Since 6.16.2 up to at least 6.16.4, IP4 connections fail randomly.

Description:

Since 6.16.2, some IP connections fail randomly, either with EACCESS or with some error saying that I'm trying to use a broadcast address (when I'm not). Retrying the exact same connection usually succeeds.

I've noticed that without running the X session, I can't reproduce it, maybe my tainted kernel has some relationship? Or maybe just the system load?

It has been reported in the forums: https://bbs.archlinux.org/viewtopic.php?pid=2258114

The linux-zen package fails exactly the same, in the same versions.

I've applied the patch in the link for the upsteam bug below, and it fixes it perfectly. I've tried reverting the linked patch (9e30ecf23b1b) and it also fixes the issue.

Additional info:

• package version(s): 6.16.4.arch1-1.
• config and/or log files: standard configuration, tainted with nvidia-open-dkms.
• link to upstream bug report, if any: https://lore.kernel.org/regressions/20250822165231.4353-4-bacs@librecast.net/

Steps to reproduce:

Reproducing the bug is not easy, according the forum posts if fails differently for different people.

In my case I can reproduce it consistently with the following Python script:

from socket import AF_INET, SOCK_DGRAM, socket

while True:
    with socket(AF_INET, SOCK_DGRAM) as sock:
        sock.connect(("1.1.1.1", 80))

But for it to fail consistently I have to run it with strace:

$ strace python test.py

In my system it will fail in between 0 and 30 seconds with PermissionError: [Errno 13] Permission denied.