Return-Path: Received: from mail-io0-f194.google.com ([209.85.223.194]:33662 "EHLO mail-io0-f194.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755550AbcLSQyW (ORCPT ); Mon, 19 Dec 2016 11:54:22 -0500 Received: by mail-io0-f194.google.com with SMTP id p13so19719578ioi.0 for ; Mon, 19 Dec 2016 08:54:22 -0800 (PST) Subject: Re: [PATCH 1/9] NFSv4: Don't invalidate the directory twice To: Trond Myklebust , linux-nfs@vger.kernel.org References: <20161217182711.10643-1-trond.myklebust@primarydata.com> <20161217182711.10643-2-trond.myklebust@primarydata.com> From: Anna Schumaker Message-ID: <2099263b-3d8e-f237-3dd8-50dc0f63fb3e@gmail.com> Date: Mon, 19 Dec 2016 11:54:20 -0500 MIME-Version: 1.0 In-Reply-To: <20161217182711.10643-2-trond.myklebust@primarydata.com> Content-Type: text/plain; charset=windows-1252 Sender: linux-nfs-owner@vger.kernel.org List-ID: Hi Trond, On 12/17/2016 01:27 PM, Trond Myklebust wrote: > If an operation that modified the directory raced with a GETATTR, then we > don't need to invalidate the directory cache more than once. This patch causes cthon basic tests to fail with: ./test6: readdir ./test6: (/nfs/all) unlinked 'file.0' dir entry read pass 1 ./test6: (/nfs/all) Test failed with 1 errors basic tests failed Thanks, Anna > > Signed-off-by: Trond Myklebust > --- > fs/nfs/nfs4proc.c | 3 +++ > 1 file changed, 3 insertions(+) > > diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c > index d33242c8d95d..713932440e07 100644 > --- a/fs/nfs/nfs4proc.c > +++ b/fs/nfs/nfs4proc.c > @@ -1088,12 +1088,15 @@ static void update_changeattr(struct inode *dir, struct nfs4_change_info *cinfo) > struct nfs_inode *nfsi = NFS_I(dir); > > spin_lock(&dir->i_lock); > + if (dir->i_version == cinfo->after) > + goto out; > nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA; > if (!cinfo->atomic || cinfo->before != dir->i_version) > nfs_force_lookup_revalidate(dir); > dir->i_version = cinfo->after; > nfsi->attr_gencount = nfs_inc_attr_generation_counter(); > nfs_fscache_invalidate(dir); > +out: > spin_unlock(&dir->i_lock); > } > >