[Gluster-users] 2.0.6/transport/socket/src/name.c af_inet_bind_to_port_lt_ceiling usage
Anand Avati
avati at gluster.com
Fri Sep 11 08:14:37 UTC 2009
> here you can see some code from above source that you really do not want if
> you like defensive programming:
>
>
> int32_t
> client_bind (transport_t *this,
> struct sockaddr *sockaddr,
> socklen_t *sockaddr_len,
> int sock)
> {
> int ret = 0;
>
> *sockaddr_len = sizeof (struct sockaddr_in6);
> switch (sockaddr->sa_family)
> {
> case AF_INET_SDP:
> case AF_INET:
> *sockaddr_len = sizeof (struct sockaddr_in);
>
> case AF_INET6:
> ret = af_inet_bind_to_port_lt_ceiling (sock, sockaddr,
> *sockaddr_len, CLIENT_PORT_CEILING);
> if (ret == -1) {
> gf_log (this->xl->name, GF_LOG_ERROR,
> "cannot bind inet socket (%d) to port less than %d (%s)",
> sock, CLIENT_PORT_CEILING, strerror (errno));
> ret = 0;
> }
> break;
>
> case AF_UNIX:
> *sockaddr_len = sizeof (struct sockaddr_un);
> ret = af_unix_client_bind (this, (struct sockaddr *)sockaddr,
> *sockaddr_len, sock);
> break;
>
> default:
> gf_log (this->xl->name, GF_LOG_ERROR,
> "unknown address family %d", sockaddr->sa_family);
> ret = -1;
> break;
> }
>
> return ret;
> }
>
> If you look closely you find that the error case of af_inet_bind_to_port_lt_ceiling
> returning -1 is ending up as ret=0, although all other error cases of this function
> return -1. Which means this function returns 0 or -1 in case of error and 0 in case
> of nonerror. No chance to make anything useful of the return value for a caller.
Unlike what you think, the code has the intended behavior. It makes
the best effort to bind to a port less than the ceiling value. If it
cannot bind to port less than 1024, it proceeds anyways with the hope
that the server is not doing a host based authentication check.
BTW, such mails are best posted to gluster-devel.
Avati
More information about the Gluster-users
mailing list