| Network Working Group | K. Moore |
| Internet-Draft | Network Heretics |
| Updates: 5231, 6409 (if approved) | C. Newman |
| Intended status: Standards Track | Oracle |
| Expires: January 20, 2016 | July 19, 2015 |
SMTP and SUBMISSION Service Extensions For Address Query
draft-moore-email-addrquery-01.txt
This document defines several mechanisms which can be used by a client such as a Mail User Agent or Mail Submission Agent, to query an SMTP server which is configured to accept incoming mail for a mail domain, to obtain information associated with an email address based in that domain. Among other purposes, these mechanisms are intended to facilitate discovery of senders' and/or recipients' public keys for use in automatic verification of whole-message digital signatures and automatic whole-message encryption of email sent to recipients.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.
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."
This Internet-Draft will expire on January 20, 2016.
Copyright (c) 2015 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
At least since the introduction of MIME [RFC1321] there has been a desire to allow message senders to discover capabilities of email recipients, so that senders could avoid sending message contents to recipients who were unable to make use of such contents. Similarly, deployment of per-message encryption (e.g. PEM [RFC1113], S/MIME [RFC5751], and OpenPGP [RFC4880]) has long been hampered for lack of a standard and widely supported means to discover and verify authenticity of senders' and recipients' public key(s).
The issue surfaced recently as part of the DANE working group discussion in Dallas, and specifically in an effort to adapt TLSA DNS records [RFC6698] for use in discovery of email recipients' public keys. The problem there was that there is no clean way to map recipient email addresses onto DNS labels, because the interpretation of a local-part of an email address is entirely left to the SMTP server(s) that accept incoming mail for that address's mail domain, and different mail domains have configured their SMTP servers to interpret their email addresses in different ways. The "local parts" of email addresses may be case-sensitive or case-insensitive, subaddresses may be allowed, there may be some sort of fuzzy matching, an address may be forwarded elsewhere, and so on. Also, having public keys for email recipients advertised in DNS would have facilitated email traffic analysis by an observer watching DNS queries and responses in cleartext.
Since the knowledge of how to interpret an email address is inherently embedded in the code and configuration of the SMTP servers that accept incoming mail for that address's email domain, it appears that the best way to advertise public keys and other information associated with email addresses is to do so using the same SMTP servers that accept such incoming mail. That way, the logic that maps from address to associated information will be the same logic that maps from recipient address to recipient mailbox (or forwarding address). A separate lookup service could be used, but this would introduce a high probability that the service would interpret the address differently than that mail domain's SMTP servers, if for no other reason than configuration errors. However as a compromise for large mail service providers, and especially those that serve large numbers of mail domains, the proposed SMTP extension also includes a "redirect" mechanism that can be used to refer a client to a separate service which then provides the requested information. Finally, this document defines an extension to the Mail Submission service which allows that service to perform an address information lookup operation on behalf of its authenticated client, which can be useful to circumvent the common practice of blocking outbound port 25 traffic.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119].
This specification expresses syntax using the Augmented Backus-Naur Form (ABNF) as described in [RFC5234], including the core rules in Appendix B and rules from [RFC5322].
In examples illustrating protocol interactions, "C:" and "S:" indicate lines sent by the client and server respectively. If a single "C:" or "S:" label applies to multiple lines, then the line breaks between those lines are for editorial clarity only and are not part of the actual protocol exchange.
This section defines a service extension to the Simple Message Transfer Protocol (SMTP) [RFC5321] which can be used by a client to query the server for information about an email address for which the server accepts incoming mail.
The AQRY SMTP command is used to query an SMTP server about an address containing a domain name for which the server is configured to act as a mail exchanger, i.e. to accept incoming mail for delivery. A SMTP server which accepts incoming mail for a domain is in a unique position to interpret email addresses containing that domain, since only such a server can reliably know whether the local part of that email address is case-sensitive (i.e. whether Joe@example.com and joe@example.com are distinct users), whether subaddressing applies to that domain (e.g. whether joe+xyz@example.com refers to the same user as joe@example.com), whether a particular recipient has mail forwarded, and so on. Therefore an SMTP server MUST reject an AQRY command which contains an address for which the server is not explicitly configured to accept incoming mail.
In addition, to ensure the integrity of the information provided to the client and to deter both passive and active attacks, any SMTP server supporting ADDRQUERY MUST also support the STARTTLS service extension, and MUST reject any AQRY command not appearing in a TLS-protected session. Clients using the AQRY command MUST support the TLS Server Name Indication (SNI) [RFC6066] extension, and MUST supply the host name of the server to which they wish to connect in the ServerNameList portion of the extension_data field of the extended client hello message. (This requirement also applies to Mail Submission servers that implement the Address Query Proxy extension.) This host name will either be the target of the MX record associated with the address being queried, or the "host" field as obtained from an AQRY or AQPX redirect response as defined below. Servers supporting the Address Query extension SHOULD support SNI and use it to provide an appropriate server certificate, if available.
aqry = "AQRY" SP "<" Mailbox ">"
[ "REFERBY=" address-literal ]
[ "RRVS=" date-time ] [ "COOKIE=" Atom ] CRLF
The syntax of the AQRY command is as follows:
where address-literal, Atom, and Mailbox are as defined in [RFC5321] (or if the SMTP server supports the SMTPUTF8 extension, Mailbox is as defined in [RFC6531]), and date-time is as defined in [RFC3339], with the added restriction that a "time-secfrac" MUST NOT be used. (XXX amend the above to use the right nonterminal ABNF symbol for servers that support SMTPUTF8. Similarly for AQPX command.)
The AQRY command requests that the SMTP server return public information about the email address ("Mailbox") specified in the command. If the optional RRVS parameter is included, it specifies that the email address must have been valid at least since that date and time. If the server knows that the address has not been valid that long, it MUST return either an error, or a redirect to a server that will return an "address not found" error.
(Note: Although the RRVS parameter to the AQRY command has the same syntax as the RRVS parameter to the RCPT command as defined in [RFC7293], the two are separate and have different purposes. An SMTP server MAY support the Address Query extension even if it does not support the RRVS extension.)
The COOKIE and REFERBY parameters are used only in redirects, as described below.
(XXX consider max length of COOKIE parameter and whether this affects minimum SMTP command line length that the server must support.)
Clients wishing to query for email address information MUST first perform a DNS [RFC1035] lookup with query type of MX, specifying the domain name that appears in the email address. The selection of SMTP servers among those returned from the DNS query follows the same algorithm used for selection of SMTP servers to be used for forwarding mail [RFC5321]: servers with lower MX precedence values are queried before servers with higher MX precedence values.
Clients MUST NOT send an AQRY command to a server that isn't listed in DNS as a mail exchanger for the mail domain of the address to be queried. Exception: a client MAY send an AQRY command to an arbitrary SMTP server without first obtaining that from a DNS MX lookup, if this is done specifically and entirely for the purpose of fault diagnosis or configuration checking and the results are not used to encrypt email nor validate a digital signature.
Note: In contrast to DNS lookups for normal mail routing, the presence of one or more MX records for the mail domain of the address being queried is REQUIRED. In particular, the AQPX command described below, when used without a SERVER argument, will not query an SMTP server if there are no MX records pointing to it, and only A or AAAA records.
Clients wishing to use AQRY MUST first negotiate use of TLS encryption using the STARTTLS command [RFC3207]. If the server does not advertise STARTTLS, or the TLS negotiation fails, the client MUST NOT attempt to use AQRY. Furthermore, the client MUST NOT attempt to use AQRY before first establishing the identity of the server using the server's certificate, and in particular, that the server's TLS certificate contains either a DNS-ID (subjectAltName of dNSName type, see [RFC5280]) or a CN-ID (CN attribute from subject name, see [RFC6125]) that matches either the DNS name that is the target of the MX record, or the DNS name appearing in the email address for which information is being requested. (Exception: the check of the TLS certificate MAY be skipped if the AQRY operation is done specifically and entirely for the purpose of fault diagnosis or configuration checking, and the results are not used to encrypt email nor validate a digital signature.)
Note: The above rule does not, by itself, establish that the SMTP server is an authoritative source of information about the address(es) to be presented to AQRY, because (for example) an MX record might have been spoofed (unless signed by DNSSEC and the signature was appropriately verified), or the DNS name associated with the MX record might not actually have an arrangement with the SMTP server. However, if the server certificate fails this test, there's no point in the client doing the AQRY at all. See Section 6.
In response to an AQRY command, the server MUST return one of: a normal response, a redirect response, or an error response.
A normal response contains information about the email address for which the request was issued which is specific to that email address, and/or information about the mail domain name which appears in that email address. A normal response MAY also contain information such as address(es) to which incoming mail will be forwarded. In some such cases the client will need to perform additional AQRY operations, perhaps of other SMTP servers serving other domains, in order to learn information about the addresses that would eventually receive mail sent to the originally queried address.
A redirect response does not contain information about the requested email address, but does contain one or more URLs which may then be queried to learn about that address and/or its mail domain.
The normal (non-redirect, non-error) response to a valid AQRY command consists of multiple lines. Each line but the last line of the response begins with "212-". The remainder of each line beginning with "212-" consists of JSON text [RFC7159] subsequently encoded in BASE64 format as defined in [RFC2045]. BASE64 is used to avoid the need for the server to produce JSON text which conforms to SMTP line-length restrictions.
A normal response is not an indication that the address supplied in the AQRY command is valid. An implementation that does not wish to disclose whether recipients are valid MAY return "fake" information in response to AQRY requests for nonexistent recipients. However the implementation MUST NOT return "fake" information for valid recipients.
The data structure encoded in the JSON object is further described in section Section 5.
"212" SP "." CRLF
The last line of the response is of the form:
To produce the normal response to an AQRY command, the server first produces or obtains the requested information in JSON format. The server then encodes the entire JSON object using the BASE64 algorithm, such that each line of the BASE64 output does not exceed 76 characters, not including the CRLF character sequence that terminates each line. The server then prepends "212-" to the beginning of each line of the BASE64 output. Finally, the server appends a single line consisting of "212 ." to the output. Per normal SMTP convention, each line of the reply MUST be terminated by CRLF.
Note: If a address is configured to forward mail to one or more other addresses, this can affect the contents of the JSON object or result in an error. See Section 5.
To recover the JSON from the AQRY reply text, the client first collects the text and ensures that the terminating "212 ." line is present. The terminating line is then discarded, and the "212-" prefix is removed from each of the preceding lines. The resulting text is then fed to the BASE64 decoder to produce a JSON object. The resulting JSON object may then be interpreted.
In the case where the SMTP server is configured to accept incoming mail for the address presented in the AQRY command, but either of the following two conditions apply:
the SMTP server MAY return a multi-line redirect response with a response code of 213. Similar in presentation format to the normal response, the redirect response consists of BASE64-encoded JSON, with each line of the BASE64 text preceded by "213-" and the last line of the response consisting entirely of "213 ." followed by CRLF. However, the data structure represented in JSON for a redirect response is different than that of a normal response. The data structure encoded in a redirect response consists of an array of objects describing SMTP servers to which the query can be referred. Each such object may contain the following elements:
There is no significance to the order in which the list items, or the elements of any of the objects in the list, appear in the JSON.
Example: A client issues a query for information about joe@example.com, and the server returns a redirect response:
C: AQRY <joe@example.com> S: 213-W3siaG9zdCI6ICJmb28uZXhhbXBsZS5jb20iLCAiY29va2llIjogImxranNl S: 213-b3J1IiwgInBvcnQiOiA5ODc2fSwgeyJob3N0IjogIjEwLjEuMi4zIiwgImNv S: 213-b2tpZSI6ICJzZndlcnYzMyJ9LCB7Imhvc3QiOiAiMjAwMTpEQjg6YWJjZDo6 S: 213-MToyIiwgImNvb2tpZSI6ICJsa2pzZW9ydSIsICJwb3J0IjogNDMyNX1d S: 213 .
The client decodes this and obtains the following data structure (formatted for readability below):
[
{ "host": "foo.example.com", "port": 9876,
"cookie": "lkjseoru" },
{ "host": "10.1.2.3", "cookie": "sfwerv33" },
{ "host": "2001:DB8:abcd::1:2", "port": 4325,
"cookie": "lkjseoru" }
]
The client could then obtain the requested information via any of the following:
In each of the above instances, the client will supply the "host" parameter from the object as the TLS Server Name Indication (SNI) HostName. Any RRVS parameter appearing in the original AQRY command is also supplied when issuing the AQRY command to the redirect servers. In addition the AQRY REFERBY parameter is supplied with its value set to the Internet Protocol (v4 or v6) address of the SMTP server from which the redirect was obtained.
