INTERNET-DRAFT Hadmut Danisch Category: Experimental Jun 2003 Expires: Dec 1, 2003 A DNS RR for simple SMTP sender authentication draft-danisch-dns-rr-smtp-02.txt Status of this Memo This document is an Internet-Draft and is subject to all provisions of Section 10 of RFC2026. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet- Drafts. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." The list of current Internet-Drafts can be accessed at http://www.ietf.org/1id-abstracts.html The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html Abstract This memo suggests a new DNS RR type to provide a very simple light-weight authentication scheme for the domain part of the sender address for SMTP mail transport. It is designed to be a simple and robust protection against e-mail fraud and especially spam. It is easy to implement and administer, compatible with all legislations known by the author, and cheap. It also focuses on security and privacy problems and requirements in context of spam defense, and touches non-technical problems of defense against spam. Hadmut Danisch Experimental [Page 1] INTERNET-DRAFT DNS RMX RR Jun 2003 Table of Contents 1. Threat description . . . . . . . . . . . . . . . . . . . . . . . 4 1.1. Mail sender forgery . . . . . . . . . . . . . . . . . . . 4 1.2. Spam . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 2. Problem analysis . . . . . . . . . . . . . . . . . . . . . . . . 4 3. Shortcomings of strong authentication mechanisms . . . . . . . . 4 4. DNS based sender address verification . . . . . . . . . . . . . 5 5. RMX entry types . . . . . . . . . . . . . . . . . . . . . . . . 5 5.1. Overall structure . . . . . . . . . . . . . . . . . . . . 5 5.1.1 Description . . . . . . . . . . . . . . . . . . . . 6 5.1.2 Zone File Syntax . . . . . . . . . . . . . . . . . 6 5.1.3 Encoding . . . . . . . . . . . . . . . . . . . . . 6 5.2. IPv4 and IPv6 address ranges (Type1/2) . . . . . . . . . . 7 5.2.1 Description . . . . . . . . . . . . . . . . . . . . 7 5.2.2 Zone File Syntax . . . . . . . . . . . . . . . . . 7 5.2.3 Encoding . . . . . . . . . . . . . . . . . . . . . 7 5.3. Hostname (Type 3) . . . . . . . . . . . . . . . . . . . . 7 5.3.1 Description . . . . . . . . . . . . . . . . . . . . 7 5.3.2 Zone File Syntax . . . . . . . . . . . . . . . . . 8 5.3.3 Encoding . . . . . . . . . . . . . . . . . . . . . 8 5.3.4 Additional Records . . . . . . . . . . . . . . . . 8 5.4. APL Reference (Type 4) . . . . . . . . . . . . . . . . . . 8 5.4.1 Description . . . . . . . . . . . . . . . . . . . . 8 5.4.2 Zone File Syntax . . . . . . . . . . . . . . . . . 8 5.4.3 Encoding . . . . . . . . . . . . . . . . . . . . . 9 5.4.4 Additional Records . . . . . . . . . . . . . . . . 9 5.5. Future Entry Types . . . . . . . . . . . . . . . . . . . . 9 6. Message Headers . . . . . . . . . . . . . . . . . . . . . . . . 9 7. Compatibility . . . . . . . . . . . . . . . . . . . . . . . . . 10 7.1. Compatibility with old mail receivers . . . . . . . . . . 10 7.2. Compatibility with old mail senders . . . . . . . . . . . 10 7.3. Compatibility with old DNS clients . . . . . . . . . . . . 10 7.4. Compatibility with old DNS servers . . . . . . . . . . . . 10 8. Message relaying and forwarding . . . . . . . . . . . . . . . . 10 8.1. Problem description . . . . . . . . . . . . . . . . . . . 10 8.2. Trusted relaying/forwarding . . . . . . . . . . . . . . . 11 8.3. Untrusted relaying/forwarding . . . . . . . . . . . . . . 11 9. Enforcement policy . . . . . . . . . . . . . . . . . . . . . . . 12 10. Security Considerations . . . . . . . . . . . . . . . . . . . . 12 10.1. Draft specific considerations . . . . . . . . . . . . . . 12 10.1.1 Authentication strength . . . . . . . . . . . . . 12 10.1.2 Where Authentication and Authorization end . . . . 13 10.1.3 Vulnerability of DNS . . . . . . . . . . . . . . . 13 10.1.4 Additional data attack? . . . . . . . . . . . . . 15 10.1.5 Open SMTP relays . . . . . . . . . . . . . . . . . 15 10.1.6 Unforged Spam . . . . . . . . . . . . . . . . . . 16 Hadmut Danisch Experimental [Page 2] INTERNET-DRAFT DNS RMX RR Jun 2003 10.1.7 Empty Sender address . . . . . . . . . . . . . . . 16 10.1.8 Reliability of Whois Entries . . . . . . . . . . . 16 10.1.9 Hazards for Freedom of Speech . . . . . . . . . . 17 10.2. General Considerations about spam defense . . . . . . . . 17 10.2.1 Action vs. reaction . . . . . . . . . . . . . . . 18 10.2.2 Content based Denial of Service attacks . . . . . 18 11. Privacy Considerations . . . . . . . . . . . . . . . . . . . . 18 11.1. Draft specific considerations . . . . . . . . . . . . . . 18 11.1.1 No content leaking . . . . . . . . . . . . . . . . 19 11.1.2 Message reception and sender domain . . . . . . . 19 11.1.3 Network structure . . . . . . . . . . . . . . . . 19 11.1.4 Owner information distribution . . . . . . . . . . 19 11.2. General Considerations about spam defense . . . . . . . . 20 11.2.1 Content leaking of content filters . . . . . . . . 20 11.2.2 Black- and Whitelists . . . . . . . . . . . . . . 20 12. Deployment Considerations . . . . . . . . . . . . . . . . . . . 20 12.1. The economical problem . . . . . . . . . . . . . . . . . 21 12.2. The POP problem . . . . . . . . . . . . . . . . . . . . . 21 12.3. The network structure problem . . . . . . . . . . . . . . 22 12.4. The mentality problem . . . . . . . . . . . . . . . . . . 22 12.5. The identity problem . . . . . . . . . . . . . . . . . . 23 12.6. The multi-legislation problem . . . . . . . . . . . . . . 23 Implementation and further Information . . . . . . . . . . . . . . . 23 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 Author's Address . . . . . . . . . . . . . . . . . . . . . . . . . . 24 Hadmut Danisch Experimental [Page 3] INTERNET-DRAFT DNS RMX RR Jun 2003 1. Threat description 1.1. Mail sender forgery Security surveillances found a significant increase in the number of e-mails with forged sender addresses. As a consequence, damage caused by such e-mails increased as well. In the majority of examined e-mails the domain name of the envelope sender address was forged, and the e-mail was sent from an IP address which does not belong to a network used by the actual owner of the domain. 