svn commit: r184851 - user/dfr/gssapi/6/sys/nfsserver

[Doug Rabson]

On 11 Nov 2008, at 19:41, Robert Watson wrote:

> On Tue, 11 Nov 2008, Doug Rabson wrote:
>
>> Modified: user/dfr/gssapi/6/sys/nfsserver/nfs_srvkrpc.c
>> =
>> =
>> =
>> =
>> =
>> =
>> =
>> =
>> =
>> =====================================================================
>> --- user/dfr/gssapi/6/sys/nfsserver/nfs_srvkrpc.c	Tue Nov 11  
>> 14:58:07 2008	(r184850)
>> +++ user/dfr/gssapi/6/sys/nfsserver/nfs_srvkrpc.c	Tue Nov 11  
>> 15:51:45 2008	(r184851)
>> @@ -227,10 +227,14 @@ nfs_rephead(int siz, struct nfsrv_descri
>> 	if (err == EBADRPC)
>> 		return (NULL);
>>
>> +	/* XXXRW: not 100% clear the lock is needed here. */
>> +	NFSD_LOCK_ASSERT();
>> +
>
> Perhaps this should now be an XXXDFR? :-)  Or alternatively, if you  
> think the locking really is needed, we should convert my old and  
> questionable comment to a more firm one.

The locking model in 6.x is somewhat over-complex - it holds the  
NFSD_LOCK while executing service procs but doesn't really need to.  
This causes lots of unlock/relock code where it calls things that  
might sleep. Its all much simpler in 7.x and current where the service  
procs run without holding any locks.