Problem with ctdb_sys_have_ip

[ronnie sahlberg]

On Sat, Aug 29, 2009 at 2:58 AM, Harald Klatte<klatte at hrz.uni-kassel.de> wrote:
>
> Hi,
>
> testing ctdb on AIX (not on Linux) I detected a misbehaviour:
> Ctdb tries to take over IP Adresses, even if they are still configured
> according to ifconfig.
>
> I determined the function "ctdb_sys_have_ip" in system_common.c as the cause.
> It always returns false no matter if the concerning address is configured or not.
>
> I wound up with inserting a debug line into the code:
>
>
> /*
>  see if we currently have an interface with the given IP
>
>  we try to bind to it, and if that fails then we don't have that IP
>  on an interface
>  */
> bool ctdb_sys_have_ip(ctdb_sock_addr *_addr)
> {
>        int s;
>        int ret;
>        ctdb_sock_addr __addr = *_addr;
>        ctdb_sock_addr *addr = &__addr;
>
>        switch (addr->sa.sa_family) {
>        case AF_INET:
>                addr->ip.sin_port = 0;

Can you try adding this code  at this point :
#ifdef HAVE_SOCK_SIN_LEN
	          addr->ip.sin_len = sizeof(addr->ip);
#endif


>                break;
>        case AF_INET6:
>                addr->ip6.sin6_port = 0;
>                break;
>        }
>
>        s = socket(addr->sa.sa_family, SOCK_STREAM, htons(IPPROTO_TCP));
>        if (s == -1) {
>                return false;
>        }
>
>        ret = bind(s, (struct sockaddr *)addr, sizeof(*addr));
>        if (ret == -1) {
>                DEBUG(DEBUG_CRIT,("  !! failed to bind address to socket (%s)\n", strerror(errno) ));
>                return false;
>        }
>
>        close(s);
>        return ret == 0;
> }
>
>
> Syslog shows:
>
> 2009/08/28 18:19:15.730807 [188748]: server/ctdb_recoverd.c:1541 Recovery - disabled recovery mode
> 2009/08/28 18:19:15.730894 [188748]: Deterministic IPs enabled. Resetting all ip allocations
> 2009/08/28 18:19:15.731116 [147676]:   !! failed to bind address to socket (Invalid argument)
> 2009/08/28 18:19:15.731210 [147676]: Takeover of IP xx.xx.xx.xx/24 on interface en1
> 2009/08/28 18:19:15.731891 [147676]:   !! failed to bind address to socket (Invalid argument)
> 2009/08/28 18:19:15.731980 [147676]: Takeover of IP xx.xx.xx.xx/24 on interface en1
> 2009/08/28 18:19:15.733006 [188748]: server/ctdb_recoverd.c:1552 Recovery - takeip finished
> 2009/08/28 18:19:15.733140 [147676]: Recovery has finished
>
>
> this shows EINVAL while there is no obviously wrong data type.
>
> Has someone seen this fault before?
> Where can be the problem?
>
>
> Thanks
>
> Harald
>
>
> --
>
> +----------+  Harald Klatte                       email: klatte at hrz.uni-kassel.de
> |Uni-Kassel|  ITS,  Universitaet Kassel                   Tel.: (49) 561/804-2280
> +----------+  Moenchebergstr. 11, 34109 Kassel            Fax:  (49) 561/804-2297
>
>