[PATCH] tcp: fix setting csum_start in tcp_gso_segment

[PATCH] tcp: fix setting csum_start in tcp_gso_segment

[Tom Herbert]

Dave Jones reported that a crash is occuring in

csum_partial
tcp_gso_segment
inet_gso_segment
? update_dl_migration
skb_mac_gso_segment
__skb_gso_segment
dev_hard_start_xmit
sch_direct_xmit
__dev_queue_xmit
? dev_hard_start_xmit
dev_queue_xmit
ip_finish_output
? ip_output
ip_output
ip_forward_finish
ip_forward
ip_rcv_finish
ip_rcv
__netif_receive_skb_core
? __netif_receive_skb_core
? trace_hardirqs_on
__netif_receive_skb
netif_receive_skb_internal
napi_gro_complete
? napi_gro_complete
dev_gro_receive
? dev_gro_receive
napi_gro_receive

It looks like a likely culprit is that SKB_GSO_CB()->csum_start is
not set correctly when doing non-scatter gather. We are using
offset as opposed to doffset.

Reported-by: Dave Jones <davej@redhat.com>
Signed-off-by: Tom Herbert <therbert@google.com>
---
 net/core/skbuff.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 9cd5344..c1a3303 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -2993,7 +2993,7 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
 							    skb_put(nskb, len),
 							    len, 0);
 			SKB_GSO_CB(nskb)->csum_start =
-			    skb_headroom(nskb) + offset;
+			    skb_headroom(nskb) + doffset;
 			continue;
 		}
 
-- 
2.0.0.526.g5318336

Re: [PATCH] tcp: fix setting csum_start in tcp_gso_segment

[Linus Torvalds]

On Tue, Jun 24, 2014 at 9:03 PM, Tom Herbert <therbert@google.com> wrote:
>
> It looks like a likely culprit is that SKB_GSO_CB()->csum_start is
> not set correctly when doing non-scatter gather. We are using
> offset as opposed to doffset.
>
> Reported-by: Dave Jones <davej@redhat.com>

DaveJ, I think you triggered this in five minutes on your box, and I
don't recall seeing anybody else reporting the oops (and google
doesn't find anything in the last month). So it's presumably somewhat
hw-specific. Does this fix the problem?

               Linus

Re: [PATCH] tcp: fix setting csum_start in tcp_gso_segment

[Dave Jones]

On Tue, Jun 24, 2014 at 09:17:25PM -0700, Linus Torvalds wrote:
 > On Tue, Jun 24, 2014 at 9:03 PM, Tom Herbert <therbert@google.com> wrote:
 > >
 > > It looks like a likely culprit is that SKB_GSO_CB()->csum_start is
 > > not set correctly when doing non-scatter gather. We are using
 > > offset as opposed to doffset.
 > >
 > > Reported-by: Dave Jones <davej@redhat.com>
 > 
 > DaveJ, I think you triggered this in five minutes on your box, and I
 > don't recall seeing anybody else reporting the oops (and google
 > doesn't find anything in the last month). So it's presumably somewhat
 > hw-specific. Does this fix the problem?

It's survived routing ~1GB of packets overnight, so I'd call this good.

thanks Tom.

	Dave

Re: [PATCH] tcp: fix setting csum_start in tcp_gso_segment

[David Miller]

From: Dave Jones <davej@redhat.com>
Date: Wed, 25 Jun 2014 10:10:52 -0400

> On Tue, Jun 24, 2014 at 09:17:25PM -0700, Linus Torvalds wrote:
>  > On Tue, Jun 24, 2014 at 9:03 PM, Tom Herbert <therbert@google.com> wrote:
>  > >
>  > > It looks like a likely culprit is that SKB_GSO_CB()->csum_start is
>  > > not set correctly when doing non-scatter gather. We are using
>  > > offset as opposed to doffset.
>  > >
>  > > Reported-by: Dave Jones <davej@redhat.com>
>  > 
>  > DaveJ, I think you triggered this in five minutes on your box, and I
>  > don't recall seeing anybody else reporting the oops (and google
>  > doesn't find anything in the last month). So it's presumably somewhat
>  > hw-specific. Does this fix the problem?
> 
> It's survived routing ~1GB of packets overnight, so I'd call this good.

Tom, please adjust the Subject line as suggested by Eric Dumazet and add
a Tested-by: for Dave.

Thanks!

Re: [PATCH] tcp: fix setting csum_start in tcp_gso_segment

[Eric Dumazet]

On Tue, 2014-06-24 at 21:03 -0700, Tom Herbert wrote:

> 
> It looks like a likely culprit is that SKB_GSO_CB()->csum_start is
> not set correctly when doing non-scatter gather. We are using
> offset as opposed to doffset.
> 
> Reported-by: Dave Jones <davej@redhat.com>
> Signed-off-by: Tom Herbert <therbert@google.com>
> ---
>  net/core/skbuff.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/net/core/skbuff.c b/net/core/skbuff.c
> index 9cd5344..c1a3303 100644
> --- a/net/core/skbuff.c
> +++ b/net/core/skbuff.c
> @@ -2993,7 +2993,7 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
>  							    skb_put(nskb, len),
>  							    len, 0);
>  			SKB_GSO_CB(nskb)->csum_start =
> -			    skb_headroom(nskb) + offset;
> +			    skb_headroom(nskb) + doffset;
>  			continue;
>  		}
>  

Yes, seems an obvious typo, but please change patch title.

