VTI gateways not adding static routes in 24.03

[OhYeah 0]

Tried something a bit more drastic.

1. Deleted everything: static routes, gateway, disabled interface, deleted assignment, deleted P2, deleted P1.
2. Restart.
3. Switch global states back to "floating" and IPSEC filter mode back to "on IPSEC tab".
4. Restart.
5. Add everything back in the same order as standard (but different names just to make sure something doesn't clash with cached or old entries).
6. Restart.

Same status. P1 comes up, routes are not added to the routing table.

[LarryFahnoe]

@OhYeah-0 And when you rebuilt, you did so with 0.0.0.0/0 correct? The rationale for that was a mixed environment if I understood.

Would it be possible to do a test using a private /30 transit network?

--Larry

[OhYeah 0]

@LarryFahnoe said in VTI gateways in 24.03:

And when you rebuilt, you did so with 0.0.0.0/0 correct? The rationale for that was a mixed environment if I understood.

Yep. With that client we have a hub-and-spoke topology with different vendor platforms (also mix of virtual and physical instances). The solution had been working flawlessly until the 24.03 update.

[stephenw10]

Ok so some of the connected clients are using policy mode and require a P2 that carries all destination traffic?

[OhYeah 0]

@stephenw10 said in VTI gateways in 24.03:

Ok so some of the connected clients are using policy mode and require a P2 that carries all destination traffic?

All endpoints are connected via the same method (0.0.0.0/0 local/remote and static routes).

I know that while it's possible to mix policy and route based IPSEC; it's really not a good idea. You lose all the benefits and there's another source of potential problems.

[stephenw10]

Right, I agree with that. So these are all route mode devices the tunnels are connected to? In which case why are you using 0/0 for the P2s?

[OhYeah 0]

@stephenw10 said in VTI gateways in 24.03:

So these are all route mode devices the tunnels are connected to? In which case why are you using 0/0 for the P2s?

Yes, all the spokes are connected to the hub via 0/0. Except for end-user remote access VPN which is a separate virtual network and then routed to the hub via parent router LAN/IPSEC (Fortinet because it offers 365/Entra integration).

As to why use 0/0 for P2s... tried it out with pfsense and a couple of ISPs/partners and found out it works incredibly well across multiple platforms.

If that mode of VPN setup is suddenly not supported anymore, I would like to hear the reasoning behind this change. At the moment it sounds more like a bug. :)

[stephenw10]

Hmm, curious. The only time I've ever seen that is when one side of the tunnel is using policy mode. Otherwise having a local interface defined as 0/0 could potentially break routing entirely.

However I'm not aware of any specific change in 24.03 that would prevent it if it worked in 23.09. It's unlikely a setup like that was ever tested though. Let me see what I can find....

[OhYeah 0]

@stephenw10 said in VTI gateways in 24.03:

However I'm not aware of any specific change in 24.03 that would prevent it if it worked in 23.09. It's unlikely a setup like that was ever tested though.

I can provide also some logs/data from routers that are running 23.09, if it would help to figure out what actually changed.

[Nikkeli]

I'm also having problems with static routes not being loaded on boot.
However they get loaded after editing and saving routes (without changes), after which the tunnel works as intended.

I have IPsec VTI with local/remote networks set to "address".
Issue appeared after upgrade from 23.09.1 with no changes to configuration between upgrades.

I can post more information if needed.

[OhYeah 0]

Maybe it's also a good idea to change the title of the topic to include the phrase "static routes"?

[LarryFahnoe]

@Nikkeli Your situation sounds a lot like mine.

Might be interesting to take a peek at your /cf/conf/config.xml and compare it to what I showed above in https://forum.netgate.com/post/1170175

Do you have a spurious <gateway_item> with a <gateway> containing an address rather than "dynamic"?

I have on my "spare time list" (ahem!) to roll back to 23.09.1, then do the upgrade again and document how the config changes. I suspect there is a bug in the upgrade process.

@stephenw10 I'd vote for adding "static routes" to the title of this thread if possible.

--Larry

[Nikkeli]

@LarryFahnoe
I actually don't have this problem, the configuration seems fine. Below is the configuration for the only (vti) gateway listed.

