openssh unix dev - Mar 2001 - Use of non-user readable (null password) private keys

Executive summary: Why can I not have a private key which is `public' ?


Gory details ....

I'm new to openssh. I've been using ssh for years. However, I'm in
the process
of investigating RH 7.* (0.91 at the moment) and am wanting to be as 
`standard' as possible, so trying openssh.

I looked on http://www.openssh.com/list.html but could not find a list for 
"general OpenSSH discussion", but this was the closest match, so
I'll call it
a bug :-)

We use a client/server model with no `user' accounts on servers.
There are certain operations which a user may require to run with certain 
privs, and we use ssh to do this. The capability may be given to an individual 
user (user-only-readable in their .ssh/), a group (using UN*X group semantics) 
or may be accessible to all users of a particular machine or set of machines 
(e.g. when a user changes their password, a process is woken up on the 
password server).

This all worked fine under ssh, but under openssh load_private_key() does a
		(st.st_mode & 077) != 0) {
and then complains that it is readable and won't use it. (it says "It
is
recommended that your private key files are NOT accessible by others." but 
appears to implement somewhat more than a `recommendation' !)

Is this bug intended as a feature ?   [ :-) ]

I can see no code to disable this test [ other than setting HAVE_CYGWIN and 
writing a check_ntsec() which returns FALSE :-) ]

On Tue, Mar 27, 2001 at 02:11:08PM +0100, Piete Brooks
wrote:
> Executive summary: Why can I not have a private key which is `public' ?
is this a good idea?
> Is this bug intended as a feature ?   [ :-) ]
feature.

many ppl are confused by private/public distinction
and are starting to change permissions for all kind
of files. however it's a bad idea to have the private
key group or world readable, this is why openssh ignore
the key. perhaps we should allow group-readable private keys?

but i really don't like the idea.

On Tue, 27 Mar 2001, Piete Brooks wrote:
> Executive summary: Why can I not have a private key which is `public' ?
>
>
Example:

* UserA has ~/.ssh/id_rsa readable by everyone on MachineA
* UserA copied ~/.ssh/id_rsa.pub to MachineB and put it in their
~/.ssh/authorized_keys2
* UserB finds ~UserA/.ssh/id_rsa being readable and steals the key.
* UserB learns that UserA logs into MachineB using non-interactive key
exchange method.
* UserB copies stolen key to his ~/.ssh/ and now does: ssh UserA at MachineB.

Security has now been compermised.

> Gory details ....
>
> I'm new to openssh. I've been using ssh for years. However, I'm
in the process
> of investigating RH 7.* (0.91 at the moment) and am wanting to be as
> `standard' as possible, so trying openssh.
>
> I looked on http://www.openssh.com/list.html but could not find a list for
> "general OpenSSH discussion", but this was the closest match, so
I'll call it
> a bug :-)
>
This list doubles as a lot of things.  And luckly the traffic is not
extremely high. =)
> We use a client/server model with no `user' accounts on servers.
> There are certain operations which a user may require to run with certain
> privs, and we use ssh to do this. The capability may be given to an
individual
> user (user-only-readable in their .ssh/), a group (using UN*X group
semantics)
> or may be accessible to all users of a particular machine or set of
machines
> (e.g. when a user changes their password, a process is woken up on the
> password server).
>
> This all worked fine under ssh, but under openssh load_private_key() does a
> 		(st.st_mode & 077) != 0) {
> and then complains that it is readable and won't use it. (it says
"It is
> recommended that your private key files are NOT accessible by others."
but
> appears to implement somewhat more than a `recommendation' !)
>
> Is this bug intended as a feature ?   [ :-) ]
>
I would not consider this a bug.  It's a safety feature to protect the
user from doing stupid things.  (As shown in the example above).  I don't
see why the 'private' key should be allowed to be made public.  Feel
free
to explain why such behavior is not correct.   I can't see how allowing
everyone to read/steal my keys is considered a Good Thing(tm). =)

- Ben

On Tue, Mar 27, 2001 at 02:11:08PM +0100, Piete Brooks
wrote:
> Executive summary: Why can I not have a private key which is `public' ?
You can perform this action by not starting ssh directly but
by starting an intermediate executable which

- Checks if the calling user is allowed to perform that specific action.
- Sets uid to the uid which owns the ssh private key for that action.
- Calls in turn ssh to perform the action.

Corinna
> 
> 
> Gory details ....
> 
> I'm new to openssh. I've been using ssh for years. However, I'm
in the process
> of investigating RH 7.* (0.91 at the moment) and am wanting to be as 
> `standard' as possible, so trying openssh.
> 
> I looked on http://www.openssh.com/list.html but could not find a list for 
> "general OpenSSH discussion", but this was the closest match, so
I'll call it
> a bug :-)
> 
> We use a client/server model with no `user' accounts on servers.
> There are certain operations which a user may require to run with certain 
> privs, and we use ssh to do this. The capability may be given to an
individual
> user (user-only-readable in their .ssh/), a group (using UN*X group
semantics)
> or may be accessible to all users of a particular machine or set of
machines
> (e.g. when a user changes their password, a process is woken up on the 
> password server).
> 
> This all worked fine under ssh, but under openssh load_private_key() does a
> 		(st.st_mode & 077) != 0) {
> and then complains that it is readable and won't use it. (it says
"It is
> recommended that your private key files are NOT accessible by others."
but
> appears to implement somewhat more than a `recommendation' !)
> 
> Is this bug intended as a feature ?   [ :-) ]
> 
> I can see no code to disable this test [ other than setting HAVE_CYGWIN and
> writing a check_ntsec() which returns FALSE :-) ]
-- 
Corinna Vinschen
Cygwin Developer
Red Hat, Inc.
mailto:vinschen at redhat.com

Piete Brooks wrote:
> Executive summary: Why can I not have a private key which is
> `public' ?
Every time I use OpenSSH I seem to get caught out by the permission
checks.  I use umask 002 and my private files are all in a private
group.  This way I don't accidentally deny others access when I work
on shared material.  But OpenSSH doesn't like mode 775...
> We use a client/server model with no `user' accounts on servers.
> There are certain operations which a user may require to run with
> certain privs, and we use ssh to do this.
You could use the agent.  I've just tried and it doesn't look as
though ssh checks permissions on the socket directory.  This would
also have the nice feature that users couldn't copy the key.  You
would therefore be able to revoke access from one user without
revoking the key for the whole group.

-- 
Pete