Patch for MTU in the routing table cache

Patch for MTU in the routing table cache

[Simeon Penev]

[-- Attachment #1: Type: text/plain, Size: 303 bytes --]

Hi,

after testing the new 3.2 kernel and the new route caching code, i've 
found a bug in net/ipv4/route.c. The route cache does not detect changes 
to the MTU of the interface. Therefore IP fragmentation will happen with 
false MTU. Please see attached the proposed patch.

Best regards,
Simeon Penev

[-- Attachment #2: route-cache-mtu.patch --]
[-- Type: text/x-patch, Size: 542 bytes --]

--- linux-3.2.7/net/ipv4/route.c.orig	2012-02-28 17:26:00.214386221 +0100
+++ linux-3.2.7/net/ipv4/route.c	2012-02-28 17:25:27.826225630 +0100
@@ -1926,7 +1926,7 @@
 	const struct rtable *rt = (const struct rtable *) dst;
 	unsigned int mtu = dst_metric_raw(dst, RTAX_MTU);
 
-	if (mtu && rt_is_output_route(rt))
+	if (mtu && rt_is_output_route(rt) && mtu == dst->dev->mtu)
 		return mtu;
 
 	mtu = dst->dev->mtu;
@@ -1940,6 +1940,8 @@
 	if (mtu > IP_MAX_MTU)
 		mtu = IP_MAX_MTU;
 
+	dst_metric_set(dst, RTAX_MTU, mtu);
+
 	return mtu;
 }
 

Re: Patch for MTU in the routing table cache

[Eric Dumazet]

Le vendredi 02 mars 2012 à 14:13 +0100, Simeon Penev a écrit :
> Hi,
> 
> after testing the new 3.2 kernel and the new route caching code, i've 
> found a bug in net/ipv4/route.c. The route cache does not detect changes 
> to the MTU of the interface. Therefore IP fragmentation will happen with 
> false MTU. Please see attached the proposed patch.
> 

Hi

Did you test this patch doesnt break things like

ip ro add 96.17.148.89 via 10.0.0.1 mtu 800

(and keep eth0 mtu being standard mtu)

Re: Patch for MTU in the routing table cache

[Simeon Penev]

Hi Eric,

no, i didn't. I'm new to the developer list and just wanted to post a 
patch proposition. I hoped that the maintainer of the code (probably 
Alan Cox) will decide whether to use it.

Best regards,
Simeon

On 03/02/2012 02:43 PM, Eric Dumazet wrote:
> Le vendredi 02 mars 2012 à 14:13 +0100, Simeon Penev a écrit :
>> Hi,
>>
>> after testing the new 3.2 kernel and the new route caching code, i've
>> found a bug in net/ipv4/route.c. The route cache does not detect changes
>> to the MTU of the interface. Therefore IP fragmentation will happen with
>> false MTU. Please see attached the proposed patch.
>>
> Hi
>
> Did you test this patch doesnt break things like
>
> ip ro add 96.17.148.89 via 10.0.0.1 mtu 800
>
> (and keep eth0 mtu being standard mtu)
>
>
>
>

Re: Patch for MTU in the routing table cache

[Eric Dumazet]

Le vendredi 02 mars 2012 à 14:49 +0100, Simeon Penev a écrit :
> Hi Eric,
> 
> no, i didn't. I'm new to the developer list and just wanted to post a 
> patch proposition. I hoped that the maintainer of the code (probably 
> Alan Cox) will decide whether to use it.
> 

David Miller is the current network maintainer, but thats not the
point :)

I didnt test your patch, all I wanted to say is that you cant assume all
routes share the same MTU (device MTU)

And this is the assumption you coded :

if (mtu && rt_is_output_route(rt) && mtu == dst->dev->mtu)
	return mtu;