Wireshark-dev: [Wireshark-dev] Improving Network Monitor (and Message Analyzer) dissection supp
From: Michael Mann <[email protected]>
Date: Sat, 9 Sep 2017 11:30:31 -0400
I started working on the problem of having Wireshark read packet comments from Network Monitor (NetMon) (.cap) files. It escalated quickly and I found myself implementing way more functionality than I had intended just because of the information I stumbled across (and being "in the neighborhood").  I thought it was worth sharing some of that information on the -dev mailing list to give guidance in case anyone else wants to expand Wireshark's support of Netmon files. I do not have any plans to further expand the dissection, but will gladly answer any questions if others want to give it a try. Note this is only related to .cap files.  Support for .etl files has NOT been added (see bug 6694), and I'm not sure how much of this information may apply to .etl files.

.cap files use NPL (NetMon Parsing Language?), which is a combination of a C-like language mixed with IDL.  It is certainly "human readable", but I think the meaning of some of the "IDL" parts would make it a bit too complex to make it worthwhile to have a "npl2wrs" tool.  However, I did notice what appears to be some work towards that in the Wireshark tools directory.  Jakub, any comments?
Message Analyzer appears to use NPL files, but also OPN files, which seem to have a similar format.  Not sure of the details/differences, but I used both for reference when adding to Wireshark functionality.

Some of the functionality I added to Wireshark for .cap support was in wiretap.  Most of the rest, while done with dissectors, could be considered "metadata" and may be "translated" into pcapng blocks.  But since .cap files were treating them as "frames", I did as well as that's why they ended up with dissectors.  One of the key "frame types" would be "NetMon events".  It has an architecture that aligns with using a Wireshark dissector table.  I believe the "event frame types" can hold protocol dissection data, but I focused on the "metadata" since that's what was present in the captures I had (retrieved from Bugzilla)
I stayed away from protocol dissection mostly because I didn't have capture files for it, but also because I know Wireshark is already pretty good at that.  I was focused on how to "connect the dots" so that data could easily be passed to existing (protocol) dissection.

My two key sources of information were:
1. https://NMParsers.svn.codeplex.com/svn (just grab it with SVN client).  It's basically the "parser source" for NetMon. There are 2 directories, "Develop_Branch" and "Verified_Build_Branch".  For what I've done, I haven't seen a difference between the two, but I've also not done a full compare.  Given that I believe work on Network Monitor has stopped, these may not be updated anymore.
2. C:\Program Files\Microsoft Message Analyzer\OPNAndConfiguration.  This is the parser information installed with MessageAnalyzer.  Bug 10556 (https://bugs.wireshark.org/bugzilla/show_bug.cgi?id=10556) seems to give some information on how to extract the information without installing Message Analyzer, but since I'm mostly a Windows developer, this wasn't an issue for me.

To me both are comparable to Wireshark's dissector directory (with a little bit of wiretap mixed in to handle "metadata").

1. frame.npl, netmon.npl and MAExported.npl (from NMParsers\Develop_Branch\NPL\common directory is information for wiretap (netmon.c).  I implemented what I thought was "most popular" and things that I found sample captures for.
2. grep "NetEvent.UserData" in the NMParsers directory and you'll see all of the potential GUIDs used for Provider IDs.  Provider IDs are the "identifier" to the layer following a "NetMon event".  Again, I focused on "metadata" dissectors, but there were over 100 GUIDs registered with "NetEvent.UserData", most of which are probably protocol related.
3. grep "EventDescriptor.DefaultKeyword" for a potential match with a GUID from NetEvent.UserData.  The "NetMon event dissector" displays the keyword (and a few other fields), but its the subsequent layer that really defines its meaning.
4. packet-netmon.c was used mostly for functionality I found from NPL code, packet-messageanalyzer.c was for code I associated with MessageAnalyzer (either from OPN files or NPL files that used "MA" or other indications to show it was for MessageAnalyzer).  Protocol dissection may go in existing dissectors (if related/similar) or there may be new protocols that justify new dissection files. packet-netmon.c, packet-messageanalyzer.c, and packet-usb.c can be used as current examples of how to hook into the ProviderID dissector table ("netmon.provider_id") used by "NetMon events"