Re: [PATCH] cifs: eliminate CONFIG_CIFS_WEAK_PW_HASH

[Jeff Layton <jlayton-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>]

On Sat, 21 Jan 2012 12:26:43 -0500
simo <idra-eUNUBHrolfbYtjvyW6yDsg@public.gmane.org> wrote:

> On Sat, 2012-01-21 at 07:37 -0500, Jeff Layton wrote: 
> > On Sat, 21 Jan 2012 15:03:31 +1100
> > Andrew Bartlett <abartlet-eUNUBHrolfbYtjvyW6yDsg@public.gmane.org> wrote:
> > 
> > > On Fri, 2012-01-20 at 14:45 -0600, Steve French wrote:
> > > > My general thinking on this is as follows:
> > > > 
> > > > If the kernel is distributed to all the workstations in an organization
> > > > with this Kconfig option disabled, it makes it harder for individual users
> > > > to make the mistake of enabling lanman (sec=lanman, or the Kconfig
> > > > option) on a public network and thus send weak password hashes
> > > > which could be discovered simply.   Most distros make the choice
> > > > of enabling broader compatibility with old pre-1997 servers but
> > > > it is a very small set of servers who would require lanman support,
> > > > and a large number of potential attackers who could benefit if
> > > > users enable lanman on a public network.  I suspect that there
> > > > are environments where removing code (via Kconfig) is preferred
> > > > to trusting all owners of all workstations running that organizations
> > > > standard linux to never enable lanman at runtime.
> > > > 
> > > > But ... the opinion of security specialists on this would be welcome.
> > > 
> > > We have been though some of this with the kerberos libs, which now allow
> > > (default?) to not even compile with weak crypto.  If the weak crypto is
> > > not compiled in, it can therefore be asserted that the weak crypto
> > > cannot be used, and this makes it easier to comply with security
> > > audits/certification etc.
> > > 
> > > I don't want to make your code more complex than it needs to be, but LM
> > > encryption really, really needs to go away.  If it is not a major
> > > bother, I would like to make it easier for that to happen if possible.
> > > 
> > 
> > The only way for it to go away completely is for all servers that
> > support only that encryption to go away completely. Unfortunately,
> > that's a tall order -- there are still at least some in the field and
> > people need to get at data on them.
> 
> Jeff, can you identify them ?
> 
> LM only servers means pre Win 95 machines, I'd be curious to know what
> servers are there that really support only LM hashes and not NT hashes.
> 

Sorry for the long delay in responding here. Yes, mostly pre-win95
machines. We occasionally get reports from people using OS/2 and I'm
pretty sure it's LM-only.

Steve, in any case...shall I consider this patch NAK'ed for now? I was
carrying it in my tree for 3.5, but it seems like there's resistance to
removing this option and I'm not particularly religious on the matter.

-- 
Jeff Layton <jlayton-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org>