Re: better coping with multiple forms of ipv6 route and address assignment

On Tue, 2017-02-14 at 14:55 -0800, Dave Taht wrote:
I see that network manager has an easy means of specifying different
metrics for routes.

The defaults I get on ubuntu are a bit screwy, in that the ipv4
default is 600 (which is much more than the default), and the ipv6
default is also 600, which is lower than the default 1024.

I just arbitrarily changed mine universally to 2048, which helps in
complex routed environment. But that raises a couple questions.


Sidenote: you can set the metric via ipv4.route-metric property, but if
you leave that unspecified, the default value depends on the device
type, e.g. 600 is Wi-Fi:

But actually, you can overwrite that default value via global configuation:


(see `man NetworkManager.conf`)

Presently NetworkManager uses "proto static" for routes derived from
dhcp. There is an entry for "dhcp" in /etc/iproute2/rt_protos (16) -
is it possible to use that rather than static?

No, it's not possible, the protocol is automatically chosen based on
the address type (basically, just DHCP, NDISC, and STATIC). Beniamino
is working on supporting more properties for routes, if you care about
the "proto", maybe it could be added too. What is your usecase? Do you
want to set "proto" for manual/static routes?

Is there a way to not install default routes at all? (source specific
routing here)

You can set ipv6.never-default=yes for the connection, that prevents
the default route from being added.
See `man nm-settings`.

And: dhcpv6-pd is proving to be a PITA to deploy and might be useful
to have it's own rt_proto due to various special requirements
(off-link ra announcements and the like). Don't know if network
manager can do pd (yet?), dhclient and friends can....

Lubomir implemented ipv6.method=shared. In that case, NM does v6 prefix
delegation. But that is not meant as a flexible, configurable setup,
more like a simple way to share a network.
This feature is new in 1.6.0+, some testing would be good.

And things like the inet homenet protocol (hnetd) are using an
unofficial proto number of 43, and some work in progress of mine is
using 44 (for lack of a better number)

And, did the concept of NLM_F_MATCH die? [1]

I don't know

Lastly: Given all the different forms of ipv6 address assignment, I
kind of wish the proto concept also applied to addresses, or scope
identifiers, or something.

And: Sorry for so many ands!



Attachment: signature.asc
Description: This is a digitally signed message part

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]