A Framework to tackle Spam and Unwanted Communication for Internet Telephony

The problem of Spam for Internet Telephony (SPIT) is an imminent challenge and only the combination of several techniques can provide a framework for dealing with unwanted communication attempts. provides four core recommendations that need to be considered for a SPIT solution, namely Strong Identity White Lists Solve the Introduction Problem Don't Wait Until its Too Late This document illustrates how existing building blocks can be put together to form a solution to deal with SPIT. New building blocks can be added to provide more efficient SPIT handling, since there is no single solution that provides 100 % protection. The main purpose of this document is largely to define a model of internal device processing, protocol interfaces, and terminology to illustrate a way in which SPIT prevention techniques can be added in a seamless fashion. We focus only on the most important solution components and consider many other aspects either outside the scope of this work (see ) and postpone them for future work.

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 RFC 2119 .

The framework considered in this document assumes that an end user uploads authorization policies to a SIP proxy of its VoIP provider. That VoIP provider enforces those authorization policies. This separation allows the end user to delegate some authorization decisions to the VoIP provider. below shows the interaction between the end host and a SIP proxy belonging to its VoIP provider. The end user, in the role of a recipient for communication attempts, may upload authorization policies. The entity that writes the rules is referred as Rule Maker. The Rule Maker does not necessarily need to be recipient of the communication but it could instead be entity that acts on behalf of him or her. Note that a combination is possible as well where different entities contribute to the set of rules. These policies are processed by corresponding module within the SIP proxy, called Authorization Engine, that interacts with the message routing functionality. A part of the rule set might, however, also be created automatically as part of interactive interactions (e.g., monitoring ongoing communication attempts).

|+-------+ v # | * | | | | ||Msg. | +-----------+| Authorization * | +--------+ | ||Routing| | Authz. || Policy (explicit) * | ^ o | ||Engine |<->| Engine |<#################+ * | * o | |+-------+ +-----------+| # * | * o | +-^--*--^-----------------+ # v | * o | o * o +-------------+ | * o | o * o | | | * +oooo|oooo+ * +ooooooooooooooooooooooooooooo>| Recipient/ | | +**************************************************>| Rule Maker | | | +-------------+ | | | | | +-------------------Domain Boundary-----------------------+ Legend: oooo: SIP message interaction ****: Protocol Interaction for authorizing the message sender ####: Management of authorization policies ]]> Assume that an arbitrary sender transmits a message towards the recipients URI that finally hits the SIP proxy on the recipients side. Information provided within that message are used as input to the rule evaluation. Any part of the message may serve as input to the evaluation process but for practical reasons only a few selected fields do most of the work. In this document, we argue that the authenticated identity of the sender is the most valuable information item. In the future, when standardization progresses then, for example, reputation information obtained from social networks may offer additional input to the authorization process. The protocol interaction for authorizing the message sender refers to the ability of the recipient or the proxy to interact with the sender to request authorization. The request for authorization is a pull model whereby the proxy or the recipient challenges the sender (e.g., via hash cash , or SIP payment , or Completely Automated Public Turing Test to Tell Computers and Humans Apart (CAPTCHA) based robot challenges ) for authorization. SIP Identity on the other hand realizes as push model whereby authentication information is pushed towards the recipient. shows this integration step. The conditions part of the rule offer a mechanisms to incrementally extend the overall framework with new SPIT prevention solution components. Depending on the rule evaluation the message may be rerouted to another entity, such as an answering machine, to the recipient, rejected or other actions are triggered. The latter aspect is particularly interesting since it allows further solution components to be executed.

