Date: Fri, 08 Sep 2006 09:01:41 -0600 From: Richard Megginson rmeggins@redhat.com
Josh Kelley wrote:
On 9/7/06, Richard Megginson rmeggins@redhat.com wrote:
I checked RFC 4513 - http://www.ietf.org/rfc/rfc4513.txt - it doesn't say anything about the correct result code to return in this case, other than it is an error if anything other than success or bindinprogress is returned. You might want to ask on ldap@umich.edu or on IRC.freenode.net #ldap if there is a standard that covers this case.
Thanks for the suggestion. I'll ask.
I skimmed RFC 4513 (sans coffee) and didn't find the section you're referring to. I did see that RFC 4422 (last paragraph of section 3.6) seems to suggest that OS X's and OpenLDAP's behavior is legitimate and useful.
Before you go any further with this, please tell us which version of OpenLDAP you're using. Current releases (since 2.3.6) return invalidCredentials for a SASL bind failure:
ldapsearch -H ldap://:9000 -Y DIGEST-MD5 SASL/DIGEST-MD5 authentication started Please enter your password: ldap_sasl_interactive_bind_s: Invalid credentials (49) additional info: SASL(-13): user not found: no secret in database
Probably we should also do something about not returning the SASL-specific error code in this case too, to adhere more to the intent of rfc4422. Logging it on the server side should be sufficient.
I just checked, and releases 2.1 and 2.2 returned error code 80 here. So it seems Apple is relying on a broken behavior.
Yes. But it seems to differ from the behavior of a simple bind (rfc4513 5.1.3). In a simple bind, the server resultCode differentiates these cases:
- Invalid bind DN results in a noSuchObject (well, not exactly
specified, but this is the usual behavior) 2) Valid bind DN but invalid password results in invalidCredentials
However, the rfc (and also rfc 4511 Appendix A LDAP Result Codes) says that other codes may be substituted for the above "to prevent unauthorized disclosures (such as substitution of noSuchObject for insufficientAccessRights, or invalidCredentials for insufficientAccessRights)."
The SASL doc (rfc4422) says:
"It is also important that the server can be configured such that the outcome message will not distinguish between a valid user with invalid credentials and an invalid user."
So it seems that SASL wants the server not to differentiate these cases, probably for security reasons. But this makes sasl binds have different semantics than simple binds.
Even if the standards permit either behavior (and even if it's slightly more secure to not reveal additional information, as David Boreham pointed out), wouldn't it be worth having FDS compatible with OpenLDAP and OS X?
Yes. And please file a bug about this at http://bugzilla.redhat.com/
On 9/8/06, Howard Chu hyc@symas.com wrote:
Before you go any further with this, please tell us which version of OpenLDAP you're using. Current releases (since 2.3.6) return invalidCredentials for a SASL bind failure:
2.2.13, as provided by RHEL 4. I had not thought to try a current release; thanks for the info.
Probably we should also do something about not returning the SASL-specific error code in this case too, to adhere more to the intent of rfc4422. Logging it on the server side should be sufficient.
I just checked, and releases 2.1 and 2.2 returned error code 80 here. So it seems Apple is relying on a broken behavior.
I guess I really don't understand here. RFC 4422 says that "outcome message provided by the server can provide a way for the client to distinguish between errors" of various sorts, which I assumed could include errors resulting from attempting to use an unconfigured authentication mechanism. And although the RFC says that it is "important that the server can be configured such that the outcome message will not distinguish between a valid user with invalid credentials and an invalid user," it doesn't say that it's necessary that servers be so configured or that it's broken for servers to not be so configured.
Furthermore, it seems to me that what Apple's trying to do - attempt a secure authentication method first, then fall back to a nonsecure authentication method if a secure method is not configured, without needlessly sending cleartext passwords if the secure authentication method is configured and rejects the user - is a good idea. Is Apple's default approach simply not permitted by the RFCs? I understand that for the server to unnecessarily give away security-related info is potentially bad, but it seems like a minor concern compared to the gains of permitting "secure by default, fall back to unsecured" behavior like Apple's default. (I guess MITM attacks are a risk with that kind of approach; are they enough of a risk to negate that approach's value?)
I hope I'm not coming across as argumentative; I just would really like to understand the issues involved.
If it is a bad idea for the server to distinguish between "invalid credentials" and "no secret in database," then what is the best way to get OS X logins to work? For now I've simply disabled CRAM-MD5 by moving those libraries out of my /usr/lib/sasl2 directory, but that seems like a hack. I guess a better solution would be to permit the SASL mech_list to be configured from within FDS; should I submit an RFE on Bugzilla for that?
Thanks.
Josh Kelley
Josh Kelley wrote:
On 9/8/06, Howard Chu hyc@symas.com wrote:
Before you go any further with this, please tell us which version of OpenLDAP you're using. Current releases (since 2.3.6) return invalidCredentials for a SASL bind failure:
2.2.13, as provided by RHEL 4. I had not thought to try a current release; thanks for the info.
Probably we should also do something about not returning the SASL-specific error code in this case too, to adhere more to the intent of rfc4422. Logging it on the server side should be sufficient.
I just checked, and releases 2.1 and 2.2 returned error code 80 here. So it seems Apple is relying on a broken behavior.
I guess I really don't understand here. RFC 4422 says that "outcome message provided by the server can provide a way for the client to distinguish between errors" of various sorts, which I assumed could include errors resulting from attempting to use an unconfigured authentication mechanism. And although the RFC says that it is "important that the server can be configured such that the outcome message will not distinguish between a valid user with invalid credentials and an invalid user," it doesn't say that it's necessary that servers be so configured or that it's broken for servers to not be so configured.
Furthermore, it seems to me that what Apple's trying to do - attempt a secure authentication method first, then fall back to a nonsecure authentication method if a secure method is not configured, without needlessly sending cleartext passwords if the secure authentication method is configured and rejects the user - is a good idea. Is Apple's default approach simply not permitted by the RFCs? I understand that for the server to unnecessarily give away security-related info is potentially bad, but it seems like a minor concern compared to the gains of permitting "secure by default, fall back to unsecured" behavior like Apple's default. (I guess MITM attacks are a risk with that kind of approach; are they enough of a risk to negate that approach's value?)
SASL is supposed to attempt to negotiate the strongest auth available.
I hope I'm not coming across as argumentative; I just would really like to understand the issues involved.
If it is a bad idea for the server to distinguish between "invalid credentials" and "no secret in database," then what is the best way to get OS X logins to work? For now I've simply disabled CRAM-MD5 by moving those libraries out of my /usr/lib/sasl2 directory, but that seems like a hack. I guess a better solution would be to permit the SASL mech_list to be configured from within FDS; should I submit an RFE on Bugzilla for that?
Yes.
Thanks.
Josh Kelley
-- Fedora-directory-users mailing list Fedora-directory-users@redhat.com https://www.redhat.com/mailman/listinfo/fedora-directory-users
On 9/8/06, Richard Megginson rmeggins@redhat.com wrote:
Josh Kelley wrote:
If it is a bad idea for the server to distinguish between "invalid credentials" and "no secret in database," then what is the best way to get OS X logins to work? For now I've simply disabled CRAM-MD5 by moving those libraries out of my /usr/lib/sasl2 directory, but that seems like a hack. I guess a better solution would be to permit the SASL mech_list to be configured from within FDS; should I submit an RFE on Bugzilla for that?
Yes.
Submitted at https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=206053; thanks.
Josh Kelley
389-users@lists.fedoraproject.org
-
Howard Chu -
Josh Kelley -
Rich Megginson