Return-Path: Received: from fieldses.org ([173.255.197.46]:48496 "EHLO fieldses.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750756AbeDXWQM (ORCPT ); Tue, 24 Apr 2018 18:16:12 -0400 Date: Tue, 24 Apr 2018 18:16:12 -0400 From: "J. Bruce Fields" To: Olga Kornievskaia Cc: Mailing List Linux NFS Subject: Re: [PATCH] Args need to be the same for replay cache Message-ID: <20180424221612.GA30120@fieldses.org> References: <1507740502-5151-1-git-send-email-Thomas.Haynes@primarydata.com> <20180410194922.GD5652@fieldses.org> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: Sender: linux-nfs-owner@vger.kernel.org List-ID: On Tue, Apr 24, 2018 at 04:10:29PM -0400, Olga Kornievskaia wrote: > Do you by any chance have a reference to this discussion? This: http://lkml.kernel.org/r/1507740502-5151-1-git-send-email-Thomas.Haynes@primarydata.com and this: http://lkml.kernel.org/r/E0161195-9F4A-4B36-A71D-6A924498C893@primarydata.com and followups. > I would like to reference it. For now I'm hijacking this thread to > bring this up. I'm still concerned about the case where client sent a > request and slot got interrupted (so by default the client doesn't > increment the seq#). Then the client re-used the slot for the same > kind of operation (WRITE is very interesting) with same arguments but > say different FH. Is the server obligated the cache the whole call to > address and check that? You have a patch to check for false retries > that checks for different creds but I don't think you have something > that would catch this case? Right. I don't believe the spec requires us to catch false retries in every possible case. That may mean we return a pretty bizarre reply that doesn't match the request, but that's the client's own fault.... --b.