Creation and Use of Email Feedback Reports: An Applicability Statement for the Abuse Reporting Format (ARF)
draft-ietf-marf-as-12

Abstract

RFC 5965 defines an extensible, machine-readable format intended for mail operators to report feedback about received email to other parties. This Applicability Statement describes common methods for utilizing this format for reporting both abuse and authentication failure events. Mailbox Providers of any size, mail sending entities, and end users can use these methods as a basis to create procedures that best suit them. Some related optional mechanisms are also discussed.

Status of this Memo

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 September 29, 2012.

Copyright Notice

Copyright (c) 2012 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.

Table of Contents

• 1. Introduction
• 1.1. Discussion
• 2. Definitions
• 3. Solicited and Unsolicited Reports
• 4. Creating and Sending Complaint-Based Solicited Abuse Reports
• 4.1. General Considerations
• 4.2. Where To Send Reports
• 4.3. What To Put In Reports
• 5. Receiving and Processing Complaint-Based Solicited Abuse Reports
• 5.1. General Considerations
• 5.2. What To Expect
• 5.3. What To Do
• 6. Generating and Handling Unsolicited Abuse Reports
• 6.1. General Considerations
• 6.2. When To Generate Reports
• 6.3. Where To Send Reports
• 6.4. What To Put In Reports
• 6.5. What To Do With Reports
• 7. Generating Automatic Authentication Failure Reports
• 8. IANA Considerations
• 9. Security Considerations
• 9.1. In Other Documents
• 9.2. Forgeries
• 9.3. Amplification Attacks
• 9.4. Automatic Generation
• 9.5. Reporting Multiple Incidents
• 10. Acknowledgements
• 11. References
• 11.1. Normative References
• 11.2. Informative References
• Authors' Addresses

1. Introduction

The Abuse Reporting Format (ARF) was initially developed for two very specific use cases. Initially, it was intended to be used for reporting feedback between large email operators, or from large email operators to end user network access operators, any of whom could be presumed to have automated abuse-handling systems. Secondarily, it is used by those same large mail operators to send those same reports to other entities, including those involved in sending bulk email for commercial purposes. In either case, the reports would be triggered by direct end user action such as clicking on a "report spam" button in their email client.

Though other uses for the format defined in [RFC5965] have been discussed (and may be documented similarly in the future), abuse remains the primary application, with a small amount of adoption of extensions that enable authentication failure reporting.

This Applicability Statement provides direction for using the Abuse Reporting Format (ARF) in both contexts. It also includes some statements about the use of ARF in conjunction with other email technologies.

The purpose for reporting abusive messages is to stop recurrences. The methods described in this document focus on automating abuse reporting as much as practical, so as to minimize the work of a site's abuse team. There are further reasons why abuse feedback generation is worthwhile, such as instruction of mail filters or reputation trackers, or to initiate investigations of particularly egregious abuses. These other applications are not discussed in this memo.

Further introduction to this topic may be found in [RFC6449].

1.1. Discussion

[RFC Editor: please remove this section prior to publication.]

This document is being discussed within the IETF MARF Working Group, on the marf@ietf.org mailing list.

2. Definitions

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], and are intended to replace the Requirement Levels described in Section 3.3 of [RFC2026].

Some of the terminology used in this document is taken from [RFC5598].

"Mailbox Provider" refers to an organization that accepts, stores, and offers access to [RFC5322] messages ("email messages") for end users. Such an organization has typically implemented SMTP ([RFC5321]), and might provide access to messages through IMAP ([RFC3501]), POP ([RFC1939]), a proprietary interface designed for HTTP ([RFC2616]), or a proprietary protocol.

3. Solicited and Unsolicited Reports

The original application of [RFC5965], and still by far the most common, is when two mail systems make a private agreement to exchange abuse reports, usually reports due to recipients manually reporting messages as spam. We refer to these as solicited reports.

Other uses for ARF involve such reports sent between parties that don't know each other. These unsolicited reports are sent without prior arrangement between the parties as to the context and meaning of the reports, so the constraints on how these unsolicited reports need to be structured such that the reports generated are likely to be useful to the recipient, to what address(es) they can usefully be sent, what issues the can be used to report, and how they can be handled by the receiver of the report are very different.

