyck@redhat.com>
>>>>>
>>>>> On recent kernels I found that TSO on gretap interfaces didn't work.
>>>>> After
>>>>> bisecting it I found that commit b884b1a4 had introduced a regression in
>>>>> which the Ethernet header was being included in the GRE header length.
>>>>>
>>>>> This change corrects that by basing the GRE header length on the inner
>>>>> mac
>>>>> header in the case of GRE tunnels using transparent Ethernet bridging,
>>>>> and
>>>>> uses the network header for all other GRE tunnel types.
>>>>>
>>>>> Fixes: b884b1a4 ("gre_offload: simplify GRE header length calculation in
>>>>> gre_gso_segment()")
>>> Hmm. There may be other protocols, either now or in the future, where
>>> we want to be able to have a mac header inside the GRE header, rather
>>> than a network header. AFAICT it would be safer to revert b884b1a4,
>>> and go back to the previous code (from c50cd357), where we parse the
>>> GRE header to figure out its length.
>>>
>>> neal
>>
>> The change is consistent with how we handle this in other spots throughout
>> the kernel.  If nothing else you can just search for ETH_P_TEB and you will
>> find multiple spots in the kernel where IP tunnels differentiate between
>> transparent Ethernet bridging and regular IP in IP tunnels by checking for
>> the protocol ETH_P_TEB.
>>
> I'm not sure I understand this. We always use inner mac header in
> __skb_udp_tunnel_segment for computing tunnel length and don't
> distinguish between Ethernet or IP encapsulation. Presumably, in the
> case of IP encapsulation inner mac header is equal to inner network
> header. Why is this different for GRE?
>
> Thanks,
> Tom

I'll dig into that a bit more and see if I can simplify this.  I just 
wasn't sure if the inner mac header was being initialized or not in the 
case of IP in IP tunnels.

Thanks,

Alex