]> consultant

+31-492474673 (Netherlands), +33-966015248 (France) consultancy@vanderstok.org www.vanderstok.org

YumaWorks

685 Cochran St. Suite #160 Simi Valley CA 93065 USA andy@yumaworks.com

Jacobs University

Campus Ring 1 Bremen 28759 Germany j.schoenwaelder@jacobs-university.de

consultant

Campus Ring 1 Bremen 28759 Germany anuj@iurs.org

Applications core This document describes a network management interface for constrained devices, called CoMI. CoMI is an adaptation of the RESTCONF protocol for use in constrained devices and networks. It is designed to reduce the message sizes, server code size, and application development complexity. The Constrained Application Protocol (CoAP) is used to access management data resources specified in YANG, or SMIv2 converted to YANG. The payload of the CoMI message is encoded in Concise Binary Object Representation (CBOR). Discussion and suggestions for improvement are requested, and should be sent to core@ietf.org.

The Constrained Application Protocol (CoAP) is designed for Machine to Machine (M2M) applications such as smart energy and building control. Constrained devices need to be managed in an automatic fashion to handle the large quantities of devices that are expected in future installations. The messages between devices need to be as small and infrequent as possible. The implementation complexity and runtime resources need to be as small as possible. The draft describes a REST-like interface called RESTCONF, which uses HTTP methods to access structured data defined in YANG . RESTCONF allows access to data resources contained in NETCONF data-stores. RESTCONF messages can be encoded in XML or JSON . The GET method is used to retrieve data resources and the POST, PUT, PATCH, and DELETE methods are used to create, replace, merge, and delete data resources. A large amount of Management Information Base (MIB) specifications already exists for monitoring purposes. This data can be accessed in RESTCONF if the server converts the SMIv2 modules to YANG, using the mapping rules defined in . The CoRE Management Interface (CoMI) is intended to work on standardized data-sets in a stateless client-server fashion. The RESTCONF protocol is adapted and optimized for use in constrained environments, using CoAP instead of HTTP. Standardized data sets promote interoperability between small devices and applications from different manufacturers. Stateless communication is encouraged to keep communications simple and the amount of state information small in line with the design objectives of 6lowpan , RPL , and CoAP . RESTCONF uses the HTTP methods HEAD, and OPTIONS, which are not available in CoAP. HTTP uses TCP which is not recommended for CoAP. The transport protocols available to CoAP are much better suited for constrained networks. CoMI is low resource oriented, uses CoAP, and only supports the methods GET, PUT, PATCH, POST and DELETE. The payload of CoMI is encoded in CBOR which is automatically generated from JSON . CBOR has a binary format and hence has more coding efficiency than JSON. To promote small packets, CoMI uses an additional "data-identifier string-to-number conversion" to minimise CBOR payloads and URI length. It is assumed that the managed device is the most constrained entity. The client might be more capable, however this is not necessarily the case. Currently, small managed devices need to support at least two protocols: CoAP and SNMP . When the MIB can be accessed with the CoAP protocol, the SNMP protocol can be replaced with the CoAP protocol. Although the SNMP server size is not huge (see ), the code for the security aspects of SMIv3 is not negligible. Using CoAP to access secured management objects reduces the code complexity of the stack in the constrained device, and harmonizes applications development. The objective of CoMI is to provide a CoAP based Function Set that reads and sets values of managed objects in devices to (1) initialize parameter values at start-up, (2) acquire statistics during operation, and (3) maintain nodes by adjusting parameter values during operation. The end goal of CoMI is to provide information exchange over the CoAP transport protocol in a uniform manner as a first step to the full management functionality as specified in .

CoMI supports discovery of resources, accompanied by reading, writing and notification of resource values. As such it is close to the device management of the Open Mobile Alliance described in . A comparison between CoMI and LWM2M management can be found in . CoMI supports MIB modules which have been translated from SMIv2 to YANG, using . This mapping is read-only so writable SMIv2 objects need to be converted to YANG using an implementation-specific mapping. CoMI uses a simple URI to access the management object resources. Complexity introduced by instance selection, or multiple object specification is expressed with uri-query attributes. The choice for uri-query attributes makes the URI structure less context dependent. The YANG data model contains a lot of information that can be exploited by automation tools and need not be transported in the request messages, ultimately leading to reduced message sizes.

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 . Readers of this specification should be familiar with all the terms and concepts discussed in , , and . The following terms are defined in the NETCONF protocol : client, configuration data, data-store, and server. The following terms are defined in the YANG data modelling language : container, data node, key, key leaf, leaf, leaf-list, and list. The following terms are defined in RESTCONF protocol : data resource, data-store resource, edit operation, query parameter, target resource, and unified data-store. The following terms are defined in this document: CoMI object identifier, which is a 30-bit numeric hash of the YANG object identifier string for the object. When a YANG hash value is printed in a request target URI, error-path or other string, then the lowercase hexadecimal representation is used. Leading zeros are used so the value uses 8 hex characters. An instance of a data-node specified in a YANG module present in the server. The instance is stored in the memory of the server. An instance of a schema node of type notification, specified in a YANG module present in the server. The instance is generated in the server at the occurrence of the corresponding event and appended to the default stream. The following list contains the abbreviations used in this document. TODO, and others to follow.

A simplified graphical representation of the data model is used in this document. The meaning of the symbols in these diagrams is as follows: Brackets "[" and "]" enclose list keys. Abbreviations before data node names: "rw" means configuration data (read-write) and "ro" state data (read-only). Symbols after data node names: "?" means an optional node, "!" means a presence container, and "*" denotes a list and leaf-list. Parentheses enclose choice and case nodes, and case nodes are also marked with a colon (":"). Ellipsis ("...") stands for contents of subtrees that are not shown.

