static default route not always pushed to kernel (2.0.4)

[Kenth Eriksson]

On Mon, 2019-04-29 at 13:01 +0000, Kenth Eriksson wrote:
> On Mon, 2019-04-29 at 14:46 +0200, Ondrej Zajicek wrote:
> > CAUTION: This email originated from outside of the organization. Do
> > not click links or open attachments unless you recognize the sender
> > and know the content is safe.
> > 
> > 
> > On Mon, Apr 29, 2019 at 12:14:56PM +0000, Kenth Eriksson wrote:
> > > Hi!
> > > 
> > > There are cases where bird does not push a static default route
> > > to
> > > the
> > > kernel. One way to see this is as follows;
> > > 
> > > Next remove both static routes, configure, add one of them back
> > > again
> > > and both are shown.
> > > 
> > > bird> show route for 0.0.0.0
> > > Table master4:
> > > 0.0.0.0/0            unicast [static1 14:13:16.762] * (255)
> > >       via 10.210.137.1 on eth1
> > >                      unicast [kernel1 13:46:07.958] (215)
> > >       via 10.210.137.1 on eth1
> > 
> > Hi
> > 
> > Yes, BIRD on Linux intentionally avoids replacing any existing non-
> > BIRD
> > (alien) routes in kernel tables.
> > 
> > The reason is: Consider route A from BIRD and alien route B, both
> > for
> > the
> > same network. A is preferred in BIRD and therefore pushed to
> > kernel,
> > where it replaces route B. Therefore, route B is overwritten and
> > disappears. Then route A is for some reason removed, but route B no
> > longer exists, so it cannot be restored and we end with no route.
> > 
> > There are two ways how to fix that: one way is that kernel would
> > remember
> > all alien routes it learned, even when they were replaced by BIRD
> > route.
> > This would solve the problem above, but has other issues (e.g. the
> > original source of alien route may want update or remove it, but it
> > is
> > no longer in kernel table).
> > 
> > The approach we use is that we expect to have dedicated kernel
> > metric
> > value (by default 32) that is not used by alien routes (as kernel
> > table
> > keeps multiple routes with different kernel metric). This works
> > well
> > with IPv6, where default kernel metric used by alien routes is
> > higher
> > (256 or 1024) but in IPv4 the default kernel metric is 0 (most
> > preferred),
> > so we cannot override such routes without replacing them. The
> > answer
> > is
> > to create alien routes with higher kernel metric (easy when created
> > with
> > 'ip' tool, perhaps harder in other cases). It would be great if
> > there
> > existed sysctl option for default IPv4 route metric.
> > 
> 
> There is no overwrite involed here. The default route in the kernel
> here has metric 100. As you said, bird pushes with metric 32 so it
> should push again. 
> 
> kenth ~ # ip route show
> default via 10.210.137.1 dev eth1  proto bird  metric 32 
> default via 10.210.137.1 dev eth1  metric 100 
> 

There is an even simpler way to reproduce this bug;

Add two static default routes that differs only on metric (no alien
kernel route this time)

route 0.0.0.0/0 via 10.210.137.1 {preference=255;};
route 0.0.0.0/0 via 10.210.137.1 {preference=254;};

bird> show static
0.0.0.0/0
        via 10.210.137.1
0.0.0.0/0
        via 10.210.137.1
bird>

Next note how it select the route with lower preference value. That is
not correct according to the docs.

bird> show route for 0.0.0.0
Table master4:
0.0.0.0/0            unicast [static1 15:49:16.285] * (254)
        via 10.210.137.1 on eth1
bird>

Remove one of the static routes. Now the route disappears from the RIB
and it is also removed from the FIB.

bird> show static
0.0.0.0/0
        via 10.210.137.1
bird> show route for 0.0.0.0
Network not found
bird>

> > --
> > Elen sila lumenn' omentielvo
> > 
> > Ondrej 'Santiago' Zajicek (email: santiago at crfreenet.org)
> > OpenPGP encrypted e-mails preferred (KeyID 0x11DEADC3,
> > wwwkeys.pgp.net)
> > "To err is human -- to blame it on a computer is even more so."