DNS Push Notifications

Protocol Operation The DNS Push Notification protocol is a session-oriented protocol and makes use of DNS Stateful Operations (DSO) . For details of the DSO message format, refer to the DNS Stateful Operations specification . Those details are not repeated here. DNS Push Notification clients and servers MUST support DSO. A single server can support DNS Queries, DNS Updates, and DNS Push Notifications (using DSO) on the same TCP port. A DNS Push Notification exchange begins with the client discovering the appropriate server, using the procedure described in , and then making a TLS/TCP connection to it. After making the TLS/TCP connection to the server, a typical DNS Push Notification client will then immediately issue a DSO Keepalive operation to establish the DSO session and request a session timeout and/or keepalive interval longer than the 15-second default values, but this is not required. A DNS Push Notification client MAY issue other requests on the session first, and only issue a DSO Keepalive operation later if it determines that to be necessary. Sending either a DSO Keepalive operation or a Push Notification subscription request over the TLS/TCP connection to the server signals the client's support of DSO and serves to establish a DSO session. In accordance with the current set of active subscriptions, the server sends relevant asynchronous Push Notifications to the client. Note that a client MUST be prepared to receive (and silently ignore) Push Notifications for subscriptions it has previously removed, since there is no way to prevent the situation where a Push Notification is in flight from server to client while the client's UNSUBSCRIBE message canceling that subscription is simultaneously in flight from client to server.

Discovery The first step in establishing a DNS Push Notification subscription is to discover an appropriate DNS server that supports DNS Push Notifications for the desired zone. The client begins by opening a DSO session to its normal configured DNS recursive resolver and requesting a Push Notification subscription. This connection is made to TCP port 853, the default port for DNS over TLS . If the request for a Push Notification subscription is successful, and the recursive resolver doesn't already have an active subscription for that name, type, and class, then the recursive resolver will make a corresponding Push Notification subscription on the client's behalf. Results received are relayed to the client. This is closely analogous to how a client sends a normal DNS query to its configured DNS recursive resolver, which, if it doesn't already have appropriate answer(s) in its cache, issues an upstream query to satisfy the request. In many contexts, the recursive resolver will be able to handle Push Notifications for all names that the client may need to follow. Use of VPN tunnels and Private DNS can create some additional complexity in the client software here; the techniques to handle VPN tunnels and Private DNS for DNS Push Notifications are the same as those already used to handle this for normal DNS queries. If the recursive resolver does not support DNS over TLS, or supports DNS over TLS but is not listening on TCP port 853, or supports DNS over TLS on TCP port 853 but does not support DSO on that port, then the DSO session establishment will fail . If the recursive resolver does support DSO on TCP port 853 but does not support Push Notification subscriptions, then when the client attempts to create a subscription, the server will return the DSO error code DSOTYPENI (11). In some cases, the recursive resolver may support DSO and Push Notification subscriptions but may not be able to subscribe for Push Notifications for a particular name. In this case, the recursive resolver should return SERVFAIL to the client. This includes being unable to establish a connection to the zone's DNS Push Notification server or establishing a connection but receiving a non-success response code. In some cases, where the client has a pre-established trust relationship with the owner of the zone (that is not handled via the usual mechanisms for VPN software), the client may handle these failures by contacting the zone's DNS Push Notification server directly. In any of the cases described above where the client fails to establish a DNS Push Notification subscription via its configured recursive resolver, the client should proceed to discover the appropriate server for direct communication. The client MUST also determine on which TCP port the server is listening for connections, which need not be, and often is not, TCP port 53 (traditionally used for conventional DNS) or TCP port 853 (traditionally used for DNS over TLS). The discovery algorithm described here is an iterative algorithm, which starts with the full name of the record to which the client wishes to subscribe. Successive SOA queries are then issued, trimming one label each time, until the closest enclosing authoritative server is discovered. There is also an optimization to enable the client to take a "short cut" directly to the SOA record of the closest enclosing authoritative server in many cases.

The client begins the discovery by sending a DNS query to its local resolver, with record type SOA for the record name to which it wishes to subscribe. As an example, suppose the client wishes to subscribe to PTR records with the name _ipp._tcp.headoffice.example.com (to discover Internet Printing Protocol (IPP) printers being advertised in the head office of Example Company). The client begins by sending an SOA query for _ipp._tcp.headoffice.example.com to the local recursive resolver. The goal is to determine the server that is authoritative for the name _ipp._tcp.headoffice.example.com. The closest enclosing DNS zone containing the name _ipp._tcp.headoffice.example.com could be example.com, or headoffice.example.com, or _tcp.headoffice.example.com, or even _ipp._tcp.headoffice.example.com. The client does not know in advance where the closest enclosing zone cut occurs, which is why it uses the iterative procedure described here to discover this information.
If the requested SOA record exists, it will be returned in the Answer Section with a NOERROR response code, and the client has succeeded in discovering the information it needs. (This language is not placing any new requirements on DNS recursive resolvers. This text merely describes the existing operation of the DNS protocol .)
If the requested SOA record does not exist, the client will get back a NOERROR/NODATA response or an NXDOMAIN/Name Error response. In either case, the local resolver would normally include the SOA record for the closest enclosing zone of the requested name in the Authority Section. If the SOA record is received in the Authority Section, then the client has succeeded in discovering the information it needs. (This language is not placing any new requirements on DNS recursive resolvers. This text merely describes the existing operation of the DNS protocol regarding negative responses .)
If the client receives a response containing no SOA record, then it proceeds with the iterative approach. The client strips the leading label from the current query name, and if the resulting name has at least two labels in it, then the client sends an SOA query for that new name and processing continues at step 2 above, repeating the iterative search until either an SOA is received or the query name consists of a single label, i.e., a Top-Level Domain (TLD). In the case of a single-label name (TLD), this is a network configuration error, which should not happen, and the client gives up. The client may retry the operation at a later time of the client's choosing, such as after a change in network attachment.
Once the SOA is known (by virtue of being seen either in the Answer Section or in the Authority Section), the client sends a DNS query with type SRV for the record name _dns&nbhy;push&nbhy;tls._tcp.<zone>, where <zone> is the owner name of the discovered SOA record.
If the zone in question is set up to offer DNS Push Notifications, then this SRV record MUST exist. (If this SRV record does not exist, then the zone is not correctly configured for DNS Push Notifications as specified in this document.) The SRV target contains the name of the server providing DNS Push Notifications for the zone. The port number on which to contact the server is in the SRV record port field. The address(es) of the target host MAY be included in the Additional Section, however, the address records SHOULD be authenticated before use as described in and in the specification for using DNS-Based Authentication of Named Entities (DANE) TLSA Records with SRV Records , if applicable.
More than one SRV record may be returned. In this case, the priority and weight values in the returned SRV records are used to determine the order in which to contact the servers for subscription requests. As described in the SRV specification , the server with the lowest priority is first contacted. If more than one server has the same priority, the weight indicates the weighted probability that the client should contact that server. Higher weights have higher probabilities of being selected. If a server is not willing to accept a subscription request, or is not reachable within a reasonable time, as determined by the client, then a subsequent server is to be contacted.

Each time a client makes a new DNS Push Notification subscription, it SHOULD repeat the discovery process in order to determine the preferred DNS server for that subscription at that time. If a client already has a DSO session with that DNS server, the client SHOULD reuse that existing DSO session for the new subscription; otherwise, a new DSO session is established. The client MUST respect the DNS TTL values on records it receives while performing the discovery process and store them in its local cache with this lifetime (as it will generally do anyway for all DNS queries it performs). This means that, as long as the DNS TTL values on the authoritative records are set to reasonable values, repeated application of the discovery process can be completed practically instantaneously by the client, using only locally stored cached data.

DNS Push Notification SUBSCRIBE After connecting, and requesting a longer idle timeout and/or keepalive interval if necessary, a DNS Push Notification client then indicates its desire to receive DNS Push Notifications for a given domain name by sending a SUBSCRIBE request to the server. A SUBSCRIBE request is encoded in a DSO message . This specification defines a DSO Primary TLV for DNS Push Notification SUBSCRIBE Requests (DSO Type Code 0x0040). DSO messages with the SUBSCRIBE TLV as the Primary TLV are permitted in TLS early data, provided that the precautions described in are followed. The entity that initiates a SUBSCRIBE request is by definition the client. A server MUST NOT send a SUBSCRIBE request over an existing session from a client. If a server does send a SUBSCRIBE request over a DSO session initiated by a client, this is a fatal error and the client MUST forcibly abort the connection immediately. Each SUBSCRIBE request generates exactly one SUBSCRIBE response from the server. The entity that initiates a SUBSCRIBE response is by definition the server. A client MUST NOT send a SUBSCRIBE response. If a client does send a SUBSCRIBE response, this is a fatal error and the server MUST forcibly abort the connection immediately.

SUBSCRIBE Request A SUBSCRIBE request begins with the standard DSO 12-byte header , followed by the SUBSCRIBE Primary TLV. A SUBSCRIBE request is illustrated in . The MESSAGE ID field MUST be set to a unique value that the client is not using for any other active operation on this DSO session. For the purposes here, a MESSAGE ID is in use on this session if either the client has used it in a request for which it has not yet received a response, or if the client has used it for a subscription that it has not yet canceled using UNSUBSCRIBE. In the SUBSCRIBE response, the server MUST echo back the MESSAGE ID value unchanged. The other header fields MUST be set as described in the DSO specification. The DNS OPCODE field contains the OPCODE value for DNS Stateful Operations (6). The four count fields must be zero, and the corresponding four sections must be empty (i.e., absent). The DSO-TYPE is SUBSCRIBE (0x0040). The DSO-LENGTH is the length of the DSO-DATA that follows, which specifies the name, type, and class of the record(s) being sought.