4. Creating and Sending Complaint-Based Solicited Abuse Reports

[The numbered items in these subsections are not intended to be in a paricular sequence. The numbers are here during document development to make it easier to identify the items for discussion, and will be removed before publication.]

The following subsections include statements applicable when establishing an abuse report relationship and generating reports in that context.

4.1. General Considerations

1. A Mailbox Provider receives reports of abusive or unwanted mail from its users, most often by providing a "report spam" button (or similar nomenclature) in the MUA. The method of transferring this message and any associated metadata from the MUA to the Mailbox Provider's ARF processing system is not defined by any standards document, but is discussed further in Section 3.2 of [RFC6449]. Policy concerns related to the collection of this data are discussed in Section 3.4 of [RFC6449].
2. To implement the recommendations of this memo, the reports are formatted per [RFC5965], and transmitted as an email message ([RFC5322]), typically using SMTP ([RFC5321]).
3. Ongoing maintenance of an ARF processing system is discussed in Section 3.6 of [RFC6449].

4.2. Where To Send Reports

1. The Mailbox Provider SHOULD NOT send reports to addresses that have not explicitly requested them. The process whereby such parties may request the reports is discussed in Section 3.5 of [RFC6449].

4.3. What To Put In Reports

1. The reports SHOULD use "Feedback-Type: abuse", but can use other types as appropriate to the nature of the abuse being reported. However, the Mailbox Provider generating the reports needs to understand that the operator receiving the reports might not treat different feedback types any differently.
2. The following fields are optional in [RFC5965], but SHOULD be used in this context when their corresponding values are available: Original-Mail-From, Arrival-Date, Source-IP, Original-Rcpt-To. Other optional fields can be included, as the implementer feels is appropriate.
3. User-identifiable data MAY be obscured as described in [I-D.IETF-MARF-REDACTION].

5. Receiving and Processing Complaint-Based Solicited Abuse Reports

[The numbered items in these subsections are not intended to be in a paricular sequence. The numbers are here during document development to make it easier to identify the items for discussion, and will be removed before publication.]

The following subsections include statements applicable when receiving abuse reports in the context of an established abuse report feedback relationship.

5.1. General Considerations

1. At the time this document is being written, for the use cases described here, mail operators need to proactively request a stream of ARF reports from Mailbox Providers. Recommendations for preparing to make that request are discussed in Section 4.1 of [RFC6449].
2. Furthermore, this document assumes that mail operators exchange abuse reports formatted per ARF [RFC5965] as email messages [RFC5322] over SMTP [RFC5321]. These and other types of email messages that can be received are discussed in Section 4.2 of [RFC6449].
3. Mail operators need to consider the idea of automating report processing. Discussion of this can be found in Section 4.4 of [RFC6449].

5.2. What To Expect

1. An automated report processing system MUST accept all Feedback-Types defined in [RFC5965] or extensions to it. However, report receivers cannot assume that Mailbox Providers will make use of any Feedback-Type other than "abuse", except with prior specific knowledge. Additional analysis might be required to separate different types of abuse reports after receipt.
2. Implementers SHOULD NOT expect all Mailbox Providers to include the same optional fields.
3. Reports may have been subjected to redaction of user-identifiable data as described in [I-D.IETF-MARF-REDACTION]. That document also discusses the handling of such reports. This technique is also discussed in Section 4.4 of [RFC6449].

5.3. What To Do

1. Actions that mail operators might take upon receiving a report (or multiple reports) are discussed in Section 4.3 of [RFC6449].

6. Generating and Handling Unsolicited Abuse Reports

[The numbered items in these subsections are not intended to be in a paricular sequence. The numbers are here during document development to make it easier to identify the items for discussion, and will be removed before publication.]

The following subsections include statements applicable when sending or receiving reports outside of the context of an established abuse report feedback relationship.

6.1. General Considerations

1. A report generator MUST provide a way for a report recipient to request no further reports be sent to that address and MAY provide a way for recipients to change the address(es) to which reports about them are sent. Details of such mechanisms are outside of the scope of [RFC5965], [RFC6449], and this document.
2. Message authentication is generally a good idea, but it is especially important to encourage credibility of and thus response to unsolicited reports. Therefore, as with any other message, report generators sending unsolicited reports SHOULD send reports that will pass SPF and/or DKIM checks.

