Re: [PATCH 2/2] mount: RPC_PROGNOTREGISTERED should not be a permanent error

[NeilBrown <neilb@suse.com>]

[-- Attachment #1: Type: text/plain, Size: 1293 bytes --]

On Wed, Nov 30 2016, Steve Dickson wrote:

> On 11/29/2016 04:36 PM, NeilBrown wrote:
>> 
>> Completely happy for distros to choose not to support it.  But there are
>> more users of Linux than we distro people hear from.
> Fair enough... So leave the code in, but off by default
> so it's not something that has to be tested.

Thanks for clarifying.  Sounds good to me.

>> 
>> The EOPNOTSUPP commit fixed a real bug.  The bug would result in a mount
>> attempt failing when it should succeed.  Reverting that is not an
>> option.
> So the bug was that fact rpcbind was not up but the NFS server was?
> Or both were on there way up?

You could say that the bug was that rpcbind starts before the NFS server
has registered... but as the NFS server *cannot* register before rpcbind
starts, that would be a little harsh.

RPC_PROGNOTREGISTERED is a very close analogy to ECONNREFUSED.
They both say "The server process isn't listening (yet)".  One at the
TCP/IP level, one at the SUNRPC level.

If we started all daemons before enabling the network interface, and if
we could register all RPC services before rpcbind responds to external
requests, then these windows could be closed.  Unfortunately our
forefathers didn't have that foresight.

Thanks,
NeilBrown

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 832 bytes --]