Wireshark · Wireshark-dev: Re: [Wireshark-dev] LUA chained dissector drops data parameter

[Peter Wu <peter@xxxxxxxxxxxxx>]

On Fri, Nov 29, 2019 at 03:25:51PM +0000, Kanstrup, Mikael wrote:
> Hi Peter,
> 
> Thanks a lot for an extensive answer with links to related discussions!
> 
> Have proto data use been considered for passing data around instead? I
> suppose there's a performance penalty to using that over directly
> passing the pointer. For my scenario the data parameter can quite
> easily be replaced with p_get_proto_data / p_add_proto_data. This
> works for cases where the LUA script itself does not need access the
> data passed around.


That appears to be a suitable API according to doc/README.dissector:

    The two most common use cases for p_add_proto_data/p_get_proto_data
    are for persistent data about the packet for the lifetime of the
    capture (file scope) and to exchange data between dissectors across
    a single packet (packet scope).

> I uploaded a change for this discussion here. I have not done any
> extensive testing but it shows the idea:
> https://code.wireshark.org/review/35260
> 
> Performance wise using callgrind with tshark -V on my packet capture
> file with 16000 packets I get around 0.5% higher numbers with proto
> data compared to directly passing the pointer.

Thanks for the quick checks. It could potentially get worse if there are
more protocols using this API, but it should only be a small constant.

How do you plan to make the data available to Lua? Is it read-only for
Lua, or writable by Lua as well? The 'private_table' API is not very
well specified in terms of lifetime and scope, and only accepts strings
at the moment. I am not sure if we want to make it a fixed part of the
public API.

Kind regards,
Peter

> /Mikael
> 
> ________________________________
> Från: Wireshark-dev för Peter Wu
> Skickat: den 26 november 2019 01:40
> Till: Developer support list for Wireshark
> Ämne: Re: [Wireshark-dev] LUA chained dissector drops data parameter
> 
> Hi Mikael,
> 
> On Mon, Nov 18, 2019 at 05:20:54PM +0000, Kanstrup, Mikael wrote:
> > Hi,
> >
> > I'm working on dissecting a proprietary protocol that extends Bluetooth HCI_ACL with a LUA dissector. As there's no heuristics dissector list registered for this particular protocol I thought something similar could be achieved with a chained dissector. I retrieve the original HCI_ACL dissector handle and replace it with my own LUA dissector. In LUA dissector apply some heuristics and if it's not my own protocol then call the original HCI_ACL dissector via the handle.
> >
> > Code looks like this:
> >
> > local proto_test = Proto("test", "Use chaining as heuristic dissector")
> > local proto_default_acl
> >
> > function is_test_proto(tvb, pinfo)
> >     -- Apply heuristics to determine if own protocol
> >     return false
> > end
> >
> > function proto_test.dissector(tvb, pinfo, tree)
> >     if not is_test_proto(tvb, pinfo) then
> >         return proto_default_acl:call(tvb, pinfo, tree)
> >     end
> >
> >     pinfo.cols.protocol = "test"
> >     tree = tree:add(proto_test, tvb)
> >     return tvb:len()
> > end
> >
> > function proto_test.init()
> >     local hci_type = DissectorTable.get("hci_h4.type")
> >     local pattern = 0x02 -- ACL
> >     proto_default_acl = hci_type:get_dissector(pattern)
> >     hci_type:add(pattern, proto_test)
> > end
> >
> > This unfortunately did not work and I was not able to find out why until I started looking at the HCI_ACL dissector code itself.
> >
> > static gint
> > dissect_bthci_acl(tvbuff_t *tvb, packet_info *pinfo, proto_tree *tree, void *data)
> > {
> > <...>
> >     /* Reject the packet if data is NULL */
> >     if (data == NULL)
> >         return 0;
> >
> > The above NULL check is hit for all calls coming from the LUA dissector. The LUA dissector function prototype does not have the data parameter and it appears it's simply lost when chaining calls through LUA.
> 
> Correct, Lua always passes a NULL parameter. Due to type-safety,
> Lua dissectors cannot pass arbitrary data.
> 
> > Any suggestions on how to approach this? Would it be possible to
> > extend the LUA dissector interface with another function prototype
> > that supports the data parameter? Just support relaying the parameter
> > in chained dissectors, not modifying or doing any fancy stuff with it.
> 
> There is unfortunately no runtime registration of types for the 'data'
> parameter. If you can somehow keep track that a C dissector was from a
> particular dissector table, and then recognize that the Lua dissector
> was registered with the same table, then hopefully it should be possible
> to propagate the data parameter as special case.
> 
> Long-term, we might have to revisit how to pass data between (multiple
> layers of) dissectors. Not just Lua, but also C. For Lua, see also:
> 
> Bug 15931 - Add Lua support for arbitrary data parameter in dissector calls.
> https://bugs.wireshark.org/bugzilla/show_bug.cgi?id=15931
> 
> (+cc Huang)
> Discussion about another mechanism to pass data between dissectors:
> https://code.wireshark.org/review/35159
> 
> Discussion about another (abandoned) new mechanism to pass data between
> dissectors:
> https://code.wireshark.org/review/34049