1.2. Spam A common and well known problem is the dramatic increase of unsolicited e-mail, commonly called "spam". Again, the majority of examined e-mails had forged sender addresses. The abused domains were mainly those of common webmailers as hotmail or yahoo, or well-known companies. 2. Problem analysis The real cause for forged e-mail is the lack of the Simple Mail Transfer Protocol SMTP[1] to provide any kind of sender authentication, authorisation, or verification. Efforts have been made to block such e-mails by requiring the sender address domain part to be resolvable (e.g. latest sendmail configurations). This method provides protection from e-mails with non-existing sender domains, and indeed, for some time it blocked most spam e-mails. However, since attackers and spam senders began to abuse existing domain names, this method was rendered ineffective. 3. Shortcomings of strong authentication mechanisms Without any doubt, SMTP needs to be improved and reengineered in the future in order to resist against these kinds of threat. Cryptographical or organisational security mechanisms have to be designed an implemented for some sort of SMTPng. Unfortunately, it is impossible to seamlessly switch to a new protocol, since most of the SMTP servers and client programs can't be replaced or updated easily and quickly. Any kind of security mechanism has to be simple and backward compatible in order to be accepted. That's why extensions like SMTP with TLS are not widely spread. Hadmut Danisch Experimental [Page 4] INTERNET-DRAFT DNS RMX RR Jun 2003 4. DNS based sender address verification To gain at least some improvement in e-mail authenticity while keeping as much SMTP compatibility as possible, a method is suggested which doesn't change SMTP at all. When the sender has issued the "MAIL FROM:" SMTP command, the receiving mail transfer agent (MTA) can - and modern MTAs do - perform some authorization checks, e.g. run a local rule database or check whether the sender domain is resolvable. The suggested method is to let the DNS server for the sender domain provide informations about which IP addresses are authorized to use the domain within a sender's address. After receiving the "MAIL FROM:" SMTP command, the receiving MTA can verify, whether the IP address of the sending MTA is authorized to send mails with this domain name. Therefore, a list of authorized IP addresses is provided by the authoritative DNS server of that domain. This version of the draft defines three distinct methods to describe IP address authorizations: - An IPv4 or IPv6 network address and mask - A fully qualified domain name referring to an A record - A fully qualified domain name referring to an APL record RMX records could look like this: rmxtest.de. IN RMX host:relay.provider.com danisch.de. IN RMX apl:relays.rackland.de relays.rackland.de. IN APL 1:213.133.101.23/32 1:1.2.3.0/24 where the machine with the example address 213.133.101.23 and the machines in the example subnet 1.2.3.0/24 are the only machines allowed to send e-mails with an envelope sender address of domain danisch.de. Since the APL records do not necessarily belong to the same domain or zone table as the RMX records, this easily allows to refer to APL records defined by someone else, e.g. the internet access or server hosting provider, thus reducing administrative overhead to a minimum. In the example given above, the domain danisch.de and several other domains are hosted by the service provider Rackland. So if the relay structure of Rackland is modified, only the zone of rackland.de needs to be modified. The domain owners don't need to care about such details. 5. RMX entry types 5.1. Overall structure Hadmut Danisch Experimental [Page 5] INTERNET-DRAFT DNS RMX RR Jun 2003 5.1.1. Description Similar to APL, an RMX record is just a concatenation of zero or more RMX entries. The entries within one record form an ordered rule base, i. e. they are processed one ofter another until the first entry matches. This entry determines the result of the query. Once a matching entry is found, the RMX processing is finished. This method is known from many other rule based security mechanisms. For any domain name there should not exist more than a single RMX entry. Due to the structure of DNS, it is nevertheless possible to have more than a single RMX entry. Multiple RMX entries are treated as a single record consisting of the concatenation of all records. While the entries in a record are ordered, the records are not ordered and may be processed in arbitrary order. If the order of the entries matters, it is the zone maintainer's responsibility to keep those entries in a single record. 5.1.2. Zone File Syntax An RMX record may consist of zero or more entries, where the entries are separated by whitespace. Each entry consists of a tag, determining the entry type, a colon, and the entry data. Example: danisch.de. IN RMX apl:relays.rackland.de ipv4:1.2.3.0/24 5.1.3. Encoding Each entry starts with an octet containting the entry type and the negation flag: +---+---+---+---+---+---+---+---+ | N | Entry Type Code | +---+---+---+---+---+---+---+---+ N If this bit (MSB) is set, an IP address matching this entry is not authorized, but explicitely rejected. See entry type descriptions for details. Entry Type A 7bit number simply determining the entry type. Currently, entries do not have an explicit length field, the entry Hadmut Danisch Experimental [Page 6] INTERNET-DRAFT DNS RMX RR Jun 2003 length is determined implicitely by the entry type. Applications are required to abort if an unknown entry type is found, instead of skipping unknown entries. 