From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ryusuke Konishi Subject: Re: [PATCH]nilfs2: Don't load/check invalid cno Date: Tue, 28 Jul 2009 02:40:35 +0900 (JST) Message-ID: <20090728.024035.102951444.ryusuke@osrg.net> References: <4A6D7273.5010906@gmail.com> <977a2be20907270226s7c7037b5v38830e1ba68cefcb@mail.gmail.com> Reply-To: NILFS Users mailing list Mime-Version: 1.0 Content-Transfer-Encoding: quoted-printable Return-path: In-Reply-To: <977a2be20907270226s7c7037b5v38830e1ba68cefcb-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org> List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: users-bounces-JrjvKiOkagjYtjvyW6yDsg@public.gmane.org Errors-To: users-bounces-JrjvKiOkagjYtjvyW6yDsg@public.gmane.org Content-Type: text/plain; charset="iso-8859-9" To: users-JrjvKiOkagjYtjvyW6yDsg@public.gmane.org, zhu.yanhai-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org Cc: qiang.z.zhang-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org, yanhai.zhu-VuQAYsv1563Yd54FQh9/CA@public.gmane.org Hi! On Mon, 27 Jul 2009 17:26:58 +0800, Zhu Yanhai wrote: > Below inconsistent error messages will be reported without this patch. > = > [root@zyh-fedora ~]# mount -t nilfs2 /dev/loop0 ./mount1 > mount.nilfs2: WARNING! - The NILFS on-disk format may change at any time. > mount.nilfs2: WARNING! - Do not place critical data on a NILFS filesystem. > [root@zyh-fedora ~]# lscp > CNO DATE TIME MODE FLG NBLKINC ICNT > 1 2009-07-26 16:14:26 cp - 11 3 > 2 2009-07-26 16:16:12 cp - 637 144 > 3 2009-07-26 16:16:47 cp - 17 143 > 4 2009-07-26 16:16:52 cp - 16 142 > 5 2009-07-26 16:16:58 cp - 11 142 > 6 2009-07-26 16:17:08 cp - 18 141 > [root@zyh-fedora ~]# mount -t nilfs2 -r -o cp=3D10 /dev/loop0 ./mount2 > mount.nilfs2: Error while mounting /dev/loop0 on ./mount2: Invalid argume= nt > [root@zyh-fedora ~]# mount -t nilfs2 -r -o cp=3D20 /dev/loop0 ./mount2 > mount.nilfs2: Error while mounting /dev/loop0 on ./mount2: Invalid argume= nt > [root@zyh-fedora ~]# mount -t nilfs2 -r -o cp=3D30 /dev/loop0 ./mount2 > mount.nilfs2: Error while mounting /dev/loop0 on ./mount2: No such > file or directory > [root@zyh-fedora ~]# mount -t nilfs2 -r -o cp=3D40 /dev/loop0 ./mount2 > mount.nilfs2: Error while mounting /dev/loop0 on ./mount2: No such > file or directory > [root@zyh-fedora ~]# mount -t nilfs2 -r -o cp=3D25 /dev/loop0 ./mount2 > mount.nilfs2: Error while mounting /dev/loop0 on ./mount2: No such > file or directory > [root@zyh-fedora ~]# mount -t nilfs2 -r -o cp=3D21 /dev/loop0 ./mount2 > mount.nilfs2: Error while mounting /dev/loop0 on ./mount2: No such > file or directory = Thank you for pointing out the inconsistency. Your patch noticed me that a few problems exist around the nilfs_cpfile_is_snapshot() function: 1) For invalid checkpoints it should return an ENOENT error, but this check is missing; a test with nilfs_checkpoint_invalid() should come before the nilfs_checkpoint_snapshot() test. 2) If nilfs_cpfile_is_snapshot() returned an ENOENT error, nilfs_fill_super() should convert it to -EINVAL, but this is missing too. (the ENOENT code is for internal use) 3) Snapshots can be mounted concurrently with a writable mount. So, the call site of nilfs_cpfile_is_snapshot() must be protected from segment writer. I mean it should hold read lock for nilfs->ns_segctor_sem during the call-out like: down_read(&nilfs->ns_segctor_sem); err =3D nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, sbi->s_snapshot_cno= ); up_read(&nilfs->ns_segctor_sem); This is needed even to protect the use of nilfs_mdt_cno(). Then, (1) and (2) would fix the inconsistency problem. But I think your patch is still needed to exclude the next checkpoint for snapshot mount. A little bit problematic.. up to four fixes. I would appreciate it if you could rewrite patches taking the above into account. Thanks in advance, Ryusuke Konishi > 2009/7/27 Zhu Yanhai : > > nilfs2: Don't load/check cp block if specified cno is larger than the > > largest exist one. > > nilfs2 would load invalid cp block, and report random inconsistent error > > message under this situation before. > > > > Signed-off-by: Zhu Yanhai > > > > --- > > =A0fs/nilfs2/cpfile.c | =A0 =A07 ++++--- > > =A01 files changed, 4 insertions(+), 3 deletions(-) > > > > diff --git a/fs/nilfs2/cpfile.c b/fs/nilfs2/cpfile.c > > index aec942c..43978a9 100644 > > --- a/fs/nilfs2/cpfile.c > > +++ b/fs/nilfs2/cpfile.c > > @@ -814,9 +814,10 @@ int nilfs_cpfile_is_snapshot(struct inode *cpfile,= __u64 cno) > > =A0 =A0 =A0 =A0struct nilfs_checkpoint *cp; > > =A0 =A0 =A0 =A0void *kaddr; > > =A0 =A0 =A0 =A0int ret; > > - > > - =A0 =A0 =A0 if (cno =3D=3D 0) > > - =A0 =A0 =A0 =A0 =A0 =A0 =A0 return -ENOENT; /* checkpoint number 0 is= invalid */ > > + > > + =A0 =A0 =A0 /* return ENOENT if cno is invalid. */ > > + =A0 =A0 =A0 if (cno =3D=3D 0 || cno >=3D nilfs_mdt_cno(cpfile)) > > + =A0 =A0 =A0 =A0 =A0 =A0 =A0 return -ENOENT; > > =A0 =A0 =A0 =A0down_read(&NILFS_MDT(cpfile)->mi_sem); > > > > =A0 =A0 =A0 =A0ret =3D nilfs_cpfile_get_checkpoint_block(cpfile, cno, 0= , &bh); > > -- > > 1.6.2.2 > >