This section describes the CoMI architecture to use CoAP for the reading and modifying of instrumentation variables used for the management of the instrumented node.

| YANG | > COAP | | |specification(2)| |specification(1)| Request(3) | | +----------------+ +----------------+[ * | +-----------------------------*-----------[---------*----------+ * [ * * [ +-----------+ mapping * security[ | Network | * (8) [ | packet(4) | * [ +-----------+ Server * [ * +-----------------------------*-----------[---------*----------+ | * [ * | | * Retrieval, | | * Modification(5) | | \*/ * | | +-------------------------------------------------*--------+ | | | +--------------+ +------------+ | | | | |configuration | |Operational | | | | | | (6b) | | state(6a) | | | | | +--------------+ +------------+ | | | | variable store (6) * | | | +-------------------------------------------------*--------+ | | * | | Variable | | Instrumentation(7)| +--------------------------------------------------------------+ ]]>

is a high level representation of the main elements of the CoAP management architecture. A client sends requests as payload in packets over the network to a managed constrained node. Objectives are: Equip a constrained node with a management server that provides information about the operational characteristics of the code running in the constrained node. The server provides this information in a variable store that contains values describing the performance characteristics and the code parameter values. The client receives the performance characteristics on a regular basis or on request. The client sets the parameter values in the server at bootstrap and intermittently when operational conditions change. The constrained network requires the payload to be as small as possible, and the constrained server memory requirements should be as small as possible. For interoperability it is required that in addition to using the Internet Protocol for data transport: The names, type, and semantics of the instrumentation variables are standardized. The instrumentation variables are described in a standard language. The signature of the CoAP request in the server is standardized. The format of the packet payload is standardized. The notification from server to client is standardized. The different numbered components of Figure 1 are discussed according to component number. contains a set of named and versioned modules. A module specifies a hierarchy of named and typed resources. A resource is uniquely identified by a sequence of its name and the names of the enveloping resources following the hierarchy order. The YANG specification serves as input to the writers of application and instrumentation code and the humans analysing the returned values (arrow from YANG specification to Variable store). The specification can be used to check the correctness of the CoAP request and do the CBOR encoding. A named module specifies a set of variables and "conceptual tables". Named variables have simple types. Conceptual tables are composed of typed named columns. The variable name and module name identify the variable uniquely. There is an algorithm to translate SMIv2 specifications to YANG specifications. The CoAP request needs a Universal Resource Identifier (URI) and the payload of the packet to send a request. The URI is composed of the schema, server, path and query and looks like coap://entry.example.com/<path>?<query>. Fragments are not supported. Allowed operations are PUT, PATCH, GET, DELETE, and POST. New variables can be created with POST when they exist in the YANG specification. The Observe option can be used to return variable values regularly or on event occurrence (notification). The path identifies the variable in the form "/mg/<hash-value>". The query parameter is used to specify additional (optional) aspects like the module name, list instance, and others. The idea is to keep the path simple and put variations on variable specification in the query. Discovery of the variables is done with standard CoAP resource discovery using /.well-known/core with ?rt=/core.mg. The payload contains the CBOR encoding of JSON objects. This object corresponds to the converted RESTCONF message payload. The server needs to parse the CBOR encoded message and identify the corresponding instances in the Variable store. In addition, this component includes the code for CoAP Observe and block options. The store is composed of two parts: Operational state and Configuration data-store (see ). CoMI does not differentiate between variable store types. The Variable store contains data-node instances. Values are stored in the appropriate instances, and or values are returned from the instances into the payload of the packet. This code depends on implementation of drivers and other node specific aspects. The Variable instrumentation code stores the values of the parameters into the appropriate places in the operational code. The variable instrumentation code reads current execution values from the operational code and stores them in the appropriate instances. The server MUST prevent unauthorized users from reading or writing any data resources. CoMI relies on DTLS which is specified to secure CoAP communication.

CoMI adapts the RESTCONF architecture so data exchange and implementation requirements are optimized for constrained devices. The RESTCONF protocol uses a unified data-store to edit conceptual data structures supported by the server. The details of transaction preparation and non-volatile storage of the data are hidden from the RESTCONF client. CoMI also uses a unified data-store, to allow stateless editing of configuration variables and the notification of operational variables. The child schema nodes of the unified data-store include all the top-level YANG data nodes in all the YANG modules supported by the server. The YANG data structures represent a hierarchy of data resources. The client discovers the list of YANG modules, and important conformance information such as the module revision dates, YANG features supported, and YANG deviations required. The individual data nodes are discovered indirectly by parsing the YANG modules supported by the server. The YANG data definition statements contain a lot of information that can help automation tools, developers, and operators use the data model correctly and efficiently. The YANG definitions and server YANG module capability advertisements provide an "API contract" that allow a client to determine the detailed server management capabilities very quickly. CoMI allows access to the same data resources as a RESTCONF server, except the messages are optimized to reduce identifier and payload size. RESTCONF uses a simple algorithmic mapping from YANG to URI syntax to identify the target resource of a retrieval or edit operation. A client can construct operations or scripts using a predictable syntax, based on the YANG data definitions. The target resource URI can reference a data resource instance, or the data-store itself (to retrieve the entire data-store or create a top-level data resource instance). CoMI uses a compression algorithm to reduce the size of the data-node instance identifier (see .

The RESTCONF protocol uses the full path of the desired data resource in the target resource URI. The JSON encoding will include the module name string to specify the YANG module. If a representation of the target resource is included in the request or response message in RESTCONF messages, then the data definition name string is used to identify each node in the message. The module namespace (or name) may also be present in these identifiers. In order to greatly reduce the size of identifiers used in CoMI, numeric object identifiers are used instead of these strings. The specific encoding of the object identifiers is not hard-wired in the protocol. YANG Hash is the default encoding for object identifiers. This encoding in considered to be "unstructured" since the particular values for each object are determined by a hash algorithm. It is possible for 2 different objects to generate the same hash value. If this occurs, then the client and server will both need to rehash the colliding object identifiers to new unused hash values. In order to eliminate the need for rehashing, CoMI allows for alternate "structured" object identifier encoding formats. Structured object identifier MUST be managed such that no object ID collisions are possible, and therefore no rehash procedures are needed. Structured object identifiers can also be selected to minimize the size of a subset of the object identifiers (e.g., the most requested objects). In the discovery of the object ID compression scheme is described.

In CoAP a group of links can constitute a Function Set. The format of the links is specified in . This note specifies a Management Function Set. CoMI end-points that implement the CoMI management protocol support at least one discoverable management resource of resource type (rt): core.mg, with path: /mg, where mg is short-hand for management. The name /mg is recommended but not compulsory (see ). The path prefix /mg has resources accessible with the following five paths: YANG-based data with path "/mg" and using CBOR content encoding format. This path represents a data-store resource which contains YANG data resources as its descendant nodes. All identifiers referring to YANG data nodes within this path are encoded as YANG hash values (see ). URI identifying the location of the server module information, with path "/mg/mod.uri" and CBOR content format. This YANG data is encoded with plain identifier strings, not YANG hash values. String identifying the module set ID in use by the server, which is defined as the 'module-set-id' leaf in the ietf-yang-library module. This resource MUST change to a new value when the set of YANG modules in use by the server changes. String identifying the object ID numbering scheme used by the CoMI server. The only value defined in this document is 'yanghash' to indicate that the YANG Hash numbering scheme defined in this document is used. It is possible for other object numbering schemes to be defined outside the scope of this document. String identifying the CoMI server type. The value 'ro' indicates that the server is a read-only server and no editing operations are supported. A read-only server is not required to provide YANG deviation statements for any writable YANG data nodes. The value 'rw' indicates that the server is a read-write server and editing operations are supported. A read-write server is required to provide YANG deviation statements for any writable YANG data nodes that are not fully implemented. URI indicating the location of the server YANG hash information if any objects needed to be re-hashed by the server. It has the path "/mg/yh.uri" and is encoded in CBOR format. The "ietf-yang-hash" module of is used to define the syntax and semantics of this data structure. This YANG data is encoded with plain identifier strings, not YANG hash values. The server will only have this resource if there are any objects that needed to be re-hashed due to a hash collision. String identifying the default stream resource to which YANG notification instances are appended. Notification support is optional, so this resource will not exist if the server does not support any notifications. The mapping of YANG data node instances to CoMI resources is as follows: A YANG module describes a set of data trees composed of YANG data nodes. Every root of a data tree in a YANG module loaded in the CoMI server represents a resource of the server. All data root descendants represent sub-resources. The resource identifiers of the instances of the YANG specifications are YANG hash values, as described in . When multiple instances of a list node exist, the instance selection is described in The profile of the management function set, with IF=core.mg, is shown in the table below, following the guidelines of : name path rt Data Type Management /mg core.mg n/a Data /mg core.mg.data application/cbor Module Set URI /mg/mod.uri core.mg.moduri application/cbor Module Set ID /mg/mod.set core.mg.modset application/cbor Numbering Type /mg/num.typ core.mg.num-type application/cbor Server Type /mg/srv.typ core.mg.srv-type application/cbor YANG Hash Info /mg/yh.uri core.mg.yang-hash application/cbor Events /mg/stream core.mg.stream application/cbor

The MG Function Set provides a CoAP interface to perform a subset of the functions provided by RESTCONF. A subset of the operations defined in RESTCONF are used in CoMI: Operation Description GET Retrieve the data-store resource or a data resource POST Create a data resource PUT Create or replace a data resource PATCH Replace a data resource partially DELETE Delete a data resource

One or more instances of data resources are retrieved by the client with the GET method. The RESTCONF GET operation is supported in CoMI. The same constraints apply as defined in section 3.3 of . The operation is mapped to the GET method defined in section 5.8.1 of . It is possible that the size of the payload is too large to fit in a single message. In the case that management data is bigger than the maximum supported payload size, the Block mechanism from is used, as explained in more detail in . There are two query parameters for the GET method. A CoMI server MUST implement the keys parameter and MAY implement the select parameter to allow common data retrieval filtering functionality. Query Parameter Description keys Request to select instances of a YANG definition select Request selected sub-trees from the target resource The "keys" parameter is used to specify a specific instance of the resource. When keys is not specified, all instances are returned. When no or one instance of the resource exists, the keys parameter is not needed.

RESTCONF uses the 'select' parameter to specify an expression which can represent a subset of all data nodes within the target resource . This parameter is useful for filtering sub-trees and retrieving only a subset that a managing application is interested in. However, filtering is a resource intensive task and not all constrained devices can be expected to have enough computing resources such that they will be able to successfully filter and return a subset of a sub-tree. This is especially likely to be true with Class 0 devices that have significantly lesser RAM than 10 KiB . Since CoMI is targeted at constrained devices and networks, only a limited subset of the 'select' parameter is used here. Unlike the RESTCONF 'select' parameter, CoMI does not use object names in "XPath" or "path-expr" format to identify the subset that needs to be filtered. Parsing XML is resource intensive for constrained devices and using object names can lead to large message sizes. Instead, CoMI utilizes the YANG hashes described in to identify the sub-trees that should be filtered from a target resource. Using these hashes ensures that a constrained node can identify the target sub-tree without expending many resources and that the messages generated are also efficiently encoded. The implementation of the 'select' parameter is already optional for constrained devices, however, even when implemented it is expected to be a best effort feature, rather than a service that nodes must provide. This implies that if a node receives the 'select' parameter specifying a set of sub-trees that should be returned, it will only return those that it is able to.

In all examples the path is expressed in readable names and as a hash value of the name (where the hash value in the payload is expressed as a hexadecimal number, and the hash value in the URL as a base64 number). The examples in this section use a JSON payload with one or more entries describing the pair (identifier, value). CoMI transports the CBOR format to transport the equivalent contents. The CBOR syntax of the payloads is specified in .