SUBSCRIBE Request HEADER | ANCOUNT (MUST BE ZERO) | | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | | NSCOUNT (MUST BE ZERO) | | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | | ARCOUNT (MUST BE ZERO) | / +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ / | DSO-TYPE = SUBSCRIBE (0x0040) | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | DSO-LENGTH (number of octets in DSO-DATA) | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ \ \ NAME \ \ +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | | TYPE | > DSO-DATA +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | | CLASS | / +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ /]]> The DSO-DATA for a SUBSCRIBE request MUST contain exactly one NAME, TYPE, and CLASS. Since SUBSCRIBE requests are sent over TCP, multiple SUBSCRIBE DSO request messages can be concatenated in a single TCP stream and packed efficiently into TCP segments. If accepted, the subscription will stay in effect until the client cancels the subscription using UNSUBSCRIBE or until the DSO session between the client and the server is closed. SUBSCRIBE requests on a given session MUST be unique. A client MUST NOT send a SUBSCRIBE message that duplicates the name, type and class of an existing active subscription on that DSO session. For the purpose of this matching, the established DNS case insensitivity for US-ASCII letters applies (e.g., "example.com" and "Example.com" are the same). If a server receives such a duplicate SUBSCRIBE message, this is a fatal error and the server MUST forcibly abort the connection immediately. DNS wildcarding is not supported. That is, an asterisk character ("*") in a SUBSCRIBE message matches only a literal asterisk character ("*") in a name and nothing else. Similarly, a CNAME in a SUBSCRIBE message matches only a CNAME record with that name in the zone and no other records with that name. A client may SUBSCRIBE to records that are unknown to the server at the time of the request (providing that the name falls within one of the zone(s) the server is responsible for), and this is not an error. The server MUST NOT return NXDOMAIN in this case. The server MUST accept these requests and send Push Notifications if and when matching records are found in the future. If neither TYPE nor CLASS are ANY (255), then this is a specific subscription to changes for the given name, type, and class. If one or both of TYPE or CLASS are ANY (255), then this subscription matches all types and/or all classes as appropriate. NOTE: A little-known quirk of DNS is that in DNS QUERY requests, QTYPE and QCLASS 255 mean "ANY", not "ALL". They indicate that the server should respond with ANY matching records of its choosing, not necessarily ALL matching records. This can lead to some surprising and unexpected results, where a query returns some valid answers, but not all of them, and makes QTYPE = 255 (ANY) queries less useful than people sometimes imagine. When used in conjunction with SUBSCRIBE, TYPE 255 and CLASS 255 should be interpreted to mean "ALL", not "ANY". After accepting a subscription where one or both of TYPE or CLASS are 255, the server MUST send Push Notification Updates for ALL record changes that match the subscription, not just some of them.

SUBSCRIBE Response A SUBSCRIBE response begins with the standard DSO 12-byte header . The QR bit in the header is set indicating it is a response. The header MAY be followed by one or more optional Additional TLVs such as a Retry Delay Additional TLV. A SUBSCRIBE response is illustrated in . The MESSAGE ID field MUST echo the value given in the MESSAGE ID field of the SUBSCRIBE request. This is how the client knows which request is being responded to. The other header fields MUST be set as described in the DSO specification. The DNS OPCODE field contains the OPCODE value for DNS Stateful Operations (6). The four count fields must be zero, and the corresponding four sections must be empty (i.e., absent). A SUBSCRIBE response message MUST NOT include a SUBSCRIBE TLV. If a client receives a SUBSCRIBE response message containing a SUBSCRIBE TLV, then the response message is processed but the SUBSCRIBE TLV MUST be silently ignored.

SUBSCRIBE Response HEADER | ANCOUNT (MUST BE ZERO) | | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | | NSCOUNT (MUST BE ZERO) | | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | | ARCOUNT (MUST BE ZERO) | / +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ / ]]> In the SUBSCRIBE response, the RCODE indicates whether or not the subscription was accepted. Supported RCODEs are as follows: SUBSCRIBE Response Codes

Mnemonic	Value	Description
NOERROR	0	SUBSCRIBE successful.
FORMERR	1	Server failed to process request due to a malformed request.
SERVFAIL	2	Server failed to process request due to a problem with the server.
NOTIMP	4	Server does not implement DSO.
REFUSED	5	Server refuses to process request for policy or security reasons.
NOTAUTH	9	Server is not authoritative for the requested name.
DSOTYPENI	11	SUBSCRIBE operation not supported.

This document specifies only these RCODE values for SUBSCRIBE Responses. Servers sending SUBSCRIBE Responses SHOULD use one of these values. Note that NXDOMAIN is not a valid RCODE in response to a SUBSCRIBE Request. However, future circumstances may create situations where other RCODE values are appropriate in SUBSCRIBE Responses, so clients MUST be prepared to accept and handle SUBSCRIBE Responses with any other nonzero RCODE error values. If the server sends a nonzero RCODE in the SUBSCRIBE response, that means:

the client is (at least partially) misconfigured, or
the server resources are exhausted, or
there is some other unknown failure on the server.

