[jdk17u-dev] RFR: 8312065: Socket.connect does not timeout when profiling [v9]

[Andrei Pangin]

On Fri, 8 Sep 2023 13:48:22 GMT, Long Yang <duke at openjdk.org> wrote:

>> Hi all,
>> 
>> This pull request contains a fix for [JDK-8312065](https://bugs.openjdk.org/browse/JDK-8312065).
>> 
>> The old SocketImpl is still present in JDK 17, can be enabled with -Djdk.net.usePlainSocketImpl.
>> 
>> I have verified that this problem exists on Linux and macOS, and I feel that it also exists on AIX, 
>> so I repaired these 3 platforms. Windows implementations are different, so don't have this problem.
>> 
>> All callers of NET_Poll() already check for EINTR, so the fix is to remove do-while loop in the implementation of NET_Poll().
>> 
>> Other methods, such as NET_Read, NET_NonBlockingRead, NET_Accept, NET_Connect, are not affected.
>> Because if Socket has timeout, it will call NET_Timeout first, and then call them. NET_Timeout has a correct timeout implementation and is not affected by profiling signals.
>> If the Socket does not have a timeout, and if the system call is interrupted by a signal, the [Linux Kernel will automatically retry](https://man7.org/linux/man-pages/man7/signal.7.html), and the application code will not be aware of it.
>> 
>> Thanks!
>
> Long Yang has updated the pull request incrementally with one additional commit since the last revision:
> 
>   use monotonically increasing time(System.nanoTime()) in place of currentTimeMillis

Marked as reviewed by apangin (no project role).

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

PR Review: https://git.openjdk.org/jdk17u-dev/pull/1639#pullrequestreview-1617667640