Wireshark-dev: Re: [Wireshark-dev] Tapping Behaviour [Was: Export PDU:s]
From: Pascal Quantin <[email protected]>
Date: Sun, 12 May 2013 18:44:11 +0200
2013/5/12 Anders Broman <[email protected]>
Evan Huus skrev 2013-05-12 15:11:

On Sun, May 12, 2013 at 8:58 AM, Anders Broman <[email protected]> wrote:
Pascal Quantin skrev 2013-05-12 11:08:

2013/5/12 Anders Broman <[email protected]>
Pascal Quantin skrev 2013-05-10 15:20:

2013/5/5 Anders Broman <[email protected]>
Hi,
I have added a basic implementation making it possible to export higher
level PDU:s to file using a USER_DLT.
The basic implementation makes it possible to export SIP traffic to a new
file adding some meta data before the actual SIP message. The idea is that
it should be possible to export the reassembled PDU:s(and mix several
protocols) removing the under laying transport protocol but retaining some
interesting data such as IP addresses and ports.

The implementation is bare bones to get the demo to work. It would be
nice to get some feedback on useful tags
to add, helper functions to load tags and if some one is willing to work
on the GUI part that'd be nice too.

Would it be feasible/useful to apply for a link-layer type from tcpdump?

Any comments welcome.
Regards
Anders

Hi Anders,

it looks interesting. I started playing a bit with it and fixed a few bugs
in r49232. Moreover I added the tags content to a subtree. Feel free to
revert it if you do not like the output.
I would find it great to have a link layer type allocated. This way the
feature could work out of the box without any configuration.

Yes
  Then how should we proceed? Can Guy allocate us a DLT or must we send a
request to someone?

I was hoping Guy would comment :-)

Any idea on how to handle the export of several protocols ? Should we
allow the user to select them in the GUI or should we export all the
protocols registering the tap and let the user select afterwards which ones
to keep with filters?

My idea is to export all the protocols registering to the tap, if you tap
with a filter only the filtered
protocols should be tapped I think.
Fine with me.

By the way, I noticed that if a dissector and sub dissector both support
the export functionality, the sub dissector message is dumped twice (once
per protocol). Not sure whether this should be considered as a feature or a
bug.

Do you mean protocol x+y in the first packet and y in the second? I would
expect that.
Yes that's what I meant. I think we should tap the packet at the beginning
of the dissection and not at the end as it is currently done in the SIP
dissector. I see two advantages:
- packet x+y will be tapped before packet y and not the opposite (I find it
awkward to have packet y displayed before packet x+y)
- a malformed packet will still be tapped

Well as long as the tap, "taps" after reassembly is done I suppose it does
not matter but if y is on top of x why tap y
and not only X as that would include Y any way?
Tangent on the tapping behaviour: this is bringing to mind bug 8321
and the regression it cause which was bug 8610. The tap was originally
after dissection which was causing statistics to be wrong if a
malformed packet threw an exception. We moved the tap before
dissection, at which point the taps were missing some information that
they needed which hadn't been dissected yet. We eventually just
wrapped the entire thing in a Try/Catch block and called the tap at
the end again, but that was a fairly ugly solution.

I think what we need is to have taps simply get queued when called by
a dissector, and then have them actually called at the bottom of
packet-frame (or somewhere like that) so that they're always run
regardless of exceptions and with the entirety of what we were able to
dissect. I'm not sure how much work this would be, but I'm hoping it
would actually be a fairly minimal change.

Thoughts?
If the tap is extracting information from the packet and a dissector bug occurs before that
information is extracted that information will always be missing from the tap so having
"correct" information from a tap is impossible if there are dissector bugs - right?

This is indeed a really good argument. If I remember correctly, the issue with bugs 8321 / 8610 was that an exception triggered by a sub dissector was preventing tapping the information for the parent dissector.
I guess there is no universal answer for this: depending on the information we are tapping, you might need or not the call to the subdissector.

Regards,
Pascal.