netdev - Re: [PATCH,RFC] skb->network_header and __vlan_put

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <486FBE5D.1020807@trash.net>
Date:	Sat, 05 Jul 2008 20:33:01 +0200
From:	Patrick McHardy <kaber@...sh.net>
To:	Lennert Buytenhek <buytenh@...tstofly.org>
CC:	netdev@...r.kernel.org
Subject: Re: [PATCH,RFC] skb->network_header and __vlan_put_tag()

Patrick McHardy wrote:
> Lennert Buytenhek wrote:
>> __vlan_put_tag() is not only substracting 4 from the skb's
>> ->mac_header pointer (which kind of makes sense[*]), but it is
>> substracting 4 from the ->network_header pointer as well.
>>
>> Conceptually, sure, VLAN is another layer of encapsulation between
>> ethernet and IP, but doesn't it make more sense to just consider the
>> VLAN tag part of the MAC header?  What good does it do to point
>> ->network_header to the _middle_ of the VLAN tag when a VLAN tag is
>> inserted?
>>
>> I'm adding VLAN support to mv643xx_eth (which does not support HW
>> insertion of a VLAN tag, but it does support HW checksumming when
>> a VLAN tag is present, you just have to tell the HW how many bytes
>> there are between the start of the packet and the IP header), and
>> I'm ending up with code like this:
>>
>>     if (skb->protocol == htons(ETH_P_8021Q))
>>         ip_header = ip_hdr(skb) + 4;
>>     else
>>         ip_header = ip_hdr(skb);
>>
>> whereas it'd be nicer if you could just have the same code for the
>> VLAN and non-VLAN case:
>>
>>     ip_header = ip_hdr(skb);
>>
>>
>> [*] But skb->mac_header seems to be mostly NULL when this function
>>     is called, causing it to end up being 0xfffffffc by the time the
>>     packet is given to the device's ->hard_start_xmit().
> 
> I agree that your patch makes sense, it seems some drivers
> (niu, gianfar) even assume this. Regarding the invalid
> mac_header pointer, that looks like a bug and it should
> probably call skb_reset_mac_header() instead of manually
> changing the potentially invalid ->mac_header.
> 
>> Index: linux-2.6.26-rc5/include/linux/if_vlan.h
>> ===================================================================
>> --- linux-2.6.26-rc5.orig/include/linux/if_vlan.h
>> +++ linux-2.6.26-rc5/include/linux/if_vlan.h
>> @@ -280,7 +280,6 @@ static inline struct sk_buff *__vlan_put
>>  
>>      skb->protocol = htons(ETH_P_8021Q);
>>      skb->mac_header -= VLAN_HLEN;
>> -    skb->network_header -= VLAN_HLEN;
>>  
>>      return skb;
>>  }
> 
> There is another spot where VLAN headers are built with an also
> incorrect looking network_header adjustment in vlan_dev_hard_header()
> that should also be changed for consisteny.
> 
> Could you try if changing both these spots and adding the
> skb_reset_mac_header() works for you and still displays
> the packets properly in tcpdump?

I've queued this patch for testing. I'll send it upstream with
my next VLAN update if everything works properly.

View attachment "x" of type "text/plain" (1591 bytes)