6.2. When To Generate Reports

1. Handling of unsolicited reports has a significant cost to the receiver. Senders of unsolicited reports, especially those sending large volumes of them automatically, need to be aware of this and do all they reasonably can to avoid sending reports that cannot be used as a basis for action by the recipient, whether this is due to the report being sent about an incident that is not abuse-related, the report being sent to an email address that won't result in action, or the content or format of the report being hard for the recipient to read or use.
2. Systems SHOULD NOT report all mail sent from a particular sender merely because some of it is determined to be abusive.
3. Mechanical reports of mail that "looks like" spam, based solely on the results of inline content analysis tools, SHOULD NOT be sent since, because of their subjective nature, they are unlikely to provide a basis for the recipient to take action. Complaints generated by end users about mail that is determined by them to be abusive, or mail delivered to "spam trap" or "honeypot" addresses, are far more likely to be accurate.
4. If a report generator applies SPF to arriving messages, a report SHOULD NOT be generated to the RFC5321.MailFrom domain if the SPF evaluation produced a "Fail", "SoftFail", "TempError" or "PermError" report, as no reliable assertion or assumption can be made that use of the domain was authorized. A valid exception would be specific knowledge that the SPF result is not definitive for that domain under those circumstances (for example, a message that is also DKIM-signed by the same domain, and that signature validates).

6.3. Where To Send Reports

1. MUAs SHOULD NOT generate abuse reports directly to entities merely because they were found in the message, or by queries to WHOIS ([RFC3912]) or other heuristic means. Rather, the MUA needs to signal, by some means, the mailbox provider to which it connects to trigger generation of such a report.
2. Report generators SHOULD NOT send reports to recipients that are uninvolved or only peripherally involved. For example, they SHOULD NOT send reports to the operator of every Autonomous System in the path between the apparent originating system and the operator generating the report. Instead, they need to send reports to recipients that are both responsible for the messages and are able to do something about them.
3. Deciding where to send an unsolicited report will typically rely on heuristics. Abuse addresses in WHOIS records of the IP address relaying the subject message and/or of the domain name found in the results of a PTR ("reverse lookup") query on that address are likely reasonable candidates, as is the abuse@domain role address (see [RFC2142]) of related domains. Unsolicited reports SHOULD NOT be sent to email addresses that are not clearly intended to handle abuse reports. Legitimate candidates include those found in WHOIS records or on a web site that either are explicitly described as an abuse contact, or are of the form "abuse@domain".
4. Where an abusive message is authenticated using a domain-level authentication technology such as DKIM ([RFC6376]) or SPF ([RFC4408]), the domain that has been verified by the authentication mechanism is often a reasonable candidate for receiving feedback about the message. For DKIM, though, while the authenticated domain has some responsibility for the mail sent, it can be a poor contact point for abuse issues (for example, it could represent the message's author but not its sender, it could identify the bad actor responsible for the message, or it could refer to a domain that cannot receive mail at all).
5. Often, unsolicited reports will have no meaning if sent to abuse reporting addresses belonging to the abusive parties themselves. In fact, it is possible that such reports might reveal information about complainants. Reports SHOULD NOT be sent to such addresses if they can be identified beforehand, except where the abusive party is known to be responsive to such reports.

6.4. What To Put In Reports

1. Reports SHOULD use "Feedback-Type: abuse", but can use other types as appropriate. However, the Mailbox Provider generating the reports cannot assume that the operator receiving the reports will treat different Feedback-Types differently.
2. Reports SHOULD include the following optional fields whenever their corresponding values are available and applicable to the report: Original-Mail-From, Arrival-Date, Source-IP, Original-Rcpt-To. Other optional fields can be included, as the implementer feels is appropriate.
3. Experience suggests use of ARF is advisable in most contexts. Automated recipient systems can handle abuse reports sent in ARF format at least as well as any other format such as plain text, with or without a copy of the message attached. That holds even for systems that did not request ARF format reports, provided that reports are generated with use by recipients not using automated ARF parsing in mind. Anyone sending unsolicited reports in ARF format can legitimately presume that some recipients will only be able to access the human readable (first, text/plain) part of it, and SHOULD include all information needed also in this part. Further, they SHOULD ensure that the report is readable when viewed as plain text, to give low-end ticketing systems as much assistance as possible. Finally, they need to be aware that the report could be discarded or ignored due to failure to take these steps in the most extreme cases.

