root@gw:/etc/quagga# ip -6 route | grep default
default via ::220.127.116.11 dev 6to4-ge01 metric 1024
The 2002::/16 route is propagated to all clients. Note that it ends up in the kernel table.
root@gw:/etc/quagga# ip -6 route | grep 6to4
2002::/16 dev 6to4-ge01 proto kernel metric 256
If I manually on the client add the default route via
ip -6 route add default via fe80::204e:7fff:fe4a:9627 dev gw11
So my guess is that the default route in the 6to4 netifd script needs to also be inserted into the kernel table in order to be passed to clients via quagga.
I have denis’s patch as well installed for RT_PROT messages, but that doesn’t appear to affect this behavior.
0) quagga is not pushing the 6to4 default route out to other babel nodes.
1) On the local interface,
the following syntax only adds one 2002:: based addresses to the listed
config interface ‘ge01’
option proto ‘6to4’
option adv_subnet ‘1’
option wan_device ‘ge00’
option metric ‘1’
option ttl ‘64’
list adv_interface ‘gw00’
list adv_interface ‘gw10’
list adv_interface ‘se00’
list adv_interface ‘sw00’
list adv_interface ‘sw10’
Interestingly, although luci will write this syntax, when it re-reads it only finds the first (gw00) interface. (see pic)
(it will also read the old syntax and convert to this)
I have been refreshing luci a lot, but somehow perhaps I’m out of sync?
my last commit was 6c6af10f0859cbe5d61f5d2d60dfc9ea5215285e
And regardless only the third (se00) of these interfaces gain a 2002 address.
2) This generates a radvd file for the one interface that works as
Perhaps I need to put in a default route statement explicitly?
The one client connected that way (haven’t verified it’s functionality)
isn’t getting out.
3) when netifd runs it deletes up the addresses gained via ahcp, wiping out the mesh.
config interface ‘gw01’
option proto ‘ahcp’
config interface ‘gw11’
option proto ‘ahcp’
But although I changed quagga to distribute static as well, it still doesn’t pick up the 6to4-ge01 default route for ipv6. I tried adding the device to the quagga “network” stanza, to no luck… It is not shown as an export from vtysh, either.
0.0.0.0/0 metric 0 (exported)
18.104.22.168/21 metric 0 (exported)
192.168.1.0/27 metric 0 (exported)
192.168.1.64⁄27 metric 0 (exported)
192.168.1.96⁄27 metric 0 (exported)
192.168.1.128⁄27 metric 0 (exported)
192.168.1.160⁄27 metric 0 (exported)
192.168.1.224⁄32 metric 0 (exported)
::22.214.171.124⁄128 metric 0 (exported)
2002::/16 metric 0 (exported)
2002:328a:a616:bab5::1⁄128 metric 0 (exported)
Perhaps I can statically add it to the zebra config?
moving on to the other issues…
the 6to4 portion of the netifd scripts does not do address assignment. I’m not sure if it should, actually, when controlled by the adv_subnet variable. Perhaps a new variable, “assign_subnet” would be more appropro, or rs solicitations enabled in this case to automagically get a slaac address…
(btw, I like the scripts they are very clean and easy to read)
and have no idea how to tell it to ignore interfaces run with ahcp.
ipv6 route ::/0 2002:c058:6301::
I’m not certain I wanted to do it that way (certainly I don’t want to announce routes via that interface)
proto/6x4: cast 6to4 adv_interface to string when saving to uci, fixes 6in4.
Jow moved netifd to install userspace routes as static
[package] netifd: bump to git head, userspace routes are installed with RTPR_STATIC
This did not actually fix quagga. I put in a workaround in #note-3 -
but I’m not happy with having to embed the
anycast route there, even if it is pretty generic…
This leaves netifd mucking with ahcpd managed interfaces when it shouldn’t be as the last big remaining issue here.
Thx jow for leaping on this.
Actually, no, the correct line would be:
set_6to4_radvd_prefix “\$sid” “\$adv_interface” \
“\$wancfg” “\$(printf “%s:%x::/64”
\$prefix6 \$adv_subnet) \
arguably “subnet6” is misnamed.
There is another bug in this script in correctly adding multiple
interfaces. It succeeds on the first, but not the rest, and thus far I
can’t figure out why.
There is another bug in that arguably the interface itself should have
a prefix lifetime (monitored somehow) and mtu set and I’d argue for
marking it secondary in the hope that someday real ipv6 addresses
would become common.
And last bug, in that with the latest quagga OR dnsmasq, radvd is not
needed, except for detecting and generating the 6to4 prefix.
I’m just happy radvd is going to go away. soon.