Re: channel access security deadlock from asInit()

[Ralph Lange <Ralph.Lange@bessy.de>]

I was able to check this a bit further before taking off to the EPICS 
meeting at DESY:

It's even worse. Not only the CA server doesn't respond to name requests 
anymore, even the IOC initialization doesn't finish.
In our case, the main communication tasks for the CAN field bus are 
started at the final init step. If the AS file doesn't exist, CA doesn't 
find the channels anymore. When I went to the IOC, I saw there was no 
CAN traffic. Checking on the IOC's console, I found these tasks were not 
started.

Looks as if IOCinit bails out somewhere in the middle of doing things if 
the AS file doesn't exist.

Again: I agree that not giving any access would be the correct behavior 
in that case. Not finishing the IOC startup is too much of a penalty.

Cheers,
Ralph


Ralph Lange wrote:
> Ned -
>
> I thought about this, too ... and I would completely understand if the 
> CA server does not grant access (neither read nor write) to any 
> channels in that case.
>
> But actually it doesn't even resolve the names anymore, which is a 
> different thing, misleading, and probably neither appropriate nor 
> intended.
>
> Ralph
>
>
> Ned D. Arnold wrote:
> > Ralph -
> >
> > I think the behavior that you observed is by design ... if one tries 
> > to specify a configuration file that does not exist, it means you 
> > intended to have *some* access control in place.  If the file doesn't 
> > exist, this is a fatal error and the safest thing to do is not allow 
> > any access at all so someone will fix the problem immediately.
> > I actually remember this discussion with Marty Kraimer about 16 years 
> > ago ...
> >
> >   Ned