Hello,

It looks like something is broken in ioctl in 3.2 when the supplied (mutable) bytearray is exactly 1024 bytes long - the result is not copied into the buffer after the ioctl succedes:

def open_tuntap(type, name):
        TUN_TYPE = {
                'TUN' : 0x0001,
                'TAP' : 0x0002
        }

        TUNSETIFF = 0x400454ca

        dev = os.open('/dev/net/tun', os.O_RDWR)
        buf = bytearray(SET_LEN_HERE)
        name = name[:16]
        buf[:len(name)] = name
        buf[16:18] = TUN_TYPE[type].to_bytes(2, sys.byteorder)
        fcntl.ioctl(dev, TUNSETIFF, buf, True)
        print(buf)
        print(len(buf))

open_tuntap('TAP', b'madtun%d')

Now try it with SET_LEN_HERE = 1024, 1023, 1025 and any other values. For < 1024 it copies to the static buffer and back, for > 1024 it operates on the buffer itself (hopefully) and for 1024 it ignores the modified buffer completely. It's probably some corner case bug.

The example was tested under Linux 2.6.35.

Python 3.2a1+ (py3k:84054, Sep  3 2010, 01:45:35) 
[GCC 4.4.2] on linux2

The attached patch fixes this.

Do you think there is a simple, generic way to test for this (see Lib/test/test_ioctl.py)?

I'd already had a look around, and my knowledge of ioctls is not sufficient to answer that question but I suspect the answer is no. 

Does someone know of a ioctl that works across platforms, doesn't require specific hardware or privileges and that does some form of read operation?

Actually, writing a test ended up quite easy (by re-using an existing test in test_ioctl and simply enlarging its buffer). I've committed an updated patch in r84589 (3.x), r84590 (3.1) and r84591 (2.7).
Thank you!