[jdk16] RFR: 8259636: Check for buffer backed by shared segment kicks in in unexpected places

Alan Bateman alanb at openjdk.java.net
Tue Jan 12 16:25:57 UTC 2021

On Tue, 12 Jan 2021 15:48:18 GMT, Maurizio Cimadamore <mcimadamore at openjdk.org> wrote:

> When support for shared segment was added, we decided to disable support for buffers derived from shared segment in certain async operations, as there's currently no way to make sure that the memory won't be reclaimed while the IO operation is still taking place.
> After looking at the code, it seemed like the best place to put the restriction would be sun.nio.ch.DirectBuffer::address() method, since this method is used in a lot of places just before jumping into some piece of JNI code.
> While I still stand by that decision, the Netty team has discovered that this decision also affected operations such as creating slices from byte buffers derived from shared segment - this is caused by the fact that one direct buffer constructor (the one for views and slices) is calling the dreaded DirectBuffer::address method.
> The fix is simple: just avoid the method call - which is very easy to do in the case of the buffer constructor: in fact this method just returns the value of the `address` field inside the `Buffer` class, so we can always cast to `Buffer` and then access `address` field from there.

Marked as reviewed by alanb (Reviewer).


PR: https://git.openjdk.java.net/jdk16/pull/110

More information about the core-libs-dev mailing list