-
Notifications
You must be signed in to change notification settings - Fork 138
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
OCSP responder to serve status check for itself using latest CRL #4504
Conversation
The first step of inverting the init order to load the store first is done. |
The init order for OCSP is modified to allow CRL retrieval before creating connection with DS or other services. Secure`connections will be verified against the CRL. Solve RHCS-4262
4db78c4
to
ecc6b01
Compare
Add new field in CMS for a callback validation of certificate instantiated by PKISocketFactory. This is useful for OCSP where the OCSP protocol cannot be enabled and the verification is done on CRLs. Solve RHCS-4262
Add a new parameter to enable the crl check for OCSP connection when acting as client. The new parameter is `ocsp.store.ldapStore.checkSubsystemConnection` and its default value is `false`. When set to `true` connection certificate are verified using the crl stored in the LDAP.
e8ed05b
to
e1e85e5
Compare
When OCSP is acting as server certificate can be verified using CRL internally stored. To verify the certificates the `LDAPStore` has to be enabled with the variable `ocsp.store.ldapStore.checkSubsystemConnection` and the variable `auths.revocationChecking.enabled` both set to true. Solve RHCS-4262
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Please see my comments below. I have one question about the loop, and the rest are just minor suggestions.
@@ -53,6 +54,8 @@ public final class CMS { | |||
|
|||
private static CMSEngine engine; | |||
|
|||
private static SSLCertificateApprovalCallback approvalCallback; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
In the newer branches we've removed the dependency on global variables in CMS
class. I'd suggest moving this variable into CMSEngine
class.
|
||
@Override | ||
public boolean approve(X509Certificate certificate, ValidityStatus currentStatus) { | ||
logger.info("CRLLdapValidator: validate of peer's certificate for the connection " + certificate.getSubjectDN().toString()); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The .toString()
can be removed since it's implicitly called by string concatenation.
} | ||
} | ||
} catch (EBaseException e) { | ||
logger.error("CRLLdapValidator: problem find CRL issuing point for " + certificate.getIssuerDN().toString()); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
To simplify troubleshooting I'd suggest displaying the exception message and optionally attach the exception (to show the stack trace):
logger.error("...<message>...: " + e.getMessage(), e);
for (X509Certificate cert: certificates) { | ||
if(crlCertValid(crlStore, cert, null)) { | ||
return false; | ||
} | ||
} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
IIUC this loop will return false
as soon as it finds the first valid cert. Should we check all certs in the chain instead?
Socket callback moved to CMSEngine to avoid dependencies on global variables.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the update! Looks good to me, but I'll defer to others for the final approval.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I have conducted a few tests with the PR on revoked DS server-cert, agent cert at webUI, and admin cert with pkiconsole and I think they work as expected. Of course more testing could be used for other scenarios but I'll leave that to QE. With regard to this PR, please see my comments. Thanks.
@@ -137,6 +139,7 @@ public void init(IConfigStore config, DBSubsystem dbSubsystem) throws EBaseExcep | |||
DEF_CA_CERT_ATTR); | |||
mByName = mConfig.getBoolean(PROP_BY_NAME, true); | |||
|
|||
mCheckConnection = mConfig.getBoolean(PROP_CHECK_SUBSYSTEM_CONNECTION, false); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Or, simply put ocsp.store.ldapStore.checkSubsystemConnection=true as default in OCSP's CS.cfg. Also, based on my latest experiment, how about adding the following to the OCSP's CS.cfg: auths.revocationChecking.enabled=true . This is so that role access at the WebUI and pkiconsole can do that by default. CA already has that by default.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This code is only used by the OCSP subsystem. Is there any case where we want this functionality disabled? If it's always going to be enabled maybe we don't need this param.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I agree with @edewata, I do not see a reason why this should be disabled so I think we could make permanent this change and not triggered by an option. However, we could just makes it the default behaviour just in case the scenario where disabling the check arise. @ladycfu I am converting to true
.
