Possible bug. Bind 9.18 blocking on dnstap

[Borja Marcos]

Hi,

I just stumbled upon a problem. It happened on FreeBSD 13.1-RC (going to update to 13.1 today).

I am running bind 9.18.3 with dnstap using a Unix socket. 

Once the socket has been opened by bind, if the process serving the Unix socket blocks and you try to 
kill named, it fails to stop, stays blocked on the dnstap socket until it is closed.


How to reproduce:

- I started bind 9.18.3

- Started dnstap, the Go program from https://github.com/dnstap/golang-dnstap. (dnstap -u /var/tmp…)

- I suspended the dnstap program (kill -STOP)

- Sent some queries to bind

- Tried to stop bind (either rndc stop or kill -TERM)

It won’t really exit unless I resume the dnstop process or I kill it.

Is it expected behavior? I am not sure there should be such a strong contract to deliver dnstap
messages to a blocked process. At least there should be a reasonable timeout maybe?

I haven’t opened a ticket because I am not sure it is considered a bug.


Cheers,






Borja.