Re: reproducible panic in netisr

On Thu, 6 Aug 2009, Julian Elischer wrote:

>> Righto, yes. So does that imply that the alignment provided by crget()
>> { which uses malloc() } is sufficient for td->td_ucred or is td->td_ucred
>> a special case?
>
> It should be enough.
>
Ok, that's good news.

>
> we should probably do the right thign refcount-wise for the ucred
> but refcount (atomic) ops are expensive.
>
In the clnt_rc.c case, it does do a crdup(), although I think a
crhold() would have been sufficient. In the nlm, there is a case
where the mount point cred gets used without a crhold() and that
does cause panics for pho_at_'s tests related to "umount -f". I have
a fix for that one, which is waiting for a Doug R. review. (It
only affects "umount -f", which I think still has other issues,
so I haven't tried to push for it getting into head.)

So, yes, I think that the refcount issue is being handled, rick