Too verbose compat_ioctl messages?

[H. Peter Anvin]

Geert Uytterhoeven wrote:
> 
> Fuse-san discovered that running the umount that's part of busybox on a PS3
> with a recent kernel causes an error message to be printed on the console:
> 
> | ioctl32(busybox:1340): Unknown cmd fd(3) cmd(00004c01){t:'L';sz:0} arg(00000000) on /dev/sda1 
> 
> On older kernels (e.g. 2.6.16), this doesn't happen.
> 
> It can easily be reproduced by installing busybox and running
> 
> | busybox umount /mountpoint
> 
> on a mounted filesystem (except when using the loop device).
> 
> Apparently Busybox uses the LOOP_CLR_FD ioctl when unmounting a file system,
> which is supported by the loop device only.
> On other block device types, this ioctl is not supported:
>   - With a 64-bit application, the block layer returns ENOTTY (Inappropriate
>     ioctl for device), while the SCSI layer returns EINVAL (Invalid argument)
>   - With a 32-bit application, the compat_ioctl code returns EINVAL (Invalid
>     argument) and prints an error on the console (for the first 50
>     occurrencies, cfr. fs/compat_ioctl.c:compat_sys_ioctl())
> 
> As I understand, compat_ioctl_error() is used to inform the user about ioctl
> values that are not yet handled by the compat_ioctl layer. However, LOOP_CLR_FD
> doesn't need to be handled (no data to convert between 32-bit and 64-bit), and
> it's perfectly valid for a block device to not implement it.
> So it's confusing to print this error message.
> 
> Is there anything we can do about this?
> 

For one thing, it looks like we're returning the wrong thing (EINVAL
rather than ENOTTY) across the board.  This was unfortunately a common
misunderstanding with non-tty-related ioctls in the early days of Linux.

	-hpa