Return-Path: Received: from mail-qk0-f178.google.com ([209.85.220.178]:34019 "EHLO mail-qk0-f178.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1756024AbdC2LBs (ORCPT ); Wed, 29 Mar 2017 07:01:48 -0400 Received: by mail-qk0-f178.google.com with SMTP id d10so9211146qke.1 for ; Wed, 29 Mar 2017 04:01:47 -0700 (PDT) Message-ID: <1490785302.2738.2.camel@poochiereds.net> Subject: Re: [PATCH] svcrdma: set XPT_CONG_CTRL flag for bc xprt From: Jeff Layton To: "J. Bruce Fields" Cc: Chuck Lever , Chuck Lever , linux-nfs@vger.kernel.org Date: Wed, 29 Mar 2017 07:01:42 -0400 In-Reply-To: <20170329011638.GA20963@fieldses.org> References: <20170326231254.1319.26075.stgit@manet.1015granger.net> <1490577699.6879.1.camel@poochiereds.net> <62622BEB-E234-4035-94FE-0C34E00693AE@gmail.com> <1490612833.2808.1.camel@poochiereds.net> <1490618353.6879.3.camel@poochiereds.net> <20170329011638.GA20963@fieldses.org> Content-Type: text/plain; charset="UTF-8" Mime-Version: 1.0 Sender: linux-nfs-owner@vger.kernel.org List-ID: On Tue, 2017-03-28 at 21:16 -0400, J. Bruce Fields wrote: > On Mon, Mar 27, 2017 at 08:39:13AM -0400, Jeff Layton wrote: > > Something like this patch maybe? Builds but is otherwise untested. It > > might not DTRT though in the (nonsensical) case where you have a server > > that is listening on UDP but doesn't support v2 or v3. Not sure I > > really care about that too much. > > I don't think this is worth the trouble. > > A client that attempts to mount NFSv4 over UDP is operating out of spec, > and we don't owe them much. > > I'm not even convinced that transport-specific high/low version returns > are correct. A client could in theory be configured to prefer UDP and > NFSv3, but to fall back to NFSv4 and TCP if NFSv3 was unavailable, and > this would break that. That would be legal but admittedly odd client > behavior. > That's fine with me. My rationale here was that we have to treat each listening socket as a different "remote", in RPC parlance, since not all versions are supported on all socket types. Again though, the version info reported here is pretty useless. > If somebody actually hits a case where this patch would help, then let's > reconsider. > Fine by me. -- Jeff Layton