On Tue, 11 Oct 2022 14:38:35 GMT, Daniel Jeliński <djelin...@openjdk.org> wrote:
> This patch fixes the issue where a thread doing SSLSocket.close() could > destroy the read cipher while it was used by another thread doing > SSLSocket.read(). > > The reported issue was triggered by SSLSocket.close() calling > inputRecord.close() -> readCipher.dispose() -> cipher.doFinal() on an AES/GCM > cipher between `updateAAD` and `doFinal`; this changed the cipher state and > caused the doFinal call to fail. > > The issue for AES/GCM (and all other AEAD ciphers) was fixed by JDK-8294848; > this patch addresses the non-AEAD ciphers. > > Tier1-3 clean. Also, no failures after 1200 repetitions of > NoInvalidateSocketException. Overall the changes look clean. Thanks for taking care of this. A couple notes: - Per our side discussion, can you please remove noreg-self from the bug and add this bug ID to NoInvalidateSocketException.java - Can you please do a large-iteration run on test/jdk/sun/security/ssl/SSLSocketImpl/ClientSocketCloseHang.java I'll approve once that multi-iteration run comes back clean. ------------- PR: https://git.openjdk.org/jdk/pull/10657
