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

[Martin Schwidefsky <schwidefsky@de.ibm.com>]

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();
        set_fs(KERNEL_DS);
        res = blkdev_ioctl(bdev->bd_inode, NULL, cmd, arg);
        set_fs(old_fs);
        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

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