Re: [v6ops] I-D Action: draft-buraglio-v6ops-ula-01.txt

[Brian E Carpenter <[e-mail]>]

Hi,

Thanks for this draft. I have a few comments (and a tiny nit at the end).

>  The core issue is the stated interpretation from gai.conf that has the following default:
> 
> #scopev4  <mask> <value>
> #    Add another rule to the RFC 6724 scope table for IPv4 addresses.

I'm not sure why this matters. RFC6724 is quite correct to indicate that
most IPv4 unicast addresses formally have global scope, but auto-config
and loopback addresses have link-local scope. IPv6 is pretty much the
same, and in particular ULAs have *global scope* even though they are
not globally reachable. RFC1918 addresses are identical to ULAs in
that respect.

Citing RFC4291 and https://www.rfc-editor.org/rfc/rfc8190.html#section-2.1
would clarify the difference between global scope (architectural) and
globally reachable (practical). What we care about here is whether an
address is globally reachable ("no" for both RFC1918 and ULA, although
they are both architecturally global). Unfortunately this distinction is
lacking in the description of gai.conf and, I suspect, in the code of
Linux getaddrinfo().

What I think is lacking in the draft is an explanation of how
getaddrinfo() works and why it matters. Here's a walkthrough that
I hope will help clarify what I mean:

Consider an end-user network with the following properties:

It is dual stacked.
It uses 10.1.0.0/16 (NATted to the Internet).
It uses (or wants to use)  fdee:face:fade::/48 for internal IPv6.
It uses 2001:db8:fade::/48 for external IPv6

We'll neglect for now whether it has a subnet structure. It shouldn't
matter.

Consider a host user.mynet.example.com, a local server printer.mynet.example.com,
and a remote server www.theirnet.example.com. Assume they have these various
addresses:

user.mynet.example.com has:

10.1.0.1
fdee:face:fade::1
2001:db8:fade::1

printer.mynet.example.com has:

10.1.0.10  (A record in local DNS)
fdee:face:fade::a  (AAAA record in local DNS)

www.theirnet.example.com has:

192.0.2.15  (A record in global DNS)
2001:db8:cafe::f  (AAAA record in global DNS)

What do we *want* to happen?

If user opens a connection to printer, we want it to choose
SA = fdee:face:fade::1
DA = fdee:face:fade::a

If user opens a connection to www, we want it to choose
SA = 2001:db8:fade::1
DA = 2001:db8:cafe::f

Now, if user does a DNS lookup, via getaddrinfo(), the results
will look like this (in the Python universe):

For printer:

(<AddressFamily.AF_INET: 2>, 0, 0, '', ('10.1.0.10', 0))
(<AddressFamily.AF_INET6: 23>, 0, 0, '', ('fdee:face:fade::a', 0, 0, 0))

For www:

(<AddressFamily.AF_INET6: 23>, 0, 0, '', ('2001:db8:cafe::f', 0, 0, 0))
(<AddressFamily.AF_INET: 2>, 0, 0, '', ('192.0.2.15', 0))

At this point, consider what RFC6724 says:

    As a consequence, we intend that implementations of APIs such as
    getaddrinfo() will use the destination address selection algorithm
    specified here to sort the list of IPv6 and IPv4 addresses that they
    return.  Separately, the IPv6 network layer will use the source
    address selection algorithm when an application or upper layer has
    not specified a source address.

Thus, to get the desired behaviour, what matters is destination
address selection: if we select DA = fdee:face:fade::a, then the
ULA source address will follow.

Of course this is a small matter of programming, and most programmers
just pick the first address. That's why we need the Section 10.6
mechanism of RFC6724, to insert an appropriate precedence like
     
    fdee:face:fade::/48 45 14

which will prioritize local use of ULAs but will change nothing
for off-site access.

At that point in my thinking, I started coding the program that
I posted yesterday.

Nit:

s/gai.cnf/gai.conf/

Regards
    Brian