I have some concerns with auths.revocationChecking.enabled=true
. For the OCSP subsystem this work only if the CA sends CRLs through an LDAP and not with direct publishing. Should we stopped, or at least log an error, if it is true but LDAP publishing is not configured?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The reason why I propose having a "switch" to go back to the previous behavior was because we want to be sure that there is no gotcha's in the new mechanism. For now, I think this switch serves its purpose, as a back up . I think it's fine to have it enabled by default. Now, wrt auths.revocationChecking.enabled=true, this "Method C" precludes the direct CA->OCSP CRL publishing because the ocsp's ssl port is not able to receive CRL from the CA due to chick-n-egg issue. The CA depends on the ocsp to start up, and the ocsp depends on the CA to start up, if replying on ca->ocsp direct publishing. The CA is to have enableOCSP=true and its internaldb's server-cert is to bear an AIA pointing to the ocsp. @fmarco76 I think you have a good suggestion there on checking whether it's ldapStore that's being set for crl update if ocsp.store.ldapStore.checkSubsystemConnection=true. btw, what is the concern you wrt auths.revocationChecking.enabled=true ?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@fmarco76 btw, why do you call this flag "checkSubsystemConnection"? To me it seems something like "ocsp.store.ldapStore.validateConnCertWithCRL" would be more telling. Actually, could you add an explanation in comment where this flag is defined what exactly it does? We need to add something in the doc. Thanks.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think you have a good suggestion there on checking whether it's ldapStore that's being set for crl update if ocsp.store.ldapStore.checkSubsystemConnection=true.
This setting is done on ldapStore of the OCSP so we do not need to verify if it is used because if it is not then this setting does not apply.
what is the concern you wrt auths.revocationChecking.enabled=true ?
In the OCSP, if this is true but the ldapStore is not used the revocation cannot be verified. I have not tested if this generate errors. Therefore I was wondering if the user should be notified that revocation will not work even it is enabled.
why do you call this flag "checkSubsystemConnection"? To me it seems something like "ocsp.store.ldapStore.validateConnCertWithCRL" would be more telling. Actually, could you add an explanation in comment where this flag is defined what exactly it does? We need to add something in the doc.
Yes, sound better, I will change the name. Where exactly should I write an explanation? A comment in github where the code is provided or a comment inside the code?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
wrt auths.revocationChecking.enabled=true, I'm still not sure why it would require ldapStore. I will take a closer look later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Because the method OCSPEngine.isRevoked()
is invoked when a new connection to OCSP is created. This method uses the LDAPStore to verify the certificate. If LDAPStore is not used, or the CRL verification is disabled I have not verified what happen.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
ok I just dug slightly deeper and saw that you had overwritten the isRevoked in the OCSPEngine.
while (eCRL.hasMoreElements() && pt == null) { | ||
ICRLIssuingPointRecord tPt = eCRL.nextElement(); | ||
logger.debug("CRLLdapValidator: CRL check issuer " + tPt.getId()); | ||
if(tPt.getId().equals(certificate.getIssuerDN().toString())) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I'm not certain about issuerDN string comparison here. If you look at the processRequest(...) method in ocsp/src/main/java/com/netscape/cms/ocsp/LDAPStore.java, it uses the CA's public key hash as key to locate the right CRL, which is more assuring, isn't it?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I have implemented this using the CheckCertServlet as reference where the ICRLIssuingPointRecord
is identified using the common name. Nevertheless, using the key hash is definitely a stronger mechanism. I will try to use the processRequest
method or extract some code in a separate method.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@ladycfu I did not find an easy way to get the issuer key hash from the certificate but to be more confident I have verified that the CA cert is the one signing the certificate. The test is done only if the DNs match to avoid the overhead of encoding/decoding certificates. I have performed the operation considering Mozilla-JSS provider is available. If this is not always the case some extra step are needed. Is this approach viable for the job?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hi @fmarco76 , I think in general, it's a better practice to avoid string comparison when it comes to DN names, due to encoding differences. Looking closely, the processRequest was for processing an ocsp request which contains the CertID, so it's not exactly what we have here. We have the peer cert, X509Certificate certificate, and we have the ca cert, X509CertImpl ca_certImpl = new X509CertImpl(tPt.getCACert()); and we can get the ca's public key PublicKey ca_pubkey = ca_certImpl.getPublicKey(); where PublicKey is java.security.PublicKey. Meanwhile, the peer cert X509Certificate certificate can have it's public key gotten with PublicKey peerCert_pubkey = certificate.getPublicKey(); So now you can compare. I'm not sure where you said you did signature verification. I could be wrong, but that sounds more expensive than the PublicKey comparison. Anyway, I think this patch is as good as it gets for the time we have, so you could leave this as an improvement for later if you want if you donn't have time for it now. Thanks.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hi @fmarco76 , I think in general, it's a better practice to avoid string comparison when it comes to DN names, due to encoding differences.
IIRC String in java have the same encoding independently of how they are stored in the certificate. The encoding is converted when they are read/write. The format could be different but this is related to the used classes so it is known (although it could change moving to different versions of Java).
