svn commit: r197584 - head/sys/rpc/rpcsec_gss

[Robert Watson]

On Mon, 28 Sep 2009, John Baldwin wrote:

> ==============================================================================
>> --- head/sys/rpc/rpcsec_gss/svc_rpcsec_gss.c	Mon Sep 28 18:54:26 2009
> (r197583)
>> +++ head/sys/rpc/rpcsec_gss/svc_rpcsec_gss.c	Mon Sep 28 18:55:29 2009
> (r197584)
>> @@ -449,6 +449,8 @@ rpc_gss_svc_getcred(struct svc_req *req,
>>  	cr->cr_uid = cr->cr_ruid = cr->cr_svuid = uc->uid;
>>  	cr->cr_rgid = cr->cr_svgid = uc->gid;
>>  	crsetgroups(cr, uc->gidlen, uc->gidlist);
>> +	cr->cr_prison = &prison0;
>> +	prison_hold(cr->cr_prison);
>>  	*crp = crhold(cr);
>>
>>  	return (TRUE);
>
> FYI, it would be nice if prison_hold() returned a pointer to the prison as 
> you can then do what crhold() does above:
>
> 	cr->cr_prison = prison_hold(&prison0);
>
> I prefer combining the refcount and assignment into one step with the goal 
> of avoiding outright assignments that don't go via foo_hold() or fooref() 
> for refcounted objects.

In the long-term, explicit references to proc0, prison0, thread0, filedesc0, 
... all make me rather nervous.  I'd rather that all these things were linked 
to either the credentials of the file system mount, the user thread, or 
perhaps gssd in some cases.  From the perspective if virtualization, the file 
system mount credential seems the most likely candidate.

Robert