The lowdown: Repeat situation description: - 2 uplinks with own public address space - 1 internal network with private address space Goal: do DNAT through address in both public ranges, converging on internal hosts: ExtIP1 > IntIP1 ExtIP2 > IntIP1 Everything is done in iptables, nothing was changed in the standard route/rules setup. Trick used: iptables -t nat -A PREROUTING -i $ExtNet1 -p tcp -d $ExtAddress1InternalHost \ -j DNAT --to-destination $IntAddressInternalHost iptables -t nat -A PREROUTING -i $ExtNet2 -p tcp -d $ExtAddress2InternalHost \ -j DNAT --to-destination $IntAddressInternalHost This part is easy and works. The return path is the problem. Solution: iptables -t mangel -A POSTROUTING -s $IntNet/IntMask -m conntrack \ --ctorigdst $ExtAddressRange1/ExtAddressRange1Mask \ --ctstate ESTABLISHED,RELATED -j ROUTE --gw GWExtNet1 --oif IFExtNet1 --continue iptables -t mangel -A POSTROUTING -s $IntNet/IntMask -m conntrack \ --ctorigdst $ExtAddressRange2/ExtAddressRange2Mask \ --ctstate ESTABLISHED,RELATED -j ROUTE --gw GWExtNet2 --oif IFExtNet2 --continue Caveats: To get routing working, you _must_ have Equal-cost-multipath routing enabled in the kernel and you must set up a multipath route over both links. If someone can explain to me why I would be very thankfull. Simply adding another default route with different metric won't work. You need a recent p-o-m and iptables to use the --continue. You _need_ the --continue or UnDNATting will not be performed (you jump out on the -j ROUTE). If you want even more flexibility, you can use the other suggested route: instead of -j ROUTE, use -j MARK x where x is different for each uplink to match. You can then proceed to use iproute2+tc and 'rules' based on fwmark to do even more funky stuff. Thanks for all the help and suggestions, I hope this helps some other poor schmuck who is at a loss. Gaby Schilders IBFD network admin
