[
https://issues.jboss.org/browse/WFLY-11007?page=com.atlassian.jira.plugin...
]
Jan Lieskovsky commented on WFLY-11007:
---------------------------------------
{noformat}
Of course, I could trim the list down manually (by using keytool) or even remove
/etc/pki/ca-trust/extracted/java/cacerts altogether but I guess this is a valid situation
we have here and it's the WF that should be monitoring the length of the payload it
sends (especially if there's a limit set by the RFC).
It is worth to mention that this happen only on client auth turned on. Only in this case,
the server sends a list of CAs.
{noformat}
Given the above research, IMHO this is EAP server bug. Of course, one can say, if you want
to get it working, don't use that much certificates. But since this might be valid
usecase / need, IMHO the server should be updated to send the list of CA certs in such
chunks, it won't overflow the TLS field length (like split them in more subsequent
packets, having the TLS field only as that much long, as required / allowed by the RFC).
[~dlofthouse] WDYT -^?
{noformat}
I hope this answers your questions Jan Lieskovsky and Darran Lofthouse
{noformat}
[~sebastian.laskawiec] Hey Seb, yes it does! Thank you for your investigation!
Using OpenShift generated certificates and client auth cause TLS
errors
-----------------------------------------------------------------------
Key: WFLY-11007
URL:
https://issues.jboss.org/browse/WFLY-11007
Project: WildFly
Issue Type: Bug
Components: Security, Web (Undertow)
Affects Versions: 13.0.0.Final
Reporter: Sebastian Ćaskawiec
Assignee: Stuart Douglas
h2. Summary
It seems that when using OpenShift generated certificates and client auth (with
{{want-client-auth="true"}}) the TLS handshake fails with {{RECV TLSv1.2 ALERT:
fatal, record_overflow}} message.
h2. Explanation
I'm using {{oc cluster up}} and deploying Keycloak (WF 13 based) on OpenShift local
cluster using the (1) template. The service in the the template uses OpenShift generated
certificates ({{"service.alpha.openshift.io/serving-cert-secret-name":
"keycloak-x509-https-secret"}}). Both files are mounted in the Keycloak pod and
translated into keystore and truststore (see the configuration after the transformation
(2)). Once the pod is up and running, I'm issuing a {{curl}} command as shown in (3).
{{curl}} fails saying that {{* error:1408F092:SSL routines:ssl3_get_record:data length too
long}}. The server logs with TLS Handshake debugging turned on might be found here (4). As
shown in the link, the server has written {{16384}} bytes.
I also did a test with manually created certificates (5). The result might be found here
(6). As shown in the link, we've written {{16050}} bytes instead of {{16384}} and the
handshake was successful.
h2. Possible solution
Perhaps we should cut the list CAs transmitted by the server when asking for client auth
when it exceeds certain number of bytes. It would be helpful to write a warn message too.
Links:
- (1) Keycloak OCP Template
https://gist.github.com/slaskawi/57ed810a7109a02a9d884b61ce2e7f13
- (2) Transformed configuration
https://gist.github.com/slaskawi/92aead6c519b867621129b640b4a3c88
- (3) curl command
https://gist.github.com/slaskawi/3bc32b8e96c2499cb7b48c3c5cb28616
- (4)
https://gist.github.com/slaskawi/b6477fe3cd65890c879cfe6f95359450#file-lo...
- (5) Keycloak and OpenShift integration demo
https://github.com/keycloak/openshift-integration/blob/master/install-key...
- (6)
https://gist.github.com/slaskawi/7fd87e1f2e6c4faf657d9e8289ed3392#file-lo...
--
This message was sent by Atlassian JIRA
(v7.5.0#75005)