On Fri, 30 Oct 2026 05:03:15 GMT, Timofei Fedotov <[email protected]> wrote:
>> DTLSReassembler can queue fragments of the same handshake message with >> different total messageLength values. Reassembly holes are initialized from >> the first fragment, while the output array is allocated using the fragment >> with the lowest offset. >> >> For the same epoch, message sequence, and HELLO_REQUEST type, submit these >> fragments in order: >> offset 5, length 50, total message length 100; >> offset 0, length 5, total message length 10; >> offset 55, length 45, total message length 100. >> >> The final fragment completes the holes, but copying the buffered fragments >> into the array sized for a ten-byte message triggers >> ArrayIndexOutOfBoundsException. SSLEngineImpl catches this runtime exception >> and reports a fatal SSLException; it does not establish a JVM crash. >> >> Expected behavior the inconsistent handshake metadata must be rejected as a >> protocol error before fragment reassembly. >> >> --------- >> - [x] I confirm that I make this contribution in accordance with the >> [OpenJDK Interim AI Policy](https://openjdk.org/legal/ai). > > Timofei Fedotov has updated the pull request incrementally with two > additional commits since the last revision: > > - Add number of bug in test > - Add test src/java.base/share/classes/sun/security/ssl/DTLSInputRecord.java line 691: > 689: > 690: // All fragments of one message must agree on its total > length. > 691: for (RecordFragment fragment : bufferedFragments) { We should avoid scanning the entire list of already-buffered fragments; it could be abused by a malicious peer. Could you store the expected message length somewhere instead? `holesMap` could be the right place. ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/33633#discussion_r4421780216
