On 28/08/2020 12:25, Alexander Scherbatiy wrote:
Hello,

Could you review the updated fix:
  http://cr.openjdk.java.net/~alexsch/8252248/webrev.01/

 - moving shared code to net_util_md.h is avoided
 - INTERRUPT_SIGNAL is defined as SIGRTMAX - 2 instead of __SIGRTMAX - 2 for Linux in NativeThread.c
 - "#include <pthread.h>" is moved out from "#ifdef" in NativeThread.c
 - sigWakeup is changed to "#define WAKEUP_SIGNAL (SIGRTMAX - 2)" in linux_close.c

At least test/jdk/java/net/Socket/asyncClose/AsyncClose.java touches all four methods where the signal changes are used:
NativeThread.c
- Java_sun_nio_ch_NativeThread_init(JNIEnv *env, jclass cl)
- Java_sun_nio_ch_NativeThread_signal(JNIEnv *env, jclass cl, jlong thread)
linux_close.c
-  __attribute((constructor)) init()
- closefd(int fd1, int fd2)
NativeThread current/signal will be exercised by the tests in java/nio/channels.

The code in linux_close.c is only used with the old (and not used by default) SocketImpl and DatagramSocketImpl implementation. Some of the tests jdk_net test group will re-run the tests with the old implementation so you should be okay.



I run test/jdk/java/net tests on Ubuntu 18.04.4 and Alpine Linux 3.11.6 with musl libc 1.1.24. There are 10 failed tests but they fail without the fix as well. I believe that it is because of some network settings of  my machine.
Sometimes there is will firewall or iptables configuration need to allow multicast datagrams to be received.

-Alan

Reply via email to