This is not "tcp: fix setting csum_start in tcp_gso_segment"

Maybe "net: fix setting csum_start in skb_segment()"

[PATCH] net: fix setting csum_start in skb_segment()

[Eric Dumazet]

From: Tom Herbert <therbert@google.com>

Dave Jones reported that a crash is occurring in

csum_partial
tcp_gso_segment
inet_gso_segment
? update_dl_migration
skb_mac_gso_segment
__skb_gso_segment
dev_hard_start_xmit
sch_direct_xmit
__dev_queue_xmit
? dev_hard_start_xmit
dev_queue_xmit
ip_finish_output
? ip_output
ip_output
ip_forward_finish
ip_forward
ip_rcv_finish
ip_rcv
__netif_receive_skb_core
? __netif_receive_skb_core
? trace_hardirqs_on
__netif_receive_skb
netif_receive_skb_internal
napi_gro_complete
? napi_gro_complete
dev_gro_receive
? dev_gro_receive
napi_gro_receive

It looks like a likely culprit is that SKB_GSO_CB()->csum_start is
not set correctly when doing non-scatter gather. We are using
offset as opposed to doffset.

Reported-by: Dave Jones <davej@redhat.com>
Tested-by: Dave Jones <davej@redhat.com>
Signed-off-by: Tom Herbert <therbert@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Fixes: 7e2b10c1e52ca ("net: Support for multiple checksums with gso")
---
 net/core/skbuff.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 9cd5344fad73..c1a33033cbe2 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -2993,7 +2993,7 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
 							    skb_put(nskb, len),
 							    len, 0);
 			SKB_GSO_CB(nskb)->csum_start =
-			    skb_headroom(nskb) + offset;
+			    skb_headroom(nskb) + doffset;
 			continue;
 		}
 

Re: [PATCH] net: fix setting csum_start in skb_segment()

[Tom Herbert]

On Wed, Jun 25, 2014 at 12:51 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> From: Tom Herbert <therbert@google.com>
>
> Dave Jones reported that a crash is occurring in
>
> csum_partial
> tcp_gso_segment
> inet_gso_segment
> ? update_dl_migration
> skb_mac_gso_segment
> __skb_gso_segment
> dev_hard_start_xmit
> sch_direct_xmit
> __dev_queue_xmit
> ? dev_hard_start_xmit
> dev_queue_xmit
> ip_finish_output
> ? ip_output
> ip_output
> ip_forward_finish
> ip_forward
> ip_rcv_finish
> ip_rcv
> __netif_receive_skb_core
> ? __netif_receive_skb_core
> ? trace_hardirqs_on
> __netif_receive_skb
> netif_receive_skb_internal
> napi_gro_complete
> ? napi_gro_complete
> dev_gro_receive
> ? dev_gro_receive
> napi_gro_receive
>
> It looks like a likely culprit is that SKB_GSO_CB()->csum_start is
> not set correctly when doing non-scatter gather. We are using
> offset as opposed to doffset.
>
Acked-by: Tom Herbert <therbert@google.com>

> Reported-by: Dave Jones <davej@redhat.com>
> Tested-by: Dave Jones <davej@redhat.com>
> Signed-off-by: Tom Herbert <therbert@google.com>
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Fixes: 7e2b10c1e52ca ("net: Support for multiple checksums with gso")
> ---
>  net/core/skbuff.c |    2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/net/core/skbuff.c b/net/core/skbuff.c
> index 9cd5344fad73..c1a33033cbe2 100644
> --- a/net/core/skbuff.c
> +++ b/net/core/skbuff.c
> @@ -2993,7 +2993,7 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
>                                                             skb_put(nskb, len),
>                                                             len, 0);
>                         SKB_GSO_CB(nskb)->csum_start =
> -                           skb_headroom(nskb) + offset;
> +                           skb_headroom(nskb) + doffset;
>                         continue;
>                 }
>
>
>

Re: [PATCH] net: fix setting csum_start in skb_segment()

[David Miller]

From: Eric Dumazet <eric.dumazet@gmail.com>
Date: Wed, 25 Jun 2014 12:51:01 -0700

> From: Tom Herbert <therbert@google.com>
> 
> Dave Jones reported that a crash is occurring in
 ...
> It looks like a likely culprit is that SKB_GSO_CB()->csum_start is
> not set correctly when doing non-scatter gather. We are using
> offset as opposed to doffset.
> 
> Reported-by: Dave Jones <davej@redhat.com>
> Tested-by: Dave Jones <davej@redhat.com>
> Signed-off-by: Tom Herbert <therbert@google.com>
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> Fixes: 7e2b10c1e52ca ("net: Support for multiple checksums with gso")

Applied, thanks everyone.