I'd like to resurrect this. Here's a quick summary:
- The TRANSUM post-dissector requires access to some decoded values during the first pass across the trace file
- Defining values (e.g. smb2.cmd) in a “fake tap” causes the values to be available in the following scenarios:
- Wireshark with a C plugin
- Wireshark with a LUA plugin
- Tshark with a LUA plugin
- Tshark with a C plugin doesn’t work i.e. the fake tap values are not available during the first pass
- The cause is identified and just due to an inconsistency in code – see
https://bugs.wireshark.org/bugzilla/show_bug.cgi?id=13393
- Guy suggested that I rewrite TRANSUM as a tap but that suffers from the same inconsistency problem
As making the code consistent has been rejected and the tap idea won’t work, where do we go from here?
Thanks and regards…Paul
-----Original Message-----
From: Guy Harris (Code Review) [mailto:code-review-do-not-reply@xxxxxxxxxxxxx]
Sent: 13 February 2017 04:03
To: Michael Mann <mmann78@xxxxxxxxxxxx>
Cc: Petri Dish Buildbot <buildbot-no-reply@xxxxxxxxxxxxx>; Paul Offord <Paul.Offord@xxxxxxxxxxxx>
Subject: Change in wireshark[master]: tshark: Always have the same conditions to determine if tree...
Guy Harris has posted comments on this change.
Change subject: tshark: Always have the same conditions to determine if tree should be created ......................................................................
Patch Set 2: Code-Review-2
OK, the problem here is that the TRANSUM "dissector" is doing both tap-like things (reading the protocol tree and processing data) and post-dissector-like things (adding stuff to the protocol tree based on the results of that processing).
Given that taps are run *after* post-dissectors, this means that it has to "cheat" and do the tap-like things in the post-dissector, but post-dissectors are, like other dissectors, handed a protocol tree only if there's a need to generate
one.
For this, we need a way to do the "read the protocol tree and process it" stuff *before* the post-dissector is run, i.e. a way to register taps to be run *before* the post-dissectors.
(In the longer term, we probably need to re-think both taps and post-dissectors in a number of ways, and make whatever changes are necessary to get done what the taps and post-dissectors are currently doing in ways that don't involve
abusing those mechanisms to get things done.
I'm a little suspicious, for example, of the Parallel Redundancy Protocol trailer post-dissector - it's a protocol that's actually processing part of the packet data; see the "This is horribly broken" comment I added to it a while ago.
We might also want a way to have taps/"post-"dissectors that act as extensions to particular protocol dissectors - that might be what TRANSUM, and possibly MATE, *really* want to be.)
--
To view, visit
https://code.wireshark.org/review/20078
To unsubscribe, visit
https://code.wireshark.org/review/settings
Gerrit-MessageType: comment
Gerrit-Change-Id: I69bb395b55b73f26665d90f341c92a978e50aa70
Gerrit-PatchSet: 2
Gerrit-Project: wireshark
Gerrit-Branch: master
Gerrit-Owner: Michael Mann <mmann78@xxxxxxxxxxxx>
Gerrit-Reviewer: Guy Harris <guy@xxxxxxxxxxxx>
Gerrit-Reviewer: Michael Mann <mmann78@xxxxxxxxxxxx>
Gerrit-Reviewer: Paul Offord <paul.offord@xxxxxxxxxxxx>
Gerrit-Reviewer: Petri Dish Buildbot <buildbot-no-reply@xxxxxxxxxxxxx>
Gerrit-HasComments: No