A request to read the values of instances of a management object or the leaf of an object is sent with a confirmable CoAP GET message. A single object is specified in the URI path prefixed with /mg. Using for example the clock container from , a request is sent to retrieve the value of clock/current-datetime specified in module system-state. The answer to the request returns a (identifier, value) pair.

The YANG hash value for 'current-datetime' is calculated by constructing the schema node identifier for the object:

The 30 bit murmur3 hash value is calculated on this string (0x15370408 and VNwQI). The request using this hash value is shown below:

The specified object can be an entire object. Accordingly, the returned payload is composed of all the leaves associated with the object. Each leaf is returned as a (YANG hash, value) pair. For example, the GET of the clock object, sent by the client, results in the following returned payload sent by the managed entity:

The YANG hash values for 'clock', 'current-datetime', and 'boot-datetime' are calculated by constructing the schema node identifier for the objects, and then calculating the 30 bit murmur3 hash values (shown in parenthesis):

The request using the hash values is shown below:

A "list" node can have multiple instances. Accordingly, the returned payload is composed of all the instances associated with the list node. Each instance is returned as a (identifier, value) pair. The "keys" query parameter is used to identify a specific list instance by specifying a given index value (see ). For example, the GET of the /interfaces/interface/ipv6/neighbor instance identified with interface index "eth0" , sent by the client, results in the following returned payload sent by the managed entity:

The YANG hash values for 'neighbor', 'ip', and 'link-layer-address' are calculated by constructing the schema node identifier for the objects, and then calculating the 30 bit murmur3 hash values (shown in parenthesis):

The request using the hash values is shown below:

The YANG translation of the SMI specifying the ipNetToMediaTable yields:

The following example shows an "ipNetToPhysicalTable" with 2 instances, using JSON encoding:

The YANG hash values for 'ipNetToPhysicalEntry' and its child nodes are calculated by constructing the schema node identifier for the objects, and then calculating the 30 bit murmur3 hash values (shown in parenthesis):

The following example shows a request for the entire ipNetToPhysicalTable. Since all the instances are requested, no "keys" query parameter is needed.

There is a mandatory query parameter that MUST be supported by servers called "keys". This parameter is used to specify the key values for an instance of an object identified by a YANG hash value. Any key leaf values of the instance are passed in order. The first key leaf in the top-most list is the first key encoded in the 'keys' parameter. The key leafs from top to bottom and left to right are encoded as a comma-delimited list. If a key leaf value is missing then all values for that key leaf are returned. Example: In this example exactly 1 instance is requested from the ipNetToPhysicalEntry (from a previous example).

An example illustrates the syntax of keys query parameter. In this example the following YANG module is used:

The path identifier for the leaf "col1" is the following string:

The YANG hash for this identifier string has values: 0xa9abdcca and pq9zK). The following string represents the RESTCONF target resource URI expression for the "col1" leaf for the key values "top", 17, and "group1":

The following string represents the CoMI target resource identifier for the same instance of the "col1" leaf:

The select parameter is used along with the GET method to provide a sub-tree filter mechanism. A list of YANG hashes that should be filtered is provided along with a list of keys identifying the instances that should be returned. When the keys parameter is used together with the select, the key values are added in brackets without using the "keys=" text. The following example shows an "ipNetToPhysicalTable" (from a previous example) with 4 instances, using JSON encoding:

Data may be retrieved using the select query parameter in the following way:

In this example exactly 2 instances are returned as response from the ipNetToPhysicalTable because both those instances match the provided keys. Supposing there were multiple YANG hashes with their own sets of keys that were to be filtered, the select query parameter can be used to retrieve results from these in one go as well. The following string represents the CoMI target resource identifier when multiple YANG hashes, with their own sets of keys are queried:

/mg/?select=hash1(hash1-key1,hash1-key2,...),hash2(hash2-key1)...

CoMI allows data-store contents to be created, modified and deleted using CoAP methods. Data-editing is an optional feature. The server will indicate its editing capability with the "/core.rg.srv-type resource type. If the value is 'rw' then the server supports editing operations. If the value is 'ro' then the server does not support editing operations.

A CoMI server is not required to support entry insertion of lists and leaf-lists that are ordered by the user (i.e., YANG statement "ordered-by user"). The 'insert' and 'point' query parameters from RESTCONF are not used in CoMI. A CoMI server SHOULD preserve the relative order of all user-ordered list and leaf-list entries that are received in a single edit request. These YANG data node types are encoded as arrays so messages will preserve their order.

Data resource instances are created with the POST method. The RESTCONF POST operation is supported in CoMI, however it is only allowed for creation of data resources. The same constraints apply as defined in section 3.4.1 of . The operation is mapped to the POST method defined in section 5.8.2 of . There are no query parameters for the POST method.

Data resource instances are created or replaced with the PUT method. The PUT operation is supported in CoMI. A request to set the values of instances of an object/leaf is sent with a confirmable CoAP PUT message. The Response is piggybacked to the CoAP ACK message corresponding with the Request. The same constraints apply as defined in section 3.5 of . The operation is mapped to the PUT method defined in section 5.8.3 of . There are no query parameters for the PUT method.

Data resource instances are partially replaced with the PATCH method . The PATCH operation is supported in CoMI. A request to set the values of instances of a subset of the values of the resource is sent with a confirmable CoAP PATCH message. The Response is piggybacked to the CoAP ACK message corresponding with the Request. The same constraints apply as defined in section 3.5 of . The operation is mapped to the PATCH method defined in . There are no query parameters for the PATCH method.

Data resource instances are deleted with the DELETE method. The RESTCONF DELETE operation is supported in CoMI. The same constraints apply as defined in section 3.7 of . The operation is mapped to the DELETE method defined in section 5.8.4 of . There are no optional query parameters for the PUT method.

