The Postfix SMTP server receives mail from the network and is exposed to the big bad world of junk email and viruses. This document introduces the built-in and external methods that control what SMTP mail Postfix will accept, what mistakes to avoid, and how to test your configuration.
Topics covered in this document:
In a distant past, the Internet was a friendly environment. Mail servers happily forwarded mail on behalf of anyone towards any destination. On today's Internet, spammers abuse servers that forward mail from arbitrary systems, and abused systems end up on anti-spammer blacklists. See, for example, the information on http://www.mail-abuse.org/ and other websites.
By default, Postfix has a moderately restrictive approach to mail relaying. Postfix forwards mail only from clients in trusted networks, or to domains that are configured as authorized relay destinations. For a description of the default policy, see the smtpd_recipient_restrictions parameter in the postconf(5) manual page, and the information that is referenced from there.
Most of the Postfix SMTP server access controls are targeted at stopping junk email.
Protocol oriented: some SMTP server access controls block mail by being very strict with respect to the SMTP protocol; these catch poorly implemented and/or poorly configured junk email software, as well as email worms that come with their own non-standard SMTP client implementations. Protocol-oriented access controls become less useful over time as spammers and worm writers learn to read RFC documents.
Blacklist oriented: some SMTP server access controls query blacklists with known to be bad sites such as open mail relays, open web proxies, and home computers that have been compromised and that are under remote control by criminals. The effectiveness of these blacklists depends on how complete and how up to date they are.
Threshold oriented: some SMTP server access controls attempt to raise the bar by either making the client do more work (greylisting) or by asking for a second opinion (SPF and sender/recipient address verification). The greylisting and SPF policies are implemented externally, and are the subject of the SMTPD_POLICY_README document. Sender/recipient address verification is the subject of the ADDRESS_VERIFICATION_README document.
Unfortunately, all junk mail controls have the possibility of falsely rejecting legitimate mail. This can be a problem for sites with many different types of users. For some users it is unacceptable when any junk email slips through, while for other users the world comes to an end when a single legitimate email message is blocked. Because there is no single policy that is "right" for all users, Postfix supports different SMTP access restrictions for different users. This is described in the RESTRICTION_CLASS_README document.
Besides the restrictions that can be made configurable per client or per user as described in the next section, Postfix implements a few restrictions that apply to all SMTP mail.
The built-in header_checks and body_checks content restrictions, as described in the BUILTIN_FILTER_README document. This happens while Postfix receives mail, before it is stored in the incoming queue.
Requiring that the client sends the HELO or EHLO command before sending the MAIL FROM or ETRN command. This may cause problems with home-grown applications that send mail. For this reason, the requirement is disabled by default ("smtpd_helo_required = no").
Disallowing illegal syntax in MAIL FROM or RCPT TO commands. This may cause problems with home-grown applications that send mail, and with ancient PC mail clients. For this reason, the requirement is disabled by default ("strict_rfc821_envelopes = no").
Disallowing RFC 822 address syntax (example: "MAIL FROM: the dude <firstname.lastname@example.org>").
Disallowing addresses that are not enclosed with <> (example: "MAIL FROM: email@example.com").
Rejecting mail from a non-existent sender address. This form of egress filtering helps to slow down worms and other malware, but may cause problems with home-grown software that sends out mail software with an unreplyable address. For this reason the requirement is disabled by default ("smtpd_reject_unlisted_sender = no").
Rejecting mail for a non-existent recipient address. This form of ingress filtering helps to keep the mail queue free of undeliverable MAILER-DAEMON messages. This requirement is enabled by default ("smtpd_reject_unlisted_recipient = yes").
Postfix allows you to specify lists of access restrictions for each stage of the SMTP conversation. Individual restrictions are described in the postconf(5) manual page.
Examples of simple restriction lists are:
/etc/postfix/main.cf: # Allow connections from trusted networks only. smtpd_client_restrictions = permit_mynetworks, reject # Don't talk to mail systems that don't know their own hostname. # With Postfix < 2.3, specify reject_unknown_hostname. smtpd_helo_restrictions = reject_unknown_helo_hostname # Don't accept mail from domains that don't exist. smtpd_sender_restrictions = reject_unknown_sender_domain # Whitelisting: local clients may specify any destination domain. smtpd_recipient_restrictions = permit_mynetworks, reject_unauth_destination # Block clients that speak too early. smtpd_data_restrictions = reject_unauth_pipelining # Enforce mail volume quota via policy service callouts. smtpd_end_of_data_restrictions = check_policy_service unix:private/policy
Each restriction list is evaluated from left to right until some restriction produces a result of PERMIT, REJECT or DEFER (try again later). The end of the list is equivalent to a PERMIT result. By placing a PERMIT restriction before a REJECT restriction you can make exceptions for specific clients or users. This is called whitelisting; the last example above allows mail from local networks but otherwise rejects mail to arbitrary destinations.
