Automatically generated nginx resolver setting is invalid when /etc/resolv.conf contains IPv6 addresses

[ldruschk]

Hi,

the code for generating the nginx resolver setting when the setting is not explicitly specified can be found here:

docker-jitsi-meet/web/rootfs/etc/cont-init.d/10-config

Line 68 in 53b2654

export NGINX_RESOLVER=$(grep nameserver /etc/resolv.conf | cut -d" " -f2 | tr "\n" " " | sed -e "s/[[:space:]]*$//")

When the /etc/resolv.conf contains IPv6 addresses, the resulting config snippet looks as follows:

resolver 127.0.0.11 2a03:4000:0:1::e1e6

This results in the following error, leading to nginx not being able to start:

nginx: [emerg] invalid port in resolver "2a03:4000:0:1::e1e6" in /config/nginx/nginx.conf:30

According to the documentation IPv6 resolvers must be specified by surrounding the addresses with brackets: https://nginx.org/en/docs/http/ngx_http_core_module.html#resolver

When manually changing the setting to the following, nginx starts up normally

resolver 127.0.0.11 [2a03:4000:0:1::e1e6]

[saghul]

What environment is that?

[cdauth]

Do I see it right that this has not made it into the stable release yet?

[saghul]

The commit has been in 9 stable release, since 8044.

[cdauth]

Ah, I got confused since I didn't see it in the release notes and the commit in the pull request didn't show any branches/commits (it seems that the pull request doesn't contain the same commit as the master branch).

Anyways, I am running into this error message with the current stable image. The /etc/resolv.conf file inside the container looks like this:

nameserver 127.0.0.11
nameserver fd63:2af7:9f8b:ef5a:1::1
nameserver fe80::216:3eff:fef4:3236%11
options ndots:0

Which results in the following nginx config:

resolver 127.0.0.11 [fd63:2af7:9f8b:ef5a:1::1] [fe80::216:3eff:fef4:3236%11];

This results in the following error message:

nginx: [emerg] invalid IPv6 address in resolver "[fe80::216:3eff:fef4:3236%11]" in /config/nginx/nginx.conf:30

I have found an nginx issue that might explain this: https://trac.nginx.org/nginx/ticket/2509

It seems like the /etc/resolv.conf file inside the container is the same as on my system, so as a workaround I guess I have to investigate how that address ends up there in the first place and get rid of it.

[saghul]

Looks like it doesn't like the scope ID.

I'd take a PR which filters those addresses while Nginx fixes it.

If you want to take a crack at it, just ignore the addresses with a % in them.