netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* GSO and IPv4 forwarding
@ 2010-10-26 20:02 Kevin Wilson
  2010-10-26 20:15 ` Eric Dumazet
  2010-10-26 20:19 ` David Miller
  0 siblings, 2 replies; 6+ messages in thread
From: Kevin Wilson @ 2010-10-26 20:02 UTC (permalink / raw)
  To: netdev

Hi,
When we set a netdevice to support forwarding, we disable LRO.
This is done because we don't want to forward an SKB which has been
processed by LRO.

This is done in inet_forward_change() in net/ipv4/devinet.c:
We call dev_disable_lro(dev) in this method, when setting forwarding
for the device.
As a result, in ip_forward(), the packet will be dropped. (because
skb_warn_if_lro(), called by this method,  returns TRUE)

My question is:
dev_disable_lro(dev) disable the LRO feature (NETIF_F_LRO) of the
device. But suppose I have a device where GRO is enabled (and LRO is
not). And let's say I set forwarding on this device.

it seems to me that is such case, calling dev_disable_lro(dev)  in
net_forward_change() to disable the LRO feature of the device (which
is already disabled) is not enough, and in such case , GRO packets,
which want to be forwarded,  will **not** be dropped in ip_forward().
(since kb_warn_if_lro() will return false in this case)

Is it so ? I am ready to send a patch fixing it, but I am a newbie in
kernel, so I want to ask first.

rgs,
Kevin

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: GSO and IPv4 forwarding
  2010-10-26 20:02 GSO and IPv4 forwarding Kevin Wilson
@ 2010-10-26 20:15 ` Eric Dumazet
  2010-10-26 20:26   ` Kevin Wilson
  2010-10-26 20:19 ` David Miller
  1 sibling, 1 reply; 6+ messages in thread
From: Eric Dumazet @ 2010-10-26 20:15 UTC (permalink / raw)
  To: Kevin Wilson; +Cc: netdev

Le mardi 26 octobre 2010 à 22:02 +0200, Kevin Wilson a écrit :
> Hi,
> When we set a netdevice to support forwarding, we disable LRO.
> This is done because we don't want to forward an SKB which has been
> processed by LRO.
> 
> This is done in inet_forward_change() in net/ipv4/devinet.c:
> We call dev_disable_lro(dev) in this method, when setting forwarding
> for the device.
> As a result, in ip_forward(), the packet will be dropped. (because
> skb_warn_if_lro(), called by this method,  returns TRUE)
> 
> My question is:
> dev_disable_lro(dev) disable the LRO feature (NETIF_F_LRO) of the
> device. But suppose I have a device where GRO is enabled (and LRO is
> not). And let's say I set forwarding on this device.
> 
> it seems to me that is such case, calling dev_disable_lro(dev)  in
> net_forward_change() to disable the LRO feature of the device (which
> is already disabled) is not enough, and in such case , GRO packets,
> which want to be forwarded,  will **not** be dropped in ip_forward().
> (since kb_warn_if_lro() will return false in this case)
> 
> Is it so ? I am ready to send a patch fixing it, but I am a newbie in
> kernel, so I want to ask first.

GRO packets can be forwarded just fine.




^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: GSO and IPv4 forwarding
  2010-10-26 20:02 GSO and IPv4 forwarding Kevin Wilson
  2010-10-26 20:15 ` Eric Dumazet
@ 2010-10-26 20:19 ` David Miller
  1 sibling, 0 replies; 6+ messages in thread
From: David Miller @ 2010-10-26 20:19 UTC (permalink / raw)
  To: wkevils; +Cc: netdev

From: Kevin Wilson <wkevils@gmail.com>
Date: Tue, 26 Oct 2010 22:02:26 +0200

> My question is:
> dev_disable_lro(dev) disable the LRO feature (NETIF_F_LRO) of the
> device. But suppose I have a device where GRO is enabled (and LRO is
> not). And let's say I set forwarding on this device.

GRO is completely different from LRO, and can remain enabled
when forwarding is turned on.

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: GSO and IPv4 forwarding
  2010-10-26 20:15 ` Eric Dumazet
@ 2010-10-26 20:26   ` Kevin Wilson
  2010-10-26 20:43     ` Eric Dumazet
  2010-10-26 20:52     ` Stephen Hemminger
  0 siblings, 2 replies; 6+ messages in thread
From: Kevin Wilson @ 2010-10-26 20:26 UTC (permalink / raw)
  To: Eric Dumazet; +Cc: netdev

Hi,
Thanks a lot for your quick answer, I appreciate it (and did not
expect it to be so quick!)

