Ethereal-dev: Re: [ethereal-dev] dup xid in rpc dissector after applying display filter
> The requirement that "XXX_init_protocol()" calls be added by hand, as
> necessary, is a bit of a botch. Protocols should, if they need an init
> routine, call a routine to register that init routine, and, in all
> places where init routines should be called, the Ethereal infrastructure
> should call all registered init routines (right now, that's in
> "open_cap_file()" and "colorize_packets()" - yes, "colorize_packets()";
> "filter_packet()" calls it, because most of the code path for colorizing
> and filtering is the same).
>
> I'll look at implementing that.
Done.
There's now a "register_init_routine()" routine; the register routine
for a protocol should, if that protocol has an "init" routine, call
"register_init_routine()", passing it a pointer to its "init" routine.
I changed the existing dissectors (AFS, NCP, SMB, and ONC RPC) to do
that.
The routines in the list of routines so registered are now called when a
new capture file is opened *and* when a filtering or colorizing pass
over the packets is about to be done.
In addition, I made the ONC RPC "init" routine zero out the table of RPC
calls, so that it completely erases any state from the previous
dissection pass - it wasn't doing that, and "rpc_call_lookup()" assumes
the list has at least one entry in it, and if there was an entry left
around from the last dissection pass, it'd be found, and possibly cause
a packet to be falsely marked as a duplicate (I saw that a couple of
times when testing my changes).
