[Security] [PATCH 00/20] world-writable files in sysfs and debugfs

[Greg KH]

On Tue, Mar 15, 2011 at 07:50:28AM -0400, James Bottomley wrote:
> On Mon, 2011-03-14 at 20:09 -0700, Greg KH wrote:
> > On Mon, Mar 14, 2011 at 10:26:05PM -0400, James Bottomley wrote:
> > > On Sat, 2011-03-12 at 23:23 +0300, Vasiliy Kulikov wrote:
> > > > > Vasiliy Kulikov (20):
> > > > >  mach-ux500: mbox-db5500: world-writable sysfs fifo file
> > > > >  leds: lp5521: world-writable sysfs engine* files
> > > > >  leds: lp5523: world-writable engine* sysfs files
> > > > >  misc: ep93xx_pwm: world-writable sysfs files
> > > > >  rtc: rtc-ds1511: world-writable sysfs nvram file
> > > > >  scsi: aic94xx: world-writable sysfs update_bios file
> > > > >  scsi: iscsi: world-writable sysfs priv_sess file
> > > > 
> > > > These are still not merged :(
> > > 
> > > OK, so I've not been tracking where we are in the dizzying ride on
> > > security systems.  However, I thought we landed up in the privilege
> > > separation arena using capabilities.  That means that world writeable
> > > files aren't necessarily a problem as long as the correct capabilities
> > > checks are in place, right?
> > 
> > There are no capability checks on sysfs files right now, so these all
> > need to be fixed.
> 
> That statement is true but irrelevant, isn't it?  There can't be
> capabilities within sysfs files because the system that does them has no
> idea what the capabilities would be.  If there were capabilities checks,
> they'd have to be in the implementing routines.

Ah, you are correct, sorry for the misunderstanding.

> I think the questions are twofold:
> 
>      1. Did anyone actually check for capabilities before assuming world
>         writeable files were wrong?

I do not think so as the majority (i.e. all the ones that I looked at)
did no such checks.

>      2. Even if there aren't any capabilities checks in the implementing
>         routines, should there be (are we going the separated
>         capabilities route vs the monolithic root route)?

I think the general consensus is that we go the monolithic root route
for sysfs files in that we do not allow them to be world writable.

Do you have any exceptions that you know of that do these checks?

thanks,

greg k-h