Eric Blake <eblake@redhat.com> wrote on 08/13/2010 03:56:12 PM:

> [image removed]

>
> On 08/13/2010 01:45 PM, Stefan Berger wrote:
> >>> -                    // skip eth. bcast and mcast addresses,
> >>> +                    // skip eth. bcast and mcast addresses (224.0.0.0
> > -
> >>> +                    // 239.255.255.255), class E (255.*)
> >>>                      // and zero address in DHCP Requests
> >>> -                    if ((ntohl(vmaddr) & 0xc0000000) || vmaddr == 0)
>
> > http://en.wikipedia.org/wiki/Classful_network
> >
> > Class D addresses have highest bits with pattern 1110 0000 -> 0xe0
> > Class E addresses have highest bits with pattern 1111 0000 -> 0xf0
> >
> > I think my masks are fine and the masking with 0xf0 00 00 00 should also
> > include 254.*  =  0xfe.*  .
>
> In that case, the comments are wrong.  Class E is more than 255.*, it is
> 240.0.0.0-255.255.255.255.  And in that case, the bit operations can be
> simplified:
>
> if ((ntohl(vmaddr) & 0xc0000000) == 0xc0000000) || vmaddr == 0)
>
> In other words, the logic bug is that we were rejecting IP addresses
> that had 1 or 2, but not all three, of the top three bits set.  The
> desired action is to reject IP packets if all three of the top bits are
> simultaneously set.

Right, the comment was not correct. I think the simplified if should be

if ((ntohl(vmaddr) & 0xe0000000) == 0xe0000000) || vmaddr == 0)

That then covers class D and E since this mask then covers 0xe0.00.00.00 - 0xff.ff.ff.ff = 225.0.0.0 - 255.255.255.255.

>
> Let's see a v2 that gets the comments right, and uses the simpler logic.

Coming soon.

   Stefan