In any case, the client shouldn't retry the subscription to this server right away. If multiple SRV records were returned as described in , a subsequent server MAY be tried immediately. If the client has other successful subscriptions to this server, these subscriptions remain even though additional subscriptions may be refused. Neither the client nor the server is required to close the connection, although either end may choose to do so. If the server sends a nonzero RCODE, then it SHOULD append a Retry Delay Additional TLV to the response specifying a delay before the client attempts this operation again. Recommended values for the delay for different RCODE values are given below. These recommended values apply both to the default values a server should place in the Retry Delay Additional TLV and the default values a client should assume if the server provides no Retry Delay Additional TLV.

For RCODE = 1 (FORMERR), the delay may be any value selected by the implementer. A value of five minutes is RECOMMENDED to reduce the risk of high load from defective clients.
For RCODE = 2 (SERVFAIL), the delay should be chosen according to the level of server overload and the anticipated duration of that overload. By default, a value of one minute is RECOMMENDED. If a more serious server failure occurs, the delay may be longer in accordance with the specific problem encountered.
For RCODE = 4 (NOTIMP), which occurs on a server that doesn't implement DNS Stateful Operations , it is unlikely that the server will begin supporting DSO in the next few minutes, so the retry delay SHOULD be one hour. Note that in such a case, a server that doesn't implement DSO is unlikely to place a Retry Delay Additional TLV in its response, so this recommended value in particular applies to what a client should assume by default.
For RCODE = 5 (REFUSED), which occurs on a server that implements DNS Push Notifications but is currently configured to disallow DNS Push Notifications, the retry delay may be any value selected by the implementer and/or configured by the operator.
If the server being queried is listed in a _dns&nbhy;push&nbhy;tls._tcp.<zone> SRV record for the zone, then this is a misconfiguration, since this server is being advertised as supporting DNS Push Notifications for this zone, but the server itself is not currently configured to perform that task. Since it is possible that the misconfiguration may be repaired at any time, the retry delay should not be set too high. By default, a value of 5 minutes is RECOMMENDED.
For RCODE = 9 (NOTAUTH), which occurs on a server that implements DNS Push Notifications but is not configured to be authoritative for the requested name, the retry delay may be any value selected by the implementer and/or configured by the operator.
If the server being queried is listed in a _dns&nbhy;push&nbhy;tls._tcp.<zone> SRV record for the zone, then this is a misconfiguration, since this server is being advertised as supporting DNS Push Notifications for this zone, but the server itself is not currently configured to perform that task. Since it is possible that the misconfiguration may be repaired at any time, the retry delay should not be set too high. By default, a value of 5 minutes is RECOMMENDED.
For RCODE = 11 (DSOTYPENI), which occurs on a server that implements DSO but doesn't implement DNS Push Notifications, it is unlikely that the server will begin supporting DNS Push Notifications in the next few minutes, so the retry delay SHOULD be one hour.
For other RCODE values, the retry delay should be set by the server as appropriate for that error condition. By default, a value of 5 minutes is RECOMMENDED.

For RCODE = 9 (NOTAUTH), the time delay applies to requests for other names falling within the same zone. Requests for names falling within other zones are not subject to the delay. For all other RCODEs, the time delay applies to all subsequent requests to this server. After sending an error response, the server MAY allow the session to remain open, or MAY follow it with a DSO Retry Delay operation (using the Retry Delay Primary TLV) instructing the client to close the session as described in the DSO specification. Clients MUST correctly handle both cases. Note that the DSO Retry Delay operation (using the Retry Delay Primary TLV) is different to the Retry Delay Additional TLV mentioned above.

DNS Push Notification Updates Once a subscription has been successfully established, the server generates PUSH messages to send to the client as appropriate. In the case that the answer set was already non-empty at the moment the subscription was established, an initial PUSH message will be sent immediately following the SUBSCRIBE Response. Subsequent changes to the answer set are then communicated to the client in subsequent PUSH messages. A client MUST NOT send a PUSH message. If a client does send a PUSH message, or a PUSH message is sent with the QR bit set indicating that it is a response, this is a fatal error and the receiver MUST forcibly abort the connection immediately.

