[lvs-users] CentOS/RHEL 6 ipvs issue with DR (different interfaces and networks)
u2sashko at gmail.com
Fri Oct 18 13:39:20 BST 2013
On Fri, Oct 18, 2013 at 4:07 AM, David Coulson <david at davidcoulson.net>wrote:
> On 10/18/13 12:50 AM, Alex Attarian wrote:
> > You need to change the settings in order for IPVS DR to work if you have
> > private and public networks or different networks on different
> > Those settings are:
> > echo 2 > /proc/sys/net/ipv4/conf/default/rp_filter
> > echo 2 > /proc/sys/net/ipv4/conf/all/rp_filter
> > or if you would like to just do the interface where loadbalancer is
> > traffic to, then:
> > echo 2 > /proc/sys/net/ipv4/conf/<interface of real server ip>/rp_filter
> > (in my case eth1, so echo 2 > /proc/sys/net/ipv4/conf/eth1/rp_filter)
> We have a large LVS environment with RHEL6 for both balancers and real
> servers. On most real servers we have rp_filter set to 1. Our network
> config is pretty much the same as yours (VIPs on lo, usually from a
> different subnet).
> The only systems where we do have rp_filter set to 0 (I need to back and
> check why we picked 0 instead of 2 - 0 does work however) are systems
> which get packets from balancers using IPIP tunnels.
> > Please note value of 1 does not have the same behavior on RHEL6 as it did
> > on RHEL5.
> Did you open a RedHat case on this behavior? Did they give you a BZ for it?
> Not saying changing rp_filter wasn't the right thing to do, but I've not
> experienced a difference in how it functions in the last 12 years,
> including with RHEL6. On the few RHEL5 systems I have default rp_filter
> value is 1, so I'd expect them to behave the same.
> Please read the documentation before posting - it's available at:
> LinuxVirtualServer.org mailing list - lvs-users at LinuxVirtualServer.org
> Send requests to lvs-users-request at LinuxVirtualServer.org
> or go to http://lists.graemef.net/mailman/listinfo/lvs-users
yes both values 0 and 2 work, however 0 does no source checking at all, 2
does a loose checking:
2 = basically if any interface can actually route any of the networks the
packet is coming in from, then allow.
rp_filter - INTEGER
0 - No source validation.
1 - Strict mode as defined in RFC3704 Strict Reverse Path
Each incoming packet is tested against the FIB and if the
is not the best reverse path the packet check will fail.
By default failed packets are discarded.
2 - Loose mode as defined in RFC3704 Loose Reverse Path
Each incoming packet's source address is also tested against
and if the source address is not reachable via any interface
the packet check will fail.
Yes, I actually found it on RHEL solutions site as well:
More information about the lvs-users