Problem upgrading on the fly when ipv6 is used

Server NGINX
1

Hi!

I’ve always upgraded nginx on the fly following this guide:
CommandLine | NGINX without problems.
I recently added IPv6 support to nginx and now, when I need to upgrade
it, it doesn’t start the new instance when I send the USR2 signal to the
master process, giving the following erros:

2010/05/29 09:05:52 [notice] 21060#0: using inherited sockets from 
"14;15;"
2010/05/29 09:05:52 [emerg] 21060#0: bind() to [2001:41d0:2:8673::1]:80 
failed (98: Address already in use)
[emerg]: bind() to [2001:41d0:2:8673::1]:80 failed (98: Address already 
in use)
2010/05/29 09:05:52 [emerg] 21060#0: bind() to [2001:41d0:2:8673::1]:80 
failed (98: Address already in use)
[emerg]: bind() to [2001:41d0:2:8673::1]:80 failed (98: Address already 
in use)
2010/05/29 09:05:52 [emerg] 21060#0: bind() to [2001:41d0:2:8673::1]:80 
failed (98: Address already in use)
[emerg]: bind() to [2001:41d0:2:8673::1]:80 failed (98: Address already 
in use)
2010/05/29 09:05:52 [emerg] 21060#0: bind() to [2001:41d0:2:8673::1]:80 
failed (98: Address already in use)
[emerg]: bind() to [2001:41d0:2:8673::1]:80 failed (98: Address already 
in use)
2010/05/29 09:05:52 [emerg] 21060#0: bind() to [2001:41d0:2:8673::1]:80 
failed (98: Address already in use)
[emerg]: bind() to [2001:41d0:2:8673::1]:80 failed (98: Address already 
in use)
2010/05/29 09:05:52 [emerg] 21060#0: still could not bind()
[emerg]: still could not bind()

I’m using nginx 0.8.37 and I was trying to upgrade to 0.8.38.
Here’s an example of my nginx configuration regarding IPv6 (all the
other vhost have the same one):

    server { # php/fastcgi
        listen [2001:41D0:2:8673::1];
        listen 80;
        server_name  img.italiansubs.net;
        access_log   logs/img.access.log  main;
        root         /var/www/img.italiansubs.net/public_html;


        location / {
            index    index.html index.htm index.php index.xml;
        }

        location /news2/data/ {
            expires 35d;
        }

        location ~ \.php$ {
            fastcgi_pass   unix:/tmp/php-fpm.sock;
#             fastcgi_pass   127.0.0.1:9000;
        }
    }

If I manually restart nginx via init script it works, but of course all
the open connections are dropped, which is something I would like to
avoid.

Thanks a lot,
Andrea

Posted at Nginx Forum:

2

Hello!

On Sat, May 29, 2010 at 03:23:54AM -0400, Usu wrote:

I’ve always upgraded nginx on the fly following this guide:
CommandLine | NGINX without problems.
I recently added IPv6 support to nginx and now, when I need to
upgrade it, it doesn’t start the new instance when I send the
USR2 signal to the master process, giving the following erros:

[code]
2010/05/29 09:05:52 [notice] 21060#0: using inherited sockets from “14;15;”
2010/05/29 09:05:52 [emerg] 21060#0: bind() to [2001:41d0:2:8673::1]:80 failed (98: Address already in use)
[emerg]: bind() to [2001:41d0:2:8673::1]:80 failed (98: Address already in use)

[…]

Attached patch should resolve this issue.

Maxim D.

3

Hello!

On Mon, May 31, 2010 at 03:57:28AM -0400, Usu wrote:

Hello,

thanks for the help, I tried patching the 0.8.38 release but it’s giving me an error:

root@localhost:/usr/src/nginx-0.8.38# patch -p1 < ipv6.diff
patching file src/core/ngx_connection.c
Hunk #1 FAILED at 96.
1 out of 1 hunk FAILED – saving rejects to file src/core/ngx_connection.c.rej

And here’s the file generated:

[…]

Am I doing something wrong?

It’s mailing list ↔ forum gateway which tried to interpret
bbcodes and mungled patch. You may obtain correct one from here:

http://nginx.org/pipermail/nginx/2010-May/020621.html

Maxim D.

4

Ok, it works like a charm thank you!
Will it be fixed in the next release of nginx?

Posted at Nginx Forum:

5

Hello!

On Mon, May 31, 2010 at 07:53:51AM -0400, Usu wrote:

Ok, it works like a charm thank you!
Will it be fixed in the next release of nginx?

It’s up to Igor, but most likely it will be fixed.

Maxim D.

6

On Mon, May 31, 2010 at 05:13:15PM +0400, Maxim D. wrote:

Hello!

On Mon, May 31, 2010 at 07:53:51AM -0400, Usu wrote:

Ok, it works like a charm thank you!
Will it be fixed in the next release of nginx?

It’s up to Igor, but most likely it will be fixed.

Yes, I will include it in 0.8.39, thank you.

–
Igor S.
http://sysoev.ru/en/

7

Hello,

thanks for the help, I tried patching the 0.8.38 release but it’s giving
me an error:

root@localhost:/usr/src/nginx-0.8.38# patch -p1 < ipv6.diff
patching file src/core/ngx_connection.c
Hunk #1 FAILED at 96.
1 out of 1 hunk FAILED – saving rejects to file
src/core/ngx_connection.c.rej

And here’s the file generated:

*************** ngx_set_inherited_sockets(ngx_cycle_t *c
*** 96,109 ****
      ls = cycle->listening.elts;
      for (i = 0; i < cycle->listening.nelts; i++) {

-         /* AF_INET only */
-
-         ls.sockaddr = ngx_palloc(cycle->pool, sizeof(struct 
sockaddr_in));
          if (ls.sockaddr == NULL) {
              return NGX_ERROR;
          }

-         ls.socklen = sizeof(struct sockaddr_in);
          if (getsockname(ls.fd, ls.sockaddr, &ls.socklen) == -1) {
              ngx_log_error(NGX_LOG_CRIT, cycle->log, ngx_socket_errno,
                            "getsockname() of the inherited "
--- 96,107 ----
      ls = cycle->listening.elts;
      for (i = 0; i < cycle->listening.nelts; i++) {

+         ls.sockaddr = ngx_palloc(cycle->pool, NGX_SOCKADDRLEN);
          if (ls.sockaddr == NULL) {
              return NGX_ERROR;
          }

+         ls.socklen = NGX_SOCKADDRLEN;
          if (getsockname(ls.fd, ls.sockaddr, &ls.socklen) == -1) {
              ngx_log_error(NGX_LOG_CRIT, cycle->log, ngx_socket_errno,
                            "getsockname() of the inherited "

Am I doing something wrong?

Thanks,
Andrea

Posted at Nginx Forum: