VLAN tagging of Ethernet frames about dpkt HOT 17 OPEN

Not currently being worked on.  Feel free to submit a patch when you're ready.

I just committed support for decoding MPLS labels, Cisco ISL VLAN tags, and 
802.2 LLC 
fields, but not for encoding them. Feel free to post what code you have and we 
can try 
to adapt it, if you want.

dugsong: Yes I think I saw that. My code extends Ethernet class to create one 
which
is adds an extra header to tag the packet. (After all, all it needs is an extra 
4
bytes after the ethernet header). Currently it's extremely crude so I 
definitely want
to tidy it up.

Also I'm not sure using a subclass is necessarily the best idea, but I couldn't 
see a
way to make __hdr__ in Ethernet be different depending on if the packets are 
meant to
be tagged or not.

Meh, here we go so far.

Like I say, it's extremely crude and was just something I knocked up as I 
needed to
quickly send out some VLAN tagged packets. It's by no means the best solution. I
really hope to get some more time to look at it soon.

So, the true fix for this...

The problem is that __hdr__ can't be dynamic depending on some condition. For 
VLAN
tagged packets we need __hdr__ to be 4 bytes longer, if we've selected the 
packet to
be tagged. So, we'd need __hdr__ to be a function which returns the headers.

Or we stick with the idea of my patch and have a separate class for it. That's 
not
all that bad and it works and stays with the same paradigm that dpkt already 
has.

What do you think?

Hello, 

Instead of modifying ethernet.py, why not add a new script for ieee802.1q 
header? By
doing that all one has to do is construct ieee802.1q header and pass it to 
ethernet
as it's data. I am attaching the script that I created (ieee8021q.py) along 
with this
comment. I have tested it and it does serve the purpose; attaching the test
script(tag150_rarpRequest.py) that makes use of my ieee8021q.py script.

What do you all think?

Sounds like a good idea to me. Although it feel like it should really be in 
Ethernet.py. We need to form the VLAN 
header field properly.

I do agree about the fact that we need to form the VLAN header field properly. 
I am
still not able to find a way to set Priority and CFI field in 802.1q header. 
May be
when passing the Tag I can set it. For instance if the tag is 4094 it's 
equivalent
hex is 0x0ffe, so if I want priority as '1' instead of '0' I will be passing 
0x2ffe
instead of 0x0ffe. 

I think your suggested change does not take that into consideration either, 
right?
That's one area where we are lacking flexibility.

Sure, you're right that my patch doesn't handle the header correctly either. 
The problem is that some bits of the 
header are not easily packed using struct.pack. The current dpkt way of doing 
things doesn't lend itself to this. 
This is why I suggest the notion of the header field being generated by calling 
a function.

Forgive me if this is a stupid question.  How do you extract the VLAN tag from 
an Ethernet packet using dpkt?  I understand that the 802.1q VLAN tag is 4 
additional bytes in the header, that you can tell a packet is tagged because 
bytes 13 and 14 contain 0x8100, and that moves the ethertype or length field 
and the payload 4 bytes back.

Thank you

Hi,

I have dpkt-1.7 and I don't see a way to extract vlan tags. Is there some way 
to do it that someone can point me to?

Thanks

Any progress on this?
If necessary, I'm willing to test/more code/unit tests.
How can I help?

Does this issue been solved?

Hey guys, 

I'd like to propose this patch for vlans. The Vlan file and ethernet/llc patch 
are included. I created it for a project I am working on, and everything is 
running smoothly.

Vlan is just another layer, so you can create stack of Ethernet > IEEE8021Q > 
IP.

This also supports 802.1QinQ natively which is nice.

Best,
Omid

Hi Omid,

I also created a patch to support 802.1q headers in dpkt for a project I'm 
working on.

It can be found here.
https://github.com/smutt/hexcap/blob/master/dpkt-read-only/dpkt/dot1q.py
https://github.com/smutt/hexcap/blob/master/dpkt-read-only/dpkt/ethernet.py
https://github.com/smutt/hexcap/blob/master/dpkt-read-only/dpkt/llc.py

ethernet.py also has changes for EDP and 802.3 SNAP.

A couple things I noticed about your patch.
1) It does not support writing.
2) I don't think your llc.py change will work.  Check mine.

I don't care what dpkt eventually implements so long as it supports arbitrary 
reading/writing of 802.1q headers at any layer of the packet.  You mention 
802.1ad(QinQ) which is good.  There is also 802.1ah and various VPN services 
that require 802.1q headers at different places in the packet.  We shouldn't 
assume 802.1q headers will always appear between the OSI data-link and network 
layers.

Thanks,
Andrew

Hello Andrew,

1. The patch I have included does supports writing, although the packing is 
done at the final stage.
2. Thank you for the second point, I fixed the LLC. I guess I forgot all about 
it while I was working on it.

About your third point,
While I agree that we shouldn't always assume that VLANs occur at L2, until 
someone implements 802.1ah or something similar we should be fine using either 
of our patches.

It all ends up whether anyone from the dpkt team would like to discuss this 
issue.

Also, in your patch, Is there any reason that you don't enforce ethernet.type 
when you have a vlan for next layer? E.g. if next layer is VLAN just enforce 
0x8100 as ethtype.

It would be a really nice feature if there was a "back pointer" to upper 
layers, just like pox.lib.packet, so some of the headers could automatically be 
inferred from next layers.

E.g. When I create

ethernet() -> ip() -> udp()

I shouldn't need to set eth.type OR ip.p. 

But I believe that's for another issue.

Best,
Omid

Hi Omid,

> 1. The patch I have included does supports writing, although the packing is 
done at the final stage.
[AM] Yes, I see it now.

> 2. Thank you for the second point, I fixed the LLC. I guess I forgot all 
about it while I was working on it. 
[AM] After looking more into it, I'm not even sure this matters.  Perhaps in a 
previous version of dpkt llc deconstruction was done in llc.py.  But I can't 
really see how llc.py gets called now.  In my ethernet.py you can see I hacked 
in support for 802.2 SNAP headers without touching llc.py.  So I suspect llc.py 
is not used anymore.  Anyone from dpkt core team want to comment on this?  
Bueller?  Bueller?

> Also, in your patch, Is there any reason that you don't enforce ethernet.type 
when you have a vlan for next layer? E.g. if next layer is VLAN just enforce 
0x8100 as ethtype. 
[AM]  Not sure exactly what you mean.  But be careful with "type" variable in 
ethernet.py.  If packet is LLC then ethernet.type actually refers to llc.pid 
since it occupies the same offset as Ethernet II etype field.

> It would be a really nice feature if there was a "back pointer" to upper 
layers, just like pox.lib.packet, so some of the headers could automatically be 
inferred from next layers.
>
> E.g. When I create
>
> ethernet() -> ip() -> udp()
>
> I shouldn't need to set eth.type OR ip.p.
[AM] I hear what you're saying, but I kinda like the ability to create broken 
packets with dpkt ;)  Even when I create packets that cannot be read back into 
dpkt successfully.  I like the control.

Thanks,
Andrew