Jo Rhett wrote: > Magnus Holmgren wrote: >> A list search for ALL_TRUSTED would have given you tons of hits. You >> could also have gone to the FAQ page and from there to the >> FixingErrors wiki page, where you'd find a reference to ALL_TRUSTED. > > Magnus, to be fair - the search will tell you that autodetection > should work unless you are behind a NAT. So a person who believes > that without testing won't realize that they're looking at the problem. > > The autodetection is totally broken actually, and needs to be fixed.
How do you propose it be fixed? This has been brought up a few dozen times, and really it boils down to breaking people with NATed MX servers (as it is now), or breaking people without NATed MX servers but with NATed internal mailservers. You can't have both work. The autodetection is broken, but it is fundamentally unfixable. You cannot fix it, you can only change between two different kinds of broken. The fundamental problem is if you work backwards in time through the headers in time, if you start off with private addresses, what do you do with the first non-private? If you trust it, as SA does now, you'll work properly for networks with a non-NAT MX server, but you'll over-trust for networks that NAT their MX. If you don't trust it, it will work for networks that NAT their MX, but you'll under-trust for those with non-NAT MXes. And under-trust is NOT better than over-trust. Both cause severe breakage in SA's accuracy. ie: under-trust will break the DUL-RBL tests, just like over-trust, but rather than causing properly relayed mail to hit a DUL it causes direct-delivered mail to miss. It also breaks whitelist_from_rcvd, SPF, and ALL_TRUSTED in various ways. Both scenarios break roughly the same number of sites, and break them for the same group of tests, but with slightly different cases causing broken behavior.
