Re: Finishing deprecation of isc-dhcp-client in ifupdown*
Le 01/11/2025 à 18:23, Vincent Danjean a écrit :
- `accept_ra 0` will prevent kernel SLAAC address and (default) route
creation entirely, leaving no address configuration method enabled
(other than link-local). Further, no stateful DHCPv6 client will be
enabled by RA flag M being advertised by the network since that's not
supported by dhclient AFAIK.
I use this config indeed to prevent the default route to be set (here again
this is due to the multi ISP settings and specific routing policy), while
preserving kernel SLAAC address allocation and requesting prefix delegation
with dhcp (IPv6).
Sorry, I made an error in my statement.
This ISP does not give me a SLAAC address. In fact, it does not
give me any routable IPv6 address directly. I just use a link-local IPv6
(hence the "inet6 auto" setup in /etc/network/interface) and
I route my IPv6 traffic to an other IPv6 link-local address (fe80::...)
I had to learn it initially by accepting RA and looking at the provided
value.
I plan to monitor RA to check that this IPv6 link-local address does not
change, but it is only on my TODO list for now. The gateway changes only
once (a few months ago) for 8 years and it was due to a hardware change
on the ISP side (so the mac address of its interface changes). When it occurs,
I had forgotten this static setup and it took me a while to find why the
IPv6 connexion was not working anymore.
As said, I disabled the kernel routing setup (accept_ra 0) in order to avoid
the kernel to mess up with the multi-ISP routing in shorewall6.
However, this ISP gives me a /48 network when requested with a IPv6 dhcp client.
It routes it to me through the IPv6 link-local address used by the dhcp client.
Note that the routing for the delegated prefix is manually set (in shorewall).
But, if the dhclient does not request it, the ISP does not route it to my interface.
Hence the need to have a IPv6 dhcp client that requests PD (even if
all the address assignment and routing is done manually (or with radvd))
Regards,
Vincent
Reply to: