RFR: 7146776: deadlock between URLStreamHandler.getHostAddress and file.Handler.openconnection

[Alan Bateman]

On Sat, 16 Jan 2021 12:40:47 GMT, Jaikiran Pai <jpai at openjdk.org> wrote:

> As noted by the reporter in the comments of https://bugs.openjdk.java.net/browse/JDK-7146776, the `URLStreamHandler` in its `getHostAddress` is incorrectly synchronizing on the `URLStreamHandler` instance, instead of synchronizing on the `URL` instance that is being passed to that method.
> 
> The implementation of `URLStreamHandler#getHostAddress` checks for the (package private) `hostAddress` member of the passed `URL` and if it isn't set then does a DNS lookup to get the address and finally sets the `hostAddress` member of the `URL` to this returned address. The access to `hostAddress` of `URL` needs to be `synchronized` since it can be updated (in a different thread) in the `URL#set` methods.
> 
> The commit here removes the synchronization from the `URLStreamHandler#getHostAddress` and moves the entire implementation into a new (package private) `URL#getHostAddress` method and then calls this new method from the `URLStreamHandler#getHostAddress`. I decided to do it this way instead of just leaving the implementation in `URLStreamHandler#getHostAddress` and synchronizing on the passed `URL` instance, since IMO, this makes it much more easier to see (in one place within the `URL` class) what the synchronization requirements are when dealing with the `hostAddress`. Furthermore, this now allows the `hostAddress` to be made a `private` member (which I have done in this commit) instead of the previous package private access. I checked that no other code tries to access this package private `URL#hostAddress`.
> 
> I can't think of an easy way to add a jtreg test for this change, so I haven't added one.

src/java.base/share/classes/java/net/URL.java line 817:

> 815:      *         while getting the host address will result in this method returning
> 816:      *         {@code null}
> 817:      */

The fix looks okay but the javadoc might be a big clearer if you move the text into the method description.

-------------

PR: https://git.openjdk.java.net/jdk/pull/2109