5.2. IPv4 and IPv6 address ranges (Type1/2) 5.2.1. Description These entry types (IPv4 = Type 1, IPv6 = Type 2) contain a bit sequence representing a CIDR address part. If that bit sequence matches the given IP address, authorization is granted or denied, depending on the negation flag. 5.2.2. Zone File Syntax The entry is prepended with the tag "IPv4" or "IPv6". The colon is followed with an IPv4 or IPv6 address in standard notation, optionally followed by a slash and a mask length. Examples: danisch.de IN RMX ipv4:213.133.101.23 ipv6:fe00::0 IN RMX ipv4:10.0.0.0/8 ipv6:fec0::0/16 IN RMX !ipv4:1.2.3.4 5.2.3. Encoding After the entry type tag as described above, one octet follows giving the length L of the bit sequence. Then a sequence of exactly as many octets follows as needed to carry L bits of information (= trunc((L+7)/8) ). +---+---+---+---+---+---+---+---+ | N | Entry Type Code (1 or 2) | +---+---+---+---+---+---+---+---+ | Length Field L | +---+---+---+---+---+---+---+---+ | Bit Field | / ((L+7)/8) Octets / +---+---+---+---+---+---+---+---+ 5.3. Hostname (Type 3) 5.3.1. Description This entry type simply contains a regular DNS name, which is to be resolved as a host name (fetch the A record or IPv6 equivalent). If the given IP address matches the result, authorization is granted or denied, depending on the negation flag. Hadmut Danisch Experimental [Page 7] INTERNET-DRAFT DNS RMX RR Jun 2003 It is still to be defined how to treat unresolvable entries. 5.3.2. Zone File Syntax The entry is prepended with the tag "host", followed by a colon and the hostname. Examples: danisch.de IN RMX host:relay.provider.de IN RMX !host:badmachine.domain.de apl:relays.domain.de 5.3.3. Encoding After the entry type tag immediately follows a DNS encoded and compressed [2] domain name. Length is determined by called the decompression function of the resolver library. +---+---+---+---+---+---+---+---+ | N | Entry Type Code (3) | +---+---+---+---+---+---+---+---+ | Encoded and Compressed DNS | / Name as described in RFC1035 / +---+---+---+---+---+---+---+---+ 5.3.4. Additional Records In order to avoid the need of a second query to resolve the given host name, a DNS server should enclose the A record for that domain name in the additional section of the additional section of the DNS reply, if the server happens to be authoritative. 5.4. APL Reference (Type 4) 5.4.1. Description This entry type simply contains a regular DNS name, which is to be resolved as an APL record index (fetch the APL record). If the given IP address positively matches the APL, authorization is granted. Details of the semantic (espially when the negation bit is set) are still to be defined. It is still to be defined how to treat unresolvable entries. 5.4.2. Zone File Syntax The entry is prepended with the tag "host", followed by a colon and the hostname. Examples: danisch.de IN RMX apl:relays.rackland.de Hadmut Danisch Experimental [Page 8] INTERNET-DRAFT DNS RMX RR Jun 2003 5.4.3. Encoding After the entry type tag immediately follows a DNS encoded and compressed domain name. Length is determined by called the decompression function of the resolver library. +---+---+---+---+---+---+---+---+ | N | Entry Type Code (4) | +---+---+---+---+---+---+---+---+ | Encoded and Compressed DNS | / Name as described in RFC1035 / +---+---+---+---+---+---+---+---+ 5.4.4. Additional Records In order to avoid the need of a second query to resolve the given host name, a DNS server should enclose the APL record for that domain name in the additional section of the additional section of the DNS reply, if the server happens to be authoritative. 5.5. Future Entry Types In future, more entry type might be defined, e.g. data used for challenge response authentication, a URL or DNS name ponting to some kind of service for authentication and authorization, entries pointing to other directory services (e.g. LDAP), or public key informations to require signatures on the message header or body. 6. Message Headers An RMX query must be followed by any kind of action depending on the RMX result. One action might be to reject the message. Another action might be to add a header line to the message body, thus allowing MUAs and delivery programs to filter or sort messages. In future, the RMX result might be melted into the Received: header line. The details of such entries are to be discussed. As a proposal the following form is suggested: X-RMX: RESULT addr ADDRESS by HOST on DATE where RESULT is one of "Granted", "Denied", "NotInRMX", "NoRMX", "TempFail", "BadData", "Trusted". Hadmut Danisch Experimental [Page 9] INTERNET-DRAFT DNS RMX RR Jun 2003 ADDRESS is the IP address where the RMX query was based on. HOST is the name of the machine performing the RMX query. DATE is the date of the query. 7. Compatibility 7.1. Compatibility with old mail receivers Since the suggested extension doesn't change the SMTP protocol at all, it is fully compatible with old mail receivers. They simply don't ask for the RMX records and don't perform the check. 7.2. Compatibility with old mail senders Since the SMTP protocol is unchanged and the SMTP sender is not involved in the check, the method is fully compatible with old mail senders. 7.3. Compatibility with old DNS clients Since the RMX is a new RR, the existing DNS protocol and zone informations remain completely untouched. 7.4. Compatibility with old DNS servers If a server can't provide RMX records or a zone doesn't contain them, the check can't be performed, while compatibility is still guaranteed. 8. Message relaying and forwarding 8.1. Problem description Message forwarding and relaying means that an MTA which received an e-mail by SMTP does not deliver it locally, but resends the message - usually unchanged except for an additional Received header line and maybe the recipient's address rewritten - to the next SMTP MTA. Message forwarding is an essential functionality of e-mail transport services, for example: - Message transport from outer MX relay to the intranet - Message forwarding and Cc-ing by .forward or .procmail-alike mechanisms - Mailing list processing Hadmut Danisch Experimental [Page 10] INTERNET-DRAFT DNS RMX RR Jun 2003 - Message reception by mail relays with low MX priority, usually provided by third parties as a stand-by service in case of relay failure or maintenance - "Forwarding" and "Bouncing" as a MUA functionality In all these cases a message is sent by SMTP from a host which is not covered by the original sender domain's RMX records. While the RMX records would forbid accepting this message, it still must be accepted. The following subsections explain how to cope with that problem. 