Skip site navigation (1)Skip section navigation (2)
Date:      Fri, 2 Apr 2010 12:04:01 +0300
From:      Kostik Belousov <kostikbel@gmail.com>
To:        Mikolaj Golub <to.my.trociny@gmail.com>
Cc:        freebsd-fs@freebsd.org
Subject:   Re: Lock leakage on rename if vop->vop_rename is NULL (kern/107439)
Message-ID:  <20100402090401.GG2415@deviant.kiev.zoral.com.ua>
In-Reply-To: <86sk7e1gad.fsf@zhuzha.ua1>
References:  <86sk7e1gad.fsf@zhuzha.ua1>

next in thread | previous in thread | raw e-mail | index | archive | help

--wQFNOvpuONRGomCw
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

[Do not want to get "YOU ARE NOT IN WHITELIST" garbage]

On Fri, Apr 02, 2010 at 10:10:02AM +0300, Mikolaj Golub wrote:
> Hi,
>=20
> Could somebody look at kern/107439? Although the pr is closed the issue s=
till
> persists: if a file system does not provide vop_rename the vop_bypass() is
> called and the vnods, which are expected to be unlocked in VOP_RENAME(), =
are
> left locked.
>=20
> I have attached there the patch that solves the problem but I am not conf=
ident
> about the solution.

Patch seems to be eaten by unnamed daemon, but survived in the PR log.
I think that ap->ap_tvp should be handled too.

Anyway, I assume that the proper solution there would be to implement
default vop_rename that is in fact identical to deadfs_rename(). The
later exists exactly to handle a very similar situation.

Please test the patch below. If it works for you, I will combine
deadfs_rename() and vop_norename().

Thanks for the report and patch.

diff --git a/sys/kern/vfs_default.c b/sys/kern/vfs_default.c
index 50bf0d2..7e278a5 100644
--- a/sys/kern/vfs_default.c
+++ b/sys/kern/vfs_default.c
@@ -67,6 +67,7 @@ __FBSDID("$FreeBSD$");
 #include <vm/vnode_pager.h>
=20
 static int	vop_nolookup(struct vop_lookup_args *);
+static int	vop_norename(struct vop_rename_args *);
 static int	vop_nostrategy(struct vop_strategy_args *);
 static int	get_next_dirent(struct vnode *vp, struct dirent **dpp,
 				char *dirbuf, int dirbuflen, off_t *off,
@@ -113,6 +114,7 @@ struct vop_vector default_vnodeops =3D {
 	.vop_poll =3D		vop_nopoll,
 	.vop_putpages =3D		vop_stdputpages,
 	.vop_readlink =3D		VOP_EINVAL,
+	.vop_rename =3D		vop_norename,
 	.vop_revoke =3D		VOP_PANIC,
 	.vop_strategy =3D		vop_nostrategy,
 	.vop_unlock =3D		vop_stdunlock,
@@ -206,6 +208,27 @@ vop_nolookup(ap)
 }
=20
 /*
+ * vop_norename:
+ *
+ * Handle unlock and reference counting for arguments of vop_rename
+ * for filesystems that do not implement rename operation.
+ */
+static int
+vop_norename(struct vop_rename_args *ap)
+{
+
+	if (ap->a_tvp !=3D NULL)
+		vput(ap->a_tvp);
+	if (ap->a_tdvp =3D=3D ap->a_tvp)
+		vrele(ap->a_tdvp);
+	else
+		vput(ap->a_tdvp);
+	vrele(ap->a_fdvp);
+	vrele(ap->a_fvp);
+	return (EOPNOTSUPP);
+}
+
+/*
  *	vop_nostrategy:
  *
  *	Strategy routine for VFS devices that have none.

--wQFNOvpuONRGomCw
Content-Type: application/pgp-signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (FreeBSD)

iEYEARECAAYFAku1swEACgkQC3+MBN1Mb4ir9gCfYzqnk4so9FZYi4MPTSW7DYt1
uXMAnjXJc7iewWVW7jrn8suOZ0HyMXGa
=DdtH
-----END PGP SIGNATURE-----

--wQFNOvpuONRGomCw--



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20100402090401.GG2415>