[PATCH net] veth: Fix vlan_features so as to be able to use stacked vlan interfaces

[PATCH net] veth: Fix vlan_features so as to be able to use stacked vlan interfaces

[Toshiaki Makita]

Even if we create a stacked vlan interface such as veth0.10.20, it sends
single tagged frames (tagged with only vid 10).
Because vlan_features of a veth interface has the
NETIF_F_HW_VLAN_[CTAG/STAG]_TX bits, veth0.10 also has that feature, so
dev_hard_start_xmit(veth0.10) doesn't call __vlan_put_tag() and
vlan_dev_hard_start_xmit(veth0.10) overwrites vlan_tci.
This prevents us from using a combination of 802.1ad and 802.1Q
in containers, etc.

Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
---
 drivers/net/veth.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/net/veth.c b/drivers/net/veth.c
index 2ec2041..5b37437 100644
--- a/drivers/net/veth.c
+++ b/drivers/net/veth.c
@@ -285,7 +285,8 @@ static void veth_setup(struct net_device *dev)
 	dev->ethtool_ops = &veth_ethtool_ops;
 	dev->features |= NETIF_F_LLTX;
 	dev->features |= VETH_FEATURES;
-	dev->vlan_features = dev->features;
+	dev->vlan_features = dev->features &
+			     ~(NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX);
 	dev->destructor = veth_dev_free;
 
 	dev->hw_features = VETH_FEATURES;
-- 
1.8.1.2

[PATCH net] tun: remove bogus hardware vlan acceleration flags from vlan_features

[Toshiaki Makita]

From: Fernando Luis Vazquez Cao <fernando@oss.ntt.co.jp>

Even though only the outer vlan tag can be HW accelerated in the transmission
path, in the TUN/TAP driver vlan_features mirrors hw_features, which happens
to have the NETIF_F_HW_VLAN_?TAG_TX flags set. Because of this, during packet
tranmisssion through a stacked vlan device dev_hard_start_xmit, (incorrectly)
assuming that the vlan device supports hardware vlan acceleration, does not
add the vlan header to the skb payload and the inner vlan tags are lost
(vlan_tci contains the outer vlan tag when userspace reads the packet from
the tap device).

Signed-off-by: Fernando Luis Vazquez Cao <fernando@oss.ntt.co.jp>
Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
---

diff -urNp linux-3.13.3-orig/drivers/net/tun.c linux-3.13.3/drivers/net/tun.c
--- linux-3.13.3-orig/drivers/net/tun.c	2014-01-20 11:40:07.000000000 +0900
+++ linux-3.13.3/drivers/net/tun.c	2014-02-18 16:23:04.461078629 +0900
@@ -1651,7 +1651,9 @@ static int tun_set_iff(struct net *net,
 				   TUN_USER_FEATURES | NETIF_F_HW_VLAN_CTAG_TX |
 				   NETIF_F_HW_VLAN_STAG_TX;
 		dev->features = dev->hw_features;
-		dev->vlan_features = dev->features;
+		dev->vlan_features = dev->features &
+				     ~(NETIF_F_HW_VLAN_CTAG_TX |
+				       NETIF_F_HW_VLAN_STAG_TX);
 
 		INIT_LIST_HEAD(&tun->disabled);
 		err = tun_attach(tun, file, false);

Re: [PATCH net] tun: remove bogus hardware vlan acceleration flags from vlan_features

[David Miller]

From: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Date: Tue, 18 Feb 2014 21:20:09 +0900

> From: Fernando Luis Vazquez Cao <fernando@oss.ntt.co.jp>
> 
> Even though only the outer vlan tag can be HW accelerated in the transmission
> path, in the TUN/TAP driver vlan_features mirrors hw_features, which happens
> to have the NETIF_F_HW_VLAN_?TAG_TX flags set. Because of this, during packet
> tranmisssion through a stacked vlan device dev_hard_start_xmit, (incorrectly)
> assuming that the vlan device supports hardware vlan acceleration, does not
> add the vlan header to the skb payload and the inner vlan tags are lost
> (vlan_tci contains the outer vlan tag when userspace reads the packet from
> the tap device).
> 
> Signed-off-by: Fernando Luis Vazquez Cao <fernando@oss.ntt.co.jp>
> Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>

Applied and queued up for -stable.