PUSH Message A PUSH unidirectional message begins with the standard DSO 12-byte header , followed by the PUSH Primary TLV. A PUSH message is illustrated in . In accordance with the definition of DSO unidirectional messages, the MESSAGE ID field MUST be zero. There is no client response to a PUSH message. The other header fields MUST be set as described in the DSO specification . The DNS OPCODE field contains the OPCODE value for DNS Stateful Operations (6). The four count fields must be zero, and the corresponding four sections must be empty (i.e., absent). The DSO-TYPE is PUSH (0x0041). The DSO-LENGTH is the length of the DSO-DATA that follows, which specifies the changes being communicated. The DSO-DATA contains one or more change notifications. A PUSH Message MUST contain at least one change notification. If a PUSH Message is received that contains no change notifications, this is a fatal error and the client MUST forcibly abort the connection immediately. The change notification records are formatted similarly to how DNS Resource Records are conventionally expressed in DNS messages, as illustrated in , and are interpreted as described below. The TTL field holds an unsigned 32-bit integer . If the TTL is in the range 0 to 2,147,483,647 seconds (0 to 2³¹ - 1, or 0x7FFFFFFF), then a new DNS Resource Record with the given name, type, class, and RDATA is added. Type and class MUST NOT be 255 (ANY). If either type or class are 255 (ANY), this is a fatal error and the client MUST forcibly abort the connection immediately. A TTL of 0 means that this record should be retained for as long as the subscription is active and should be discarded immediately the moment the subscription is canceled. If the TTL has the value 0xFFFFFFFF, then the DNS Resource Record with the given name, type, class, and RDATA is removed. Type and class MUST NOT be 255 (ANY). If either type or class are 255 (ANY), this is a fatal error and the client MUST forcibly abort the connection immediately. If the TTL has the value 0xFFFFFFFE, then this is a 'collective' remove notification. For collective remove notifications, RDLEN MUST be zero, and consequently, the RDATA MUST be empty. If a change notification is received where TTL = 0xFFFFFFFE and RDLEN is not zero, this is a fatal error and the client MUST forcibly abort the connection immediately. There are three types of collective remove notification. For collective remove notifications:

If CLASS is not 255 (ANY) and TYPE is not 255 (ANY), then for the given name, this removes all records of the specified type in the specified class.
If CLASS is not 255 (ANY) and TYPE is 255 (ANY), then for the given name, this removes all records of all types in the specified class.
If CLASS is 255 (ANY), then for the given name, this removes all records of all types in all classes. In this case, TYPE MUST be set to zero on transmission and MUST be silently ignored on reception.

Summary of change notification types:

Remove all RRsets from a name in all classes:
TTL = 0xFFFFFFFE, RDLEN = 0, CLASS = 255 (ANY).
Remove all RRsets from a name in given class:
TTL = 0xFFFFFFFE, RDLEN = 0, CLASS gives class, TYPE = 255 (ANY).
Remove specified RRset from a name in given class:
TTL = 0xFFFFFFFE, RDLEN = 0,
CLASS and TYPE specify the RRset being removed.
Remove an individual RR from a name:
TTL = 0xFFFFFFFF,
CLASS, TYPE, RDLEN, and RDATA specify the RR being removed.
Add individual RR to a name:
TTL >= 0 and TTL <= 0x7FFFFFFF,
CLASS, TYPE, RDLEN, RDATA, and TTL specify the RR being added.

Note that it is valid for the RDATA of an added or removed DNS Resource Record to be empty (zero length). For example, an Address Prefix List Resource Record may have empty RDATA. Therefore, a change notification with RDLEN = 0 does not automatically indicate a remove notification. If RDLEN = 0 and TTL is in the range 0 to 0x7FFFFFFF, this change notification signals the addition of a record with the given name, type, class, and empty RDATA. If RDLEN = 0 and TTL = 0xFFFFFFFF, this change notification signals the removal specifically of that single record with the given name, type, class, and empty RDATA. If the TTL is any value other than 0xFFFFFFFF, 0xFFFFFFFE, or a value in the range 0 to 0x7FFFFFFF, then the receiver SHOULD silently ignore this particular change notification record. The connection is not terminated and other valid change notification records within this PUSH message are processed as usual. In the case where a single change affects more than one active subscription, only one PUSH message is sent. For example, a PUSH message adding a given record may match both a SUBSCRIBE request with the same TYPE and a different SUBSCRIBE request with TYPE = 255 (ANY). It is not the case that two PUSH messages are sent because the new record matches two active subscriptions. The server SHOULD encode change notifications in the most efficient manner possible. For example, when three AAAA records are removed from a given name, and no other AAAA records exist for that name, the server SHOULD send a "Remove specified RRset from a name in given class" PUSH message, not three separate "Remove an individual RR from a name" PUSH messages. Similarly, when both an SRV and a TXT record are removed from a given name, and no other records of any kind exist for that name in that class, the server SHOULD send a "Remove all RRsets from a name in given class" PUSH message, not two separate "Remove specified RRset from a name in given class" PUSH messages. For efficiency, when generating a PUSH message, rather than sending each change notification as a separate DSO message, a server SHOULD include as many change notifications as it has immediately available to send to that client, even if those change notifications apply to different subscriptions from that client. Conceptually, a PUSH message is a session-level mechanism, not a subscription-level mechanism. Once it has exhausted the list of change notifications immediately available to send to that client, a server SHOULD then send the PUSH message immediately rather than waiting speculatively to see if additional change notifications become available. For efficiency, when generating a PUSH message a server SHOULD use standard DNS name compression, with offsets relative to the beginning of the DNS message . When multiple change notifications in a single PUSH message have the same owner name, this name compression can yield significant savings. Name compression should be performed as specified in the Multicast DNS specification; namely, owner names should always be compressed, and names appearing within RDATA should be compressed for only the RR types listed below:

