Content of /proc/net/tcp not what it is supposed to be

[robinmonjo]

I've noticed something weird I couldn't reproduce on other docker images such as ubuntu or on a "real" distribution (i.e: not running in a container engine).

When I launch the command: nc -l 9999, I expect to see the in /proc/net/tcp something like this:

0: 00000000:270F 00000000:0000 0A 00000000:00000000 00:00000000 00000000     0        0 132769 1 ffff880078ff4040 100 0 0 10 0

where the 00000000:270F part corresponds to <local_addr>:<local_port>. In this sample, I see 0.0.0.0:9999 (once converted to decimal). However on alpine docker I see random number for the port part instead of the 9999 I expect.

Here is a screenshot of my session:

[andyshinn]

I'll do some investigation. It is beyond my initial knowledge of the musl libc TCP implementation. What is this actually affecting?

[robinmonjo]

I guess it's affecting every tools that use this file to gather system information such as lsof or netcat. I noticed this while working on a tool I'm writing that must be able to check if a PID as bound a given port (I don't want any dependencies, that is why I do it "manually").

[robinmonjo]

Just FYI, it works as expected when spawning a http server with the go net/http library. Just saw this with nc so far (I use nc mainly for testing)

[hanimustafa]

I just hit this "bug" as well. You need to do "nc -l -p 9999"