inconsistent null checking in ipx_ioctl()

[Dan Carpenter <dan.carpenter@oracle.com>]

Hi, I'm working on some new Smatch stuff and going through some warnings
in old code.

----
This is a semi-automatic email about new static checker warnings.

The patch b0d0d915d1d1: "ipx: remove the BKL" from Jan 25, 2011, 
leads to the following Smatch complaint:

net/ipx/af_ipx.c:1928 ipx_ioctl()
	 error: we previously assumed 'sk' could be null (see line 1913)

net/ipx/af_ipx.c
  1912			rc = -EINVAL;
  1913			if (sk)
                           ^^^^
Check.

  1914				rc = sock_get_timestamp(sk, argp);
  1915			break;
  1916		case SIOCGIFDSTADDR:
  1917		case SIOCSIFDSTADDR:
  1918		case SIOCGIFBRDADDR:
  1919		case SIOCSIFBRDADDR:
  1920		case SIOCGIFNETMASK:
  1921		case SIOCSIFNETMASK:
  1922			rc = -EINVAL;
  1923			break;
  1924		default:
  1925			rc = -ENOIOCTLCMD;
  1926			break;
  1927		}
  1928		release_sock(sk);
                ^^^^^^^^^^^^^^^^^
The lock and release functions dereference "sk".  Probably the check
can be removed.  The rest of the function dereferences "sk" without
checking.  A lot of this code goes back to 2.6.12.

  1929	
  1930		return rc;

regards,
dan carpenter