CentOS - Jun 2015 - ssh -X versus -Y

On Fri, 26 Jun 2015 at 03:16 -0000, Alexandru Chiscan wrote:
> On 06/25/2015 11:51 PM, Stuart Barkley wrote:
> > Then from your desktop (assuming Linux already running X) in a
> > local xterm do something like:
> >
> >      ssh -Y remote-system
>
> Do not use that because any user logged on the server can connect to
> your X server display and snoop what you are doing, open windows
> etc.
>
> -Y disables all the X server authentication mechanisms
> (http://www.x.org/wiki/Development/Documentation/Security/)
I believe this is incorrect.  The authentication protocols are still
used (thus the need for the xorg-x11-xauth package on CentOS).

This is not the same as 'xhost +' which should never be used.

Both -X and -Y require read access to the ~/.Xauthority file on the
remote system in order to connect back to the X server.  You can see
this by using the 'xauth remove' command to remove the authentication
token for the display and clients can no longer connect.
> > Note about -X versus -Y with ssh:
> >
> > -X enables basic X forwarding, It disables some X functionality
> > making it "safer" to allow.  -X also stops working after
about 20
> > minutes (this is by design but not well documented).  I only
> > recently learned why it would stop working after pulling out the
> > last of my hair.
>
> I have been using ssh X forwarding for current work use (local
> betwork) for more than 15 years and never got into this kind of
> problem from RH 7 to Centos 7, AIX and Solaris.
Likewise, I've used ssh/X for 20+ years on a variety of systems.  In
most cases -X is sufficient, but some applications seem to require -Y
and I have not dug into all of the reasons.
> Maybe it is some other issue that is closing your ssh connection
> (maybe you should use the KeepAlive options on the ssh
> server/client); just guessing.
On Debian and FreeBSD 'man ssh_config' now shows:

     ForwardX11Timeout
	     Specify a timeout for untrusted X11 forwarding using
	     the format described in the TIME FORMATS section of
	     sshd_config(5).  X11 connections received by ssh(1)
	     after this time will be refused.  The default is to
	     disable untrusted X11 forwarding after twenty minutes
	     has elapsed.

This option seems to have appeared in OpenSSH 6.0 [See more at the end
of this email].
> > -Y allows the full X protocol which might be a security risk.
> > Some applications will only work with -Y.  With this, remote X
> > applications can grab keyboard interactions, grab passwords, put
> > windows on top of other windows (obscuring security messages),
> > etc.
> >
> > For my own choice I use -Y (although I only enable it occasionally
> > to specific systems).
This is a recent behaviour change on my part with limited use to
system I trust.  Now that I know about the timeout I can probably just
live with -X and will consider moving back to -X for my occasional
use,

The documentation of the practical differences between -X and -Y is
pretty obscure (mostly defering to the X Security extension
documentation).  I would like to see better clarification of the
differences.

...some additional research looking at the source code and
revision history...

The ForwardX11Timeout change was 5 years ago in OpenSSH 6.0.  CentOS 6
still has OpenSSH 5.3 without this change (or if a patch was applied
the documentation was not also patched).  CentOS 7 has OpenSSH 6.6
which includes this change.

   
http://cvsweb.openbsd.org/cgi-bin/cvsweb/src/usr.bin/ssh/clientloop.c.diff?r1=1.220&r2=1.221

Prior to that there was an intended hard coded 20 minute timeout since
at least 2005:

   
http://cvsweb.openbsd.org/cgi-bin/cvsweb/src/usr.bin/ssh/clientloop.c.diff?r1=1.137&r2=1.138

Based upon the comments in the June 2010 revision it appears that the
timeout was not working correctly and perhaps was falling back to
trusted authentication after 20 minutes:

    Add X11ForwardTimeout option to specify timeout for untrusted
    X11 authentication cookies to avoid fallback in X11 code to
    fully-trusted implicit authentication using SO_PEERCRED described
    at: http://lists.x.org/archives/xorg-devel/2010-May/008636.html

    After the X11ForwardTimeout has expired the client will now
    refuse incoming X11 channel opens.

I will need to see it this is an unpatched security issue on
CentOS/RedHat 6.  If so, I claim credit for observing it as a
possibility.

Stuart
-- 
I've never been lost; I was once bewildered for three days, but never lost!
                                        --  Daniel Boone

I stand corrected.