8.2. Trusted relaying/forwarding In some cases the receiving MTA trusts the sending MTA to not fake messages and to already have checked the RMX records at message reception. As a typical example, a company might have an outer mail relay which receives messages from the Internet and checks the RMX records. This relay then forwards the messages to the different department's mail servers. It does not make sense for these department mail servers to check the RMX record, since the RMX records have already been checked and - since the message was relayed by the outer relay - always would deny the message. In this case there is a trust relationship between the department relays and the outer relay. So RMX checking is turned off for trusted relays. In this example, the department relays would not check messages from the outer relay (but for intranet security, they could still check RMX records of the other departments sub-domains to avoid internal forgery between departments). When a relay forwards a message to a trusting machine, the envelope sender address should remain unchanged. 8.3. Untrusted relaying/forwarding If the receiving MTA does not trust the forwarding MTA, then there is no chance to leave the sender envelope address unchanged. At a first glance this might appear impracticable, but this is absolutely necessary. If an untrusted MTA could claim to have forwarded a message from a foreign sender address, it could have forged the message as well. Spammers and forgers would just have to act as such a relay. Therefore, it is required that, when performing untrusted forwarding, the envelope sender address has to be replaced by the sender address of someone responsible for the relaying mechanism, e.g. the owner of the mailing list or the mail address of the user who's .forward caused the transmission. It is important to stress that untrusted relaying/forwarding means taking over responsibility Hadmut Danisch Experimental [Page 11] INTERNET-DRAFT DNS RMX RR Jun 2003 for the message. It is the idea of RMX records to tie responsibility to message transmission. Untrusted relaying without replacing the sender address would mean to transmit without taking responsibility. The disadvantage is that the original sender address is lost. Therefore, whenever a sender address replacement happens, the Received-Line must contain the old address. Many of today's MTAs already insert the envelope recipient address, but not the sender address into the Received header line. It seems reasonable to require every Received line to include both the sender and recipient address of the incoming SMTP connection. 9. Enforcement policy Obviously, for reasons of backward compatibility and smooth introduction of this scheme, RMX records can't be required immediately. Domains without RMX records must temporarily be treated the same way as they are treated right now, i.e. e-mail must be accepted from anywhere. But once the scheme becomes sufficiently widespread, mail relays can start to refuse e-mails with sender addresses from domains without RMX records, thus forcing the owner of the domain to include a statement of authorization into the domain's zone table. Domain owners will still be free to have an RMX record with a network and mask 0.0.0.0/0, i.e. to allow e-mails with that domain from everywhere. On the other hand, mail receivers will be free to refuse mails from domains without RMX records or RMX records which are too loose. Advanced MTAs might have a configuration option to set the maximum number of IP addresses authorized to use a domain. E-mails from a domain, which's RMX records exceed this limit, would be rejected. For example, a relay could reject e-mails from domains which authorize more than 8 IP addresses. That allows to accept e-mails only from domains with a reasonable security policy. 10. Security Considerations 10.1. Draft specific considerations 10.1.1. Authentication strength It is important to stress, that the suggested method does not provide high level security and does not completely block forged e- mails or spam. It is not a reliable and completely secure security mechanism. It is to be considered as a very cheap and simple light weight mechanism, which can nevertheless provide a significant improvement in mail security against a certain class of attacks, until a successor of SMTP has been defined and commonly accepted. Hadmut Danisch Experimental [Page 12] INTERNET-DRAFT DNS RMX RR Jun 2003 10.1.2. Where Authentication and Authorization end RMX records do not cover the local part of the e-mail address, i.e. what's on the left side of the @ sign. Authentication and authorization are limited to the sending MTA's IP address. The authentication is limited to the TCP functionality, which is sufficient for light weight authentication. The RMX records authorize the IP address of the sending host only, not the particular sender of the message. So if a machine is authorized to use sender addresses of more than a single domain, the authentication scheme does not prevent that any user on this machine can send with any of these domains. RMX is not a substitute for the host security of the involved machines. The proposed authentication scheme can be seen as a "half way authentication": It does not track back an e-mail to the effective sender. It tracks only half of the way, i. e. it tracks back to the domain and it's DNS administrators who authorized that particular sender IP address to use it for sending e-mail. How the party responsible for that domain performs user authentication, whom it grants access to, how it helds people responsible for abuse, is completely left as the private business of those who are in charge of that domain. So this draft does not interfere with the domain's individual security policy or any legislation about such policies. On the other hand, the proposed authentication scheme does not give any statement about the nature and quality of the domain's security policy. This is an essential feature of the proposal: E-mail authentication must be deployed world wide, otherwise it won't do the job. Any security