Internet X.509 Public Key Infrastructure -- HTTP Transfer for the Certificate Management Protocol (CMP)

Introduction The Certificate Management Protocol (CMP) requires a well- defined transfer mechanism to enable End Entities (EEs), Registration Authorities (RAs), and Certification Authorities (CAs) to pass PKIMessage sequences between them. The first version of the CMP specification included a brief description of a simple transfer protocol layer on top of TCP. Its features were simple transfer-level error handling and a mechanism to poll for outstanding PKI messages. Additionally, it was mentioned that PKI messages could also be conveyed using file-, E-mail-, and HTTP-based transfer, but those were not specified in detail. The current version of the CMP specification incorporated its own polling mechanism, and thus the need for a transfer protocol providing this functionality vanished. The remaining features CMP requires from its transfer protocols are connection and error handling. Before this document was published as an RFC, the draft version underwent drastic changes during the long-lasting work process. The so-called "Direct TCP-Based Management Protocol" specified in was enhanced, and at some point a version existed where this protocol was again transferred over HTTP. As both approaches proved to be needless and cumbersome, implementers preferred to use plain HTTP transfer following or . This document now reflects that by exclusively describing HTTP as the transfer protocol for CMP. The usage of HTTP for transferring CMP messages exclusively uses the POST method for requests, effectively tunneling CMP over HTTP. While this is generally considered bad practice and should not be emulated, there are good reasons to do so for transferring CMP. HTTP is used as it is generally easy to implement and it is able to traverse network borders utilizing ubiquitous proxies. Most importantly, HTTP is already commonly used in existing CMP implementations. Other HTTP request methods, such as GET, are not used because PKI management operations can only be triggered using CMP's PKI messages, which need to be transferred using a POST request. With its status codes, HTTP provides needed error reporting capabilities. General problems on the server side, as well as those directly caused by the respective request, can be reported to the client. As CMP implements a transaction ID, identifying transactions spanning over more than just a single request/response pair, the statelessness of HTTP is not blocking its usage as the transfer protocol for CMP messages.

Conventions Used in This Document 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 .

HTTP-Based Protocol For direct interaction between two entities, where a reliable transport protocol like TCP is available, HTTP SHOULD be utilized for conveying CMP messages.

HTTP Versions Implementations MUST support HTTP/1.0 and SHOULD support HTTP/1.1 .

Persistent Connections HTTP persistent connections allow multiple interactions to take place on the same HTTP connection. However, neither HTTP nor the protocol specified in this document are designed to correlate messages on the same connection in any meaningful way; persistent connections are only a performance optimization. In particular, intermediaries can do things like mix connections from different clients into one "upstream" connection, terminate persistent connections, and forward requests as non-persistent requests, etc. As such, implementations MUST NOT infer that requests on the same connection come from the same client (e.g., for correlating PKI messages with ongoing transactions); every message is to be evaluated in isolation.

General Form A DER-encoded PKIMessage is sent as the entity-body of an HTTP POST request. If this HTTP request is successful, the server returns the CMP response in the body of the HTTP response. The HTTP response status code in this case MUST be 200; other "Successful 2xx" codes MUST NOT be used for this purpose. HTTP responses to pushed CMP Announcement messages (i.e., CA Certificate Announcement, Certificate Announcement, Revocation Announcement, and Certificate Revocation List (CRL) Announcement) utilize the status codes 201 and 202 to identify whether the received information was processed. While "Redirection 3xx" status codes MAY be supported by implementations, clients should only be enabled to automatically follow them after careful consideration of possible security implications. As described in , "301 Moved Permanently" could be misused for permanent denial of service. All applicable "Client Error 4xx" or "Server Error 5xx" status codes MAY be used to inform the client about errors.

Media Type The Internet Media Type "application/pkixcmp" MUST be set in the HTTP Content-Type header field when conveying a PKIMessage.

Communication Workflow In CMP, most communication is initiated by the EEs where every CMP request triggers a CMP response message from the CA or RA. The CMP Announcement messages described in are an exception. Their creation may be triggered by certain events or done on a regular basis by a CA. The recipient of the Announcement only replies with an HTTP status code acknowledging the receipt or indicating an error, but not with a CMP response. If the receipt of an HTTP request is not confirmed by receiving an HTTP response, it MUST be assumed that the transferred CMP message was not successfully delivered to its destination.