<gateway_item>
<interface>opt10</interface>
<gateway></gateway>
<name>IPSEC_VT13_VT10_VTIV4</name>
<weight>1</weight>
<ipprotocol>inet</ipprotocol>
<descr><![CDATA[Interface IPSEC_VT13_VT10_VTIV4 Gateway]]></descr>
<gw_down_kill_states></gw_down_kill_states>
</gateway_item>

[stephenw10]

So no additional gateways? No disabled gateways?

[Nikkeli]

@stephenw10
The only other gateway is WAN gateway. No gateways are disabled.

[stephenw10]

Hmm, any errors in the routing or system logs at boot?

[Nikkeli]

@stephenw10
On System/General I can actually see some errors/warnings that seem to be relevant. On other logs I could not find anything relevant.
IPsec logging has too much log noise but I can turn that down aswell and reboot, if you think it could help.

Here is System/General logging after booting, with the relevant lines.

May 24 10:11:27 	php-cgi 	685 	rc.bootup: route_add_or_change: Invalid gateway and/or network interface ipsec1
May 24 10:11:27 	php-cgi 	685 	rc.bootup: route_add_or_change: Invalid gateway and/or network interface ipsec1
May 24 10:11:27 	php-cgi 	685 	rc.bootup: route_add_or_change: Invalid gateway and/or network interface ipsec1
May 24 10:11:27 	php-cgi 	685 	rc.bootup: Gateway, NONE AVAILABLE
May 24 10:11:27 	syslogd 		kernel boot file is /boot/kernel/kernel
May 24 10:11:27 	syslogd 		exiting on signal 15
May 24 10:11:26 	kernel 		done.
May 24 10:11:26 	php-cgi 	685 	rc.bootup: Creating rrd update script
May 24 10:11:24 	kernel 		.done.
May 24 10:11:24 	check_reload_status 	650 	Restarting IPsec tunnels
May 24 10:11:24 	kernel 		...
May 24 10:11:15 	kernel 		done.
May 24 10:11:15 	check_reload_status 	650 	Updating all dyndns
May 24 10:11:14 	kernel 		done.
May 24 10:11:14 	php-cgi 	685 	rc.bootup: NTPD is starting up.
May 24 10:11:08 	kernel 		done.
May 24 10:11:08 	kernel 		done.
May 24 10:11:08 	php-cgi 	685 	rc.bootup: sync unbound done.
May 24 10:11:07 	kernel 		done.
May 24 10:11:07 	php-cgi 	685 	rc.bootup: route_add_or_change: Invalid gateway and/or network interface ipsec1
May 24 10:11:07 	php-cgi 	685 	rc.bootup: route_add_or_change: Invalid gateway and/or network interface ipsec1
May 24 10:11:07 	php-cgi 	685 	rc.bootup: route_add_or_change: Invalid gateway and/or network interface ipsec1
May 24 10:11:07 	php-cgi 	685 	rc.bootup: Gateway, NONE AVAILABLE
May 24 10:11:07 	php-cgi 	685 	rc.bootup: Default gateway setting as default.

[OhYeah 0]

Rebooted device, went through the logs to see if I catch something that might be relevant (Netgate 4100).

May 24 13:53:46	php-cgi	678	rc.bootup: The command '/sbin/ifconfig 'ipsec1' inet '0.0.0.0/0' '0.0.0.0'' returned exit code '1', the output was 'ifconfig: ioctl (SIOCAIFADDR): Destination address required'
May 24 13:53:46	php-cgi	678	rc.bootup: Gateway, NONE AVAILABLE
May 24 13:53:46	php-cgi	678	rc.bootup: Gateway, NONE AVAILABLE
May 24 13:53:46	kernel		route: message indicates error: Invalid argument

[stephenw10]

Ah, there we go. Yup that's pretty much what I'd expect when trying to use 0/0. It tries to apply it to the interfaces and fails because it's invalid there.

The interesting thing is how that ever worked in 23.09.

[OhYeah 0]

And these are similar messages from a Netgate 4100 running 23.09:

May 24 19:26:59	php-cgi	466	rc.bootup: The command '/sbin/ifconfig 'ipsec2' inet '0.0.0.0/0' '0.0.0.0/0'' returned exit code '1', the output was 'ifconfig: 0.0.0.0/0: bad value'
May 24 19:26:59	php-cgi	466	rc.bootup: Gateway, NONE AVAILABLE

The message is very slightly different, so I assume it must be meaningful in some way.

I also got offered 24.03_1 on the same device but no release notes yet?