Unfortunately, not all XMPP entities are well-behaved -- they may send spam of various kinds, harrass chat rooms, generate large amounts of traffic, etc. Currently, if an XMPP entity (the "attacker") sends abusive stanzas to another XMPP entity (the "victim"), there is no way for the victim or the victim's server to inform the attacker's server that the attacker is generating abusive traffic. In current practice, the victim's server may have no choice but to terminate the server-to-server connection rather than continue to accept the abusive traffic.

This situation is far from desirable. Therefore, this specification defines several protocol functions that can help to discourage abuse on the XMPP network:

1. A method by which a victim or the victim's server can send an abuse report to the attacker's server.
2. A method by which the attacker's server can inform other servers (or dedicated reporting services) that the IP address from which the attacker connected may be compromised.
3. A method by which the victim's server can inform other servers (or dedicated reporting services) that the attacker's server may be a rogue server.
4. An application-specific stanza error condition that can be combined with the standard ¬acceptable; stanza error condition to inform the attacker's server that a particular XMPP stanza is considered abusive.
5. An application-specific stream error condition that can be combined with the standard &policy; stream error condition to inform the attacker's server about the reason for termination of an XML stream (if necessary).

An abuse report shall be sent in an IQ stanza of type "set" containing an <abuse/> element qualified by the 'urn:xmpp:tmp:abuse' namespace &NSNOTE;.

The children of the <abuse/> element are as follows:

• A <condition/> element containing a child element that specifies the abuse condition (see list of conditions).
• Optionally a <description/> that provides a natural-language description of the abusive incident(s).
• A <jid/> element that specifies the JID of the abusive sender.
• Optionally a <pointer/> element that specifies a URI at which further information can be found (e.g., a web page that contains a server log).
• Optionally a <stanzas/> element that contains one or more XMPP &PRESENCE;, &IQ;, or &MESSAGE; stanzas.

This specification intentionally does not define exactly what constitutes abuse, since "abuse is in the eye of the beholder". However, the following machine-readable conditions are defined as children of the <reason/> element.

Note: The foregoing list of conditions is not exhaustive. The list may be augmented or otherwise modified in a future version of this specification as a result of implementation and deployment experience.

An abuse report can be generated by the victim or the victim's server. If the report is generated by the victim's server, the report MUST be sent to the attacker's server. If the report is generated by the victim, the report MUST be sent to the attacker's server and SHOULD also be sent to the victim's server as well (since the victim may not know if the attacker's server is a rogue server).

The following rules and guidelines apply to the act of reporting abuse.

1. The recipient SHOULD NOT report the abuse stanza to a server or service until it determines that the server or service supports the Abuse Reporting protocol (see the Discovering Support section of this document).

2. The recipient SHOULD report the abuse to the suspected abuser's server.

3. If the recipient's home server (i.e., the server with which it has a registered account or other trust relationship) supports the abuse Reporting protocol, the recipient SHOULD also report the offending stanza to its own server.

4. If the recipient's home server does not support the Abuse Reporting protocol, the recipient SHOULD report the abuse stanza to one or more dedicated abuse reporting services if available.

5. The recipient SHOULD NOT report the abuse stanza to the suspected abuser.

Upon receiving the abuse report, the attacker's server MUST proceed as follows.

If the sending server does not understand the abuse reporting protocol, it MUST return a &unavailable; error.

If the JID contained in the abuse report does not exist at the attacker's server, it MUST return an ¬found; error.

Otherwise, if the sending server accepts the abuse report, it MUST return an IQ stanza of type "result".

This specification does not define how a sending server shall behave when it receives an abuse report. In general it is expected that the sending server (1) will notify the human administrators of the server in some implementation-specific or deployment-specific fashion, and (2) may employ the abuse report in an automated fashion (e.g., as input to a rate-limiting algorithm, reputation system, or decision about temporarily suspending the privileges of the sending entity based on JabberID or IP address).

