Re: [PATCH 3/3] NFSD: Enable NFS server use of PF_INET6

["J. Bruce Fields" <bfields@fieldses.org>]

On Fri, Jan 15, 2010 at 06:09:17PM -0500, Chuck Lever wrote:
>
> On Jan 15, 2010, at 5:35 PM, J. Bruce Fields wrote:
>
>> On Wed, Jan 13, 2010 at 04:10:48PM -0500, Chuck Lever wrote:
>>> Try to use an PF_INET6 listener for NFSD if IPv6 is enabled in the
>>> kernel.
>>>
>>> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
>>> ---
>>>
>>> fs/nfsd/nfsctl.c |   36 +++++++++++++++++++++++++++++-------
>>> fs/nfsd/nfssvc.c |   27 ++++++++++++++++++++++-----
>>> 2 files changed, 51 insertions(+), 12 deletions(-)
>>>
>>> diff --git a/fs/nfsd/nfsctl.c b/fs/nfsd/nfsctl.c
>>> index 2604c3e..7ebb7a5 100644
>>> --- a/fs/nfsd/nfsctl.c
>>> +++ b/fs/nfsd/nfsctl.c
>>> @@ -981,6 +981,26 @@ static ssize_t __write_ports_delfd(char *buf)
>>> 	return len;
>>> }
>>>
>>> +static ssize_t __write_ports_create_xprt(struct svc_serv *serv,
>>> +					 const char *transport,
>>> +					 const int family,
>>> +					 const unsigned short port)
>>> +{
>>> +	int err;
>>> +
>>> +	err = svc_create_xprt(serv, transport, family, port,
>>> +						SVC_SOCK_ANONYMOUS);
>>> +
>>> +	if (err < 0) {
>>> +		/* Give a reasonable perror msg for bad transport string */
>>> +		if (err == -ENOENT)
>>> +			err = -EPROTONOSUPPORT;
>>
>> I realize you're just moving this code, not writing it, so this is a
>> separate question, but: it looks like it's svc_create_xprt() itself  
>> that
>> chooses ENOENT for this case.  Is there any reason it shouldn't just  
>> use
>> EPROTONOSUPPORT from the start?
>
> I guess this little transformation is for the benefit of rpc.nfsd?  I  
> don't know of a reason offhand why you couldn't change  
> svc_create_xprt().  I don't immediately see any of svc_create_xprt's  
> callers that might depend on getting ENOENT, but I didn't look very  
> hard.

Could I talk you into writing a patch that does that, as long as you're
touching this code?

--b.