Looking closely, the processRequest was for processing an ocsp request which contains the CertID, so it's not exactly what we have here. We have the peer cert, X509Certificate certificate, and we have the ca cert, X509CertImpl ca_certImpl = new X509CertImpl(tPt.getCACert()); and we can get the ca's public key PublicKey ca_pubkey = ca_certImpl.getPublicKey(); where PublicKey is java.security.PublicKey. Meanwhile, the peer cert X509Certificate certificate can have it's public key gotten with PublicKey peerCert_pubkey = certificate.getPublicKey(); So now you can compare. I'm not sure where you said you did signature verification. I could be wrong, but that sounds more expensive than the PublicKey comparison.
The problem here is that you get the PublicKey of the CA and the PublicKey of the peer certificate. They are different so there is no way to compare. The OCSP request contain the hash of the issuer PublicKey so it is possible to get the hash of the CA PublicKey and compare.
From the peer certificate I found no way to get the issuer PublicKey to compare with the CA.
The approach I implemented is in line 61-63 of CRLLdapValidator.
It is definitely more expensive of the PublicKey comparison and this is the reason I do only if the DN match. However, considering a low number of CA for an OCSP we could skip the DN match and do always the signature check. Is this assumption correct?
Anyway, I think this patch is as good as it gets for the time we have, so you could leave this as an improvement for later if you want if you don't have time for it now. Thanks.
I can work on this until the end of the sprint so we still have several days before merge.
while (eCRL.hasMoreElements() && pt == null) { | ||
ICRLIssuingPointRecord tPt = eCRL.nextElement(); | ||
logger.debug("OCSPEngine: CRL check issuer " + tPt.getId()); | ||
if(certificate.getIssuerX500Principal().equals(new X500Principal(tPt.getId()))) { |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
same comment as above regarding using CA's public key hash.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Same approach as above.
The parameter `ocsp.store.ldapStore.checkSubsystemConnection` default value has been modified to `true` so when LDAPStore is used certificates are verified against the CRL. Additionally, during the certificate verification the certificate signer is verified with the CA certificate providing the CRL to be sure it is the real issuer.
With last commit there is an error on TPS tests but I do not see changes in this PR that could generate the error |
The option `ocsp.store.ldapStore.validateConnCertWithCRL` enables the revocation verification of peer certificates using the CRL stored in the LDAP shared with the CA. When it is set to `true` (default value), the peer certificate of all the outcome connections from the OCSP subsystem are verified with the CRL. If the option `auths.revocationChecking.enabled` is also set to `true` the peer certificate ot all the income connections to the OCSP subsystem are verified with the CRL.
@@ -82,6 +81,7 @@ public class LDAPStore implements IDefStore, IExtendedPluginInfo { | |||
private static final String DEF_CA_CERT_ATTR = "cACertificate;binary"; | |||
private static final String PROP_HOST = "host"; | |||
private static final String PROP_PORT = "port"; | |||
private static final String PROP_VALIDATE_CONNECTION_WITH_CRL = "validateConnCertWithCRL"; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The option ocsp.store.ldapStore.validateConnCertWithCRL
enables the revocation verification of peer certificates using the CRL stored in the LDAP shared with the CA.
When it is set to true
(default value), the peer certificate of all the outcome connections from the OCSP subsystem are verified with the CRL.
If the option auths.revocationChecking.enabled
is also set to true
the peer certificate ot all the income connections to the OCSP subsystem are verified with the CRL.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
How about put the info right here in the code where PROP_VALIDATE_CONNECTION_WITH_CRL = "validateConnCertWithCRL" is assigned? Comments in code are longer lasting and provide info others later. I will also be using it for the doc changes.
Identification of CRL issuing point done by matching Authority Key Identifier with Subject Key Identifier instead of DN matching. This should make more reliable the check because not affected of encoding or format changes in the DN.
try { | ||
X509CertImpl peerCert = new X509CertImpl(certificate.getEncoded()); | ||
Enumeration<ICRLIssuingPointRecord> eCRL = crlStore.searchAllCRLIssuingPointRecord(-1); | ||
AuthorityKeyIdentifierExtension aPeeExt = (AuthorityKeyIdentifierExtension) peerCert.getExtension(PKIXExtensions.AuthorityKey_Id.toString()); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
May I suggest that you rename the variable "ePeeExt" to something else? lol. How about "peerAKIExt"?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hi Marco, thanks for the update. I think the only things I have for you now is just to address the couple minor requests I have regarding adding a comment explaining the new param and changing one variable name.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the update. LGTM
Modify OCSP responder to perform peer certificate checks with the CRL instead of using OCSP, avoiding the chicken and egg issue.
Solve RHCS-4262