The following reply was made to PR kern/33409; it has been noted by GNATS.

From: Chapman Flack <nblists@anastigmatix.net>
To: gnats-bugs@NetBSD.org, Darren Reed <darrenr@pobox.com>
Cc: 
Subject: Re: kern/33409 ipnat fails with EIO
Date: Sun, 28 May 2006 22:01:35 -0400

 I can repeat the behavior in NetBSD 3.99.20, just as Roland describes:
 any use of ipnat(8) fails with
   ioctl(SIOCGNATS): Input/output error
 UNLESS 'ipf -E' has been done first.
 
 This seems to be at the very least a doc bug, as our ipnat(8) man page
 claims explicitly:
 
   Note that if ipf(8) is not enabled when NAT is configured, it will
   be enabled automatically, ...                              ^^ ^^^^
   ^^ ^^^^^^^ ^^^^^^^^^^^^^
 
 I will copy this message to Darren Reed for an opinion as to whether
 to treat this as a doc bug (and simply change the man page to say
 ipf -E is required first) or as a sw bug and change the code to do
 what the man page says.
 
 I do not see any code that appears even to try to do what the manpage
 says, preceding the offending ioctl, so it seems more like a doc bug,
 so if not advised otherwise I'll change the man page.
 
 Darren, the ipf version in question seems to be 4.1.13.
 
 -Chap