scheme interfering with the local legislations or the domain's security policy will not be accepted and can't effectively deployed. Therefore, the security policy must remain the domain's private business, no matter how lousy the policy might be. In order to achieve this and to make use of the only existing world wide Internet directory scheme (DNS), the approach of this proposal is to just ignore the local part of the sender address (i.e. what's left of the @ part) and limit view to the domain part. After all, that's what we do anyway when delivering to a given address with SMTP. 10.1.3. Vulnerability of DNS DNS is an essential part of the proposed authentication scheme, since it requires any directory service, and DNS is currently the only one available. Unfortunately, DNS is vulnerable and can be spoofed and poisoned. This flaw is commonly known and weakens many network services, but for reasons beyond that draft DNS has not Hadmut Danisch Experimental [Page 13] INTERNET-DRAFT DNS RMX RR Jun 2003 been significantly improved yet. After the first version of this draft, I received several comments who asked me not to use DNS because of its lack of security. I took this into consideration, but came to the conclusion that this is unfeasible: Any authentication scheme linked to some kind of symbolic identity (in this case the domain name) needs some kind of infrastructure and trusted assignment. There are basically two ways to do it: Do it yourself and trust nobody else, or let someone else do it. There are methods to do it the former way, e.g. to give someone some kind of authentication information after a first successful e-mail exchange, e.g. some kind of cookie or special e-mail address. This is certainly interesting and powerful, but it does not solve the problem on a world wide scale and is far to complicated and error prone for the average user, i. e. 99% of the users. The latter method to let someone else do the symbolic name assignment and create the authentication framework is well known. It context of public key cryptography, this is called a Public Key Infrastructure (PKI). On of the best known facts about PKIs is that, until now, we don't have any covering a significant part of the Internet. And we won't have any in near future. The complexity is far too high, it is too expensive, and it involves cooperation of every single user, which is simply unrealistic and extremely error prone. So what do we have we can use? All we have is the DNS and the Whois database. And we have countries who don't allow cryptography. So the proposal was designed to use DNS without cryptography. It does not avoid DNS because of its vulnerability, it asks for a better DNS, but accepts the DNS as it is for the moment. Currently there are two main threats caused by the DNS weakness: - A spammer/forger could spoof DNS in order to gain false authorization to send fake e-mails. - An attacker could spoof DNS in order to block delivery from authorized machines, i. e. perform a Denial of Service attack. The first one is rather unrealistic, because it would require an average spammer to poison a significant part of the DNS servers of its victims. A spammer sending messages to one million receipients would need to poison at least 1-10% which is 10,000 to 100,000 receipient's DNS servers. This should be unfeasible in most cases. In contrast, the second threat is a severe one. If an attacker wanted to block messages from one company to another, he just needs to poison the recipients DNS server with a wrong RMX record in order to make the recipient's SMTP machine reject all messages. And this is feasible since the attacker needs to poison only a single Hadmut Danisch Experimental [Page 14] INTERNET-DRAFT DNS RMX RR Jun 2003 DNS server. But does this make SMTP more vulnerable? No. Because the attacker can already do even more without RMX. By poisoning the sender's DNS server with wrong MX records, the attacker can also block message delivery or even redirect the messages to the attacker's machine, thus preventing any delivery error messages and furthermore getting access to the messages. As a consequence, e-mail delivery by SMTP requires a better DNS anyway. The requirements are not significantly expanded by RMX. 10.1.4. Additional data attack? While writing a test implementation, a certain kind of attack came into my mind. I'm still not sure, whether this attack is possible on any DNS server, but I believe it should be mentioned: Imagine an unauthorized sender is sending a forged mail (e.g. spam). At connection time, before querying the RMX record, the receiving MTA usually performs a PTR query for the IP address of the sending MTA. If the sender has control over the authoritative name server for that particular IP address, the sender could give a normal PTR answer, but could append a wrong RMX, APL, or A record in the additional section of the query. A subsequent RMX query could receive wrong DNS data if the DNS server used by the receiving MTA accepted those forged records. 10.1.5. Open SMTP relays Open SMTP relays (i.e. machines who accept any e-mail message from anyone and deliver to the world) abused by spammers are a one of the main problems of spam defense and sender backtracking. In most cases this problem just vanishes because foreign open relay machines will not be covered by the RMX records of the forged sender address. But there are two special cases: If the spammer knows about a domain which authorizes this particular machine, that domain can be used for forgery. But in this case, the IP address of the relay machine and the RMX records of the domain track back to the persons responsible. Both can be demanded to fix the relay or remove the RMX record for this machine. An open relay is a security flaw like leaving the machine open for everybody to login and send random mails from inside. Once the administrative persons refuse to solve the problem, they can be identified as spammers and held responsible. The second special case is when a domain authorizes all IP addresses by having the network 0.0.0.0/0 in the RMX/APL record. In this case, open relays don't make things worse. It's up to the Hadmut Danisch Experimental [Page 15] INTERNET-DRAFT DNS RMX RR Jun 2003 recipient's MTA to reject mails from domains with loose security policies. 