(usagi-users 02475) Kernel patches changing behavior of socket bind()

[Mark Huth <mhuth@xxxxxxxxxx>]

I notice that certain of the usagi patches to 2.4.18+ kernels modify the 
behavior of get_port in the various protocols, causing the isc dhcp 
server and relay to fail.  I've examined the code and can readily fix 
either the server or the kernel to work.  However, I would like to 
understand the motivation for some of the kernel changes.  I need to 
decide what "flavor" of SO_REUSEADDR to support, and consequently, what 
behavior the damons should expect.

The stock kernel basically allows SO_REUSEADDR to allow unlimited 
bindings to the same socket tuple.  This behavior is basically carried 
forward into the 2.5 snd 2.6.test kernels.  The usagi IPV6 patches 
change the behavior to be more selective and configurable - SO_REUSEADDR 
is mapped into the SO_REUSEPORT equivalent for multicast addresses, 
device and address bindings must be specified for both sockets to avoid 
a reuse error, MODERATE_REUSE is allowed to be configured to protect a 
DOS attack, and perhaps a couple of other nits.

We need to release a distribution supporting IPV6, and we generally use 
the usagi patches.  However, I need your input to decide what sort of 
reuse we need to support.

Comments, please.

Thanks,
Mark Huth