[lxc-devel] lxc_container->start fork keeping server sockets alive?

[Serge Hallyn]

Quoting John Keiser (jkeiser at getchef.com):
> Hi all,
> 
> I am trying to use lxc_container->start() from within a process that
> happens to have started a TCP server, and even when my program exits, the
> server socket stays alive.  This can be tested by running this program
> (which starts a server and then starts a container) twice--the second time
> resulting in a bind() error:
> 
> #include <sys/socket.h>
> #include <netinet/in.h>
> #include <stdio.h>
> #include <errno.h>
> #include <strings.h>
> #include <stdlib.h>
> #include <lxc/lxccontainer.h>
> 
> int main(int argc, char**argv)
> {
>   struct lxc_container *c;
>   int listenfd,connfd,n;
>   struct sockaddr_in servaddr;
>   int port;
> 
>   // Start server
>   port = atoi(argv[2]);
>   printf("Listening on port %d ...\n", port);
>   listenfd=socket(AF_INET,SOCK_STREAM,0);
>   bzero(&servaddr,sizeof(servaddr));
>   servaddr.sin_family = AF_INET;
>   servaddr.sin_addr.s_addr=htonl(INADDR_ANY);
>   servaddr.sin_port=htons(port);
>   if (bind(listenfd,(struct sockaddr *)&servaddr,sizeof(servaddr)))
>   {
>     perror("Error calling bind()");
>     exit(1);
>   }
>   listen(listenfd,1024);
> 
>   // Start
>   printf("Starting container ...\n");
>   c = lxc_container_new(argv[1], NULL);
>   c->want_daemonize(c, true);
>   c->want_close_all_fds(c, true);
>   c->start(c, 0, NULL);
> 
>   return 0;
> }
> 
> 
> *> gcc test.c -llxc -o test*
> *> lxc-create -n blarghle -t ubuntu*
> 
> *> ./test blarghle 9050*
> Listening on port 9050 ...
> Starting container ...
> *> ./test blarghle 9050*
> Listening on port 9050 ...
> Error calling bind(): Address already in use
> 
> As you can see, even though ./test finished the first time, the second
> ./test cannot bind to the port because it is still in use.
> 
> This appears to be because the lxc init process has a ghost "./test" as its
> parent:
> 
> *> ps -ef | grep 30276*
> root     30276     1  0 18:37 ?        00:00:00 ./test blarghle 9050
> root     30285 30276  0 18:37 ?        00:00:00 /sbin/init
> *> lsof | grep 9050*
> test      30276              root    3u     IPv4             366649
>  0t0        TCP *:9050 (LISTEN)

Hm.  That's the container monitor.  It doesn't close all fds like
container init does.

> I am stymied on how to investigate further.  My best guess is it has to do
> with the double forking in
> lxccontainer.h<https://github.com/lxc/lxc/blob/master/src/lxc/lxccontainer.c#L608-L621>,
> though I don't really understand what that is for yet :)  Any thoughts?
> 
> Thanks!
> 
> --John

> _______________________________________________
> lxc-devel mailing list
> lxc-devel at lists.linuxcontainers.org
> http://lists.linuxcontainers.org/listinfo/lxc-devel