Editing multiple data resources at once can allow a client to use fewer messages to make a configuration change. It also allows multiple edits to all be applied or none applied, which is not possible if the data resources are edited one at a time. It is easy to add multiple entries at once. The "PATCH" method can be used to simply patch the parent node(s) of the data resources to be added. If multiple top-level data resources need to be added, then the data-store itself ('/mg') can be patched. If other operations need to be performed, or multiple operations need to be performed at once, then the YANG Patch media type can be used with the PATCH method. A YANG patch is an ordered list of edits on the target resource, which can be a specific data node instance, or the data-store itself. The resource type used by YANG Patch is 'application/yang.patch'. A status message is returned in the response, using resource type 'application/yang.patch.status'. The following YANG tree diagram describes the YANG Patch structure, Each 'edit' list entry has its own operation, sub-resource target, and new value (if needed).

The YANG Hash values for the YANG Patch request objects are calculated as follows:

Refer to for more details on the YANG Patch request and response contents.

Notification by the server to a selection of clients when an event occurs in the server is an essential function for the management of servers. CoMI allows events specified in YANG to be notified to a selection of requesting clients. There is one, so-called "default", stream in a CoMI server. The /mg/stream resource identifies the default stream. When a CoMI server generates an internal event, it is appended to the default stream, and the contents of a notification instance is ready to be sent to all CoMI clients which observe the default stream resource. Reception of generated notification instances is enabled with the CoAP Observe function. The client subscribes to the notifications by sending a GET request with an "Observe" option, specifying the /mg/stream resource. Every time an event is generated, the default stream is cleared, and the generated notification instance is appended to the stream. After appending the instance, the contents of the instance is sent to all observing clients. Suppose the server generates the event specified with:

The YANG Hash values for this notification are assigned as follows:

By executing a GET on the /mg/stream resource the client receives the following response:

In the example, the request returns a success response with the contents of the last generated event. Consecutively the server will regularly notify the client when a new event is generated. To check that the client is still alive, the server MUST send confirmable notifications once in a while. When the client does not confirm the notification from the server, the server will remove the client from the list of observers . In the registration request, the client MAY include a "Response-To-Uri-Host" and optionally "Response-To-Uri-Port" option as defined in . In this case, the observations SHOULD be sent to the address and port indicated in these options. This can be useful when the client wants the managed device to send the trap information to a multicast address.

The CoAP protocol provides reliability by acknowledging the UDP datagrams. However, when large pieces of text need to be transported the datagrams get fragmented, thus creating constraints on the resources in the client, server and intermediate routers. The block option allows the transport of the total payload in individual blocks of which the size can be adapted to the underlying fragment sizes such as: (UDP datagram size ~64KiB, IPv6 MTU of 1280, IEEE 802.15.4 payload of 60-80 bytes). Each block is individually acknowledged to guarantee reliability. The block size is specified as exponents of the power 2. The SZX exponent value can have 7 values ranging from 0 to 6 with associated block sizes given by 2**(SZX+4); for example SZX=0 specifies block size 16, and SZX=3 specifies block size 128. The block number of the block to transmit can be specified. There are two block options: Block1 option for the request payload transported with PUT, POST or PATCH, and the block2 option for the response payload with GET. Block1 and block2 can be combined. Examples showing the use of block option in conjunction with observer options are provided in . Notice that the Block mechanism splits the data at fixed positions, such that individual data fields may become fragmented. Therefore, assembly of multiple blocks may be required to process the complete data field.

The presence and location of (path to) the management data are discovered by sending a GET request to "/.well-known/core" including a resource type (RT) parameter with the value "core.mg" . Upon success, the return payload will contain the root resource of the management data. It is up to the implementation to choose its root resource, but it is recommended that the value "/mg" is used, where possible. The example below shows the discovery of the presence and location of management data.

; rt="core.mg" ]]>

Management objects MAY be discovered with the standard CoAP resource discovery. The implementation can add the hash values of the object identifiers to /.well-known/core with rt="core.mg.data". The available objects identified by the hash values can be discovered by sending a GET request to "/.well-known/core" including a resource type (RT) parameter with the value "core.mg.data". Upon success, the return payload will contain the registered hash values and their location. The example below shows the discovery of the presence and location of management data.

; rt="core.mg.data", ; rt="core.mg.data" ]]>

Lists of hash values may become prohibitively long. It is discouraged to provide long lists of objects on discovery. Therefore, it is recommended that details about management objects are discovered following the RESTCONF protocol. The YANG module information is stored in the "ietf-yang-library" module . The resource "/mg/mod.uri" is used to retrieve the location of the YANG module library. Since many constrained servers within a deployment are likely to be similar, the module list can be stored locally on each server, or remotely on a different server.

Within the YANG module library all information about the module is stored such as: module identifier, identifier hierarchy, grouping, features and revision numbers. The hash identifier is obtained as specified in . When a collision occurred in the name space of the target server, a rehash is executed as explained in .

The RESTCONF return status codes defined in section 6 of the RESTCONF draft are used in CoMI error responses, except they are converted to CoAP error codes. TODO: complete RESTCONF to CoAP error code mappings TODO: assign an error cpde for a rehash-error. RESTCONF Status Line CoAP Status Code 100 Continue none? 200 OK 2.05 201 Created 2.01 202 Accepted none? 204 No Content ? 304 Not Modified 2.03 400 Bad Request 4.00 403 Forbidden 4.03 404 Not Found 4.04 405 Method Not Allowed 4.05 409 Conflict none? 412 Precondition Failed 4.12 413 Request Entity Too Large 4.13 414 Request-URI Too Large 4.00 415 Unsupported Media Type 4.15 500 Internal Server Error 5.00 501 Not Implemented 5.01 503 Service Unavailable 5.03

A mapping for the encoding of YANG data in CBOR is necessary for the efficient transport of management data in the CoAP payload. Since object names may be rather long and may occur repeatedly, CoMI allows for association of a given object path identifier string value with an integer, called a "YANG hash".

