Re: [PATCH 1/4] add compat_ioctl methods to dasd

From: Martin Schwidefsky
Date: Thu Nov 17 2005 - 07:42:56 EST

On Wed, 2005-11-16 at 09:45 +0100, Christoph Hellwig wrote:
> On Tue, Nov 15, 2005 at 06:24:38PM +0100, Christoph Hellwig wrote:
> > > and doesn't work after fixing the compile problem. It's a
> > > problem with the bdev->bd_disk->private_data which is NULL at the time
> > > the partition detection code calls the BIODASDINFO and HDIO_GETGEO ioctl
> > > with ioctl_by_bdev. I don't see an easy way to fix this right now.
> >
> > my patch doesn't change anything related to dereferencing those fields.
> >
> > I see the problem that you're probably having: ioctl_by_bdev calls
> > ->ioctl without ensuring ->open has been called previously. But I don't
> > see why this couldn't have happened previously.
> So looking at it again I found a bug: we return EINVAL on an invalid
> ioctl, but the compat layer expects ENOIOCTLCMD so it returns using the
> generic compat bits. Returning ENOIOCTLCMD from unlocked_ioctl is fine
> aswell as the ioctl layer turns it back. The patch below fix this issue
> and the missing semicolon after lock_kernel()

Still doesn't work. The reason is ioctl_by_bdev:

int ioctl_by_bdev(struct block_device *bdev,
unsigned cmd, unsigned long arg)
int res;
mm_segment_t old_fs = get_fs();
res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
return res;

blkdev_ioctl is explicitly called with a NULL file pointer. That can't
work with block device drivers that use unlocked ioctls. We'd need to
create a struct file with at least a valid f_dentry->d_inode->i_bdev
chain in ioctl_by_bdev to make it work with unlocked ioctls. Not nice ..

blue skies,

Martin Schwidefsky
Linux for zSeries Development & Services
IBM Deutschland Entwicklung GmbH

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at
Please read the FAQ at