6.5. What To Do With Reports

1. Receivers of unsolicited reports can take advantage of the standardized parts of the ARF format to automate processing. Independent of the sender of the report, they can improve processing by separating valid from invalid reports by, for example, looking for references to IP address ranges, domains, and mailboxes for which the recipient organization is responsible in the copy of the reported message, and by correlating multiple reports of similar messages to identify bulk senders.
2. Per Section 4.4 of [RFC6449], a network service provider MAY use ARF data for automated forwarding of feedback messages to the originating customer.
3. Published abuse mailbox addresses SHOULD NOT reject messages not in the ARF format, as generation of ARF messages can occasionally be unavailable or not applicable.
4. Although [RFC6449] suggests that replying to feedback is not useful, in the case of receipt of ARF reports where no feedback arrangement has been established, a non-automated reply might be desirable to indicate what action resulted from the complaint, heading off more severe filtering by the report generator. In addition, using an address that cannot receive replies precludes any requests for additional information, and increases the likelihood that further reports will be discarded or blocked. Thus, a report generator sending unsolicited reports SHOULD NOT generate reports for which a reply cannot be received. Where an unsolicited report results in the establishment of contact with a responsible and responsive party, this can be saved for future complaint handling and possible establishment of a formal (solicited) feedback arrangement. See Section 3.5 of [RFC6449] for a discussion of establishment of feedback arrangements.

7. Generating Automatic Authentication Failure Reports

[These numbered items are not intended to be in a paricular sequence. The numbers are here during document development to make it easier to identify the items for discussion, and will be removed before publication.]

There are some cases where report generation is caused by automation rather than user request. A specific example of this is reporting, using the ARF format (or extensions to it), of messages that fail particular message authentication checks. Examples of this include [I-D.IETF-MARF-DKIM-REPORTING] and [I-D.IETF-MARF-SPF-REPORTING]. The considerations presented below apply in those cases.

The applicability statement for this use case is somewhat smaller as many of the issues associated with abuse reports are not relevant to reports about authentication failures.

1. Selection of the recipient(s) for reports that are automatically generated MUST be done based on data provided by the report recipient, and MUST NOT be done heuristically. Therefore these reports are always solicited, such as the mechanisms defined in the examples listed above.
2. If the message under evaluation by the Verifier is an ARF ([RFC5965]) message, a report MUST NOT be automatically generated.
3. When sending a new report via SMTP, it is necessary to construct the message so as to avoid amplification attacks, deliberate or otherwise. The envelope sender address of the report needs to be chosen so that these reports will not generate mail loops. Similar to Section 2 of [RFC3464], the envelope sender address of the report needs to be chosen to ensure that no feedback reports will be issued in response to the report itself. Therefore, when an SMTP transaction is used to send a report, the MAIL FROM command SHOULD use the NULL reverse-path, i.e., "MAIL FROM:<>".
4. Reports SHOULD use "Feedback-Type: auth-failure", but MAY use other types as appropriate. However, the Mailbox Provider generating the reports cannot assume that the operator receiving the reports will treat different Feedback-Types differently.
5. These reports SHOULD include the following optional fields, although they are optional in [RFC5965], whenever their corresponding values are available: Original-Mail-From, Arrival-Date, Source-IP, Original-Rcpt-To. Other optional fields can be included, as the implementer feels is appropriate.

8. IANA Considerations

[RFC Editor: please remove this section prior to publication.]

This document has no IANA actions.

9. Security Considerations

9.1. In Other Documents

Implementers are strongly urged to review, at a minimum, the Security Considerations sections of [RFC5965] and [RFC6449].

9.2. Forgeries

Report generators that relay user complaints directly, rather than by reference to a stored message (e.g., IMAP or POP), could be duped into sending a complaint about a message that the complaining user never actually received, as an attack on the purported originator of the falsified message. Report generators need to be resilient to such attack methods.