The following rules and guidelines apply to the processing of an abuse report:

1. Before processing the report, the processor MAY respond to the report by sending a challenge to the sender (e.g., using &xep0158;) in order to make sure that the sender is not sending spurious reports or otherwise abusing the abuse reporting system.
2. The processor MUST add the report to a list or database of pending abuse reports.
3. If the suspected abuser is not already on the processor's list of known abusers, the processor SHOULD use the report in determining whether the suspected abuser is an actual abuser (see next section).
4. The processor MAY report the abuse stanza to one or more dedicated abuse reporting services if available.
5. The processor MAY report the abuse stanza to other servers it trusts.
6. The processor MAY report the abuse stanza to the suspected abuser's server (if the server is otherwise trusted, i.e., is not considered a rogue server).
7. The processor SHOULD NOT report the abuse stanza to the suspected abuser.

Not all abuse reports are valid, and not all JIDs that send stanzas reported as abuse are abusers. Care must be taken in correctly determining if a suspected abuser is an actual abuser. The following rules apply:

1. The processor SHOULD NOT add a suspected abuser to its list of known abusers until it has received at least three (3) valid abuse reports with regard to that suspected abuser, or until it has independently verified the veracity of the report.

2. If the processor determines that the suspected abuser is an actual abuser, the processor:

   1. MUST add that JID to its list of known abusers.
   2. SHOULD add the abuser's IP address to its list of known bad IPs.
   3. MAY send an abuser report to abuse reporting services or reputation services.
   4. MAY send an abuser report to other servers it trusts.
   5. SHOULD NOT send an abuser report to the abuser.

If the attacker's server determines that the suspected attacker is an actual attacker, it might decide that both the JabberID and the IP address associated with the attacker's JabberID are compromised. If it does so, it SHOULD report its conclusions to appropriate other entities (e.g., trusted peer servers or dedicated abuse reporting services). The protocol is quite simple: include the JabberID and IP address as children of an <abuser/> element qualified by the 'urn:xmpp:tmp:abuse' namespace and send an IQ stanza of type "set" to the entity that shall receive the report. This protocol SHOULD NOT be used directly by victims of abusive stanzas and if an entity receives such a report from an end user it SHOULD ignore it. The following is an example:

If the victim's server determines that the attacker's server is a bad actor on the network (i.e., a rogue server), it SHOULD report its conclusions to appropriate other entities (e.g., trusted peer servers or dedicated abuse reporting services). This is done by including the domain name (and optionally IP address) of the rogue server in a <rogue/> element qualified by the 'urn:xmpp:tmp:abuse' namespace and send an IQ stanza of type "set" to the entity that shall receive the report. This protocol SHOULD NOT be used directly by victims of abusive stanzas and if an entity receives such a report from an end user it SHOULD ignore it. The following is an example:

The receiving server MAY report that a particular stanza is considered abusive. The stanza error condition MUST be ¬acceptable; and the error stanza MUST include an application-specific error condition of <abuse/> qualified by the 'urn:xmpp:tmp:abuse' &NSNOTE;. The <abuse/> element MUST include one or more <jid/> elements whose XML character data specifies the JID(s) of the abusive sender(s).

If the sending entity continues to generate abusive stanzas via the sending server, the receiving server MAY close the stream between the receiving server and the sending server. The stream error condition MUST be &policy; and the stream error MUST include an application-specific error condition of <abuse/> qualified by the 'urn:xmpp:tmp:abuse'. The <abuse/> element MUST include one or more <jid/> elements whose XML character data specifies the JID(s) of the abusive sender(s).

The receiving entity then SHOULD terminate the TCP connection between the receiving server and the sending server.

If a server supports the abuse reporting protocol, it MUST report that fact by including a service discovery feature of "urn:xmpp:tmp:abuse" &NSNOTE; in response to a &xep0030; information request:

This document requires no interaction with &IANA;.