[PATCH V1 net-next 1/1] hyperv: Properly handle checksum offload

[PATCH V1 net-next 1/1] hyperv: Properly handle checksum offload

[K. Y. Srinivasan]

Do checksum offload only if the client of the driver wants checksum to be
offloaded.

In this version of the patch, I have addressed comments from
Stephen Hemminger <stephen@networkplumber.org> and
Eric Dumazet <eric.dumazet@gmail.com>.

This patch fixes a bug that is exposed in gateway scenarios.


Signed-off-by: K. Y. Srinivasan <kys@microsoft.com>
Reviewed-by: Haiyang Zhang <haiyangz@microsoft.com>
Cc: stable@kernel.org
---
 drivers/net/hyperv/netvsc_drv.c |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)

diff --git a/drivers/net/hyperv/netvsc_drv.c b/drivers/net/hyperv/netvsc_drv.c
index 939e3af..7ca9daa 100644
--- a/drivers/net/hyperv/netvsc_drv.c
+++ b/drivers/net/hyperv/netvsc_drv.c
@@ -467,6 +467,9 @@ static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 	if (skb_is_gso(skb))
 		goto do_lso;
 
+	if (skb->ip_summed != CHECKSUM_PARTIAL)
+		goto do_send;
+
 	rndis_msg_size += NDIS_CSUM_PPI_SIZE;
 	ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
 			    TCPIP_CHKSUM_PKTINFO);
-- 
1.7.4.1

Re: [PATCH V1 net-next 1/1] hyperv: Properly handle checksum offload

[David Miller]

From: "K. Y. Srinivasan" <kys@microsoft.com>
Date: Fri, 25 Apr 2014 20:55:23 -0700

> @@ -467,6 +467,9 @@ static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
>  	if (skb_is_gso(skb))
>  		goto do_lso;
>  
> +	if (skb->ip_summed != CHECKSUM_PARTIAL)
> +		goto do_send;
> +

There are many possible values of ip_summed, why would you go to do_send if
for example it was set to CHECKSUM_COMPLETE?

I think you are just rushing this change, take your time and implement the
fix properly.

For example, if you only expect two possible values here (CHECKSUM_PARTIAL
and something else), design your test so that it only allows those two
values and therefore you'll be documenting this invariant.

RE: [PATCH V1 net-next 1/1] hyperv: Properly handle checksum offload

[KY Srinivasan]

> -----Original Message-----
> From: David Miller [mailto:davem@davemloft.net]
> Sent: Monday, April 28, 2014 10:24 AM
> To: KY Srinivasan
> Cc: netdev@vger.kernel.org; linux-kernel@vger.kernel.org;
> devel@linuxdriverproject.org; olaf@aepfle.de; apw@canonical.com;
> jasowang@redhat.com; stable@kernel.org
> Subject: Re: [PATCH V1 net-next 1/1] hyperv: Properly handle checksum
> offload
> 
> From: "K. Y. Srinivasan" <kys@microsoft.com>
> Date: Fri, 25 Apr 2014 20:55:23 -0700
> 
> > @@ -467,6 +467,9 @@ static int netvsc_start_xmit(struct sk_buff *skb,
> struct net_device *net)
> >  	if (skb_is_gso(skb))
> >  		goto do_lso;
> >
> > +	if (skb->ip_summed != CHECKSUM_PARTIAL)
> > +		goto do_send;
> > +
> 
> There are many possible values of ip_summed, why would you go to
> do_send if for example it was set to CHECKSUM_COMPLETE?
> 
> I think you are just rushing this change, take your time and implement the fix
> properly.
> 
> For example, if you only expect two possible values here
> (CHECKSUM_PARTIAL and something else), design your test so that it only
> allows those two values and therefore you'll be documenting this invariant.

David,

I based this check on the check I see in tcp_v4_send_check() (and tcp_v6_send_check).
Looking at the code some more, it looks like if ip_summed is set to  CHECKSUM_NONE, the
software has computed the checksum and we can bypass the code for offloading the checksum.

Regards,

K. Y