The association between string value and string number is done through a hash algorithm. The 30 least significant bits of the "murmur3" 32-bit hash algorithm are used. This hash algorithm is described online at http://en.wikipedia.org/wiki/MurmurHash. Implementation are available online, including at https://code.google.com/p/smhasher/wiki/MurmurHash. When converting 4 input bytes to a 32-bit integer in the hash algorithm, the Little-Endian convention MUST be used. The hash is generated for the string representing the object path identifier. A canonical representation of the path identifier is used. Prefix values are used on every node. The prefix values defined in the YANG module containing the data object are used for the path expression. For external modules, this is the value of the 'prefix' sub-statement in the 'import' statement for each external module. Path expressions for objects which augment data nodes in external modules are calculated in the augmenting module, using the prefix values in the augmenting module. Choice and case node names are not included in the path expression. Only 'container', 'list', 'leaf', 'leaf-list', and 'anyxml' nodes are listed in the path expression. The "murmur3_32" hash function is executed for the entire path string. The value '42' is used as the seed for the hash function. The YANG hash is subsequently calculated by taking the 30 least significant bits. The resulting 30-bit number is used by the server, unless the value is already being used for a different object by the server. In this case, the re-hash procedure in the following section is executed.

A hash collision occurs if two different path identifier strings have the same hash value. If the server has over 30,000 objects in its YANG modules, then the probability of a collision is 10% or higher. If a hash collision occurs on the server, then the object that is causing the conflict has to be altered, such that the new hash value does not conflict with any value already in use by the server. In most cases, the hash function is expected to produce unique values for all the objects supported by a constrained device. Given a known set of YANG modules, both server and client can calculate the YANG hashes independently, and offline. Even though collisions are expected to happen rather rarely, they need to be considered. Collisions can be detected before deployment, if the vendor knows which modules are supported by the server, and hence all YANG hashes can be calculated. Collisions are only an issue when they occur at the same server. The client needs to discover any re-hash mappings on a per server basis. If the server needs to re-hash any object identifiers, then it MUST create a "rehash-map" entry for all its rehashed objects, as described in the following YANG module.

The "ietf-yang-hash" YANG module is used by the server to report any objects that have been mapped to produce a new hash value that does not conflict with any other YANG hash values used by the server. YANG tree diagram for "ietf-yang-hash" module:

file "ietf-yang-hash@2015-06-06.yang" module ietf-yang-hash { namespace "urn:ietf:params:xml:ns:yang:ietf-yang-hash"; prefix "yh"; organization "IETF CORE (Constrained RESTful Environments) Working Group"; contact "WG Web: WG List: WG Chair: Carsten Bormann WG Chair: Andrew McGregor Editor: Peter van der Stok Editor: Andy Bierman Editor: Juergen Schoenwaelder Editor: Anuj Sehgal "; description "This module contains re-hash information for the CoMI protocol. Copyright (c) 2015 IETF Trust and the persons identified as authors of the code. All rights reserved. Redistribution and use in source and binary forms, with or without modification, is permitted pursuant to, and subject to the license terms contained in, the Simplified BSD License set forth in Section 4.c of the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info). This version of this YANG module is part of RFC XXXX; see the RFC itself for full legal notices."; // RFC Ed.: replace XXXX with actual RFC number and remove this // note. // RFC Ed.: remove this note // Note: extracted from draft-vanderstok-core-comi-07.txt // RFC Ed.: update the date below with the date of RFC publication // and remove this note. revision 2015-06-06 { description "Initial revision."; reference "RFC XXXX: CoMI Protocol."; } container yang-hash { config false; description "Contains information on the YANG Hash values used by the server."; list rehash { key hash; description "Each entry describes an re-hash mapping in use by the server."; leaf hash { type uint32; description "The hash value that has a collision. This hash value cannot be used on the server. The rehashed value for each affected object must be used instead."; } list object { min-elements 2; description "Each entry identifies one of the objects involved in the hash collision and contains the rehash information for that object."; leaf module { type string; mandatory true; description "The module name for this object."; } leaf newhash { type uint32; mandatory true; description "The new hash value for this object."; } leaf pathlen { type uint32; description "The length of the path expression of the object with this hash value. This object MUST be included for any objects in the rehash entry with the same 'module' value."; } leaf path { type string; description "The path expression of the object with this hash value. This object MUST be included for any objects in the rehash entry with the same 'module' and 'pathlen' values."; } } } } } ]]>

In this example there are three YANG modules, "foo", "bar", and "bar1".

This set of 3 YANG modules containing a total of 7 objects produces the following object list. Note that actual hash values are not shown, since these modules do not actually cause the YANG Hash clashes described in the examples.

In this example, assume that the following 3 objects produce the same hash value, so 'h3', 'h6', and 'h7' have the same value (e.g. '1234'): The client might retrieve the container "/f:A" which could cause its sub-nodes to be returned. Instead, the server will return a message with the resource type "core.mg.", representing the "yang-hash" data structure.

In this example, assume that the following 4 objects produce the same hash value, so 'h3', 'h5', 'h6', and 'h7' all have the same value (e.g. '1234'): The client might retrieve the list "/f:A/f:B" which would cause its sub-nodes to be returned. Instead, the server will return a message with the resource type "core.mg.yanh-hash", representing the "yang-hash" data structure. Note that the "pathlen" field is not needed for the 'h6' and 'h7' objects.

In this example, assume that the following 5 objects produce the same hash value, so 'h3', 'h4', 'h5', 'h6', and 'h7' all have the same value (e.g. '1234'): The client might retrieve the list "/f:A/f:B" which would cause its sub-nodes to be returned. Instead, the server will return a message with the resource type "core.mg.yang-hash", representing the "yang-hash" data structure. The "path" leaf is included 2 entries because the "module" and "pathlen" values are the same for the objects.

When a URL contains a YANG hash, it is encoded using base64url "URL and Filename safe" encoding as specified in . The hash H is represented as a 30-bit integer, divided into five 6-bit integers as follows:

> 24 B2 = (H & 0xfc0000) >> 18 B3 = (H & 0x03f000) >> 12 B4 = (H & 0x000fc0) >> 6 B5 = H & 0x00003f ]]>

Subsequently, each 6-bit integer Bx is translated into a character Cx using Table 2 from , and a string is formed by concatenating the characters in the order C1, C2, C3, C4, C5. For example, the YANG hash 0x29abdcca is encoded as "pq9zK".

When encoding YANG variables in CBOR, the CBOR encodings entry is a map. The key is the YANG hash of entry variable, whereas the value contains its value. For encoding of the variable values, a CBOR datatype is used. provides the mapping between YANG datatypes and CBOR datatypes.

defines the mapping between YANG datatypes and CBOR datatypes. Elements of types not in this table, and of which the type cannot be inferred from a type in this table, are ignored in the CBOR encoding by default. Examples include the "description" and "key" elements. However, conversion rules for some elements to CBOR MAY be defined elsewhere. YANG type CBOR type Specification int8, int16, int32, int64, uint16, uint32, uint64, decimal64 unsigned int (major type 0) or negative int (mayor type 1) The CBOR integer type depends on the sign of the actual value. boolean either "true" (major type 7, simple value 21) or "false" (major type 7, simple value 20) string text string (major type 3) enumeration unsigned int (major type 0) bits array of text strings Each text string contains the name of a bit value that is set. binary byte string (major type 2) empty null (major type 7, simple value 22) TBD: This MAY not be applicable to true MIBs, as SNMP may not support empty variables... union Similar to the JSON transcription from , the elements in a union MUST be determined using the procedure specified in section 9.12 of . leaf-list array (major type 4) The array is encapsulated in the map associated with the YANG variable. list array (major type 4) of maps (major type 5) Each array element contains a map of associated YANG hash - value pairs. container map (major type 5) The map contains YANG hash - value pairs corresponding to the elements in the container. smiv2:oid array of integers Each integer contains an element of the OID, the first integer in the array corresponds to the most left element in the OID.

In case a request is received which cannot be processed properly, the managed entity MUST return an error message. This error message MUST contain a CoAP 4.xx or 5.xx response code, and SHOULD include additional information in the payload. Such an error message payload is encoded in CBOR, using the following structure: TODO: Adapt RESTCONF <errors> data structure for use in CoMI. Need to select the most important fields like <error-path>.

The variable "errorCode" has one of the values from the table below, and the OPTIONAL "errorText" field contains a human readable explanation of the error. CoMI Error Code CoAP Error Code Description 0 4.00 General error 1 4.00 Malformed CBOR data 2 4.00 Incorrect CBOR datatype 3 4.00 Unknown MIB variable 4 4.00 Unknown conversion table 5 4.05 Attempt to write read-only variable 0..2 5.01 Access exceptions 0..18 5.00 SMI error status The CoAP error code 5.01 is associated with the exceptions defined in and CoAP error code 5.00 is associated with the error-status defined in .

For secure network management, it is important to restrict access to MIB variables only to authorised parties. This requires integrity protection of both requests and responses, and depending on the application encryption. CoMI re-uses the security mechanisms already available to CoAP as much as possible. This includes DTLS for protected access to resources, as well suitable authentication and authorisation mechanisms. Among the security decisions that need to be made are selecting security modes and encryption mechanisms (see ). This requires a trade-off, as the NoKey mode gives no protection at all, but is easy to implement, whereas the X.509 mode is quite secure, but may be too complex for constrained devices. In addition, mechanisms for authentication and authorisation may need to be selected. CoMI avoids defining new security mechanisms as much as possible. However some adaptations may still be required, to cater for CoMI's specific requirements.

'rt="core.mg.data"' needs registration with IANA. 'rt="core.mg.moduri"' needs registration with IANA. 'rt="core.mg.modset"' needs registration with IANA. 'rt="core.mg.yang-hash"' needs registration with IANA. 'rt="core.mg.yang-stream"' needs registration with IANA. Content types to be registered: application/comi+cbor

We are very grateful to Bert Greevenbosch who was one of the original authors of the CoMI specification and specified CBOR encoding and use of hashes. Mehmet Ersue and Bert Wijnen explained the encoding aspects of PDUs transported under SNMP. Carsten Bormann has given feedback on the use of CBOR. The draft has benefited from comments (alphabetical order) by Dee Denteneer, Esko Dijk, Michael van Hartskamp, Zach Shelby, Michel Veillette, Michael Verschoor, and Thomas Watteyne. The CBOR encoding borrows extensively from Ladislav Lhotka's description on conversion from YANG to JSON. This material is based upon work supported by Philips Research, Huawei, and The Space & Terrestrial Communications Directorate (S&TCD); the latter under Contract No. W15P7T-13-C-A616. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of Philips Research, Huawei, or The Space & Terrestrial Communications Directorate (S&TCD). Juergen Schoenwaelder and Anuj Sehgal were partly funded by Flamingo, a Network of Excellence project (ICT-318488) supported by the European Commission under its Seventh Framework Programme.