10.1.6. Unforged Spam This proposal does not prevent spam (which is, by the way, not yet exactly defined), it prevents forgery. Since spam is against law and violates the recipients rights, spam depends on untracability of the sender. In practice the sender forges the sender address (other cases see below). This proposal is designed to detect such forgeries. However, the RMX approach is rendered ineffective, if the sender doesn't forge. If the sender uses just a normal address of it's own domain, this is just a plain, normal e-mail, which needs to be let through. Since it is up to the human's taste whether this is spam or not, there's no technical way to reliably identify this as spam. But since the sender domain is known, this domain can be blacklisted or legal steps can be gone into. 10.1.7. Empty Sender address There is a design flaw of current SMTP and SMTP programs: The empty sender envelope address. This is used for delivery of error messages, and most MTAs accept messages with an empty sender address. Obviously, RMX records can't be checked for empty sender addresses. It is a design flaw to inherently allow anonymous mails. This flaw must either be fixed, or other methods have to be developed. For example MTAs could accept such mails only if they reference the Message-ID of another e-mail which has recently been delivered. 10.1.8. Reliability of Whois Entries Once the RMX infrastructure gets deployed, what's the security gain? It allows to determine the domain which's DNS zone authorized the sending machine. What's that good for? There are some immediate uses of the domain name, e.g. in black- and whitelisting. But in most cases this is just the starting point of further investigations, either performed automatically before message acceptance, or manually after spam has been received and complainted about. The next step after determining the domain is determining the people responsible for this domain. This can sometimes be achieved by querying the Whois databases. Unfortunately, many whois entries are useless because they are incomplete, wrong, obsolete, or in Hadmut Danisch Experimental [Page 16] INTERNET-DRAFT DNS RMX RR Jun 2003 uncommon languages. Furthermore, there are several formats of address informations which make it difficult to automatically extract the address. Sometimes the whois entry identifies the provider and not the owner of the domain. Whois servers are not built for high availability and sometimes unreachable. Therefore, a mandatory standard is required about the contents and the format of whois entries, and the availability of the servers. After receiving the MAIL FROM SMTP command with the sender envelope address, the receiving MTA could check the RMX record and Whois entry. If it doesn't point to a real human, the message could be rejected and an error message like "Ask your provider to fix your Whois entry" could be issued. Obviously, domain providers must be held responsible for wrong entries. It might still be acceptable to allow anonymous domains, i. e. domains which don't point to a responsible human. But it is the receivers choice to accept e-mails from such domains or not. 10.1.9. Hazards for Freedom of Speech Currently, some governments try to enforce limitations of internet traffic in order to cut unwanted content providers from the network. Some of these governments try to hide a whole country behind firewalls, others try to force Internet providers to poison DNS servers with wrong A records for web servers, e.g. one county administration in Germany tries to do so. If message reception depends on DNS entries, the same governments will try to block not only HTTP, but SMTP also. However, since most MTAs already reject messages from unresolvable domain names this is not a new threat. 10.2. General Considerations about spam defense After discussing security requirements of the proposal, now the security advantages of the RMX approach over content based filters will be explained. Basically, there are three kinds of content filters: - Those who upload the message or some digest to an external third party and ask "Is this spam"? - Those who download a set of patterns and rules from a third party and apply this set to incoming messages in order to determine whether it is spam. - Those who are independent and don't contact any third party, but try to learn themselves what is spam and what isn't. Hadmut Danisch Experimental [Page 17] INTERNET-DRAFT DNS RMX RR Jun 2003 The message filters provided by some e-mail service providers are usually not a kind of their own, but a combination of the first two kinds. 10.2.1. Action vs. reaction Content filters suffer from a fundamental design problem: They are late. They need to see some content of the same kind before in order to learn and to block further distribution. This works for viruses and worms, which redistribute. This doesn't work for spam, since spam is usually not redistributed after the first delivery. When the filters have learned or downloaded new pattern sets, it's too late. This proposal does not have this problem. 10.2.2. Content based Denial of Service attacks All three kinds of content filters, but especially the second and the third kind are vulnerable to content based Denial of Service attacks. If some kind of third party (e.g. non-democratic government, intellectual property warriors, religious groups, military, secret services, patriots, public relation agents, etc.) wants certain contents not to be distributed, they could either poison the pattern/rule databases or feed wrong sets to particular receivers. Such pattern/rule sets are the perfect tool for censoring e-mail traffic and denial of service attacks by governments and other parties, and a similar threat are virus filters. E. g. the content industry could demand to teach all virus and spam filters to delete all e-mails containing the URL of an MP3 web server outside the legislations. Software manufacturers could try to block all e-mails containing software license keys, thus trying to make unallowed distribution more difficult. Governments could try to block distribution of unwanted informations. This proposal does not have this problem. 