Re: [PATCH net] veth: Fix vlan_features so as to be able to use stacked vlan interfaces

[Flavio Leitner]

On Tue, Feb 18, 2014 at 09:20:08PM +0900, Toshiaki Makita wrote:
> Even if we create a stacked vlan interface such as veth0.10.20, it sends
> single tagged frames (tagged with only vid 10).
> Because vlan_features of a veth interface has the
> NETIF_F_HW_VLAN_[CTAG/STAG]_TX bits, veth0.10 also has that feature, so
> dev_hard_start_xmit(veth0.10) doesn't call __vlan_put_tag() and
> vlan_dev_hard_start_xmit(veth0.10) overwrites vlan_tci.
> This prevents us from using a combination of 802.1ad and 802.1Q
> in containers, etc.
> 
> Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
> ---
>  drivers/net/veth.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/net/veth.c b/drivers/net/veth.c
> index 2ec2041..5b37437 100644
> --- a/drivers/net/veth.c
> +++ b/drivers/net/veth.c
> @@ -285,7 +285,8 @@ static void veth_setup(struct net_device *dev)
>  	dev->ethtool_ops = &veth_ethtool_ops;
>  	dev->features |= NETIF_F_LLTX;
>  	dev->features |= VETH_FEATURES;
> -	dev->vlan_features = dev->features;
> +	dev->vlan_features = dev->features &
> +			     ~(NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX);
>  	dev->destructor = veth_dev_free;
>  
>  	dev->hw_features = VETH_FEATURES;
> -- 
> 1.8.1.2
> 

Why that isn't a problem with another software device?
Although this patch might fix the issue, it seems to me that
the middle devices shouldn't use the same feature flags.
I mean, vlan.20 should add the header, then vlan.10 should
offload the tag to veth.  Otherwise, for any vlan on top of
veth there will be an unneeded memmove().

fbl

Re: [PATCH net] veth: Fix vlan_features so as to be able to use stacked vlan interfaces

[Toshiaki Makita]

(2014/02/19 13:13), Flavio Leitner wrote:
> On Tue, Feb 18, 2014 at 09:20:08PM +0900, Toshiaki Makita wrote:
>> Even if we create a stacked vlan interface such as veth0.10.20, it sends
>> single tagged frames (tagged with only vid 10).
>> Because vlan_features of a veth interface has the
>> NETIF_F_HW_VLAN_[CTAG/STAG]_TX bits, veth0.10 also has that feature, so
>> dev_hard_start_xmit(veth0.10) doesn't call __vlan_put_tag() and
>> vlan_dev_hard_start_xmit(veth0.10) overwrites vlan_tci.
>> This prevents us from using a combination of 802.1ad and 802.1Q
>> in containers, etc.
>>
>> Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
>> ---
>>  drivers/net/veth.c | 3 ++-
>>  1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/net/veth.c b/drivers/net/veth.c
>> index 2ec2041..5b37437 100644
>> --- a/drivers/net/veth.c
>> +++ b/drivers/net/veth.c
>> @@ -285,7 +285,8 @@ static void veth_setup(struct net_device *dev)
>>  	dev->ethtool_ops = &veth_ethtool_ops;
>>  	dev->features |= NETIF_F_LLTX;
>>  	dev->features |= VETH_FEATURES;
>> -	dev->vlan_features = dev->features;
>> +	dev->vlan_features = dev->features &
>> +			     ~(NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX);
>>  	dev->destructor = veth_dev_free;
>>  
>>  	dev->hw_features = VETH_FEATURES;
>> -- 
>> 1.8.1.2
>>
> 
> Why that isn't a problem with another software device?
> Although this patch might fix the issue, it seems to me that
> the middle devices shouldn't use the same feature flags.
> I mean, vlan.20 should add the header, then vlan.10 should
> offload the tag to veth.  Otherwise, for any vlan on top of
> veth there will be an unneeded memmove().

In this case with this patch, vlan_dev_hard_start_xmit(veth0.10.20) set
vlan_tci, dev_hard_start_xmit(veth0.10) put it into skb->data, and
vlan_dev_hard_start_xmit(veth0.10) set vlan_tci again.
dev_hard_start_xmit(veth0) doesn't put it into skb->data because veth0
has NETIF_F_HW_VLAN_*TAG_TX feature.

Similarly, in not stacked vlan case, for example if veth0.10 has no vlan
intarface on it, vlan_dev_hard_start_xmit(veth0.10) set vlan_tci and
dev_hard_start_xmit(veth0) doesn't put it into skb->data.
There will be no unnecessary memmove().

