According to the upstream security report [1] the following flaw was found in OpenVPN 2.3.0 and earlier:

OpenVPN 2.3.0 and earlier running in UDP mode are subject to chosen ciphertext injection due to a non-constant-time HMAC comparison function. Plaintext recovery may be possible using a padding oracle attack on the CBC mode cipher implementation of the crypto library, optimistically at a rate of about one character per 3 hours. PolarSSL seems vulnerable to such an attack; the vulnerability of OpenSSL has not been verified or tested.

OpenVPN servers are typically configured to silently drop packets with the wrong HMAC. For this reason measuring the processing time of the packets is not trivial without a MITM position. In practice, the attack likely needs some target-specific information to be effective.

The severity of this vulnerability can be considered low. Only if OpenVPN is configured to use a null-cipher, arbitrary plain-text can be injected which can completely open up this attack vector.

A fix has been committed to git [2]; this is corrected in OpenVPN 2.3.1 and later.

[1] https://community.openvpn.net/openvpn/wiki/SecurityAnnouncement-f375aa67cc [2] https://github.com/OpenVPN/openvpn/commit/11d21349a4e7e38a025849479b36ace7c2eec2ee

OpenVPN 2.x before 2.0.11, 2.1.x, 2.2.x before 2.2.3, and 2.3.x before 2.3.6 allows remote authenticated users to cause a denial of service (server crash) via a small control channel packet.