Hello, I have a server with multiple IPv4 routes to the internet (multipath over tunnels, plus the default route). The multipath route is picked for outgoing connections based on the IP the client socket is bound to: ip rule from <ip> lookup <table> ip route add table <table> default nexthop via ... [nexthop via ... [...]]
I have setup postfix to bind to <ip> for some emails that should appear as having entered the net from the ends of the multipath route (because these IPs are in SPF records). I did a mistake and accidentally removed <ip> from that system. Postfix complains (rightly) in the logs that it cannot bind to this address: postfix/smtp: warning: smtp_connect_addr: bind <ip>: Cannot assign requested address But then, it proceeded to send the email anyway, so it entered the internet from the main default route. This made some recipients' SMTP servers unhappy, which delayed some emails, but others continued to accept the mails (and could have done all sort of bad things to them, like file them in a spam box, ...). The custom postfix settings I have for this setup are: smtp_address_preference = ipv4 smtp_bind_address = <ip> I am on postfix 3.4.14 (Debian oldstable). I would rather postfix just stop sending emails altogether in such case, than send them from an unexpected ip: a delay is preferable to me to uncertainty as to how the emails were processed by recipient SMTPs. Is there something else I should set so postfix stops falling back to no binding ? Regards, -- Vincent Pelletier
