Breakage from patch: Only root should be able to set the N_MOUSE line discipline.

Breakage from patch: Only root should be able to set the N_MOUSE line discipline.

[Alan Cox]

On Gwe, 2005-01-28 at 16:12, Linux Kernel Mailing List wrote:
> ChangeSet 1.1977.1.2, 2005/01/28 17:12:20+01:00, vojtech@suse.cz
> 
> 	input: Only root should be able to set the N_MOUSE line discipline.
> 	

I finally had a chance to trace down why my mouse code for a little gui
library started working differently and causing problems. This broken
change breaks apps that use framebuffer in unpriviledged process form
and want to use the mouse support in kernel and forces them to become
setuid root or to revert to 2.4 style user space mouse drivers. If this
functonality is root only kernel space it might as well be entirely
deleted IMHO.

I can see no reason for this change - the ldisc is supposed to be
configurable by non root users. It is reset on close/hangup in Linux so
a user cannot jam a port up.

Can someone please justify this change. If not can it be reverted

Alan

Re: Breakage from patch: Only root should be able to set the N_MOUSE line discipline.

[Vojtech Pavlik]

On Sat, Feb 26, 2005 at 11:20:44AM +0000, Alan Cox wrote:

> On Gwe, 2005-01-28 at 16:12, Linux Kernel Mailing List wrote:
> > ChangeSet 1.1977.1.2, 2005/01/28 17:12:20+01:00, vojtech@suse.cz
> > 
> > 	input: Only root should be able to set the N_MOUSE line discipline.
> > 	
> 
> I finally had a chance to trace down why my mouse code for a little gui
> library started working differently and causing problems. This broken
> change breaks apps that use framebuffer in unpriviledged process form
> and want to use the mouse support in kernel and forces them to become
> setuid root or to revert to 2.4 style user space mouse drivers. If this
> functonality is root only kernel space it might as well be entirely
> deleted IMHO.
> 
> I can see no reason for this change - the ldisc is supposed to be
> configurable by non root users. It is reset on close/hangup in Linux so
> a user cannot jam a port up.
> 
> Can someone please justify this change. If not can it be reverted
 
A nonprivileged user could inject mouse movement and/or keystrokes
(using the sunkbd driver) into the input subsystem, taking over the
console/X, where another user is logged in.

Simply using a slightly modified inputattach on a PTY will do the trick.

-- 
Vojtech Pavlik
SuSE Labs, SuSE CR

Re: Breakage from patch: Only root should be able to set the N_MOUSE line discipline.

[Alan Cox]

On Maw, 2005-03-01 at 11:47, Vojtech Pavlik wrote:
> A nonprivileged user could inject mouse movement and/or keystrokes
> (using the sunkbd driver) into the input subsystem, taking over the
> console/X, where another user is logged in.

Ouch. Ok that explains much.

Re: Breakage from patch: Only root should be able to set the N_MOUSE line discipline.

[Linus Torvalds]

On Tue, 1 Mar 2005, Vojtech Pavlik wrote:
>  
> A nonprivileged user could inject mouse movement and/or keystrokes
> (using the sunkbd driver) into the input subsystem, taking over the
> console/X, where another user is logged in.
> 
> Simply using a slightly modified inputattach on a PTY will do the trick.

Might an alternative be to just make writes to N_MOUSE require privileges?

Ie "reading is ok, and changing to N_MOUSE is ok, but tryign to write a 
mouse packet is not"? The check should be easy enough to add to the 
ldisc.write thing?

		Linus

Re: Breakage from patch: Only root should be able to set the N_MOUSE line discipline.

[Vojtech Pavlik]

On Tue, Mar 01, 2005 at 08:17:47AM -0800, Linus Torvalds wrote:

> On Tue, 1 Mar 2005, Vojtech Pavlik wrote:
> >  
> > A nonprivileged user could inject mouse movement and/or keystrokes
> > (using the sunkbd driver) into the input subsystem, taking over the
> > console/X, where another user is logged in.
> > 
> > Simply using a slightly modified inputattach on a PTY will do the trick.
> 
> Might an alternative be to just make writes to N_MOUSE require privileges?
> 
> Ie "reading is ok, and changing to N_MOUSE is ok, but tryign to write a 
> mouse packet is not"? The check should be easy enough to add to the 
> ldisc.write thing?
 
No, since you wouldn't write anything to the device, the writes would
happen on the other end of the pty.

-- 
Vojtech Pavlik
SuSE Labs, SuSE CR