Hello!

I'd like to propose a change to vps-functions, to allow for more
complex routing setups (with multiple VLANs bound on VE0, etc.).

The change would modify vzaddrouting and vzdelrouting to always add
the VE0 source routing to the "local" table. This way, all routing
decisions regarding _local_ VEs will always be done at the very top
in the routing stack.
Therefore you can do other routing decisions, which would affect the
reachability of the local VEs lower in the routing stack, without
affecting the local VEs.
Now this all sounds very complicated, but the patch is very simple, 
and it should not affect "normal" setups.


I'm attaching the patch which we are currently running in production
on 5 HNs.

Everything tested with IPv4 only, though; I'm also not so sure that 
modifying the "local" table is the best choice -- OTOH the VEs are 
local to the HN.

Because of the iproute table usage, the kernel needs to have
'Advanced Routing' set, but I'd think the OpenVZ kernels have this
on / this is not a new requirement.


  - Christian





------- example setup & further explanations -------

Example setup (done on a Debian etch host, vzctl 3.0.22, 
 kernel 2.6.18-028stab053, custom config):

VE0 has got multiple VLAN devices:
  eth0.110 -> 10.10.110.62/24 (this is used for management of VE0)
  eth0.150 -> 10.10.150.249/24 (used for VEs)
  eth0.152 -> 10.10.152.249/24 (used for VEs)

Please note that VLAN150 + 152 are not dedicated to this HN, other
nodes also run VEs in these VLANs.
The VLANs are connected together by a single router, which does
strict source IP filtering (i.e. packets from 10.10.110.0/24 are not
allowed to come from VLAN110).

Main routing table on HN looks like this:
Destination     Gateway       Iface
10.10.152.0     0.0.0.0       eth0.152
10.10.150.0     0.0.0.0       eth0.150
10.10.110.0     0.0.0.0       eth0.110
0.0.0.0         10.10.110.1   eth0.110

Routing rules on HN:
# ip rule ls
0:      from all lookup 255
32763:  from 10.10.152.0/24 lookup 152
32764:  from 10.10.150.0/24 lookup 150
32765:  from 10.10.110.0/24 lookup 110
32766:  from all lookup main
32767:  from all lookup default

# ip route ls table 150
10.10.150.0/24 dev eth0.150  scope link
default via 10.10.150.1 dev eth0.150


Example VE2:
cat /etc/vz/conf/2.conf | grep IP_
IP_ADDRESS="10.10.150.244"


On VE2 startup, with the original vps-functions, source routes will
be configured in the "main" routing table. The "main" routing table
will not be considered in this setup, because table 150 will be
used, which already contains a (correct) default gateway. This also
implies that Proxy ARP requests for VE2 will not be handled, because
the kernel does not find the IP address of VE2 in its routing table.


With the patched vps-functions, the source route will be added to
the local table instead, and Proxy ARP requests can be handled,
because the kernel will see the IP address of VE2. The rules for
10.10.150.0/24 will be ignored during Proxy ARP (lookup can be
fulfilled already in the "local" table), but outgoing packets will
still use the rules for 10.10.150.0/24.

------- end of example -------

-- 
christian hofstaedtler

--- vps-functions	2008-03-05 15:42:02.000000000 +0100
+++ vps-functions	2008-03-05 16:30:03.000000000 +0100
@@ -193,14 +193,14 @@
 			vzerror "Unable to get source ip [${VE_ROUTE_SRC_DEV}]" $VZ_CANT_ADDIP
 		src_addr="src $src_addr"
 	fi
-	${IP_CMD} route add "$1" dev venet0 $src_addr ||
-		vzerror "Unable to add route ${IP_CMD} route add $1 dev venet0 $src_addr" $VZ_CANT_ADDIP
+	${IP_CMD} route add "$1" dev venet0 $src_addr table local ||
+		vzerror "Unable to add route ${IP_CMD} route add $1 dev venet0 $src_addr table local" $VZ_CANT_ADDIP
 }
 
 vzaddrouting6()
 {
-	${IP_CMD} route add "$1" dev venet0 ||
-		vzerror "Unable to add route ${IP_CMD} route add $1 dev venet0" $VZ_CANT_ADDIP
+	${IP_CMD} route add "$1" dev venet0 table local ||
+		vzerror "Unable to add route ${IP_CMD} route add $1 dev venet0 table local" $VZ_CANT_ADDIP
 }
 
 # Sets VE0 source routing for given IP
@@ -228,9 +228,9 @@
 	local arg
 
 	if [ "${1%%:*}" = "$1" ]; then
-		arg="route del $1 dev venet0"
+		arg="route del $1 dev venet0 table local"
 	else
-		arg="-6 route flush $1 dev venet0"
+		arg="-6 route flush $1 dev venet0 table local"
 	fi
 	${IP_CMD} $arg ||
 		vzwarning "vzdelrouting: ${IP_CMD} $arg failed"