On Sat, Jun 07, 2014 at 01:20:24PM +0200, Jurjen Bokma wrote:
> On 06/06/2014 10:24 AM, Sumit Bose wrote:
>> On Thu, Jun 05, 2014 at 04:54:50PM +0200, Jurjen Bokma wrote:
>>> Hi,
>>>
>>> I believe the closing of
>>>
https://fedorahosted.org/sssd/ticket/1871#comment:2
>>> to be wrong. Jakub Hrozek asked me to bring this up here after I brought it
to his attention through
>>>
https://bugs.launchpad.net/bugs/1274543
>>>
>>> When it logs in to the AD server, sssd uses the first key in the keytab. In
my case, that first key is of a 'join account': a special purpose AD account that
has permission to join machines to the domain, and nothing else. In particular, it lacks
the permissions to list users. Hence it is unfit to be used by sssd for lookups. As soon
as I remove the offending key, sssd works fine, by using another key, which happens to
work. When I put it back in the keytab (as the first key), sssd stops working again. I
second Jakub in the opinion that sssd should select the 'right' principal to use:
ldap_sasl_authid or shortname$@realm.
>>> I believe the cause of the problem to be in src/providers/krb5/krb5_child.c
in the loop immediately following the comment:
>>> /* We look for the first entry from our realm or take the last one */
>>> Indeed, if I make that loop skip the first key found, everything works as
expected, whether the ADJoiner key is in the keytab or not.
>>>
>>> This is just an ad-hoc fix for my case of course.
>> Thank you for the report. I reopened
>>
https://fedorahosted.org/sssd/ticket/1871 and it would be nice if you
>> can add your use-case to the ticket. The key point here is that your
> Thanks for reopening! I pleaded my case in Trac.
>> first principal cannot be used as a service principal and hence cannot
>> be used for validation.
>>
>> Another workaround would be to put the principal at some other place in
>> the keytab or to disable validation be setting 'krb5_validate = false'
>> in sssd.conf.
> Thank you for the suggestions. Yet another workaround is to have the
> application (i.c. msktutil) that put the offending key in the keytab use
> another keytab entirely. That's what I did. My latest solution is even
> nicer: have the join account on a puppet server, and never even use it from
> the client. So I'm not even affected by the bug any more.
> But more applications use the system keytab by default, reconfiguring each
> to use another keytab is tedious. Moreover, when an offending key is first
> in the keytab, sssd fails silently unless logging is sufficiently verbose,
> and the message is rather cryptic. The admin doesn't suspect keytab problems
> because the proper key *is* in the keytab too, and will only grow suspicious
> when he starts sniffing the network connection for Kerberos traffic.
> IMHO, the key point is that the first key will often work, but the
> assumption that it'll *always* work is just plain wrong. Please add code to
> pick the right principal.
yes, that's why I reopened the ticket.
Nevertheless about 'more applications use the system keytab by default'.
IMO this is bad practice and if you have a chance please complain at
vendors doing this. The reason is that other than e.g the global storage
for public SSL/TLS certificates the keytab entries should be compared
with passwords. If multiple applications share the keytab they allow the
other applications to access and use their passwords.
I'd say that is true: not
just any application should be allowed to use
the system keytab. And indeed, msktutil has an option that allows the
systems administrator to do just that. It can also run as a
non-privileged user.
Still, running msktutil as root is common practice, and I see nothing
wrong with root processes being able to see passwords/keys (unless your
security demands are strict). If they do, they have to be prepared to
encounter each other's keys though, at which point sssd fails. Moreover,
I don't see see why, if sssd is allowed to use the system keytab,
msktutil shouldn't be allowed to do the same. Msktutil joins the machine
to the domain and enables sssd in the first place.
By your reckoning, the right solution is for sssd to support a different
keytab than the system keytab. That is workable, but I wouldn't pick it
as the best solution, because there are situations (which I've
encountered) where sssd actually could (and IMHO should) use the
msktutil's key from the system keytab. But if that key doesn't fit the
lock, sssd shouldn't just break.
'gards
Jurjen