From mboxrd@z Thu Jan 1 00:00:00 1970 From: "J. Bruce Fields" Subject: Re: [PATCH 5/5] rpc: add new gssd upcall pipe Date: Sat, 14 Jun 2008 13:36:11 -0400 Message-ID: <20080614173611.GD27041@fieldses.org> References: <1213397442-15611-1-git-send-email-bfields@citi.umich.edu> <1213397442-15611-2-git-send-email-bfields@citi.umich.edu> <1213397442-15611-3-git-send-email-bfields@citi.umich.edu> <1213397442-15611-4-git-send-email-bfields@citi.umich.edu> <1213397442-15611-5-git-send-email-bfields@citi.umich.edu> <1213397442-15611-6-git-send-email-bfields@citi.umich.edu> <1213459638.7149.16.camel@localhost> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: aglo@citi.umich.edu, kwc@citi.umich.edu, linux-nfs@vger.kernel.org To: Trond Myklebust Return-path: Received: from mail.fieldses.org ([66.93.2.214]:58105 "EHLO fieldses.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752546AbYFNRgN (ORCPT ); Sat, 14 Jun 2008 13:36:13 -0400 In-Reply-To: <1213459638.7149.16.camel@localhost> Sender: linux-nfs-owner@vger.kernel.org List-ID: On Sat, Jun 14, 2008 at 12:07:18PM -0400, Trond Myklebust wrote: > On Fri, 2008-06-13 at 18:50 -0400, J. Bruce Fields wrote: > > This adds the new text-based gss upcall. We only allow an open from > > either the new pipe or the old one, and use the pipe_open method to > > enforce this. > > > > If no pipes are open, we provisionally queue messages for the new pipe; > > a subsequent open of the old pipe will cause all those messages to be > > purged. > > > > Signed-off-by: J. Bruce Fields > > --- > > I'm not happy about allowing the gss_auth layer to control timeouts and > cancel upcalls either: that caused us way too much grief and bugs in > earlier revisions of rpc_pipefs. Understood. In this case the only cancel occurs before any open has come--it seems pretty simple. But perhaps I'm missing something--I'll think about it some more. > Why not rather disallow queueing (i.e. put the task to sleep, and maybe > print out a reminder on the console every minute or so) until you are > notified by the rpc_pipefs layer that someone has opened the file for > reading? I have another version of these patches that did that (in fact, I just turned off the RPC_PIPE_WAIT_FOR_OPEN stuff and relied entirely on timeouts in gss code). I'll take another look at that. --b.