Hey, Frédéric Massot wrote: > > These 3 ISP are different bandwith, 2 ADSL (8192/256 and 1024/256) and 1 > leased line (512/512), sometimes DSL lines are down. > > DSL1 on eth1 with IP 192.168.0.1 > DSL2 on eth0 with IP 192.168.1.1 > LL on eth2 with IP 192.168.2.1 > > I use a kernel 2.6.17 with a routes patch from Julian Anastasov. > > The leased line (eth2) is used for the sending of the emails. The > reception of the emails is done on the three external connections. The > access to the Web is preferably routed towards ADSL lines. > > For a few days the sending of emails has been done intermittently. While > using tcptraceroute I realized that sometimes the IP source is badly > selected.
Which machine sends email? The router itself, or a mailserver that's behind the router? An alternative to using iproute and fwmarks is to specify the address to bind when opening a smtp connection to another server. E.g. I use postfix on a host with multiple IP's, and specify the address touse for outgoing mail in master.cf like: smtp unix - - - - - smtp -o smtp_bind_address=a.b.c.d When sending server is placed behind the router... routing should do the trick... I use the same setup here in the building, to send mail via a dsl line that's not the default outgoing route. > # LEASED LINE > ip rule add prio 203 from $NETWORK_LL/$PREFIX_LL table 203 > ip route add default via $GATEWAY_LL dev $INTERFACE_LL src $IP_LL proto > static table 203 > ip route append prohibit default table 203 metric 1 proto static The src $IP_LL is specified in the ip route add line for table 203. So the source address for the packet must be known and match $IP_LL, otherwise this route will be ignored. I seems you used this nice howto: http://www.ssi.bg/~ja/nano.txt Read section '2.3.3 Setup of the default routes' again, carefully: "The tricky part is, what the patches are doing here: They look at the local source address" etc... is what I mean. > # SMTP MARK > ip rule add prio 103 fwmark 0x25 table 203 So table 203 gets looked up, but the default route to $INTERFACE_LL won't get used. Instead one route of the multipath default route will be chosen. > ip rule add prio 222 table 222 > ip route add default table 222 proto static \ > nexthop via $GATEWAY_DSL1 dev $INTERFACE_DSL1 weight 10 \ > nexthop via $GATEWAY_DSL2 dev $INTERFACE_DSL2 weight 10 \ > nexthop via $GATEWAY_LL dev $INTERFACE_LL weight 1 I solved this problem by creating two extra routing tables. By directing traffic which routing is forced to an outgoing line to a routing table without the src address specified it matches a connectino that get initiated. I put some parts of my scripts online. (I don't promise they will be there forever) In this example I use a DSL line (xs4all) and a PPPoE connection (university campus network). Outgoing SMTP traffic from behind the router will be forwarded to the DSL line. http://od11.fttd-s.tudelft.nl/~knorrie/iproute/ (It has some dutch comments in it.) Also look at the use of -j CONNMARK --save-mark and --restore-mark for NAT traffic, using those can prevent headaches. HTH, Have fun, Hans van Kranenburg -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
