IP fastforwarding in RELENG_4 and CURRENT/RELENG_6

[Dmitry Pryanishnikov]

Hello!

  What is the current status of the fast IP forwarding in RELENG_4 and in
modern versions (CURRENT/RELENG_6)? I see that this code (either ip_flow.*
in RELENG_4 or ip_fastfwd.c in RELENG_6) is always included into kernel
(no separate option for it), but is disabled by default. What are drawbacks
from enabling it (pure-IPv4 environment, heavy use of ipfw+divert+dummynet,
occasionally use of IPSEC)? I haven't found any documentation for this
option besides comments in ip_fastfwd.c, and those comments rose several
questions:

  * Else if something is not pure IPv4 unicast forwarding we fall back to
  * the normal ip_input processing path. We should only be called from
----------------------------------------^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  * interfaces connected to the outside world.
---^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

How to achieve this aim? I see no fastforwarding-specific options in
ifconfig.

  * IPSEC is not supported if this host is a tunnel broker. IPSEC is
  * supported for connections to/from local host.

Is it true for FAST_IPSEC? Am I understand 'tunnel broker' correctly:
it's the host that wraps other host's traffic into the ESP using
IPSEC tunnel mode? How about IPSEC transport mode?

And the main question: does this description stands for ip_flow implementation
in RELENG_4? If not, what are the differences?

Sincerely, Dmitry
-- 
Atlantis ISP, System Administrator
e-mail:  dmitry at atlantis.dp.ua
nic-hdl: LYNX-RIPE