Hi Eric, it seems to me that if this is the best we can do, we're fixing it at the wrong layer. Surely putting the fix in the Linux kernel would be much smaller in code size, and higher performing, as we wouldn't have to filter packets in user space. --Juan