addition to ipfw..

[Julian Elischer]

Max Laier wrote:
> On Friday 15 December 2006 22:20, Julian Elischer wrote:
>> Max, further to your comment..
>>
>> Max Laier wrote:
>>> On Monday 11 December 2006 23:58, Julian Elischer wrote:
>>>> Andre Oppermann wrote:
>>>>> Julian Elischer wrote:
>>>>>> in ipfw layer 2 processing, the packet is passed to the firewall
>>>>>> as if it was a layer 3 IP packet but the ether header is also made
>>>>>> available.
>>>>>>
>>>>>> I would like  to add something similar in the case where a vlan
>>>>>> tag is also on the packet..
>>>>>>
>>>>>> basically I have a change where:
>>>>>>
>>>>>> If we are processing layer 2 packets (in ether or bridge code)
>>>>>> AND a sysctl says to do it,
>>>>>> and it is a vlan packet,
>>>>>>
>>>>>> Then the vlan header is also held back so that the packet can be
>>>>>> processed and examined as an IP packet. It is
>>>>>> (in the same way the ether header is) reattached when the packet
>>>>>> is accepted.
>>>>>>
>>>>>> This allows me to filter packets that are traversing my bridge,
>>>>>> even though they are encapsulated in a vlan.
>>>>>>
>>>>>> I have patches to allow this. I need this function. does anyone
>>>>>> else?
>>>>> Please have the ipfw code examine the vlan tag in the mbuf instead
>>>>> of fiddling with the mbuf contents.
>>>> The ipfw will be ignoring the vlan contents.. the patch is to move
>>>> the 'start of ip header' pointer past the vlan header.. (if asked)
>>>> so that it can identifu the IP packet.
>>>>
>>>> part of the patch is to make sure all the code uses this pointer
>>>> instead of the case now where some code uses it and some uses
>>>> mtod().
>>>>
>>>> This could be used in conjunction with vlan keyword that would look
>>>> at the vlan header, but that is a different feature..
>>> I understand you do have a patch?  Let's see it, so we are clear what
>>> we are talking about.  I think that w/o a ipfw feature to identify
>>> the vlan number, it is pretty useless.  Of course, it would enable
>>> you to do some basic sanity checks, but real filtering needs to know
>>> the vlan it is concerned with.  BTW, what speaks against plugging the
>>> bridge into the vlan on either side and bridge the vlan interfaces
>>> together?
>> I have placed the following patch files:
>> http://www.freebsd.org/~julian/vlstrip-7.diff
>> http://www.freebsd.org/~julian/vlstrip-6.diff
>>
>> which implement the ability to look within vlans when being used
>> on a bridge.
>>
>> I have done SOME testing with this but would certainly appreciate
>> another set of eyes..
>> the next change would be lyered on top of this change and would be the
>> addition of a rule:
>>
>> ipfw add 100 {operation} ip from any to any vlan {vlan_id}[-{vlan_id}]
>>
>> e.g.
>> ipfw add 1000 skipto 4000 ip from any to any vlan 100-200
>>
>> This, as it is will probably not work for the cases where vlans are
>> decoded by the hardware. I'm guessing that at some stage we need to
>> add the ability to cope with that too.. I remember that someone added
>> some capacity to do that to bpf recently.. (?) I think..
> 
> There is M_VLANTAG and m_pkthdr.ether_vtag for hardware support.  You 
> could even reuse those for this. i.e. emulate hardware support for ipfw 
> in the pfil hook.  If you want to look at the vlan tag later, you can 
> always use those then.

maybe..

> 
>> I hope I've found all the places where the old code cared that the ip
>> header was teh first thing in the mbuf..
>> if you see any places where that is stil assumed, let me know.
> 
> I don't like the implementation for this reason.  It feels hackish to me.  
> What is the reason that you didn't duplicate the ethernet header approach 
> in ip_fw_pfil.c?  Speed?  Did you measure?  It is certainly easier to 
> properly strip off the vlan header in the pfil hook code and reattach it 
> when done (or trust the hardware to do it - if M_VLANTAG was set in the 
> first place).


The big trick is that the mbuf must nt get midified in the packet filter
if there is any chance that it will be taken out and resent.

For example in a bridge.. I'm filtering packets.. they are vlan packets 
traversing the bridge. If I 'allow' the packet then it needs to 
re-enter the bridge with the same headers etc, that it came in with.
so I really want to move to having teh mbuf unchanged and having a 
pointer saying where the IP header starts.

> 
> As an aside, I agree that the mtod mania isn't that great either and we 
> should probably do away with it.  But that's orthogonal to the vlan 
> handling - I just don't like that to be pulled into *IP*fw.  This might 
> just be me, however.
> 
>> It's working for my testing here but I'm only using it to monitor
>> traffic on a tap, so the packets are discarded anyhow.
>