In I2P up to 2.2.1 there is a context-confusion bug which occurred in the bloom filter. An attacker crafts an I2NP message containing a unique messageID, and sends that messageID to a client. The message, after passing through the bloom filter, is not allowed to be re-used in a second message. The attacker then sends the same message directly to the router. The router passes the message to the bloom filter, and is dropped. This leaks the information that the messageID has been seen before, giving the attacker a strong reason to believe that the router is hosting the client. This has been fixed by separting the bloom filter's functionality into different contexts based on whether a message came down a client tunnel, an exploratory tunnel, was sent to the router directly. Under normal circumstances, this attack takes several days to perform successfully and may be confounded by several factors such as routers restarting during the attack phase and sensitivity to false-positives. Users of Java I2P are recommended to update immediately to avoid the attack.