11. Privacy Considerations (It was proposed on the 56th IETF meeting to have a privacy section in drafts and RFCs.) 11.1. Draft specific considerations Hadmut Danisch Experimental [Page 18] INTERNET-DRAFT DNS RMX RR Jun 2003 11.1.1. No content leaking Since the RMX approach doesn't touch the contents of a message in any way, there is obviously no way of leaking out any information about the content of the message. RMX is based solely on the envelope recipient address. However, methods to fix problems not covered by RMX might allow content leaking, e.g. if the acceptance of a message with an empty sender address requires the reference to the message id of an e-mail recently sent, this allows an attacker to verify whether a certain message was delivered from there. 11.1.2. Message reception and sender domain Message delivery triggers RMX and APL requests by the recipient. Thus, the admin of the DNS server or an eavesdropper could learn that the given machine has just received a message with a sender from this address, even if the SMTP traffic itself had been encrypted. However, most of today's MTAs do query the MX and A records of the domain after the MAIL FROM command, so this is not a real new threat. 11.1.3. Network structure Since RMX and its associated APL records provide a complete list of all IP addresses of hosts authorized to send messages from this address, they do reveal informations about the network structure and maybe the lifestyle of the domain owner, since a growing number of domains are owned by single persons or families. E.g. the RMX records could reveal where someone has his job or spends his time at weekends. If such informations are to be kept secret, it is the user's job to not sent e-mails from there and to relay them from non-compromising IP addresses. 11.1.4. Owner information distribution As described above, RMX depends partly on the reliability of the whois database entries. It does not make anonymous domains impossible, but it requires to keep the database entries "true", i. e. if a whois entry does not contain informations about the responsible person, this must be unambigously labeled as anonymous. It must not contain fake names and addresses to pretend a non- existing person. However, since most Internet users on the world feel extremely annoyed by spam, they will urge their MTA admin to reject messages from anonymous domains. The domain owner will have Hadmut Danisch Experimental [Page 19] INTERNET-DRAFT DNS RMX RR Jun 2003 the choice to either remain anonymous but be not able to send e- mail to everyone in the world, or to be able but to reveal his identity to everyone on the world. It would be possible to provide whois-like services only to recipients of recent messages, but this would make things too complicated to be commonly adopted. 11.2. General Considerations about spam defense 11.2.1. Content leaking of content filters As described above in the Security chapter, there are spam filters which inherently allow leakage of the message body. Those filters upload either the message body, or in most cases just some kind of checksum to a third party, which replies whether this is to be seen as spam or not. The idea is to keep a databases of all digests of all messages. If a message is sent more often than some threshold, it is to be considered as a mass mail and therefore tagged as spam. While the digest itself does not reveal the content of the message, it perfectly reveals where a particular message has been delivered to. If a government finds just a single unwanted message, if a software manufacturer finds a single message with a stolen product license key, if someone finds a message with unpatriotic content, it takes just a single database lookup to get a list of all people who received this particular message. Content filters with digest upload are the perfect "Big Brother". 11.2.2. Black- and Whitelists Some proposals against spam are based on a central database of white- or blacklisted IP addresses, Sender names, Message IDs or whatever. Again, there is a central database which learns who has received which e-mail or from which sender with every query. This allows tracking relations between persons, which is also a breach of privacy. 12. Deployment Considerations Is there a concise technical solution against spam? Yes. Will it be deployed? Certainly not. Why not? Because of the strong non-technical interests of several parties against a solution to the problem, as described below. Since these are non-technical reasons, they might be beyond the Hadmut Danisch Experimental [Page 20] INTERNET-DRAFT DNS RMX RR Jun 2003 scope of such a draft. But since they are the main problems that prevent fighting spam, it is unavoidable to address them. This chapter exists temporarily only and should support the discussion of solutions. It is not supposed to be included in a later RFC. 12.1. The economical problem As has been recently illustrated in the initial session of the IRTF's Anti Spam Research Group (ASRG) on the 56th IETF meeting, sending spam is a business with significant revenues. But a much bigger business is selling Anti-Spam software. This is a billion dollar market, and it is rapidly growing. Any simple and effective solution against spam would defeat revenues and drive several companies into bankrupt, would make consultants jobless. Therefore, spam is essential for the Anti-Spam business. If there is no spam, then no Anti-Spam software can be sold, similar to the Anti-Virus business. There are extremely strong efforts to keep this market growing. Viruses, Worms, and now spam are just perfect to keep this market alive: It is not sufficient to just buy a software. Databases need to be updated continuously, thus making the cash flow continuously. Have a single, simple, and permanent solution to the problem and - boom - this billion dollar market is dead. That's one of the reasons why people are expected to live with spam. They have to live with it to make them buy Anti-Spam software. Content filters are perfect products to keep this market alive. 