Enhance liberty to recover from java.io.IOException: EDC5122I Input/output error. (errno2=0x12AA044F)

TCP socket the Liberty is listening on becomes unusable. This causes connection attempts to the Liberty server to fail. This can happen when the z/OS system is running in a CINET environment (to be able run multiple TCPIP stacks) versus an INET environment (where only 1 TCPIP stack can run) and the Liberty server is started after TCPIP is up, but before PAGENT is up (with AT-TLS). It can also happen if TCPIP is restarted.

When TCPIP initializes and AT-TLS is configured it will register to USS as a socket transport provider but does not send the final notification to USS until PAGENT starts and installs the AT-TLS policy definitions (and outputs message EZZ4250I). It is this final notification that causes USS to fail the accept call with the EIO if SO_EioIfNewTP has been set.

We found that Java always sets SO_EioIfNewTP in an CINET environment by design.

To recover from such EIO error, it becomes responsibility of the Liberty to restart the socket and automatically recover from this condition, rather than forcing the end user to recover it.

Our understanding is that some customers use CINET to separate internal and external IP traffic into different TCP/IP stacks.

We are aware that starting z/OSMF after all TCP/IP ingredients are active avoids the problem, however there is a catch. Our z/OS component has a function that does not require AT-TLS and works fine without it. Now we are providing new function which does require AT-TLS to be configured and running. Our concern is that customers that did not use AT-TLS will try our new function, will bring up AT-TLS and that will break the active z/OSMF server (that is based on the Liberty). It may seem a minor issue, but given the time it took to investigate the problem (original defect was open over 5 months before root cause was established, after collecting dozens of dumps and trace files), it could be frustrating for a customer.