Since the SMTP servers returned in a referral response are not expected to be able to process incoming mail, they are not required to implement the full SMTP protocol. They need only implement the following commands: EHLO (advertising STARTTLS and ADDRQUERY), STARTTLS, AQRY, and QUIT. Such a server SHOULD also implement the PIPELINING extension. [RFC2920]
In addition to reply codes defined in [RFC5321], the following reply codes SHOULD be used to indicate the error conditions described below. In each case below the enhanced status code [RFC5248] that appears immediately following the 3-digit SMTP reply code is suggested for use by server implementations supporting the SMTP ENHANCEDSTATUSCODES extension [RFC2034]. However, the appropriate status code may depend to some degree on the nature of the SMTP server implementation or configuration, and there may be cases in which a different enhanced status code is appropriate. (The SMTP reply code and enhanced status code serve distinct purposes: The reply code is intended for use by SMTP clients, and in particular signals transitions in the SMTP client's state machine. The enhanced status code is intended for use in Delivery Status Notifications [RFC3464] and serves as an indication of the likely nature of a problem with the mail system or network. The relationship between the two is loose rather than strict.)
IANA NOTE: Some of these codes need to be assigned; these are marked with IANA- followed by some number. See Section 8. (RFC Editor: please remove this paragraph on publication.)
This section defines a service extension to the Mail Submission Protocol [RFC6409] which can be used by an authenticated, authorized client to query an SMTP server on port 25 for information about an email address. This is intended only as a workaround for port 25 blocking, so the extension is minimally tailored for that purpose.
The AQPX command is used to query an Submission server for information about an email address. The client user MUST have already been authenticated and verified to be authorized to use that Submission server. Use of this command by a mail client (such as a Mail User Agent) is OPTIONAL; this specification does not prohibit a client directly contacting an SMTP server. However, it is expected that clients will often need a service as a workaround for the common practice of blocking outbound traffic on TCP port 25.
The AQRY command requires a TLS-protected session, either by using a server port that automatically establishes TLS on connect, or by using a cleartext port and the STARTTLS command. Clients MUST NOT attempt to use the AQRY command if the session is not protected with TLS; and servers MUST refuse an AQRY command that appears in a session not protected with TLS.
When this command is received, the Submission server will then:
If some error occurs in the process of performing the above, the Submission server will return an appropriate response code.
aqpx = "AQPX" SP "<" Mailbox ">"
[ "SERVER=" ( Domain
/ IPv4-address-literal
/ IPv6-address-literal) ]
[ "RRVS=" date-time ] [ "COOKIE=" Atom ] CRLF
The syntax of the AQPX command is as follows: [RFC5321] (or if the Submission server supports the SMTPUTF8 extension, Domain and Mailbox are as defined in [RFC6531]), and date-time is as defined in [RFC3339] with the added restriction that a "time-secfrac" MUST NOT be used.
The optional SERVER parameter specifies an SMTP server to consult. Since this may be any server included in either a response to a DNS MX query, or a server returned in a redirect from a previous query to an SMTP server, the Submission server SHOULD NOT restrict the servers to which a client may issue a query. There is no provision for specifying the port at which the SMTP server is to be contacted; the client is assumed to be able to directly contact servers on ports other than 25. If no SERVER parameter is supplied, the Submission server will perform an MX lookup of the domain portion of the address, and attempt to issue the AQRY command to one or more servers (if any are found) in order of increasing precedence until it either receives a result that is not a temporary failure; that result is returned to the Submission client. Regardless of whether a SERVER parameter is specified, Submission servers SHOULD implement a reasonable timeout for obtaining the information necessary to respond to the AQPX command. If the timeout expires, the server should return a 431 error (see below).
The RRVS and COOKIE parameters are passed to the AQRY command issued to the SMTP server.
Since this is a proxy service that is intended to return a response from a remote SMTP server, any valid response to the SMTP AQRY command (including a normal response, redirect response, or error response) is also a valid response to a Submission service AQPX command.
The submission service SHOULD NOT follow redirects returned by an SMTP server, and MUST return the SMTP server's response intact and without modification.