12.2. The POP problem Another problem is the history of mail delivery. Once upon a time, there used to be very few SMTP relays which handled the e-mail traffic of all the world, and everybody was happy with that. Then odd things like Personal Computers, which are sometimes switched off, portable computers, dynamicly assigned IP addresses, IP access from hotel rooms, etc. was invented, and people became unhappy, because SMTP does not support delivery to such machines. To make them happy again, the Post Office Protocol[3] was invented, which turned the last part of message delivery from SMTP's push style into a pull style, thus making virtually every computer on the world with any random IP address a potential receiver of mails for random domains. Unfortunately, only receiving e-mail was covered, but sending e-mail was left to SMTP. The result is that today we have only very few SMTP relays pointed Hadmut Danisch Experimental [Page 21] INTERNET-DRAFT DNS RMX RR Jun 2003 to by MX records, but an extreme number of hosts sending e-mail with SMTP from any IP address with sender addresses from any domain. Mail delivery has become very asymmetric. Insecurity, especially forgeability, has become an essential part of mail transport. That problem could easily be fixed: Use protocols which allow uploading of messages to be delivered. If a host doesn't receive messages by SMTP, it shouldn't deliver by SMTP. Mail delivery should go the same way back that incoming mail went in. This is not a limitation to those people on the road who plug their portable computer in any hotel room's phone plug and use any provider. If there is a POP server granting download access from anywhere, then the same server should be ready to accept uploading of outgoing messages. But as I saw from the comments on the first version of this draft, people religiously insist on sending e-mail with their domain from any computer with any IP address in the world, e.g. when visiting a friend using her computer. It appears to be impossible to convince people that stopping mail forgery requires every one of them to give up forging. 12.3. The network structure problem A subsequent problem is that many organisations failed to implement a proper mail delivery structure and heavily based their network on this asymmetry. I received harsh comments from Universities who were unable to give their network a good structure. While they do have a central mail relay for incoming mail to the universities domain, they developed a structure where every member of the University randomly sends e-mails with that University's domain as a sender address from home or everywhere in the world with any dynamically assigned IP address from any provider. So this domain is to be used from every possible IP address on earth, and they are unable to operate any authentication scheme. Furthermore, they were unable to understand that such a policy heavily supports spam and that they have to expect that people don't accept such e-mails anymore once they become blacklisted. As long as organisations insist on having such policies, spammers will have a perfect playground. 12.4. The mentality problem Another problem is the mentality of many internet users of certain countries. I received harsh comments from people who strongly insisted on the freedom to send any e-mail with any sender address Hadmut Danisch Experimental [Page 22] INTERNET-DRAFT DNS RMX RR Jun 2003 from anywhere, and who heavily refused any kind of authentication step or any limitation, because they claimed that this would infringe their constitutional "Freedom of speech". They are undeviatingly convinced that "Freedom of speech" guarantees their right to talk to everybody with any sender address, and that is has to be kept the recipient's own problem to sort out what he doesn't want to read - on the recipient's expense. It requires a clear statement that the constitutional "Freedom of Speech" does not cover molesting people with unsolicited e-mail with forged sender address. 12.5. The identity problem How does one fight against mail forgery? With authentication. What is authentication? In simple words: Making sure that the sender's real identity meets the recipients idea of who is the sender, based on the sender address which came with the message. What is identity? It is the main problem. Several countries have different ideas of "identity", which turn out to be somehow incompatible. In some countries people have identity cards and never change their name and birthday. Identities are created by human birth, not by identity changes. Other countries do not have such a tight idea about identity. People's temporary identity is based on nothing more than a driving license and a social security number. With this background, it is virtually impossible to create a trustworthy PKI covering all Internet users. I learned that it is extremely difficult to convince some people to give up random e- mail sending. 12.6. The multi-legislation problem Many proposals about fighting spam are feasible under certain legislations only, and are inacceptable under some of the legislations. But a world wide applicable method is required. That's why the approach to ask everone on the world to sign messages with cryptographic keys is not feasible. Implementation and further Information Further informations and a test implementation are available at http://www.danisch.de/work/security/antispam.html and http://www.danisch.de/software/rmx/ Hadmut Danisch Experimental [Page 23] INTERNET-DRAFT DNS RMX RR Jun 2003 References 1. J. Klensin, "Simple Mail Transfer Protocol," RFC 2821 (April 2001). 2. P. Mockapetris, "DOMAIN NAMES - IMPLEMENTATION AND SPECIFICATION," RFC 1035 (November 1987). 3. J. Myers, M. Rose, "Post Office Protocol - Version 3," RFC 1939 (May 1996). Author's Address Hadmut Danisch Tennesseeallee 58 76149 Karlsruhe Germany Phone: ++49-721-843004 E-Mail: rfc@danisch.de Comments Please send comments to rfc@danisch.de. Expiry This drafts expires on Dec 1, 2003 Hadmut Danisch Experimental [Page 24]