RANDOM_IP_ID sysctl?

[David Malone]

> > On Tue, Jun 29, 2004 at 02:13:38PM +0100, David Malone wrote:
> > > It seems to me that RANDOM_IP_ID might be better as a sysctl rather
> > > than a kernel option. Would anyone mind if I changed this?

> I personally think that RANDOM_IP_ID is something that should be tweakable on
> a per-interface basis (at least). I usually want randomized IDs on my uplink
> interface while it could harm my GigE internal network due to faster ID reuse
> cycles.

Though, on your GigE internal network, you can probably use jumbograms
to avoid fragmentation and avoid having to pay any attention to IP
IDs ;-) There was code to only bother setting the IP ID for packets
with DF set, but that caused certain odd pieces of network kit that
ignored DF to be sick...

> FYI, pf(4) can set randomized IDs on a per interface (and even on a per
> connection) basis.

pf is probably the best place for this sort of more flexable IP ID
setting.

> David, I'd appreciate to review your patches in order to avoid breakage of pf,
> thanks.

Certainly.

> One would clearly transform the now present "#ifdef" with "if (sysctlvar)"
> and hence this will not incur overhead (one compare is nothing to worry
> about).

Yes - this is exactly what I had in mind. I'm certain that the
overhead of the extra "if ()" will be lost in the cost of the IP
stack.

Currently RANDOM_IP_ID also controls other things, such as flow id
generation for IPv6. Making it a sysctl would also give us independent
control over these things.

	David.