net: tcp: Context still open after timeout on connect

[ssharks]

Describe the bug
When a TCP connect fails with a timeout, because there is no communication possible, the context is not cleaned up properly.

To Reproduce
Use the testcase as submitted in the pull request
#44501

Run the test using
./scripts/twister -T tests/net/socket/tcp -p qemu_x86

Expected behavior
When a socket connect timeouts, I would expect the context to close, no connection has been build, so why do we need to close it?

Impact
Could cause resource leakage in case of a bad network connection.

Environment:

• OS: (Ubuntu 20.04)
• Target: Qemu x86
• Toolchain: Zephyr SDK 0.13.2
• Commit: Pull request tests: net: socket: tcp: Add test to validate connect timeout #44501

[rlubos]

Hi @ssharks. I can't reproduce the issue you describe. The test case you provide indeed fail, but the reason for this is not that net_context leaks somehow on connect() timeout, but because you don't close the server socket. When both, client and server sockets are closed after the failed connection attempt, there is no context left open.

[ssharks]

So if I understand you correctly, a socket should be closed even when the connect fails?

[rlubos]

So if I understand you correctly, a socket should be closed even when the connect fails?

That's right, once opened the only way to clean up its resources is to close it. Anyway, it's generally expected that you always close a socket that is no longer used, it's not only about net_context but also the file description needs to be freed etc.

[ssharks]

The suggested changes have been implemented in the pull request and now the tests pass. So there is no issue in the stack on this point.