The table below summarizes the purpose of each SMTP access restriction list. All lists use the exact same syntax; they differ only in the time of evaluation and in the effect of a REJECT or DEFER result.
Restriction list name Status Effect of REJECT or DEFER result smtpd_client_restrictions Optional Reject all client commands smtpd_helo_restrictions Optional Reject HELO/EHLO information smtpd_sender_restrictions Optional Reject MAIL FROM information smtpd_recipient_restrictions Required Reject RCPT TO information smtpd_data_restrictions Optional Reject DATA command smtpd_end_of_data_restrictions Optional Reject END-OF-DATA command smtpd_etrn_restrictions Optional Reject ETRN command
Early Postfix versions evaluated SMTP access restrictions lists as early as possible. The client restriction list was evaluated before Postfix sent the "220 $myhostname..." greeting banner to the SMTP client, the helo restriction list was evaluated before Postfix replied to the HELO (EHLO) command, the sender restriction list was evaluated before Postfix replied to the MAIL FROM command, and so on. This approach turned out to be difficult to use.
Current Postfix versions postpone the evaluation of client, helo and sender restriction lists until the RCPT TO or ETRN command. This behavior is controlled by the smtpd_delay_reject parameter. Restriction lists are still evaluated in the proper order of (client, helo, etrn) or (client, helo, sender, recipient, data, or end-of-data) restrictions. When a restriction list (example: client) evaluates to REJECT or DEFER the other restriction lists (example: helo, sender, etc.) are skipped.
Around the time that smtpd_delay_reject was introduced, Postfix was also changed to support mixed restriction lists that combine information about the client, helo, sender and recipient or etrn command.
Benefits of delayed restriction evaluation, and of restriction mixing:
Some SMTP clients do not expect a negative reply early in the SMTP session. When the bad news is postponed until the RCPT TO reply, the client goes away as it is supposed to, instead of hanging around until a timeout happens, or worse, going into an endless connect-reject-connect loop.
Postfix can log more useful information. For example, when Postfix rejects a client name or address and delays the action until the RCPT TO command, it can log the sender and the recipient address. This is more useful than logging only the client hostname and IP address and not knowing whose mail was being blocked.
Mixing is needed for complex whitelisting policies. For example, in order to reject local sender addresses in mail from non-local clients, you need to be able to mix restrictions on client information with restrictions on sender information in the same restriction list. Without this ability, many per-user access restrictions would be impossible to express.
By now the reader may wonder why we need smtpd client, helo or sender restrictions, when their evaluation is postponed until the RCPT TO or ETRN command. Some people recommend placing ALL the access restrictions in the smtpd_recipient_restrictions list. Unfortunately, this can result in too permissive access. How is this possible?
The purpose of the smtpd_recipient_restrictions feature is to control how Postfix replies to the RCPT TO command. If the restriction list evaluates to REJECT or DEFER, the recipient address is rejected; no surprises here. If the result is PERMIT, then the recipient address is accepted. And this is where surprises can happen.
Here is an example that shows when a PERMIT result can result in too much access permission:
1 /etc/postfix/main.cf: 2 smtpd_recipient_restrictions = 3 permit_mynetworks 4 check_helo_access hash:/etc/postfix/helo_access 5 reject_unknown_helo_hostname 6 reject_unauth_destination 7 8 /etc/postfix/helo_access: 9 localhost.localdomain PERMIT
Line 5 rejects mail from hosts that don't specify a proper hostname in the HELO command (with Postfix < 2.3, specify reject_unknown_hostname). Lines 4 and 9 make an exception to allow mail from some machine that announces itself with "HELO localhost.localdomain".
The problem with this configuration is that smtpd_recipient_restrictions evaluates to PERMIT for EVERY host that announces itself as "localhost.localdomain", making Postfix an open relay for all such hosts.
In order to avoid surprises like these with smtpd_recipient_restrictions, you should place non-recipient restrictions AFTER the reject_unauth_destination restriction, not before. In the above example, the HELO based restrictions should be placed AFTER reject_unauth_destination, or better, the HELO based restrictions should be placed under smtpd_helo_restrictions where they can do no harm.
Postfix has several features that aid in SMTP access rule testing:
This is a safety net that changes SMTP server REJECT actions into DEFER (try again later) actions. This keeps mail queued that would otherwise be returned to the sender. Specify "soft_bounce = yes" in the main.cf file to prevent the Postfix SMTP server from rejecting mail permanently, by changing all 5xx SMTP reply codes into 4xx.
This is a different safety net that changes SMTP server REJECT actions into warnings. Instead of rejecting a command, Postfix logs what it would reject. Specify "warn_if_reject" in an SMTP access restriction list, before the restriction that you want to test without actually rejecting mail.
With this Postfix 2.1 feature, authorized SMTP clients can impersonate other systems, so that you can do realistic SMTP access rule tests. Examples of how to impersonate other systems for access rule testing are given at the end of the XCLIENT_README document.