From mboxrd@z Thu Jan 1 00:00:00 1970 From: Nick Piggin Subject: Re: Status of buffered write path (deadlock fixes) Date: Tue, 12 Dec 2006 01:20:52 +1100 Message-ID: <457D6944.4010703@yahoo.com.au> References: <45751712.80301@yahoo.com.au> <20061207195518.GG4497@ca-server1.us.oracle.com> <4578DBCA.30604@yahoo.com.au> <20061208234852.GI4497@ca-server1.us.oracle.com> <457D20AE.6040107@yahoo.com.au> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Cc: Linux Memory Management , linux-fsdevel@vger.kernel.org, linux-kernel , OGAWA Hirofumi , Andrew Morton Return-path: Received: from smtp107.mail.mud.yahoo.com ([209.191.85.217]:36639 "HELO smtp107.mail.mud.yahoo.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with SMTP id S933568AbWLKOVq (ORCPT ); Mon, 11 Dec 2006 09:21:46 -0500 To: Mark Fasheh In-Reply-To: <457D20AE.6040107@yahoo.com.au> Sender: linux-fsdevel-owner@vger.kernel.org List-Id: linux-fsdevel.vger.kernel.org Nick Piggin wrote: > Mark Fasheh wrote: >> If we make the change I described above (looking for BH_New buffers >> outside >> the range passed), then zero length or partial shouldn't matter, but zero >> length instead of partial would be nicer imho just for the sake of >> reducing >> the total number of cases down to the entire range or zero length. > > > We don't want to do zero length, because we might make the theoretical > livelock much easier to hit (eg. in the case of many small iovecs). But > yes we can restrict ourselves to zero-length or full-length. On second thoughts, I think I'm wrong about that. Consider the last page of a file, which is uptodate. A full length commit, which extends the file, will expose transient zeroes if the usercopy fails. -- SUSE Labs, Novell Inc. Send instant messages to your online friends http://au.messenger.yahoo.com