: NS, CNAME, PTR, DNAME, SOA, MX, AFSDB, RT, KX, RP, PX, SRV, NSEC

Servers may generate PUSH messages up to a maximum DNS message length of 16,382 bytes, counting from the start of the DSO 12-byte header. Including the two-byte length prefix that is used to frame DNS over a byte stream like TLS, this makes a total of 16,384 bytes. Servers MUST NOT generate PUSH messages larger than this. Where the immediately available change notifications are sufficient to exceed a DNS message length of 16,382 bytes, the change notifications MUST be communicated in separate PUSH messages of up to 16,382 bytes each. DNS name compression becomes less effective for messages larger than 16,384 bytes, so little efficiency benefit is gained by sending messages larger than this. If a client receives a PUSH message with a DNS message length larger than 16,382 bytes, this is a fatal error and the client MUST forcibly abort the connection immediately.

PUSH Message HEADER | ANCOUNT (MUST BE ZERO) | | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | | NSCOUNT (MUST BE ZERO) | | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | | ARCOUNT (MUST BE ZERO) | / +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ / | DSO-TYPE = PUSH (0x0041) | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | DSO-LENGTH (number of octets in DSO-DATA) | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ \ \ NAME \ \ +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | | TYPE | | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | | CLASS | | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | | TTL | | | (32-bit unsigned big-endian integer) | > DSO-DATA +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | | RDLEN (16-bit unsigned big-endian integer) | | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | \ RDATA (sized as necessary) \ | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | : NAME, TYPE, CLASS, TTL, RDLEN, RDATA : | : Repeated As Necessary : / +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ /]]> When processing the records received in a PUSH Message, the receiving client MUST validate that the records being added or removed correspond with at least one currently active subscription on that session. Specifically, the record name MUST match the name given in the SUBSCRIBE request, subject to the usual established DNS case-insensitivity for US-ASCII letters. For individual additions and removals, if the TYPE in the SUBSCRIBE request was not ANY (255), then the TYPE of the record must either be CNAME or match the TYPE given in the SUBSCRIBE request, and if the CLASS in the SUBSCRIBE request was not ANY (255), then the CLASS of the record must match the CLASS given in the SUBSCRIBE request. For collective removals, at least one of the records being removed must match an active subscription. If a matching active subscription on that session is not found, then that particular addition/removal record is silently ignored. The processing of other additions and removal records in this message is not affected. The DSO session is not closed. This is to allow for the unavoidable race condition where a client sends an outbound UNSUBSCRIBE while inbound PUSH messages for that subscription from the server are still in flight. The TTL of an added record is stored by the client. While the subscription is active the TTL is not decremented, because a change to the TTL would produce a new update. For as long as a relevant subscription remains active, the client SHOULD assume that when a record goes away, the server will notify it of that fact. Consequently, a client does not have to poll to verify that the record is still there. Once a subscription is canceled (individually, or as a result of the DSO session being closed), record aging for records covered by the subscription resumes and records are removed from the local cache when their TTL reaches zero.

DNS Push Notification UNSUBSCRIBE To cancel an individual subscription without closing the entire DSO session, the client sends an UNSUBSCRIBE message over the established DSO session to the server. The entity that initiates an UNSUBSCRIBE message is by definition the client. A server MUST NOT send an UNSUBSCRIBE message over an existing session from a client. If a server does send an UNSUBSCRIBE message over a DSO session initiated by a client, or an UNSUBSCRIBE message is sent with the QR bit set indicating that it is a response, this is a fatal error and the receiver MUST forcibly abort the connection immediately.

UNSUBSCRIBE Message An UNSUBSCRIBE unidirectional message begins with the standard DSO 12-byte header , followed by the UNSUBSCRIBE Primary TLV. An UNSUBSCRIBE message is illustrated in . In accordance with the definition of DSO unidirectional messages, the MESSAGE ID field MUST be zero. There is no server response to an UNSUBSCRIBE message. The other header fields MUST be set as described in the DSO specification. The DNS OPCODE field contains the OPCODE value for DNS Stateful Operations (6). The four count fields must be zero, and the corresponding four sections must be empty (i.e., absent). The DSO-TYPE is UNSUBSCRIBE (0x0042). The DSO-LENGTH field contains the value 2, the length of the 2-octet MESSAGE ID contained in the DSO-DATA. The DSO-DATA contains the value previously given in the MESSAGE ID field of an active SUBSCRIBE request. This is how the server knows which SUBSCRIBE request is being canceled. After receipt of the UNSUBSCRIBE message, the SUBSCRIBE request is no longer active. It is allowable for the client to issue an UNSUBSCRIBE message for a previous SUBSCRIBE request for which the client has not yet received a SUBSCRIBE response. This is to allow for the case where a client starts and stops a subscription in less than the round-trip time to the server. The client is NOT required to wait for the SUBSCRIBE response before issuing the UNSUBSCRIBE message. Consequently, it is possible for a server to receive an UNSUBSCRIBE message that does not match any currently active subscription. This can occur when a client sends a SUBSCRIBE request, which subsequently fails and returns an error code, but the client sent an UNSUBSCRIBE message before it became aware that the SUBSCRIBE request had failed. Because of this, servers MUST silently ignore UNSUBSCRIBE messages that do not match any currently active subscription.