| |----------------> Additional | | Spam marked msg Msg fields | Authorization | -------------->| Engine |----------------> Other SPIT | | Re-routed msg Prevention | | Components | |----------------> -------------->+---------------+ Forwarded to | | original recipient | | <-----------+ +----------->|| Politely blocked Blocked ]]> Note that some traffic analysis and consequently some form of content filtering (e.g., of MESSAGEs) can be applied locally within the VoIP provider's domain also under the control of the end user. For example, consider a Voice over IP provider that wants to utilize a statistical analysis tool for Spam prevention. It is not necessary to standardized the algorithms; the impact for the authorization policies is mainly the ability to allow the Rule Maker to enable or to disable the usage of these statistical techniques for SPIT filtering and potentially to map the output of the analysis process to value range from 0 (i.e., the message is not classified as Spam) and 100 (i.e., the message was classified as Spam). Conveying Spam marking is proposed in , in and in . A Rule Maker may decide to act with an appropriate action on a certain level of Spam marking. In a minimalistic SPIT framework only authenticated identities in combination with authorization policies are used. This should serve as a starting point for future work. SIP Identity is assumed to be used to provide the receiver of a communication attempt with the authenticated identity. SIP Identity is a reasonable simple specification and does not rely on a huge amount of infrastructure support. Note: SIP Identity is comparable to DomainKeys Identified Mail (DKIM) used for associating a "responsible" identity with an email message and provides a means of verifying that the association is legitimate. When the white lists are stored and managed only at the end host then the authorization policies and the protocol to modify the policies do not need to be standardized since they are purely implementation specific details. Even if the authorization policies are managed centrally or some amount of policy enforcement is done by trusted intermediaries then still there might not be a need to standardize an authorization policy language if the policies can be modified via a webpage. This type of policy handling is done in many cases today already for various applications. Unfortunately, this approach tends to become cumbersome to manage for end users and therefore it is better to hide a lot of policy details from the end user itself and to make use of context information, for example, address books and authorization policies available already created for presence based systems. In some cases the above-described approach is not sufficient whereby it is necessary to define a standardized protocol, for example, if policies are used by different entities, created and modified in an automatic way and when multiple entities manipulate policies (potentially on behalf of the person affected by the policies), e.g., the user may have multiple devices. An example solution for authorization policies providing Spam prevention capabilities are described in with the requirements detailed in . The white list needs to be created somehow and hence there is an introduction problem. discusses this aspect in more details.

When communication takes place then at least three types of groups can be identified.

People in this group communicate only with the peers in their group. They do not appreciate communication attempts from outside. Communication is possible only for people within this list. Here is an example of a closed group: Consider parents that do not want their children from getting contacted by strangers. Hence, they may want to create a white list containing the identifies of known friends, parents and other relatives on behalf of their kids. The usage of authorization policies for usage with closed groups is straight forward. The introduction problem is also not considered very large given that the identities are typically known.

In a semi-open environment all members of the same group are allowed to get in contact with everyone else (e.g., for example in a company environment). For members outside the company the communication patters depend on the role of the person (e.g., standardization people, sales people, etc.) and on the work style of the person. For this category we distinguish a number of (non-spam) message sources based on their characteristics: "friends" or "acquaintances", i.e., those we have communicated with before. strangers, divided into 'interesting' and 'uninteresting'. The latter are messages from people that someone does not care to have a conversation with or respond to, at least at that particular moment. Strangers can be defined by individual names or whole domains. A special class of 'stranger' messages are transaction-related communications, such as Instant Messaging or automated calls from an airline or shipping company. One way to deal with the introduction problem is to make use of techniques like hash cash or Completely Automated Public Turing Test to Tell Computers and Humans Apart (CAPTCHA) based robot challenges . Alternatively, a communication attempt may also be forwarded to an answering maschine or alternative ways of establishing the initial interaction may be proposed. The usage of authorization policies for usage with Semi-Open Groups is challenging but is considered manageable.

People in this type of group are not allowed to limit communication attempts. Help desks, certain people in governmental agencies, banks, insurance companies, etc. For open groups a solution for providing SPIT prevention is far more complicated. Consider a person working on a customer support helpdesk. Ideally, they would like to receive only calls from friendly customers (although the motivation for calling is most likely a problem) and the topic of the calls only relates to problems they are able to solve. Without listening to the caller they will have a hard time to know whether the call could be classified as SPIT. Another extreme case is a Public Safety Answering Point where emergency service personell is not allowed to reject calls either. Many SPIT prevention techniques might not be applicable since blocking callers is likely not possible and applying other techniques, such as turing tests, might not be ideal in an case of open groups. Statistical analysis may be helpful in some cases to deliver additional information about the calling party. Social networks might provide similar techniques based on reputation based systems.

Based on the discussions regarding communication patters and groups the following observations can be made: A single person very likely has many roles and they may have an impact on the communication patterns. For example, consider a person who is working in a company but also want to be available for family members. The context in which a person is may change at any time. For example, a person might be available for family members while at work except during an important meeting where communication attempts may be rejected. Switching a context has an impact for reachability and the means for communicating with a specific recipient, based on enabled rule sets. From an authorization policy point of view it is important to be able to express a sphere, i.e., the state a user is in and to switch between different spheres easily by thereby switching to a different rule set.

An important aspect in the usage of authorization policies is to assist the user when creating policies. Ideally, the policies should be established automatically. Below, there are a couple of examples to illustrate the idea given that these aspects are largely implementation issues: It must be possible for the proxy to automatically add addresses on outbound messages and calls to the rule set. This approach is similar to stateful packet filtering firewalls where outbound packets establish state at the firewall to allow inbound packets to traverse it again. Already available information in the address book can be used for building the policy rules there is quite likely already a relationship available with these persons existent. A large amount of email is non-personal, automated communication, such as newsletters, confirmations and legitimate advertisements. These are often tagged as spam by content filters. This type of correspondence is usually initiated by a transaction over the web, such as a purchase or signing up for a service. , for example, defines an HTTP header for conveying future correspondence addresses that can be integrated in the rule set.

