Hello!

On Sun, Dec 04, 2016 at 09:39:59PM +0000, Steven Hartland wrote:

> We've used nginx for years and never had an issue with nginx upgrade
> until today where the upgrade command ran but almost instantly after the
> new process exited.
>
> /usr/local/etc/rc.d/nginx upgrade
> Performing sanity check on nginx configuration:
> nginx: the configuration file /usr/local/etc/nginx/nginx.conf syntax is ok
> nginx: configuration file /usr/local/etc/nginx/nginx.conf test is successful
> Upgrading nginx binary:
> Stopping old binary:
>
> In the default nginx log we had:
> 2016/12/04 21:18:22 [emerg] 25435#0: bind() to 127.0.0.1:81 failed (48:
> Address already in use)
> nginx: [emerg] bind() to 127.0.0.1:81 failed (48: Address already in use)
> 2016/12/04 21:18:22 [emerg] 25435#0: bind() to 127.0.0.1:81 failed (48:
> Address already in use)
> nginx: [emerg] bind() to 127.0.0.1:81 failed (48: Address already in use)
> 2016/12/04 21:18:22 [emerg] 25435#0: bind() to 127.0.0.1:81 failed (48:
> Address already in use)
> nginx: [emerg] bind() to 127.0.0.1:81 failed (48: Address already in use)
> 2016/12/04 21:18:22 [emerg] 25435#0: bind() to 127.0.0.1:81 failed (48:
> Address already in use)
> nginx: [emerg] bind() to 127.0.0.1:81 failed (48: Address already in use)
> 2016/12/04 21:18:22 [emerg] 25435#0: bind() to 127.0.0.1:81 failed (48:
> Address already in use)
> nginx: [emerg] bind() to 127.0.0.1:81 failed (48: Address already in use)
> 2016/12/04 21:18:22 [emerg] 25435#0: still could not bind()
> nginx: [emerg] still could not bind()
>
> Running the start just after resulted in a running version but is
> obviously unexpected to have upgrade result in a failure.
>
> I believe the change to add a localhost bind to the server in question
> was relatively recent so I suspect it has something to do with that.
>
> The config for this is simply:
> server {
> listen 127.0.0.1:81;
> server_name localhost;
>
> location /status {
> stub_status;
> }
> }
>
> The upgrade in this case was:
> nginx: 1.10.1_1,2 -> 1.10.2_2,2
>
> Now this server is running under FreeBSD in a jail (10.2-RELEASE) and it
> has 127.0.0.1 available yet it seems nginx has incorrectly bound the
> address:
> netstat -na | grep LIST | grep 81
> tcp4 0 0 10.10.96.146.81 *.* LISTEN

In a FreeBSD jail with a single IP address any listening address
is implicitly converted to the jail address. As a result, if you
write in config "127.0.0.1" - upgrade won't work, as it will see
inherited socket listening on the jail address (10.10.96.146 in
your case) and will try to create a new listening socket with the
address from the configuration and this will fail.

There are two possible solutions for this problem:

- configure listening on the jail IP address to avoid this
implicit conversion;
- configure listening on "*" and use multiple addresses in the jail.

In both cases there will be no implicit conversion and as a result
everything will work correctly.

--
Maxim Dounin
http://nginx.org/
_______________________________________________
nginx mailing list
nginx@nginx.org
http://mailman.nginx.org/mailman/listinfo/nginx