Regards,
Lec

On 06/26/2015 07:22 PM, Stuart Barkley wrote:
> On Fri, 26 Jun 2015 at 03:16 -0000, Alexandru Chiscan wrote:
>
>> On 06/25/2015 11:51 PM, Stuart Barkley wrote:
>>> Then from your desktop (assuming Linux already running X) in a
>>> local xterm do something like:
>>>
>>>       ssh -Y remote-system
>> Do not use that because any user logged on the server can connect to
>> your X server display and snoop what you are doing, open windows
>> etc.
>>
>> -Y disables all the X server authentication mechanisms
>> (http://www.x.org/wiki/Development/Documentation/Security/)
> I believe this is incorrect.  The authentication protocols are still
> used (thus the need for the xorg-x11-xauth package on CentOS).
>
> This is not the same as 'xhost +' which should never be used.
>
> Both -X and -Y require read access to the ~/.Xauthority file on the
> remote system in order to connect back to the X server.  You can see
> this by using the 'xauth remove' command to remove the
authentication
> token for the display and clients can no longer connect.
>
>>> Note about -X versus -Y with ssh:
>>>
>>> -X enables basic X forwarding, It disables some X functionality
>>> making it "safer" to allow.  -X also stops working after
about 20
>>> minutes (this is by design but not well documented).  I only
>>> recently learned why it would stop working after pulling out the
>>> last of my hair.
>> I have been using ssh X forwarding for current work use (local
>> betwork) for more than 15 years and never got into this kind of
>> problem from RH 7 to Centos 7, AIX and Solaris.
> Likewise, I've used ssh/X for 20+ years on a variety of systems.  In
> most cases -X is sufficient, but some applications seem to require -Y
> and I have not dug into all of the reasons.
>
>> Maybe it is some other issue that is closing your ssh connection
>> (maybe you should use the KeepAlive options on the ssh
>> server/client); just guessing.
> On Debian and FreeBSD 'man ssh_config' now shows:
>
>       ForwardX11Timeout
> 	     Specify a timeout for untrusted X11 forwarding using
> 	     the format described in the TIME FORMATS section of
> 	     sshd_config(5).  X11 connections received by ssh(1)
> 	     after this time will be refused.  The default is to
> 	     disable untrusted X11 forwarding after twenty minutes
> 	     has elapsed.
>
> This option seems to have appeared in OpenSSH 6.0 [See more at the end
> of this email].
>
>>> -Y allows the full X protocol which might be a security risk.
>>> Some applications will only work with -Y.  With this, remote X
>>> applications can grab keyboard interactions, grab passwords, put
>>> windows on top of other windows (obscuring security messages),
>>> etc.
>>>
>>> For my own choice I use -Y (although I only enable it occasionally
>>> to specific systems).
> This is a recent behaviour change on my part with limited use to
> system I trust.  Now that I know about the timeout I can probably just
> live with -X and will consider moving back to -X for my occasional
> use,
>
> The documentation of the practical differences between -X and -Y is
> pretty obscure (mostly defering to the X Security extension
> documentation).  I would like to see better clarification of the
> differences.
>
> ...some additional research looking at the source code and
> revision history...
>
> The ForwardX11Timeout change was 5 years ago in OpenSSH 6.0.  CentOS 6
> still has OpenSSH 5.3 without this change (or if a patch was applied
> the documentation was not also patched).  CentOS 7 has OpenSSH 6.6
> which includes this change.
>
>     
http://cvsweb.openbsd.org/cgi-bin/cvsweb/src/usr.bin/ssh/clientloop.c.diff?r1=1.220&r2=1.221
>
> Prior to that there was an intended hard coded 20 minute timeout since
> at least 2005:
>
>     
http://cvsweb.openbsd.org/cgi-bin/cvsweb/src/usr.bin/ssh/clientloop.c.diff?r1=1.137&r2=1.138
>
> Based upon the comments in the June 2010 revision it appears that the
> timeout was not working correctly and perhaps was falling back to
> trusted authentication after 20 minutes:
>
>      Add X11ForwardTimeout option to specify timeout for untrusted
>      X11 authentication cookies to avoid fallback in X11 code to
>      fully-trusted implicit authentication using SO_PEERCRED described
>      at: http://lists.x.org/archives/xorg-devel/2010-May/008636.html
>
>      After the X11ForwardTimeout has expired the client will now
>      refuse incoming X11 channel opens.
>
> I will need to see it this is an unpatched security issue on
> CentOS/RedHat 6.  If so, I claim credit for observing it as a
> possibility.
>
> Stuart

On 2015-06-26, Stuart Barkley <stuartb at 4gh.net> wrote:

[...]
> The documentation of the practical differences between -X and -Y is
> pretty obscure (mostly defering to the X Security extension
> documentation).  I would like to see better clarification of the
> differences.
One practical difference I have seen is the improved performance of -Y
over -X. I have long attributed that to the relaxation of security
controls in the former case.

-- 

Liam

On 07/05/2015 04:51 AM, Liam O'Toole wrote:
> One practical difference I have seen is the improved performance of -Y
> over -X. I have long attributed that to the relaxation of security
> controls in the former case.
When and how did you measure that?

The -Y change was introduced in Fedora Core 3, in November 2004.  The 
default was changed to ForwardX11Trusted=yes just a month or two later. 
  I'm not sure -X and -Y ever behaved differently on Enterprise Linux or 
CentOS.

At this point, I don't think it's even possible to set 
ForwardX11Trusted=no any more.  The X SECURITY extension was replaced 
with "X Access Control Extension" several years ago.