svn commit: r197145 - in head: etc/defaults share/man/man5

[Doug Barton]

[Please direct follow-ups to freebsd-rc@]

Hiroki,

I realize that you've posted your patches in the past, and I
definitely had it in mind to review them in detail and give you
feedback on them. However I got focused on my own projects for the
pending release, and then since we were so close to the release I did
not think you would be committing these changes until after it was
done so I let review of these patches slip down my list of priorities.
Therefore I ask you to accept my apologies for this "after the fact"
review.

Before I forget, you keep putting "mfc after 3 days" in your commit
messages. You don't actually plan to MFC these changes to RELENG_8
prior to the 8.0-RELEASE do you? I would not be supportive of this
given the sweeping nature of the changes and the (unfortunately) small
percentage of our userbase that uses and tests IPv6. I think shaking
this code out in HEAD for several weeks at least would be a good thing.

That said I very much like the idea of integrating IPv6 support into
the overall network code, and I am very supportive of that. I do have
some small problems with the direction that you've taken in a few
areas, which happen to be neatly summarized in this post.

Hiroki Sato wrote:
> Author: hrs
> Date: Sat Sep 12 22:22:31 2009
> New Revision: 197145
> URL: http://svn.freebsd.org/changeset/base/197145
> 
> Log:
>   The following changes are added because of
>   network_ipv6->rc.d/netif integration:
>   
>   - $ipv6_enable is now obsolete.  Instead, IPv6 is enabled by
>     default if the kernel supports it, and $ipv6_network_interfaces
>     is "none" by default.  If you want to use IPv6, define
>     $ipv6_network_interfaces and $ifconfig_xxx_ipv6.

In general I have a problem with the idea of drastically changing the
semantics of the current code when I can't see any real value in doing
so. I object to this change specifically because on my laptop I really
like having the ability to easily disable IPv6 when I am not on my
home network.

My preferred scenario would be something similar to what we have now,
which is that if ipv6_enable is set that it takes the same list of
interfaces as ipv4 (defaults to AUTO) and that rtadv is enabled by
default for each of those interfaces. My feeling is that this not only
significantly reduces POLA it will also more precisely fit the way
that the vast majority of our users will actually use IPv6.

On a "marketing" note I really think it would be valuable to make it
as easy as possible for the average user to get IPv6 working. We have
IPv4 down to it more or less "just works," I think IPv6 should be the
same way. (On a side note, I'd actually like to see DHCP be the
default for IPv4 such that if you have DHCP available on a network you
wouldn't have to do any configuration at all to get FreeBSD on line,
but that's a whole other topic.)

>     An interface which is in $network_interfaces and not in
>     $ipv6_network_interfaces will be marked as "inet6
>     -auto_linklocal ifdisabled" (see ifconfig(8)).

I think that if a user specifies both lists, and that they differ,
that this is a very good way to handle it.

>   - $ipv6_ifconfig_xxx is renamed to ifconfig_xxx_ipv6 for
>     consistency with other address families.  The old variables
>     still work but can be removed in the future.  Note that
>     ipv6_ifconfig_xxx="..." should be replaced with
>     ifconfig_xxx_ipv6="inet6 ...".

I see that you are giving a warn'ing when the old format is detected,
which is good.

>   - Receiving ICMPv6 Router Advertisement is not automatically
>     enabled even if there is no manual configuration of IPv6 in
>     rc.conf.  If you want it, define
>     ifconfig_xxx_ipv6="inet6 ... accept_rtadv".

What is the reason for this change? While I am definitely in favor of
making it easier to disable rtadv, I still think it should be the
default.

>   - The rc.d/ip6addrctl now chooses address selection policy based
>     on $ipv6_prefer, not $ipv6_enable.  The default is
>     ipv6_prefer=NO.

Once again, what is the reason for this change? My read of the
IPv6-using community is that if they have it available they want to
use it as a first choice. I know that is certainly my preference.


Once again my apologies for the late review, but hopefully this is useful.


Doug

-- 

    This .signature sanitized for your protection