Can someone please explain in 2-3 short sentences Why GRO can be
forwarded and LRO cannot be forwarded ?
rgs,
Kevin

On Tue, Oct 26, 2010 at 10:15 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> Le mardi 26 octobre 2010 à 22:02 +0200, Kevin Wilson a écrit :
>> Hi,
>> When we set a netdevice to support forwarding, we disable LRO.
>> This is done because we don't want to forward an SKB which has been
>> processed by LRO.
>>
>> This is done in inet_forward_change() in net/ipv4/devinet.c:
>> We call dev_disable_lro(dev) in this method, when setting forwarding
>> for the device.
>> As a result, in ip_forward(), the packet will be dropped. (because
>> skb_warn_if_lro(), called by this method,  returns TRUE)
>>
>> My question is:
>> dev_disable_lro(dev) disable the LRO feature (NETIF_F_LRO) of the
>> device. But suppose I have a device where GRO is enabled (and LRO is
>> not). And let's say I set forwarding on this device.
>>
>> it seems to me that is such case, calling dev_disable_lro(dev)  in
>> net_forward_change() to disable the LRO feature of the device (which
>> is already disabled) is not enough, and in such case , GRO packets,
>> which want to be forwarded,  will **not** be dropped in ip_forward().
>> (since kb_warn_if_lro() will return false in this case)
>>
>> Is it so ? I am ready to send a patch fixing it, but I am a newbie in
>> kernel, so I want to ask first.
>
> GRO packets can be forwarded just fine.
>
>
>
>

^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: GSO and IPv4 forwarding
  2010-10-26 20:26   ` Kevin Wilson
@ 2010-10-26 20:43     ` Eric Dumazet
  2010-10-26 20:52     ` Stephen Hemminger
  1 sibling, 0 replies; 6+ messages in thread
From: Eric Dumazet @ 2010-10-26 20:43 UTC (permalink / raw)
  To: Kevin Wilson; +Cc: netdev

Le mardi 26 octobre 2010 à 22:26 +0200, Kevin Wilson a écrit :
> Hi,
> Thanks a lot for your quick answer, I appreciate it (and did not
> expect it to be so quick!)
> 
> Can someone please explain in 2-3 short sentences Why GRO can be
> forwarded and LRO cannot be forwarded ?

Well, GRO is a pure software thing, completely handled in linux stack,
not driver specific. Its design included forwarding ability, LRO did
not.

http://lwn.net/Articles/311357/




^ permalink raw reply	[flat|nested] 6+ messages in thread

* Re: GSO and IPv4 forwarding
  2010-10-26 20:26   ` Kevin Wilson
  2010-10-26 20:43     ` Eric Dumazet
@ 2010-10-26 20:52     ` Stephen Hemminger
  1 sibling, 0 replies; 6+ messages in thread
From: Stephen Hemminger @ 2010-10-26 20:52 UTC (permalink / raw)
  To: Kevin Wilson; +Cc: Eric Dumazet, netdev

On Tue, 26 Oct 2010 22:26:07 +0200
Kevin Wilson <wkevils@gmail.com> wrote:

> Hi,
> Thanks a lot for your quick answer, I appreciate it (and did not
> expect it to be so quick!)
> 
> Can someone please explain in 2-3 short sentences Why GRO can be
> forwarded and LRO cannot be forwarded ?
> rgs,
> Kevin

LRO merges packets together creating one large skb.  This is a layering
violation for forwarding or bridging (it violates end to end principle).

GRO maintains the headers of each packet and passes them as
a cluster.

^ permalink raw reply	[flat|nested] 6+ messages in thread

end of thread, other threads:[~2010-10-26 20:52 UTC | newest]

Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2010-10-26 20:02 GSO and IPv4 forwarding Kevin Wilson
2010-10-26 20:15 ` Eric Dumazet
2010-10-26 20:26   ` Kevin Wilson
2010-10-26 20:43     ` Eric Dumazet
2010-10-26 20:52     ` Stephen Hemminger
2010-10-26 20:19 ` David Miller

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).