Also, these reports may be forged as easily as ordinary Internet electronic mail. User agents and automatic mail handling facilities (such as mail distribution list exploders) that wish to make automatic use of reports of any kind should take appropriate precautions to minimize the potential damage from denial-of-service attacks.

Perhaps the simplest means of mitigating this threat is to assert that these reports should themselves be signed with something like DKIM or authorized by SPF. On the other hand, if there is a problem with the DKIM infrastructure at the Verifier, signing DKIM failure reports may produce reports that aren't trusted or even accepted by their intended recipients. Similar issues could exist with SPF evaluation. Use of both technologies can mitigate this risk to a degree.

9.3. Amplification Attacks

Failure to comply with the recommendations regarding selection of the envelope sender can lead to amplification denial-of-service attacks.

9.4. Automatic Generation

ARF ([RFC5965]) reports have historically been generated individually as a result of some kind of human request, such as someone clicking a "Report Abuse" button in a mail reader. In contrast, the mechanisms described in some extension documents (i.e., [I-D.IETF-MARF-DKIM-REPORTING] and [I-D.IETF-MARF-SPF-REPORTING]) are focused around automated reporting. This obviously implies the potential for much larger volumes or frequency of messages, and thus greater mail system load (both for report generators and report receivers).

Those mechanisms are primarily intended for use in generating reports to aid implementers of DKIM ([RFC6376]), ADSP ([RFC5617]), and SPF ([RFC4408]), and other related protocols during development and debugging. They are not generally intended for prolonged forensic use, specifically because of these load concerns. However, extended use is possible by ADMDs that want to keep a close watch for fraud or infrastructure problems. It is important to consider the impact of doing so on both report generators and the requesting ADMDs.

A sender requesting these reports can cause its mail servers to be overwhelmed if it sends out signed messages whose signatures fail to verify for some reason, provoking a large number of reports from report generators. Similarly, a report generator could be overwhelmed by a large volume of messages requesting reports whose signatures fail to validate, as those now need to send reports back to the signer.

Limiting the rate of generation of these messages may be appropriate but threatens to inhibit the distribution of important and possibly time-sensitive information.

In general ARF feedback loop terms, it is often suggested that report generators only create these (or any) ARF reports after an out-of-band arrangement has been made between two parties. These extension mechanisms then become ways to adjust parameters of an authorized abuse report feedback loop that is configured and activated by private agreement rather than starting to send them automatically based solely on data found in the messages, which may have unintended consequences.

9.5. Reporting Multiple Incidents

If it is known that a particular host generates abuse reports upon certain incidents, an attacker could forge a high volume of messages that will trigger such a report. The recipient of the report could then be innundated with reports. This could easily be extended to a distributed denial-of-service attack by finding a number of report-generating servers.

The incident count referenced in ARF ([RFC5965]) provides a limited form of mitigation. The host generating reports can elect to send reports only periodically, with each report representing a number of identical or nearly-identical incidents. One might even do something inverse-exponentially, sending reports for each of the first ten incidents, then every tenth incident up to 100, then every 100th incident up to 1000, etc., until some period of relative quiet after which the limitation resets.

The use of this for "nearly-identical" incidents in particular causes a degradation in reporting quality, however. If for example a large number of pieces of spam arrive from one attacker, a reporting agent could decide only to send a report about a fraction of those messages. While this averts a flood of reports to a system administrator, the precise details of each incident are similarly not sent.

Other rate limiting provisions might be considered, including detection of a temporary failure response from the report destination and thus halting report generation to that destination for some period, or simply imposing or negotiating a hard limit on the number of reports to be sent to a particular receiver in a given time frame.

10. Acknowledgements

The author and editor wish to thank Steve Atkins, John Levine, Shmuel Metz, S. Moonesamy, and Alessandro Vesely for their contributions to this memo.

All of the Best Practices referenced by this document are found in [RFC6449], written within the Collaboration Committee of the Messaging Anti-Abuse Working Group (MAAWG).

Finally, the original author wishes to thank the doctors and staff at the University of Texas MD Anderson Cancer Center for doing what they do.

11. References

11.1. Normative References

11.2. Informative References

Authors' Addresses