Re: [PATCH] fully enable rpctrace to trace multitask programs.

[Da Zheng]

olafBuddenhagen@gmx.net wrote:
> > > > mach_reply_port() cannot be traced by the rpctrace at all since it's a
> > > > system trap.
> > > >         
> > > I see... So there is no way to divert it? That seems strange :-(
> > >
> > > Anyways, while this means that we won't see a port created by
> > > mach_reply_port() immediately, we still can start fully tracking it as
> > > soon as it is used in any way, can't we?...
> > >   
> > >       
> > yes, by using discover_receive_right(). Otherwise, I don't know how.
> >     
>
> Well, AIUI, there are only two ways how a receive right could be used:
> either mach_port_insert_right() (or mach_port_extract_right()) is
> invoked, in which case we can look at the parameters, so we know exactly
> where it is.
>
> The other option is that it's sent to another task via an RPC. In this
> case indeed we don't know what name it ends up as, so I guess we
> actually have to search for it -- but at least we know in which task.
>
> Unless I'm missing something essential, there is never any need to
> search through all the tasks...
>   
We can look at the parameters of mach_port_extract_right() to find the 
port name of the receive right, but it doesn't always work for 
mach_port_insert_right().
The parameter name of mach_port_insert_right() is the port name where we 
are going to insert the port right and isn't necessary to be the port 
name of the receive right (though in many cases they are).
But you are right, we can always know which task has the receive right.
I decide not to treat mach_port_insert_right() or 
mach_port_extract_right() specially, but search for the receive right in 
a specific task.

Zheng Da