"[RFC PATCH net-next 2/2] Reduce localhost to 127.0.0.0/16"

"[RFC PATCH net-next 2/2] Reduce localhost to 127.0.0.0/16"

[Mark Smith]

(Not subscribed to the ML)

Hi,

I've noticed this patch. I don't think it should be applied, as it
contradicts RFC 1122, "Requirements for Internet Hosts --
Communication Layers":

"(g)  { 127, <any> }

                 Internal host loopback address.  Addresses of this form
                 MUST NOT appear outside a host."

RFC 1122 is one of the relatively few Internet Standards, specifically
Standard Number 3:

https://www.rfc-editor.org/standards


Regards,
Mark.

Re: "[RFC PATCH net-next 2/2] Reduce localhost to 127.0.0.0/16"

[Dave Taht]

On Fri, Sep 13, 2019 at 9:54 AM Mark Smith <markzzzsmith@gmail.com> wrote:
>
> (Not subscribed to the ML)
>
> Hi,
>
> I've noticed this patch. I don't think it should be applied, as it
> contradicts RFC 1122, "Requirements for Internet Hosts --
> Communication Layers":

Yea!  I kicked off a discussion!

> "(g)  { 127, <any> }
>
>                  Internal host loopback address.  Addresses of this form
>                  MUST NOT appear outside a host."

That 1984 (89) definition of a "host" has been stretched considerably
in the past few decades. We now have
a hypervisor, multiple cores, multiple vms, vms stacked within vms,
and containers with virtual interfaces on them, and a confusing
plethora of rfc1918 and nat between them and the wire.

This RFC-to-netdev's proposed reduction to a /16 was sufficient to
cover the two main use cases for loopback in Linux,
127.0.0.1 - loopback
127.0.1.1 - dns

We'd also seen some usages of things like 127.0.0.53 and so on, and in
the discussion at linuxconf last week,
it came out that cumulus and a few others were possibly using high
values of 127.x for switch chassis addressing, but we haven't got any
documentation on how that works yet.

The 1995 IPv6 standard and later has only one loopback address.
127.0.0.0/8 is 16m wasted internal to the host addresses.

> RFC 1122 is one of the relatively few Internet Standards, specifically
> Standard Number 3:
>
> https://www.rfc-editor.org/standards

We have been exploring the solution space here:

https://github.com/dtaht/unicast-extensions/blob/master/rfcs/draft-gilmore-taht-v4uniext.txt

If you would like to file more comments and bugs - or discuss here!
that would be great.

>
> Regards,
> Mark.



-- 

Dave Täht
CTO, TekLibre, LLC
http://www.teklibre.com
Tel: 1-831-205-9740

Re: "[RFC PATCH net-next 2/2] Reduce localhost to 127.0.0.0/16"

[David Ahern]

On 9/13/19 10:14 AM, Dave Taht wrote:
> it came out that cumulus and a few others were possibly using high
> values of 127.x for switch chassis addressing, but we haven't got any
> documentation on how that works yet.

Not Cumulus.

I noted I am aware of 2 products from my history that use 127.x
addresses for communications within a box - e.g., to a bmc - that your
patch could break. Really it was meant as a data point that there are
released products that would be affected.