We run a web application with a JVM backend (Java 7 update 75; code is in Scala, but I don't believe this is relevant). We use Google for authentication via Oauth.
There have been a handful of days over the last couple of months on which we have been intermittently unable to authenticate users.
The redirect to and from Google is successful, but when we try to call the token_endpoint
at https://www.googleapis.com/oauth2/v4/token to validate the authentication we sometimes get the following exception: javax.net.ssl.SSLHandshakeException: server certificate change is restrictedduring renegotiation
.
This comment on another question led me to find a JDK bug that can manifest as this exception (What means "javax.net.ssl.SSLHandshakeException: server certificate change is restrictedduring renegotiation" and how to prevent it?).
My working hypothesis is:
The bug (https://bugs.openjdk.java.net/browse/JDK-8072385) means that only the first entry in the Subject Alternative Name (SAN) list is checked. The exception above is thrown when the hostname being verified is in the SAN list, but not at the top of the list.
Yesterday (27th May 2015 from ) we saw two different certificates being intermittently served from www.googleapis.com. The first (serial 67:1a:d6:10:cd:1a:06:cc
) had an SAN list of DNS:*.googleapis.com, DNS:*.clients6.google.com, DNS:*.cloudendpointsapis.com, DNS:cloudendpointsapis.com, DNS:googleapis.com
whilst the second (serial 61:db:c8:52:b4:77:cf:78
) had a SAN list of: DNS:*.storage.googleapis.com, DNS:*.commondatastorage.googleapis.com, DNS:*.googleapis.com
.
Due to the bug in the JVM, we can validate the first certificate, but the exception is thrown with the second (despite being perfectly valid) as *.googleapis.com
is not the first entry in the SAN list.
The fix is in the yet to be released 7u85 (no mention of when this will be available).
I've downgraded a single node of our cluster to 7u65, but the certificate seemed to be reverted at around the time we did this (last error we saw was 22:20GMT) so it's hard to pin down an affirmative fix.
Has anyone else experienced this or something similar and have any other workaround other than downgrading (downgrading removes various other SSL/TLS checks)?