On Fri, 2006-08-25 at 17:19 +0100, Matt Bernstein wrote:
> At 23:55 +0800 Ian Kent wrote:
>
> > I think I know why this isn't working.
> > It's due to the conversion to a threaded model.
> > And I fairly sure it can't work within v5 if we don't make changes.
> > It's actually quite difficult.
>
> I appreciate this--thanks for taking the time to understand it :)
>
> > Basically, the daemon doesn't ever trigger a mount itself and since
> > there is only one daemon now you can't cause a mount to happen in
> > another mount point.
>
> Yes.. (that's why the accept() patch doesn't help); it may need some kind
> of ability to recurse.
>
> For bonus points, the following scenario is a quick route to a nasty
> deadlock in autofs{3,4}: /a/1 binds to :/b/2; /b/2 binds to :/a/1. Having
> some kind of cycle-detection might prevent some difficult-to-diagnose
> crashes. (It happened to us..)
That's really hard to detect.
Mainly because of the autofs4 kernel module doesn't know much about the
user space daemon context and visa versa.
I'll think about that one.
Ian
_______________________________________________
autofs mailing list
[email protected]
http://linux.kernel.org/mailman/listinfo/autofs
