Re: [uml-user] Local Devices question

[Jeff D. <jd...@ad...>]

On Tue, Mar 18, 2008 at 12:30:53PM +0100, Flavio wrote:
> >
> >  As a sanity-check, you can do this again, look for the line above, and
> >  make sure the file descriptor (16 in this case) refers to /dev/cdrom.
> >  Look in /proc/<pid>/fd for the same pid that you strace.
> I don't know why I don't have the /proc/16 directory.
> As a matter of fact I don't know how to exactly verify if the fd 16
> refers to /dev/cdrom (maybe there's some file to cat).

You forgot the pid.

> >
> >  As another sanity-check, strace eject running on the host.  It's
> >  possible that CDROMEJECT fails the same way and one of its fallbacks
> >  works.  In that case, I just need to make UML use the same fallbacks.
> As I said before, eject perfectly works on the host.
> This is the strace 'eject /dev/cdrom' output related to the above line:
> ioctl(3, CDROMEJECT, 0x1)               = 0

I just noticed something.  I forgot to pass down the ioctl argument.
See if the patch below makes any differece.

> And the directory /proc/3/ exists on the host.

/proc/3 is irrelevant.  You want /proc/<eject-pid>/3.

	   		    Jeff

-- 
Work email - jdike at linux dot intel dot com

Index: linux-2.6.22/arch/um/drivers/ubd_kern.c
===================================================================
--- linux-2.6.22.orig/arch/um/drivers/ubd_kern.c	2008-03-06 08:57:08.000000000 -0500
+++ linux-2.6.22/arch/um/drivers/ubd_kern.c	2008-03-18 11:29:55.000000000 -0400
@@ -1162,6 +1162,10 @@ static int ubd_ioctl(struct inode * inod
 		if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
 			return -EFAULT;
 		return 0;
+	case CDROMEJECT:
+		if (ubd_dev->fd == -1)
+			return -ENXIO;
+		return os_ioctl_generic(ubd_dev->fd, CDROMEJECT, arg);
 	}
 	return -EINVAL;
 }