From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932905AbXBEOeA (ORCPT ); Mon, 5 Feb 2007 09:34:00 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S932929AbXBEOd7 (ORCPT ); Mon, 5 Feb 2007 09:33:59 -0500 Received: from mx1.redhat.com ([66.187.233.31]:52718 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932905AbXBEOd6 (ORCPT ); Mon, 5 Feb 2007 09:33:58 -0500 Subject: [GFS2] Fix recursive locking attempt with NFS [43/54] From: Steven Whitehouse To: linux-kernel@vger.kernel.org Cc: cluster-devel@redhat.com In-Reply-To: <1170684425.11001.267.camel@quoit.chygwyn.com> References: <1170684425.11001.267.camel@quoit.chygwyn.com> Content-Type: text/plain Organization: Red Hat (UK) Ltd Date: Mon, 05 Feb 2007 14:41:58 +0000 Message-Id: <1170686518.11001.357.camel@quoit.chygwyn.com> Mime-Version: 1.0 X-Mailer: Evolution 2.2.2 (2.2.2-5) Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org >>From 57c070ec9322815caaa33f4a669b86fac9002075 Mon Sep 17 00:00:00 2001 From: Steven Whitehouse Date: Thu, 25 Jan 2007 17:14:59 +0000 Subject: [PATCH] [GFS2] Fix recursive locking attempt with NFS In certain cases, its possible for NFS to call the lookup code while holding the glock (when doing a readdirplus operation) so we need to check for that and not try and lock the glock twice. This also fixes a typo in a previous NFS related GFS2 patch. Signed-off-by: Steven Whitehouse diff --git a/fs/gfs2/inode.c b/fs/gfs2/inode.c index f7c8d31..88fcfb4 100644 --- a/fs/gfs2/inode.c +++ b/fs/gfs2/inode.c @@ -395,8 +395,10 @@ struct inode *gfs2_lookup_simple(struct inode *dip, const char *name) * @is_root: If 1, ignore the caller's permissions * @i_gh: An uninitialized holder for the new inode glock * - * There will always be a vnode (Linux VFS inode) for the d_gh inode unless - * @is_root is true. + * This can be called via the VFS filldir function when NFS is doing + * a readdirplus and the inode which its intending to stat isn't + * already in cache. In this case we must not take the directory glock + * again, since the readdir call will have already taken that lock. * * Returns: errno */ @@ -409,8 +411,9 @@ struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name, struct gfs2_holder d_gh; struct gfs2_inum_host inum; unsigned int type; - int error = 0; + int error; struct inode *inode = NULL; + int unlock = 0; if (!name->len || name->len > GFS2_FNAMESIZE) return ERR_PTR(-ENAMETOOLONG); @@ -422,9 +425,12 @@ struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name, return dir; } - error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); - if (error) - return ERR_PTR(error); + if (gfs2_glock_is_locked_by_me(dip->i_gl) == 0) { + error = gfs2_glock_nq_init(dip->i_gl, LM_ST_SHARED, 0, &d_gh); + if (error) + return ERR_PTR(error); + unlock = 1; + } if (!is_root) { error = permission(dir, MAY_EXEC, NULL); @@ -439,10 +445,11 @@ struct inode *gfs2_lookupi(struct inode *dir, const struct qstr *name, inode = gfs2_inode_lookup(sb, &inum, type); out: - gfs2_glock_dq_uninit(&d_gh); + if (unlock) + gfs2_glock_dq_uninit(&d_gh); if (error == -ENOENT) return NULL; - return inode; + return inode ? inode : ERR_PTR(error); } static int pick_formal_ino_1(struct gfs2_sbd *sdp, u64 *formal_ino) diff --git a/fs/gfs2/ops_inode.c b/fs/gfs2/ops_inode.c index 747c731..5591f89 100644 --- a/fs/gfs2/ops_inode.c +++ b/fs/gfs2/ops_inode.c @@ -1018,7 +1018,7 @@ static int gfs2_getattr(struct vfsmount *mnt, struct dentry *dentry, } generic_fillattr(inode, stat); - if (unlock); + if (unlock) gfs2_glock_dq_uninit(&gh); return 0; -- 1.4.4.2