Changes from version 00 to version 01 Focus on MIB only Introduced CBOR, JSON, removed BER defined mappings from SMI to xx Introduced the concept of addressable table rows Changes from version 01 to version 02 Focus on CBOR, used JSON for examples, removed XML and EXI added uri-query attributes mod and con to specify modules and contexts Definition of CBOR string conversion tables for data reduction use of Block for multiple fragments Error returns generalized SMI - YANG - CBOR conversion Changes from version 02 to version 03 Added security considerations Changes from version 03 to version 04 Added design considerations section Extended comparison of management protocols in introduction Added automatic generation of CBOR tables Moved lowpan table to Appendix Changes from version 04 to version 05 Merged SNMP access with RESTCONF access to management objects in small devices Added CoMI architecture section Added RESTCONf NETMOD description Rewrote section 5 with YANG examples Added server and payload size appendix Removed Appendix C for now. It will be replaced with a YANG example. Changes from version 04 to version 05 Extended examples with hash representation Added keys query parameter text Added select query parameter text Better separation between specification and instance Section on discovery updated Text on rehashing introduced Elaborated SMI MIB example Yang libary use described use of BigEndian/LittleEndian in Hash generation specified Changes from version 05 to version 06 Hash values in payload as hexadecimal and in URL in base64 numbers Streamlined CoMI architecture text Added select query parameter text Data editing optional Text on Notify added Text on rehashing improved with example Changes from version 06 to version 07 reduced payload size by removing JSON hierachy changed rehash handling to support small clients added LWM2M comparison Notification handling as specified in YANG Added Patch function Rehashing completely reviewed Discover type of YANG name encoding Added new resource types Read-only servers introduced Multiple updates explained

&RFC2119; &RFC4648; &RFC5277; &RFC6020; &RFC7049; &RFC7159; &RFC7252; &I-D.becker-core-coap-sms-gprs; &I-D.ietf-core-block; &I-D.ietf-core-observe; &I-D.ietf-netmod-yang-json; &I-D.ietf-netconf-restconf; &I-D.vanderstok-core-patch; &RFC2578; &RFC3410; &RFC3411; &RFC3414; &RFC3416; &RFC3418; &RFC4293; &RFC4944; &RFC6241; &RFC6347; &RFC6643; &RFC6650; &RFC6690; &RFC6775; &RFC7223; &RFC7228; &RFC7317; &I-D.ietf-core-interfaces; &I-D.ersue-constrained-mgmt; &I-D.ietf-lwig-coap; This document describes a method for applying patches to NETCONF data-stores using data defined with the YANG data modeling language.

This section provides information on code sizes and payload sizes for a set of management servers. Approximate code sizes are: Code processor Text Data reference Observe agent erbium 800 n/a CoAP server MSP430 1K 6 SNMP server ATmega128 9K 700 Secure SNMP ATmega128 30K 1.5K DTLS server ATmega128 37K 2K NETCONF ATmega128 23K 627 JSON parser CC2538 4.6K 8 CBOR parser CC2538 1.5K 2.6K DTLS server ARM7 15K 4 DTLS server MSP430 15K 4 Certificate MSP430 23K Crypto MSP430 2-8K Thomas says that the size of the CoAP server is rather arbitrary, as its size depends mostly on the implementation of the underlying library modules and interfaces. Payload sizes are compared for the following request payloads, where each attribute value is null (N.B. these sizes are educated guesses, will be replaced with generated data). The identifier are assumed to be a string representation of the OID. Sizes for SysUpTime differ due to preambles of payload. "CBOR opt" stands for CBOR payload where the strings are replaced by table numbers. Request BERR SNMP JSON CBOR CBOR opt IPnetTOMediaTable 205 327 ~327 ~51 lowpanIfStatsTable 710 614 121 sysUpTime 29 13 ~13 20 RESTCONF example

To express CBOR structures , this document uses the following conventions: A declaration of a CBOR variable has the form:

where "name" is the name of the variable, and "datatype" its CBOR datatype. The name of the variable has no encoding in the CBOR data. "datatype" can be a CBOR primitive such as: A text string (major type 3) An unsigned integer (major type 0) A map (major type 5), where each first element of a pair is of datatype x, and each second element of datatype y. A '.' character for either x or y means that all datatypes for that element are valid. A datatype can also be a CBOR structure, in which case the variable's "datatype" field contains the name of the CBOR structure. Such CBOR structure is defined by a character sequence consisting of first its name, then a '{' character, then its subfields and finally a '}' character. A CBOR structure can be encapsulated in an array, in which case its name in its definition is preceded by a '*' character. Otherwise the structure is just a grouping of fields, but without actual encoding of such grouping. The name of an optional field is preceded by a '?' character. This means, that the field may be omitted if not required.

CoMI and LWM2M, both, provide RESTful device management services over CoAP. Differences between the designs are highlighted in this section. Unlike CoMI, which enables the use of SMIv2 and YANG data models for device management, LWM2M defines a new object resource model. This means that data models need to be redefined in order to use LWM2M. In contrast, CoMI provides access to a large variety of SMIv2 and YANG data modules that can be used immediately. Objects and resources within CoMI are identified with a YANG hash value, however, each object is described as a link in the CoRE Link Format by LWM2M. This approach by LWM2M can lead to larger complex URIs and more importantly payloads can grow large in size. Using a hash value to represent the objects and resources allows URIs and payloads to be smaller in size, which is important for constrained devices that may not have enough resources to process large messages. LWM2M encodes payload data in Type-length-value (TLV), JSON or plain text formats. While the TLV encoding is binary and can result in reduced message sizes, JSON and plain text are likely to result in large message sizes when lots of resources are being monitored or configured. Furthermore, CoMI's use of CBOR gives it an advantage over the LWM2M's TLV encoding as well since this too is more efficient [citation needed]. CoMI is aligned with RESTCONF for constrained devices and uses YANG data models that have objects containing resources organized in a tree-like structure. On the other hand, LWM2M uses a very flat data model that follows the "object/instance/resouce" format, with no possibility to have subresouces. Complex data models are, as such, harder to model with LWM2M. In situations where resources need to be modified, CoMI uses the CoAP PATCH operation when resources are modified partially. However, LWM2M uses the CoAP PUT and POST operations, even when a subset of the resource needs modifications.