Although I haven't looked over all, other drivers don't seem to have
NETIF_F_HW_VLAN_*TAG_TX in vlan_features (at least, bridge, vxlan,
e1000e, and bnx2x don't).

Thanks,
Toshiaki Makita

Re: [PATCH net] veth: Fix vlan_features so as to be able to use stacked vlan interfaces

[Flavio Leitner]

On Wed, Feb 19, 2014 at 02:22:06PM +0900, Toshiaki Makita wrote:
> (2014/02/19 13:13), Flavio Leitner wrote:
> > On Tue, Feb 18, 2014 at 09:20:08PM +0900, Toshiaki Makita wrote:
> >> Even if we create a stacked vlan interface such as veth0.10.20, it sends
> >> single tagged frames (tagged with only vid 10).
> >> Because vlan_features of a veth interface has the
> >> NETIF_F_HW_VLAN_[CTAG/STAG]_TX bits, veth0.10 also has that feature, so
> >> dev_hard_start_xmit(veth0.10) doesn't call __vlan_put_tag() and
> >> vlan_dev_hard_start_xmit(veth0.10) overwrites vlan_tci.
> >> This prevents us from using a combination of 802.1ad and 802.1Q
> >> in containers, etc.
> >>
> >> Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
> >> ---
> >>  drivers/net/veth.c | 3 ++-
> >>  1 file changed, 2 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/drivers/net/veth.c b/drivers/net/veth.c
> >> index 2ec2041..5b37437 100644
> >> --- a/drivers/net/veth.c
> >> +++ b/drivers/net/veth.c
> >> @@ -285,7 +285,8 @@ static void veth_setup(struct net_device *dev)
> >>  	dev->ethtool_ops = &veth_ethtool_ops;
> >>  	dev->features |= NETIF_F_LLTX;
> >>  	dev->features |= VETH_FEATURES;
> >> -	dev->vlan_features = dev->features;
> >> +	dev->vlan_features = dev->features &
> >> +			     ~(NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_STAG_TX);
> >>  	dev->destructor = veth_dev_free;
> >>  
> >>  	dev->hw_features = VETH_FEATURES;
> >> -- 
> >> 1.8.1.2
> >>
> > 
> > Why that isn't a problem with another software device?
> > Although this patch might fix the issue, it seems to me that
> > the middle devices shouldn't use the same feature flags.
> > I mean, vlan.20 should add the header, then vlan.10 should
> > offload the tag to veth.  Otherwise, for any vlan on top of
> > veth there will be an unneeded memmove().
> 
> In this case with this patch, vlan_dev_hard_start_xmit(veth0.10.20) set
> vlan_tci, dev_hard_start_xmit(veth0.10) put it into skb->data, and
> vlan_dev_hard_start_xmit(veth0.10) set vlan_tci again.
> dev_hard_start_xmit(veth0) doesn't put it into skb->data because veth0
> has NETIF_F_HW_VLAN_*TAG_TX feature.
> 
> Similarly, in not stacked vlan case, for example if veth0.10 has no vlan
> intarface on it, vlan_dev_hard_start_xmit(veth0.10) set vlan_tci and
> dev_hard_start_xmit(veth0) doesn't put it into skb->data.
> There will be no unnecessary memmove().
> 
> Although I haven't looked over all, other drivers don't seem to have
> NETIF_F_HW_VLAN_*TAG_TX in vlan_features (at least, bridge, vxlan,
> e1000e, and bnx2x don't).
> 
> Thanks,
> Toshiaki Makita
> 

Alright, the code looks good and I didn't notice anything
different on my testing env.

Acked-by: Flavio Leitner <fbl@redhat.com>

Thanks!
fbl

Re: [PATCH net] veth: Fix vlan_features so as to be able to use stacked vlan interfaces

[David Miller]

From: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>
Date: Tue, 18 Feb 2014 21:20:08 +0900

> Even if we create a stacked vlan interface such as veth0.10.20, it sends
> single tagged frames (tagged with only vid 10).
> Because vlan_features of a veth interface has the
> NETIF_F_HW_VLAN_[CTAG/STAG]_TX bits, veth0.10 also has that feature, so
> dev_hard_start_xmit(veth0.10) doesn't call __vlan_put_tag() and
> vlan_dev_hard_start_xmit(veth0.10) overwrites vlan_tci.
> This prevents us from using a combination of 802.1ad and 802.1Q
> in containers, etc.
> 
> Signed-off-by: Toshiaki Makita <makita.toshiaki@lab.ntt.co.jp>

Applied and queued up for -stable.