This section describes the necessary building blocks that are necessary to tie the framework together. We will describe two different environments, namely one where rule enforcement happens at the end host and another one where a trusted network intermediary performs the actions.

SIP Identity is mandatory to implement at the end host and used to determine the authenticated identity of the sending side. Authorization policies are purely implementation specific matter.

Since a purely end host based rule enforcement suffers from a number of drawbacks, rule enforcement by a trusted intermediary is also offered.

SIP Identity or a corresponding mechanism is mandatory to implement at the trusted intermediary (e.g., the immediate VoIP provider) and it determines the authenticated identity of the sending side. Authorization Policies based on the Common Policy framework , as extended in for the purpose of SPIT prevention, are mandatory to implement at the end host side and at the trusted intermediary. The implementation of the rule evaluation engine might only be necessary on the trusted VoIP proxy. Harmonization with the work done for presence authorization , which is based on Common Policy , can be accomplished and is highly desirable. XML Configuration Access Protocol (XCAP) is used to create, modify and delete authorization policies and is mandatory to implement at the end host side and at the trusted intermediary.

An important property is incremental deployment of additional solution components that can be added and used when they become available. This section aims to illustrate how the extensibility is accomplished, based on an example. Consider a VoIP provider that provides authorization policies that provide the following functionality equivalent to the Common Policy framework, i.e., identity-based, sphere and validity based conditions initially. For actions only 'redirection' and 'blocking' is provided. In our example we give this basic functionality the AUID 'new-spit-policy-example' with the namespace 'urn:ietf:params:xml:ns:new-spit-policy-example'. When a client queries the capabilities of a SIP proxy in the VoIP providers network using XCAP the following exchange may take place.

new-spit-policy-example xcap-caps urn:ietf:params:xml:ns:xcap-caps urn:ietf:params:xml:ns:spit-policy urn:ietf:params:xml:ns:common-policy ]]> As shown in the example above, Common Policy and the example SPIT extension is implemented and the client can upload rules according to the definition of the rule set functionality. Later, when the VoIP provider updates the functionality of authorization policies as more sophisticated mechanisms become available and get implemented the functionality of the authorization policy engine is enhanced with, for example, hashcash and the ability to perform statistical analysis of signaling message. The latter functionality comes with the ability to mark messages are Spam and the ability for end users to enable/disable this functionality. We use the namespaces 'urn:ietf:params:xml:ns:hashcash' and 'urn:ietf:params:xml:ns:statistical-analysis' for those. A end user could now make use of these new functions and a capability query of the SIP proxy would provide the following response.

spit-policy xcap-caps hashcash statistical-analysis urn:ietf:params:xml:ns:spit-policy urn:ietf:params:xml:ns:common-policy urn:ietf:params:xml:ns:hashcash urn:ietf:params:xml:ns:statistical-analysis ]]> New SPIT handling functionality may extend condition, actions and/or transformation elements of a rule.

This document does not propose to distribute the user's authorization policies to other VoIP providers nor is the configuration of policies at SIP proxies other than the trusted user's VoIP provider necessary. Furthemore, if blocking or influencing of the message processing is executed by the VoIP provider then they have to be explicitly enabled by the end user. Blocking of messages, even if it is based on "super-clever" machine learning techniques often introduces unpredictability. Legal norms from fields of law can take regulative effects in the context of SPIT processing, such as constitutional law, data protection law, telecommunication law, teleservices law, criminal law, and possibly administrative law. See, for example, , and . For example, it is mandatory to pass full control of SPIT filtering to the end user, as this minimises legal problems. An overview about regulatory aspects can be found in .

This section shows an example whereby we consider a user Bob@company-example.com that writes (most likely via a nice user interface) the following policies. We use a high-level language to show the main idea of the policies.

THEN REDIRECT sip:voicebox@company-example.com RULE 5: IF THEN block ]]> At some point in time Bob uploads his policies to the SIP proxy at his VoIP providers SIP proxy.

>>> [Editor's Note: In a future version an XML example policy document will be listed here.] ]]> When BoB receives a call from his friends, alice@foo.example and tony@bar.example.com, then all the rules related to the spit policy are checked. Only the first rule (rule 1) matches and is applied. Thus, the call is forwarded without any further checks based on Rule 1. The rules assume that the authenticated identity of the caller has been verified. When Bob receives a call from a co-worker, Charlie@company-example.com, Rule 2 is applied since the domain part in the rule matches the domain part of Charlie's identity. Now, when Bob receives a contact from an unknown user, called Mallice in this example. Rule 3 indicates that an extended return-routability test using hashcash is used with the call being redirected to Bob's voicebox afterwards. This exchange is shown in .

|Puzzle: work=15; | | |pre="VgVGYixbRg0mdSwTY3YIfCBuAAA="; | | 419 with Puzzle |image="NhhMQ2l7SE0VBmZFKksUC19ia04="; | | |value=160 | |<-------------------------| | | | | | ACK | | |------------------------->| | | |Puzzle: work=0; | | |pre="VgVGYixbRg0mdSwTY3YIfCBuYmg="; | | |image="NhhMQ2l7SE0VBmZFKksUC19ia04=" | | INVITE with Solution |value=160 | |------------------------->| INVITE | | |------------------------------------->| | | | | | 180 Ringing | | 180 Ringing |<-------------------------------------| |<-------------------------| | | | 200 OK | | 200 OK |<-------------------------------------| |<-------------------------| | | | ACK | |---------------------------------------------------------------->| | | | ]]> Depending on the outcome of the exchange the call is forwarded to a mailbox sip:voicebox@company-example.com (in case Malory returned the correct solution, see Rule 4) or blocked in case an incorrect response was provided. It might be quite easy to see how this rule set can be extended to support other SPIT handling mechanisms as well (e.g., CAPTCHAs, SIP Pay, etc.).

This document aims to describe a framework for addressing Spam for Internet Telephony (SPIT) in order to make it simple for users to influence the behavior of SIP message routing with an emphasis on SPIT prevention. The framework relies on three building blocks, namely SIP Identity, authorization policies based on Common Policy and Presence Authorization Policy, and XCAP. As a high-level overview, the framework allows the user to control end-to-end connectivity at the SIP message routing level whereby the glue that lets all parts fit together is based on authorization policies. Several other solution components can be developed independently and can be plugged into the framework as soon as available. It must be avoided to introduce Denial of Service attacks against the recipient by misguiding him or her to install authorization policies that allow senders to bypass the policies although that was never intended by the recipient. Additionally, it must not be possible by extensions to the authorization policy framework to create policies to block legitimate senders or to stall the processing of the authorization policy engine.

We would like to thank Jeremy Barkan, Dan York, Alexey Melnikov, Thomas Schreck, Eva Leppanen, Cullen Jennings, Marit Hansen and Markus Hansen for their review comments to a pre-00 version. Jeremy Barkan, Eva Leppanen, Michaela Greiler, Joachim Charzinski, Saverio Niccolini, Albert Caruana, and Juergen Quittek for their comments to the 00 version.

Key words for use in RFCs to Indicate Requirement Levels Harvard University &RFC4474; &RFC4745; &RFC4825; &I-D.ietf-sipping-spam; &I-D.ietf-simple-presence-rules; &I-D.jennings-sip-hashcash; &I-D.wing-sipping-spam-score; &I-D.ietf-sip-consent-framework; &I-D.ietf-dkim-overview; &I-D.tschofenig-sipping-spit-policy; &I-D.schwartz-sipping-spit-saml; &I-D.shacham-http-corr-uris; &I-D.jennings-sipping-pay; &I-D.froment-sipping-spit-requirements; &I-D.niccolini-sipping-feedback-spit; &I-D.tschofenig-sipping-captcha; Developing a Legally Compliant Reachability Management System as a Countermeasure against SPIT, Third Annual VoIP Security Workshop, Berlin, available at https://tepin.aiki.de/blog/uploads/spit-al.pdf Bundesnetzagentur: Eckpunkte der regulatorischen Behandlung von Voice over IP (VoIP), available at http://www.bundesnetzagentur.de/media/archive/3186.pdf 70. Konferenz der Datenschutzbeauftragten des Bundes und der Länder: Entschließung Telefonieren mit Internettechnologie (Voice over IP – VoIP), available at http://www.datenschutzzentrum.de/material/themen/press e/20051028-dsbk-voip.htm Working Party 29 Opinion 2/2006 on privacy issues related to the provision of email screening services, WP 118, available at http://ec.europa.eu/justice_home/fsj/privacy/docs/wpdocs/2006/wp118_en.pdf

We consider the following aspects outside the scope of this document: Mechanisms to publish SPIT causing parties on the Internet, i.e., information about domains that create SPIT. Determining the source of unwanted traffic in real-time. Pushing packet filters and authorization policies towards the SPIT sending domain.