DNS Push Notification RECONFIRM Sometimes, particularly when used with a Discovery Proxy , a DNS Zone may contain stale data. When a client encounters data that it believes may be stale (e.g., an SRV record referencing a target host+port that is not responding to connection requests), the client can send a RECONFIRM message to ask the server to re-verify that the data is still valid. For a Discovery Proxy, this causes it to issue new Multicast DNS queries to ascertain whether the target device is still present. How the Discovery Proxy causes these new Multicast DNS queries to be issued depends on the details of the underlying Multicast DNS implementation being used. For example, a Discovery Proxy built on Apple's dns_sd.h API responds to a DNS Push Notification RECONFIRM message by calling the underlying API's DNSServiceReconfirmRecord() routine. For other types of DNS server, the RECONFIRM operation is currently undefined and SHOULD result in a NOERROR response, but it need not cause any other action to occur. Frequent use of RECONFIRM operations may be a sign of network unreliability, or some kind of misconfiguration, so RECONFIRM operations MAY be logged or otherwise communicated to a human administrator to assist in detecting and remedying such network problems. If, after receiving a valid RECONFIRM message, the server determines that the disputed records are in fact no longer valid, then subsequent DNS PUSH Messages will be generated to inform interested clients. Thus, one client discovering that a previously advertised device (like a network printer) is no longer present has the side effect of informing all other interested clients that the device in question is now gone. The entity that initiates a RECONFIRM message is by definition the client. A server MUST NOT send a RECONFIRM message over an existing session from a client. If a server does send a RECONFIRM message over a DSO session initiated by a client, or a RECONFIRM message is sent with the QR bit set indicating that it is a response, this is a fatal error and the receiver MUST forcibly abort the connection immediately.

RECONFIRM Message A RECONFIRM unidirectional message begins with the standard DSO 12-byte header , followed by the RECONFIRM Primary TLV. A RECONFIRM message is illustrated in . In accordance with the definition of DSO unidirectional messages, the MESSAGE ID field MUST be zero. There is no server response to a RECONFIRM message. The other header fields MUST be set as described in the DSO specification. The DNS OPCODE field contains the OPCODE value for DNS Stateful Operations (6). The four count fields must be zero, and the corresponding four sections must be empty (i.e., absent). The DSO-TYPE is RECONFIRM (0x0043). The DSO-LENGTH is the length of the data that follows, which specifies the name, type, class, and content of the record being disputed. A DNS Push Notifications RECONFIRM message contains exactly one RECONFIRM Primary TLV. The DSO-DATA in a RECONFIRM Primary TLV MUST contain exactly one record. The DSO-DATA in a RECONFIRM Primary TLV has no count field to specify more than one record. Since RECONFIRM messages are sent over TCP, multiple RECONFIRM messages can be concatenated in a single TCP stream and packed efficiently into TCP segments. Note that this means that DNS name compression cannot be used between different RECONFIRM messages. However, when a client is sending multiple RECONFIRM messages this indicates a situation with serious network problems, and this is not expected to occur frequently enough that optimizing efficiency in this case is important. TYPE MUST NOT be the value ANY (255) and CLASS MUST NOT be the value ANY (255). DNS wildcarding is not supported. That is, an asterisk character ("*") in a RECONFIRM message matches only a literal asterisk character ("*") in a name and nothing else. Similarly, a CNAME in a RECONFIRM message matches only a CNAME record with that name in the zone and no other records with that name. Note that there is no RDLEN field, since the length of the RDATA can be inferred from DSO-LENGTH, so an additional RDLEN field would be redundant. Following the same rules as for PUSH messages, DNS name compression SHOULD be used within the RDATA of the RECONFIRM message, with offsets relative to the beginning of the DNS message .

RECONFIRM Message HEADER | ANCOUNT (MUST BE ZERO) | | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | | NSCOUNT (MUST BE ZERO) | | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | | ARCOUNT (MUST BE ZERO) | / +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ / | DSO-TYPE = RECONFIRM (0x0043) | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | DSO-LENGTH (number of octets in DSO-DATA) | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ \ \ NAME \ \ +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | | TYPE | | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ > DSO-DATA | CLASS | | +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ | \ RDATA \ / +--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+ /]]>

DNS Stateful Operations TLV Context Summary This document defines four new DSO TLVs. As recommended in the DNS Stateful Operations specification, the valid contexts of these new TLV types are summarized below. The client TLV contexts are:

C-P:: Client request message, Primary TLV
C-U:: Client Unidirectional message, primary TLV
C-A:: Client request or unidirectional message, Additional TLV
CRP:: Response back to client, Primary TLV
CRA:: Response back to client, Additional TLV