In addition, the following AQPX-specific response codes are permitted:
Note: This section is preliminary and is expected to require considerable work, and to be moved to a separate document.
XXX consider using JSON Web Signature (JWS) as an optional means of authenticating returned information.
The response to the AQRY command is a single JSON object. This JSON object contains zero or more members, each of which is itself an object. The members of the top-level object either supply information about a mail domain, or a specific email address. Mail domain objects are named using the DNS name of their mail domain (which does not contain an "@"), while email address objects are named for their email address (which does contain an "@"). In either case the domain or email address used to name the second-level objects are in the same format as would be presented to a SMTP MAIL command. (i.e. If the SMTP server supports the SMTPUTF8 extension [RFC6531], the address MAY be in UTF-8; otherwise the address MUST be in ASCII).
Both mail domain objects and email address objects are "flat", that is to say, the members of these objects are either strings, numbers, booleans, or arrays whose members consist exclusively of one or more of these. For ease of use in some programming languages, the names of the elements of both mail domain and email address objects MUST begin with an ASCII letter ("a"-"z" or "A"-"Z") and MUST consist only of letters, digits, and underscore ("_").
The requirement for a flat structure is to discourage creation of complex data models to represent mail domain and address information. (Note in draft: this is subject to change, but it seems to one of the authors that one result of the ability to define very complex structures to represent information is that there is a resulting tendency to model information using more complexity than is useful or helpful. Having a relatively simple data model for representation of such information may also make it easier to store and manipulate such data in existing SMTP implementations that are implemented in a variety of programming langauges, and in existing databases that are used to store address validity and forwarding information.
The objects included in the AQRY response are expected to provide information about the domain and/or email address supplied in the AQRY command. However, multiple domain and/or email address objects MAY be included if the information is relevant and potentially useful to the client, and the server is authoritative for such information. For instance, if joe@a.example.com has his mail forwarded to bob@b.example.com, the response to AQRY of joe@a.example.com MAY include domain objects for both "a.example.com" and "b.example.com", and email address objects for both "joe@a.example.com" and "bob@b.example.com". However, for this information to be useful to the client, the server's TLS certificate SHOULD include DNS-ID attributes matching both "a.example.com" and "b.example.com" (whether or not wildcard objects are used), and the returned information for "joe@a.example.com" SHOULD reveal that that address is being forwarded to "bob@b.example.com". (Note: Encryption of mail to be forwarded is tricky to get right for various reasons, including that a recipient may not wish to publicly reveal his forwarding address(es), and also that a sender may not wish his encrypted mail to be encrypted for, and forwarded to, one or more different persons or addresses than the originally-specified recipient.)
Examples of attributes that might appear within mail domain objects might include:
Examples of attributes that might appear within mail address objects include:
As described above, the JSON object returned in a normal AQRY response may itself contain multiple member objects, each providing information about a separate email address or mail domain. The trustworthiness of each member MUST be evaluated separately.
A member object of a normal AQRY response MUST NOT be considered trustworthy for any purpose, unless the TLS server certificate used to authenticate the session in which the information was obtained contained a DNS-ID identifier (subjectAltName of dNSName type [RFC5280]) or a CN-ID (CN attribute from subject name, [RFC6125]) specifying a dNSName matching either the domain used to name the section, or the domain portion of the email address used to name the section.
A redirect AQRY response that does not meet the above criteria (i.e neither any DNS-ID nor the CN-ID from the server's certificate matches the domain name of the address presented to AQRY) MAY be used to identify redirect servers. However, the above certificate checks MUST be applied when consulting redirect servers and determining the trustworthiness of their results. In other words, while it's acceptable for the mail exchangers for a mail domain that are listed in DNS to not have certificates that match that mail domain, it's not acceptable for the redirect AQRY servers for that mail domain to have have certificates that match that mail domain.
A Submission server implementing the AQPX extension MUST evaluate the trustworthiness of each named object in the response and only return those sections which are verified to be trustworthy according to the above rule.
A Submission client using the AQPX extension MUST follow the certificate checking rules in [I-D.ietf-uta-email-tls-certs].
(this section requires elaboration.)
(to be written)
(to be written)
XXX
XXX
XXX
This section is not normative.