Return-Path: Received: from zeniv.linux.org.uk ([195.92.253.2]:54029 "EHLO ZenIV.linux.org.uk" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750940AbcD2H6P (ORCPT ); Fri, 29 Apr 2016 03:58:15 -0400 Date: Fri, 29 Apr 2016 08:58:12 +0100 From: Al Viro To: Jeff Layton Cc: linux-nfs@vger.kernel.org, linux-fsdevel@vger.kernel.org, Trond Myklebust , Linus Torvalds Subject: Re: parallel lookups on NFS Message-ID: <20160429075812.GY25498@ZenIV.linux.org.uk> References: <20160424023453.GK25498@ZenIV.linux.org.uk> <1461501975.5219.40.camel@poochiereds.net> <20160424191835.GL25498@ZenIV.linux.org.uk> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: <20160424191835.GL25498@ZenIV.linux.org.uk> Sender: linux-nfs-owner@vger.kernel.org List-ID: On Sun, Apr 24, 2016 at 08:18:35PM +0100, Al Viro wrote: > What we get out of that is fully parallel lookup/readdir/sillyunlink - all > exclusion is on per-name basis (nfs_prime_dcache() vs. nfs_lookup() vs. > nfs_do_call_unlink()). It will require a bit of care in atomic_open(), > though... > > I'll play with that a bit and see what can be done... OK, a bunch of atomic_open cleanups (moderately tested) + almost untested sillyunlink patch are in vfs.git#untested.nfs. It ought to make lookups (and readdir, and !O_CREAT case of atomic_open) on NFS really execute in parallel. Folks, please hit that sucker with NFS torture tests. In particular, the stuff mentioned in commit 565277f6 would be interesting to try.