DSO TLV Client Context Summary

TLV Type	C-P	C-U
SUBSCRIBE	X
PUSH
UNSUBSCRIBE		X
RECONFIRM		X

The server TLV contexts are:

S-P:: Server request message, Primary TLV
S-U:: Server Unidirectional message, primary TLV
S-A:: Server request or unidirectional message, Additional TLV
SRP:: Response back to server, Primary TLV
SRA:: Response back to server, Additional TLV

DSO TLV Server Context Summary

TLV Type	S-P	S-U	S-A	SRP	SRA
SUBSCRIBE
PUSH		X
UNSUBSCRIBE
RECONFIRM

Client-Initiated Termination An individual subscription is terminated by sending an UNSUBSCRIBE TLV for that specific subscription, or all subscriptions can be canceled at once by the client closing the DSO session. When a client terminates an individual subscription (via UNSUBSCRIBE) or all subscriptions on that DSO session (by ending the session), it is signaling to the server that it is no longer interested in receiving those particular updates. It is informing the server that the server may release any state information it has been keeping with regards to these particular subscriptions. After terminating its last subscription on a session via UNSUBSCRIBE, a client MAY close the session immediately or it may keep it open if it anticipates performing further operations on that session in the future. If a client wishes to keep an idle session open, it MUST respect the maximum idle time required by the server . If a client plans to terminate one or more subscriptions on a session and doesn't intend to keep that session open, then as an efficiency optimization, it MAY instead choose to simply close the session, which implicitly terminates all subscriptions on that session. This may occur because the client computer is being shut down, is going to sleep, the application requiring the subscriptions has terminated, or simply because the last active subscription on that session has been canceled. When closing a session, a client should perform an orderly close of the TLS session. Typical APIs will provide a session close method that will send a TLS close_notify alert as described in the TLS 1.3 specification. This instructs the recipient that the sender will not send any more data over the session. After sending the TLS close_notify alert, the client MUST gracefully close the underlying connection using a TCP FIN so that the TLS close_notify is reliably delivered. The mechanisms for gracefully closing a TCP connection with a TCP FIN vary depending on the networking API. For example, in the BSD Sockets API, sending a TCP FIN is achieved by calling "shutdown(s,SHUT_WR)" and keeping the socket open until all remaining data has been read from it. If the session is forcibly closed at the TCP level by sending a RST from either end of the connection, data may be lost.

Client Fallback to Polling There are cases where a client may exhaust all avenues for establishing a DNS Push Notification subscription without success. This can happen if the client's configured recursive resolver does not support DNS over TLS, or supports DNS over TLS but is not listening on TCP port 853, or supports DNS over TLS on TCP port 853 but does not support DSO on that port, or for some other reason is unable to provide a DNS Push Notification subscription. In this case, the client will attempt to communicate directly with an appropriate server, and it may be that the zone apex discovery fails, or there is no _dns&nbhy;push&nbhy;tls._tcp.<zone> SRV record, or the server indicated in the SRV record is misconfigured, overloaded, or is unresponsive for some other reason. Regardless of the reason for the failure, after being unable to establish the desired DNS Push Notification subscription, it is likely that the client will still wish to know the answer it seeks, even if that answer cannot be obtained with the timely change notifications provided by DNS Push Notifications. In such cases, it is likely that the client will obtain the answer it seeks via a conventional DNS query instead, repeated at some interval to detect when the answer RRset changes. In the case where a client responds to its failure to establish a DNS Push Notification subscription by falling back to polling with conventional DNS queries instead, the polling rate should be controlled to avoid placing excessive burden on the server. The interval between successive DNS queries for the same name, type, and class SHOULD be at least the minimum of 900 seconds (15 minutes) or two seconds more than the TTL of the answer RRset. The reason that for TTLs up to 898 seconds the query should not be reissued until two seconds after the answer RRset has expired, is to ensure that the answer RRset has also expired from the cache on the client's configured recursive resolver. Otherwise (particularly if the clocks on the client and the recursive resolver do not run at precisely the same rate), there's a risk of a race condition where the client queries its configured recursive resolver just as the answer RRset has one second remaining in the recursive resolver's cache. The client would receive a reply telling it that the answer RRset has one second remaining; the client would then requery the recursive resolver again one second later. If by this time the answer RRset has actually expired from the recursive resolver's cache, the recursive resolver would then issue a new query to fetch fresh data from the authoritative server. Waiting until the answer RRset has definitely expired from the cache on the client's configured recursive resolver avoids this race condition and any unnecessary additional queries it causes. Each time a client is about to reissue its query to discover changes to the answer RRset, it should first make a new attempt to establish a DNS Push Notification subscription using previously cached DNS answers as appropriate. After a temporary misconfiguration has been remedied, this allows a client that is polling to return to using DNS Push Notifications for asynchronous notification of changes.

Name	Value	Early Data	Status
SUBSCRIBE	0x0040	OK	Standards Track
PUSH	0x0041	NO	Standards Track
UNSUBSCRIBE	0x0042	NO	Standards Track
RECONFIRM	0x0043	NO	Standards Track