HTTP Request-URI The Request-URI is formed as specified in . A server implementation MUST handle Request-URI paths, with or without a trailing slash, as identical. An example of a Request-Line and a Host header field in an HTTP/1.1 header, sending a CMP request to a server, located in the "/cmp" path of the host "example.com", would be POST /cmp HTTP/1.1 Host: example.com or in the absoluteURI form POST http://example.com/cmp/ HTTP/1.1 Host: example.com

Pushing of Announcements A CMP server may create event-triggered announcements or generate them on a regular basis. It MAY utilize HTTP transfer to convey them to a suitable recipient. In this use case, the CMP server acts as an HTTP client, and the recipient needs to utilize an HTTP server. As no request messages are specified for those announcements, they can only be pushed to the recipient. If an EE wants to poll for a potential CA Key Update Announcement or the current CRL, a PKI Information Request using a General Message as described in Appendix E.5 of can be used. When pushing Announcement messages, PKIMessage structures are sent as the entity-body of an HTTP POST request. Suitable recipients for CMP announcements might, for example, be repositories storing the announced information, such as directory services. Those services listen for incoming messages, utilizing the same HTTP Request-URI scheme as defined in . The following PKIMessages are announcements that may be pushed by a CA. The prefixed numbers reflect ASN.1 numbering of the respective element. CMP Announcement messages do not require any CMP response. However, the recipient MUST acknowledge receipt with an HTTP response having an appropriate status code and an empty body. When not receiving such a response, it MUST be assumed that the delivery was not successful. If applicable, the sending side MAY try sending the Announcement again after waiting for an appropriate time span. If the announced issue was successfully stored in a database or was already present, the answer MUST be an HTTP response with a "201 Created" status code and an empty message body. In case the announced information was only accepted for further processing, the status code of the returned HTTP response MAY also be "202 Accepted". After an appropriate delay, the sender may then try to send the Announcement again and may repeat this until it receives a confirmation that it has been successfully processed. The appropriate duration of the delay and the option to increase it between consecutive attempts should be carefully considered. A receiver MUST answer with a suitable 4xx or 5xx HTTP error code when a problem occurs.

HTTP Considerations While all defined features of the HTTP protocol are available to implementations, they SHOULD keep the protocol utilization as simple as possible. For example, there is no benefit in using chunked Transfer-Encoding, as the length of an ASN.1 sequence is known when starting to send it. There is no need for the clients to send an "Expect" request-header field with the "100-continue" expectation and wait for a "100 Continue" status as described in Section 8.2.3 of . The CMP payload sent by a client is relatively small, so having extra messages exchanged is inefficient, as the server will only seldom reject a message without evaluating the body.

Implementation Considerations Implementors should be aware that implementations might exist that use a different approach for transferring CMP over HTTP, because this document has been under development for more than a decade. Further, implementations based on earlier drafts of this document might use an unregistered "application/pkixcmp-poll" MIME type.

Security Considerations The following aspects need to be considered by implementers and users:

There is the risk for denial-of-service attacks through resource consumption by opening many connections to an HTTP server. Therefore, idle connections should be terminated after an appropriate timeout; this may also depend on the available free resources. After sending a CMP Error Message, the server should close the connection, even if the CMP transaction is not yet fully completed.
Without being encapsulated in effective security protocols, such as Transport Layer Security (TLS) , there is no integrity protection at the HTTP protocol level. Therefore, information from the HTTP protocol should not be used to change state of the transaction.
Client users should be aware that storing the target location of an HTTP response with the "301 Moved Permanently" status code could be exploited by a man-in-the-middle attacker trying to block them permanently from contacting the correct server.
If no measures to authenticate and protect the HTTP responses to pushed Announcement messages are in place, their information regarding the Announcement's processing state may not be trusted. In that case, the overall design of the PKI system must not depend on the Announcements being reliably received and processed by their destination.
CMP provides inbuilt integrity protection and authentication. The information communicated unencrypted in CMP messages does not contain sensitive information endangering the security of the PKI when intercepted. However, it might be possible for an eavesdropper to utilize the available information to gather confidential technical or business critical information. Therefore, users of the HTTP transfer for CMP might want to consider using HTTP over TLS according to or virtual private networks created, for example, by utilizing Internet Protocol Security according to . Compliant implementations MUST support TLS with the option to authenticate both server and client.

IANA Considerations The IANA has already registered the MIME media type "application/pkixcmp" for identifying CMP sequences due to an request made in connection with . No further action by the IANA is necessary for this document or any anticipated updates.

Acknowledgments Amit Kapoor and Ronald Tschlaer were the original authors of this document, and their version focused on the so-called "TCP-Based Management Protocol", which has been removed from this document. Their contact data, as originally stated by them, is as follows: The authors gratefully acknowledge the contributions of various members of the IETF PKIX working group and the ICSA CA-talk mailing list (a list solely devoted to discussing CMP interoperability efforts). By providing ideas, giving hints, and doing invaluable review work, the following alphabetically listed individuals have significantly contributed to this document: