All SecurityProtocols shall implement the OpenSecureChannel and CloseSecureChannel services defined in OPC 10000-4. These Services specify how to establish a SecureChannel and how to apply security to Messages exchanged over that SecureChannel. The Messages exchanged and the security algorithms applied to them are shown in Figure 10.

SecurityProtocols shall support three SecurityModes: None, SignOnly and SignAndEncrypt. If the SecurityMode is None then no security is used and the security handshake shown in Figure 10 is not required. However, a SecurityProtocol implementation shall still maintain a logical channel and provide a unique identifier for the SecureChannel.

Figure 10 – Security handshake

Each SecurityProtocol mapping specifies exactly how to apply the security algorithms to the Message. A set of security algorithms that shall be used together during a security handshake is called a SecurityPolicy. OPC 10000-7 defines standard SecurityPolicies as parts of the standard Profiles which OPC UA applications are expected to support. OPC 10000-7 also defines a URI for each standard SecurityPolicy.

A Stack is expected to have built in knowledge of the SecurityPolicies that it supports. applications specify the SecurityPolicy they wish to use by passing the URI to the Stack.

Table 35 defines the contents of a SecurityPolicy. Each SecurityProtocol mapping specifies how to use each of the parameters in the SecurityPolicy. A SecurityProtocol mapping may not make use of all of the parameters.

Table 35 – SecurityPolicy

Name	Description
PolicyUri	The URI assigned to the SecurityPolicy.
SymmetricSignatureAlgorithm	The symmetric signature algorithm to use.
SymmetricEncryptionAlgorithm	The symmetric encryption algorithm to use.
AsymmetricSignatureAlgorithm	The asymmetric signature algorithm to use.
AsymmetricEncryptionAlgorithm	The asymmetric encryption algorithm to use.
MinAsymmetricKeyLength	The minimum length, in bits, for an asymmetric key.
MaxAsymmetricKeyLength	The maximum length, in bits, for an asymmetric key.
KeyDerivationAlgorithm	The key derivation algorithm to use.
DerivedSignatureKeyLength	The length in bits of the derived key used for Message authentication.
CertificateSignatureAlgorithm	The asymmetric signature algorithm used to sign certificates.
SecureChannelNonceLength	The length, in bytes, of the Nonces exchanged when creating a SecureChannel.

The KeyDerivationAlgorithm is used to create the keys used to secure Messages sent over the SecureChannel. The length of the keys used for encryption is implied by the SymmetricEncryptionAlgorithm. The length of the keys used for creating Signatures are specified by the DerivedSignatureKeyLength.

The CertificateSignatureAlgorithm is used to sign the Certificates used for asymmetric cryptography. OPC 10000-7 specifies the bit lengths that need to be supported for each SecurityPolicy.

The SecureChannelNonceLength specifies the length of the Nonces exhanged when establishing a SecureChannel (see 6.7.4).

OPC UA applications use Certificates to store the Public Keys needed for Asymmetric Cryptography operations. All SecurityProtocols use X.509 v3 Certificates (see X.509 v3) encoded using the DER format (see X690). Certificates used by OPC UA applications shall also conform to RFC 3280 which defines a profile for X.509 v3 Certificates when they are used as part of an Internet based application.

The ServerCertificate and ClientCertificate parameters used in the abstract OpenSecureChannel service are instances of the Application Instance Certificate Data Type. 6.2.2 describes how to create an X.509 v3 Certificate that can be used as an Application Instance Certificate.

An Application Instance Certificate is a ByteString containing the DER encoded form (see X690) of an X.509 v3 Certificate. This Certificate is issued by certifying authority and identifies an instance of an application running on a single host. The X.509 v3 fields contained in an Application Instance Certificate are described in Table 36. The fields are defined completely in RFC 3280.

Table 36 also provides a mapping from the RFC 3280 terms to the terms used in the abstract definition of an Application Instance Certificate defined in OPC 10000-4.

Table 36 – Application Instance Certificate

Name	OPC 10000-4 Parameter Name	Description
Application Instance Certificate		An X.509 v3 Certificate.
version	version	shall be “V3”
serialNumber	serialNumber	The serial number assigned by the issuer.
signatureAlgorithm	signatureAlgorithm	The algorithm used to sign the Certificate.
signature	signature	The signature created by the Issuer.
issuer	issuer	The distinguished name of the Certificate used to create the signature. The issuer field is completely described in RFC 3280.
validity	validTo, validFrom	When the Certificate becomes valid and when it expires.
subject	subject	The distinguished name of the application Instance. The Common Name attribute shall be specified and should be the productName or a suitable equivalent. The Organization Name attribute shall be the name of the Organization that executes the application instance. This organization is usually not the vendor of the application. Other attributes may be specified. The subject field is completely described in RFC 3280.
subjectAltName	applicationUri, hostnames	The alternate names for the application Instance. Shall include a uniformResourceIdentifier which is equal to the applicationUri. The URI shall be a valid URL (see RFC 1738) or a valid URN (see RFC 2141). Servers shall specify a partial or a fully qualified dNSName or a static IPAddress which identifies the machine where the application Instance runs. Additional dNSNames may be specified if the machine has multiple names. The subjectAltName field is completely described in RFC 3280.
publicKey	publicKey	The public key associated with the Certificate.
keyUsage	keyUsage	Specifies how the Certificate key may be used. Shall include digitalSignature, nonRepudiation, keyEncipherment and dataEncipherment. Other key uses are allowed.
extendedKeyUsage	keyUsage	Specifies additional key uses for the Certificate. Shall specify 'serverAuth and/or clientAuth. Other key uses are allowed.
authorityKeyIdentifier	(no mapping)	Provides more information about the key used to sign the Certificate. It shall be specified for Certificates signed by a CA. It should be specified for self-signed Certificates.

Any X.509 v3 Certificate may be signed by CA which means that validating the signature requires access to the X.509 v3 Certificate belonging to the signing CA. Whenever an application validates a signature it must recursively build a chain of Certificates by finding the issuer Certificate, validating the Certificate and then repeat the process for the issuer Certificate. The chain ends with a self-signed Certificate.

The number of CAs used in a system should be small so it is common to install the necessary CAs on each machine with an OPC UA application. However, applications have the option of including a partial or complete chain whenever they pass a Certificate to a peer during the SecureChannel negotiation and during the CreateSession/ActivateSession handshake. All OPC UA applications shall accept partial or complete chains in any field that contains a DER encoded Certificate.

Chains are stored in a ByteString by simply appending the DER encoded form of the Certificates. The first Certificate shall be the end Certificate followed by its issuer. If the root CA is sent as part of the chain it is last Certificate appended to the ByteString.

Chains are parsed by extracting the length of each Certificate from the DER encoding. For Certificates with lengths less than 65 535 bytes it is a MSB encoded UInt16 starting at the 3rd byte.

All SecurityProtocols require that system clocks on communicating machines be reasonably synchronized in order to check the expiry times for Certificates or Messages. The amount of clock skew that can be tolerated depends on the system security requirements and applications shall allow administrators to configure the acceptable clock skew when verifying times. A suitable default value is 5 minutes.

The Network Time Protocol (NTP) provides a standard way to synchronize a machine clock with a time server on the network. Systems running on a machine with a full featured operating system like Windows or Linux will already support NTP or an equivalent. Devices running embedded operating systems should support NTP.

If a device operating system cannot practically support NTP then an OPC UA application can use the Timestamps in the ResponseHeader (see OPC 10000-4) to synchronize its clock. In this scenario, the OPC UA application will have to know the URL for a Discovery Server on a machine known to have the correct time. The OPC UA application or a separate background utility would call the FindServers Service and set its clock to the time specified in the ResponseHeader. This process will need to be repeated periodically because clocks can drift over time.

All times in OPC UA are in UTC, however, UTC can include discontinuities due to leap seconds or repeating seconds added to deal with variations in the earth’s orbit and rotation. Servers that have access to source for International Atomic Time (TAI) may choose to use this instead of UTC. That said, Clients must always be prepared to deal with discontinuities due to the UTC or simply because the system clock is adjusted on the Server machine.

Kerberos UserIdentityTokens can be passed to the Server using the IssuedIdentityToken. The body of the token is an XML element that contains the WS-Security token as defined in the Kerberos Token Profile (Kerberos) specification.

Servers that support Kerberos authentication shall provide a UserTokenPolicy which specifies what version of the Kerberos Token Profile is being used, the Kerberos Realm and the Kerberos Principal Name for the Server. The Realm and Principal name are combined together with a simple syntax and placed in the issuerEndpointUri as shown in Table 37.

Table 37 – Kerberos UserTokenPolicy

Name	Description
tokenType	ISSUEDTOKEN_3
issuedTypeType	http://docs.oasis-open.org/wss/oasis-wss-kerberos-token-profile-1.1
issuerEndpointUri	A string with the form \\<realm>\<server principal name> where <realm> is the Kerberos realm name (e.g. Windows Domain); <server principal name> is the Kerberos principal name for the OPC UA Server.

The interface between the Client and Server applications and the Kerberos Authentication Service is application specific. The realm is the DomainName when using a Windows Domain controller as the Kerberos provider.

JSON Web Token (JWT) UserIdentityTokens can be passed to the Server using the IssuedIdentityToken. The body of the token is a string that contains the JWT as defined in RFC 7159.

Servers that support JWT authentication shall provide a UserTokenPolicy which specifies the Authorization Service which provides the token and the parameters needed to access that service. The parameters are specified by a JSON object specified as the issuerEndpointUrl. The contents of this JSON object are described in Table 39. The general UserTokenPolicy settings for JWT are defined in Table 38.

Table 38 – JWT UserTokenPolicy

Name	Description
tokenType	ISSUEDTOKEN_3
issuedTokenType	http://opcfoundation.org/UA/UserToken#JWT
issuerEndpointUrl	For JWTs this is a JSON object with fields defined in Table 39.

Table 39 – JWT IssuerEndpointUrl Definition

Name	Type	Description
IssuerEndpointUrl	JSON object	Specifies the parameters for a JWT UserIdentityToken.
ua:resourceId	String	The URI identifying the Server to the Authorization Service. If not specified, the Server’s ApplicationUri is used.
ua:authorityUrl	String	The base URL for the Authorization Service. This URL may be used to discover additional information about the authority. This field is equivalent to the “issuer” defined in OpenID-Discovery.
ua:authorityProfileUri	String	The profile that defines the interactions with the authority. If not specified, the URI is “http://opcfoundation.org/UA/Authorization#OAuth2”.
ua:tokenEndpoint	String	A path relative to the base URL used to request Access Tokens. This field is equivalent to the “token_endpoint” defined in OpenID-Discovery.
ua:authorizationEndpoint	String	A path relative to the base URL used to validate user credentials. This field is equivalent to the “authorization_endpoint” defined in OpenID-Discovery.
ua:requestTypes	JSON array String	The list of request types supported by the authority. The possible values depend on the authorityProfileUri. OPC 10000-7 specifies the default for each authority profile defined.
ua:scopes	JSON array String	A list of Scopes that are understood by the Server. If not specified, the Client may be able to access any Scope supported by the Authorization Service. This field is equivalent to the “scopes_supported” defined in OpenID-Discovery.

The OAuth2 Authorization Framework (see RFC 6749) provides a web based mechanism to request claims based Access Tokens from an Authorization Service (AS) that is supported by many major companies providing cloud infrastructure. These Access Tokens are passed to by a Client to a Server in a UserIdentityToken as described in OPC 10000-4.

The OpenID Connect specification (see OpenID) builds on the OAuth2 specification by defining the contents of the Access Tokens more strictly.

The OAuth2 specification supports a number of use cases (called ‘flows’) to handle different application requirements. The use cases that are relevant to OPC UA are discussed below.

The JSON Web Token is the Access Token format which this specification requires when using OAuth2. The JWT supports signatures using asymmetric cryptography which implies that Servers which accept the Access Token must have access to the Certificate used by the Authorization Service (AS). The OpenID Connect Discovery specification is implemented by many AS products and provides a mechanism to fetch the AS Certificate via an HTTP request. If the AS does not support the discovery specification, then the signing Certificate will have to be provided to the Server when the location of the AS is added to the Server configuration.

Access Tokens expire and all Servers should revoke any privileges granted to the Session when the Access Token expires. If the Server allows for anonymous users, the Server should allow the Session to stay open but treat it as an anonymous user. If the Server does not allow anonymous users, it should close the Session immediately.

Clients know when the Access Token will expire and should request a new the Access Token and call ActivateSession before the old Access Token expires.

The JWT format allows the Authorization Service to insert any number of fields. The mandatory fields are defined in RFC 7159. Some additional fields are defined in Table 40 (see RFC 7523).

Table 40 – Access Token Claims

Field	Description
sub	The subject for the token. Usually the client_id which identifies the Client. If returned from an Identity Provider it may be a unique identifier for the user.
aud	The audience for the token. Usually the resource_id which identifies for the Server or the Server ApplicationUri.
name	A human readable name for the Client application or user.
scp	A list of Scopes granted to the subject. Scopes apply to the Access Token and restrict how it may be used. Usually permissions or other restriction which limit access rights.
nonce	A nonce used to mitigate replay attacks. Shall be the value provided by the Client in the request.
groups	A list of groups which assigned to the subject. Usually a list of unique identifiers for platform specific security groups. For example, Azure AD user account groups may be returned in this claim.
roles	A list of roles which assigned to the subject. Roles apply to the requestor and described what the requestor can do with the resource. Usually a list of unique identifiers for roles known to the Authorization Service. These values are typically mapped to the Roles defined in OPC 10000-3.

The authorization code flow is available to Clients which allow interaction with a human user. The Client application displays a window with a web browser which sends an HTTP GET to the Identity Provider. When the human user enters credentials that the Identity Provider validates the Identity Provider returns an authorization code which is passed to the Authorization Service. The Authorization Service validates the code and returns an Access Token to the Client.

The complete flow is described in RFC 6749 Clause 4.1.

A requestType of “authorization_code” in the UserTokenPolicy (see 6.5.2) means the Authorization Service supports the authorization code flow.

The refresh token flow applies when a Client application has access to a refresh token returned in a previous response to an authorization code request. The refresh token allows applications to skip the step that requires human interaction with the Identity Provider. This flow is initiated when the Client sends the refresh token to Authorization Service which validates it and returns an Access Token. A Client that saves the refresh token for later use shall use encryption or other means to ensure the refresh token cannot be accessed by unauthorized parties.

The complete flow is described in RFC 6749 Clause 6.

A requestType is not defined since support for refresh token is determined by checking the response to an authorization code request.

The client credentials flow applies when a Client application cannot prompt a human user for input. This flow requires a secret know to the Authorization Service which the Client application can protect. This flow is initiated when the Client sends the client_secret to Authorization Service which validates it and returns an Access Token.

The complete flow is described in RFC 6749 Clause 4.4.

A requestType of “client_credentials” in the UserTokenPolicy (see 6.5.2) means the Authorization Service supports the client credentials flow.

Note: Deprecated in Version 1.03 because WS-SecureConversation has not been widely adopted by industry. …..

OPC UA Secure Conversation (UASC) allows secure communication using binary encoded Messages.

UASC is designed to operate with different TransportProtocols that may have limited buffer sizes. For this reason, OPC UA Secure Conversation will break OPC UA Messages into several pieces (called ‘MessageChunks’) that are smaller than the buffer size allowed by the TransportProtocol. UASC requires a TransportProtocol buffer size that is at least 8 192 bytes.

All security is applied to individual MessageChunks and not the entire OPC UA Message. A Stack that implements UASC is responsible for verifying the security on each MessageChunk received and reconstructing the original OPC UA Message.

All MessageChunks will have a 4-byte sequence assigned to them. These sequence numbers are used to detect and prevent replay attacks.

UASC requires a TransportProtocol that will preserve the order of MessageChunks, however, a UASC implementation does not necessarily process the Messages in the order that they were received.

Figure 11 shows the structure of a MessageChunk and how security is applied to the Message.

Figure 11 – OPC UA Secure Conversation MessageChunk

Every MessageChunk has a Message header with the fields defined in Table 41.

Table 41 – OPC UA Secure Conversation Message header

Name	Data Type	Description
MessageType	Byte [3]	A three byte ASCII code that identifies the Message type. The following values are defined at this time: MSGA Message secured with the keys associated with a channel. OPN OpenSecureChannel Message. CLO CloseSecureChannel Message.
IsFinal	Byte	A one byte ASCII code that indicates whether the MessageChunk is the final chunk in a Message. The following values are defined at this time: C An intermediate chunk. F The final chunk. A The final chunk (used when an error occurred and the Message is aborted). This field is only meaningful for MessageType of ‘MSG’ This field is always ‘F’ for other MessageTypes.
MessageSize	UInt32	The length of the MessageChunk, in bytes. The length starts from the beginning of the MessageType field.
SecureChannelId	UInt32	A unique identifier for the SecureChannel assigned by the Server. If a Server receives a SecureChannelId which it does not recognize it shall return an appropriate transport layer error. When a Server starts the first SecureChannelId used should be a value that is likely to be unique after each restart. This ensures that a Server restart does not cause previously connected Clients to accidently ‘reuse’ SecureChannels that did not belong to them.

The Message header is followed by a security header which specifies what cryptography operations have been applied to the Message. There are two versions of the security header which depend on the type of security applied to the Message. The security header used for asymmetric algorithms is defined in Table 42. Asymmetric algorithms are used to secure the OpenSecureChannel Messages. PKCS #1 defines a set of asymmetric algorithms that may be used by UASC implementations. The AsymmetricKeyWrapAlgorithm element of the SecurityPolicy structure defined in Table 35 is not used by UASC implementations.

Table 42 – Asymmetric algorithm Security header

Name	Data Type	Description
SecurityPolicyUriLength	Int32	The length of the SecurityPolicyUri in bytes. This value shall not exceed 255 bytes. If a URI is not specified this value may be 0 or -1. Other negative values are invalid.
SecurityPolicyUri	Byte [*]	The URI of the Security Policy used to secure the Message. This field is encoded as a UTF-8 string without a null terminator.
SenderCertificateLength	Int32	The length of the SenderCertificate in bytes. This value shall not exceed MaxSenderCertificateSize bytes. If a certificate is not specified this value may be 0 or -1. Other negative values are invalid.
SenderCertificate	Byte [*]	The X.509 v3 Certificate assigned to the sending application Instance. This is a DER encoded blob. The structure of an X.509 v3 Certificate is defined in X.509 v3. The DER format for a Certificate is defined in X690 This indicates what Private Key was used to sign the MessageChunk. The Stack shall close the channel and report an error to the application if the SenderCertificate is too large for the buffer size supported by the transport layer. This field shall be null if the Message is not signed. If the Certificate is signed by a CA, the DER encoded CA Certificate may be appended after the Certificate in the byte array. If the CA Certificate is also signed by another CA this process is repeated until the entire Certificate chain is in the buffer or if MaxSenderCertificateSize limit is reached (the process stops after the last whole Certificate that can be added without exceeding the MaxSenderCertificateSize limit). Receivers can extract the Certificates from the byte array by using the Certificate size contained in DER header (see X.509 v3). Receivers that do not handle Certificate chains shall ignore the extra bytes.
ReceiverCertificateThumbprintLength	Int32	The length of the ReceiverCertificateThumbprint in bytes. If encrypted the length of this field is 20 bytes. If not encrypted the value may be 0 or -1. Other negative values are invalid.
ReceiverCertificateThumbprint	Byte [*]	The thumbprint of the X.509 v3 Certificate assigned to the receiving application Instance. The thumbprint is the CertificateDigest of the DER encoded form of the Certificate. This indicates what public key was used to encrypt the MessageChunk. This field shall be null if the Message is not encrypted.

The receiver shall close the communication channel if any of the fields in the security header have invalid lengths.

The SenderCertificate, including any chains, shall be small enough to fit into a single MessageChunk and leave room for at least one byte of body information. The maximum size for the SenderCertificate can be calculated with this formula:

MaxSenderCertificateSize =

MessageChunkSize –

12 - // Header size

4 - // SecurityPolicyUriLength

SecurityPolicyUri -// UTF-8 encoded string

4 - // SenderCertificateLength

4 - // ReceiverCertificateThumbprintLength

20 - // ReceiverCertificateThumbprint

8 - // SequenceHeader size

1 - // Minimum body size

1 - // PaddingSize if present

Padding - // Padding if present

ExtraPadding - // ExtraPadding if present

AsymmetricSignatureSize// If present

The MessageChunkSize depends on the transport protocol but shall be at least 8 192 bytes. The AsymmetricSignatureSize depends on the number of bits in the public key for the SenderCertificate. The Int32FieldLength is the length of an encoded Int32 value and it is always 4 bytes.

The security header used for symmetric algorithms defined in Table 43. Symmetric algorithms are used to secure all Messages other than the OpenSecureChannel Messages. FIPS 197 define symmetric encryption algorithms that UASC implementations may use. FIPS 180-2 and HMAC define some symmetric signature algorithms.

Table 43 – Symmetric algorithm Security header

Name	Data Type	Description
TokenId	UInt32	A unique identifier for the SecureChannel SecurityToken used to secure the Message. This identifier is returned by the Server in an OpenSecureChannel response Message. If a Server receives a TokenId which it does not recognize it shall return an appropriate transport layer error.

The security header is always followed by the sequence header which is defined in Table 44. The sequence header ensures that the first encrypted block of every Message sent over a channel will start with different data.

Table 44 – Sequence header

Name	Data Type	Description
SequenceNumber	UInt32	A monotonically increasing sequence number assigned by the sender to each MessageChunk sent over the SecureChannel.
RequestId	UInt32	An identifier assigned by the Client to OPC UA request Message. All MessageChunks for the request and the associated response use the same identifier.

A SequenceNumber may not be reused for any TokenId. The SecurityToken lifetime should be short enough to ensure that this never happens; however, if it does the receiver should treat it as a transport error and force a reconnect.

The SequenceNumber shall also monotonically increase for all Messages and shall not wrap around until it is greater than 4 294 966 271 (UInt32.MaxValue – 1 024). The first number after the wrap around shall be less than 1 024. Note that this requirement means that a SequenceNumber does not reset when a new TokenId is issued. The SequenceNumber shall be incremented by exactly one for each MessageChunk sent unless the communication channel was interrupted and re-established. Gaps are permitted between the SequenceNumber for the last MessageChunk received before the interruption and the SequenceNumber for first MessageChunk received after communication was re-established. Note that the first MessageChunk after a network interruption is always an OpenSecureChannel request or response. If gaps occur in any other case the receiver shall close the SecureChannel.

The sequence header is followed by the Message body which is encoded with the OPC UA Binary encoding as described in 5.2.9. The body may be split across multiple MessageChunks.

Each MessageChunk also has a footer with the fields defined in Table 45.

Table 45 – OPC UA Secure Conversation Message footer

Name	Data Type	Description
PaddingSize	Byte	The number of padding bytes (not including the byte for the PaddingSize).
Padding	Byte [*]	Padding added to the end of the Message to ensure length of the data to encrypt is an integer multiple of the encryption block size. The value of each byte of the padding is equal to PaddingSize.
ExtraPaddingSize	Byte	The most significant byte of a two-byte integer used to specify the padding size when the key used to encrypt the message chunk is larger than 2 048 bits. This field is omitted if the key length is less than or equal to 2 048 bits.
Signature	Byte [*]	The signature for the MessageChunk. The signature includes the all headers, all Message data, the PaddingSize and the Padding.

The formula to calculate the amount of padding depends on the amount of data that needs to be sent (called BytesToWrite). The sender shall first calculate the maximum amount of space available in the MessageChunk (called MaxBodySize) using the following formula:

MaxBodySize = PlainTextBlockSize * Floor ((MessageChunkSize –

HeaderSize - 1)/CipherTextBlockSize) –

SequenceHeaderSize – SignatureSize;

The HeaderSize includes the MessageHeader and the SecurityHeader. The SequenceHeaderSize is always 8 bytes.

During encryption a block with a size equal to PlainTextBlockSize is processed to produce a block with size equal to CipherTextBlockSize. These values depend on the encryption algorithm and may be the same.

The OPC UA Message can fit into a single chunk if BytesToWrite is less than or equal to the MaxBodySize. In this case the PaddingSize is calculated with this formula:

PaddingSize = PlainTextBlockSize –

((BytesToWrite + SignatureSize + 1) % PlainTextBlockSize);

If the BytesToWrite is greater than MaxBodySize the sender shall write MaxBodySize bytes with a PaddingSize of 0. The remaining BytesToWrite – MaxBodySize bytes shall be sent in subsequent MessageChunks.

The PaddingSize and Padding fields are not present if the MessageChunk is not encrypted.

The Signature field is not present if the MessageChunk is not signed.

MessageChunks are sent as they are encoded. MessageChunks belonging to the same Message shall be sent sequentially. If an error occurs creating a MessageChunk then the sender shall send a final MessageChunk to the receiver that tells the receiver that an error occurred and that it should discard the previous chunks. The sender indicates that the MessageChunk contains an error by setting the IsFinal flag to ‘A’ (for Abort). Table 46 specifies the contents of the Message abort MessageChunk.

Table 46 – OPC UA Secure Conversation Message abort body

Name	Data Type	Description
Error	UInt32	The numeric code for the error. This shall be one of the values listed in Table 55.
Reason	String	A more verbose description of the error. This string shall not be more than 4 096 bytes. A Client shall ignore strings that are longer than this.

The receiver shall check the security on the abort MessageChunk before processing it. If everything is ok, then the receiver shall ignore the Message but shall not close the SecureChannel. The Client shall report the error back to the application as StatusCode for the request. If the Client is the sender, then it shall report the error without waiting for a response from the Server.

Most Messages require a SecureChannel to be established. A Client does this by sending an OpenSecureChannel request to the Server. The Server shall validate the Message and the ClientCertificate and return an OpenSecureChannel response. Some of the parameters defined for the OpenSecureChannel service are specified in the security header (see 6.7.2) instead of the body of the Message. Table 47 lists the parameters that appear in the body of the Message.

Note that OPC 10000-4 is an abstract specification which defines interfaces that can work with any protocol. This specification provides a concrete implementation for specific protocols. This document is the normative reference for all protocols and takes precedence if there are differences with OPC 10000-4.

Table 47 – OPC UA Secure Conversation OpenSecureChannel Service

Name	Data Type
Request
RequestHeader	RequestHeader
ClientProtocolVersion	UInt32
RequestType	SecurityTokenRequestType
SecurityMode	MessageSecurityMode
ClientNonce	ByteString
RequestedLifetime	UInt32
Response
ResponseHeader	ResponseHeader
ServerProtocolVersion	UInt32
SecurityToken	ChannelSecurityToken
SecureChannelId	UInt32
TokenId	UInt32
CreatedAt	UtcTime
RevisedLifetime	UInt32
ServerNonce	ByteString

The ClientProtocolVersion and ServerProtocolVersion parameters are not defined in OPC 10000-4 and are added to the Message to allow backward compatibility if OPC UA-SecureConversation needs to be updated in the future. Receivers always accept numbers greater than the latest version that they support. The receiver with the higher version number is expected to ensure backward compatibility.

If OPC UA-SecureConversation is used with the OPC UA-TCP protocol (see 7.1) then the version numbers specified in the OpenSecureChannel Messages shall be the same as the version numbers specified in the OPC UA-TCP protocol Hello/Acknowledge Messages. The receiver shall close the channel and report a Bad_ProtocolVersionUnsupported error if there is a mismatch.

The Server shall return an error response as described in OPC 10000-4 if there are any errors with the parameters specified by the Client.

The RevisedLifetime tells the Client when it shall renew the SecurityToken by sending another OpenSecureChannel request. The Client shall continue to accept the old SecurityToken until it receives the OpenSecureChannel response. The Server has to accept requests secured with the old SecurityToken until that SecurityToken expires or until it receives a Message from the Client secured with the new SecurityToken. The Server shall reject renew requests if the SenderCertificate is not the same as the one used to create the SecureChannel or if there is a problem decrypting or verifying the signature. The Client shall abandon the SecureChannel if the Certificate used to sign the response is not the same as the Certificate used to encrypt the request. Note that datatype is a UInt32 value representing the number of milliseconds instead of the Double (Duration) defined in OPC 10000-4. This optimization is possible because sub-millisecond timeouts are not supported.

The OpenSecureChannel Messages are signed and encrypted if the SecurityMode is not None (even if the SecurityMode is Sign).

The Nonces shall be cryptographic random numbers with a length specified by the SecureChannelNonceLength of the SecurityPolicy.

See OPC 10000-2 for more information on the requirements for random number generators. The OpenSecureChannel Messages are not signed or encrypted if the SecurityMode is None. The Nonces are ignored and should be set to null. The SecureChannelId and the TokenId are still assigned but no security is applied to Messages exchanged via the channel. The SecurityToken shall still be renewed before the RevisedLifetime expires. Receivers shall still ignore invalid or expired TokenIds.

If the communication channel breaks the Server shall maintain the SecureChannel long enough to allow the Client to reconnect. The RevisedLifetime parameter also tells the Client how long the Server will wait. If the Client cannot reconnect within that period it shall assume the SecureChannel has been closed.

The AuthenticationToken in the RequestHeader shall be set to null.

If an error occurs after the Server has verified Message security it shall return a ServiceFault instead of a OpenSecureChannel response. The ServiceFault Message is described in OPC 10000-4.

If the SecurityMode is not None then the Server shall verify that a SenderCertificate and a ReceiverCertificateThumbprint were specified in the SecurityHeader.

Once the SecureChannel is established the Messages are signed and encrypted with keys derived from the Nonces exchanged in the OpenSecureChannel call. These keys are derived by passing the Nonces to a pseudo-random function which produces a sequence of bytes from a set of inputs. A pseudo-random function is represented by the following function declaration:

Byte[] PRF(

Byte[] secret,

Byte[] seed,

Int32 length,

Int32 offset)

Where length is the number of bytes to return and offset is a number of bytes from the beginning of the sequence.

The lengths of the keys that need to be generated depend on the SecurityPolicy used for the channel. The following information is specified by the SecurityPolicy:

1. SigningKeyLength (from the DerivedSignatureKeyLength);
2. EncryptingKeyLength (implied by the SymmetricEncryptionAlgorithm);
3. EncryptingBlockSize (implied by the SymmetricEncryptionAlgorithm).

The pseudo random function requires a secret and a seed. These values are derived from the Nonces exchanged in the OpenSecureChannel request and response. Table 48 specifies how to derive the secrets and seeds when using RSA based SecurityPolicies.

Table 48 – PRF inputs for RSA based SecurityPolicies

Name	Derivation
ClientSecret	The value of the ClientNonce provided in the OpenSecureChannel request.
ClientSeed	The value of the ClientNonce provided in the OpenSecureChannel request.
ServerSecret	The value of the ServerNonce provided in the OpenSecureChannel response.
ServerSeed	The value of the ServerNonce provided in the OpenSecureChannel response.

The parameters passed to the pseudo random function are specified in Table 49.

Table 49 – Cryptography key generation parameters

Key	Secret	Seed	Length	Offset
ClientSigningKey	ServerSecret	ClientSeed	SigningKeyLength	0
ClientEncryptingKey	ServerSecret	ClientSeed	EncryptingKeyLength	SigningKeyLength
ClientInitializationVector	ServerSecret	ClientSeed	EncryptingBlockSize	SigningKeyLength+ EncryptingKeyLength
ServerSigningKey	ClientSecret	ServerSeed	SigningKeyLength	0
ServerEncryptingKey	ClientSecret	ServerSeed	EncryptingKeyLength	SigningKeyLength
ServerInitializationVector	ClientSecret	ServerSeed	EncryptingBlockSize	SigningKeyLength+ EncryptingKeyLength

The Client keys are used to secure Messages sent by the Client. The Server keys are used to secure Messages sent by the Server.

The SSL/TLS specification defines a pseudo random function called P_HASH which is used for this purpose. The function is iterated until it produces enough data for all of the required keys. The Offset in Table 49 references to the offset from the start of the generated data.

The P_ hash algorithm is defined as follows:

P_HASH(secret, seed) = HMAC_HASH(secret, A(1) + seed) +

HMAC_HASH(secret, A(2) + seed) +

HMAC_HASH(secret, A(3) + seed) + ...

Where A(n) is defined as:

A(0) = seed

A(n) = HMAC_HASH(secret, A(n-1))

+ indicates that the results are appended to previous results.

Where ‘HASH’ is a hash function such as SHA256. The hash function to use depends on the SecurityPolicyUri.

The contents of the MessageChunk shall not be interpreted until the Message is decrypted and the signature and sequence number verified.

If an error occurs during Message verification the receiver shall close the communication channel. If the receiver is the Server, it shall also send a transport error Message before closing the channel. Once the channel is closed the Client shall attempt to re-open the channel and request a new SecurityToken by sending an OpenSecureChannel request. The mechanism for sending transport errors to the Client depends on the communication channel.

The receiver shall first check the SecureChannelId. This value may be 0 if the Message is an OpenSecureChannel request. For other Messages, it shall report a Bad_SecureChannelUnknown error if the SecureChannelId is not recognized. If the Message is an OpenSecureChannel request and the SecureChannelId is not 0 then the SenderCertificate shall be the same as the SenderCertificate used to create the channel.

If the Message is secured with asymmetric algorithms, then the receiver shall verify that it supports the requested Securi tyPolicy. If the Message is the response sent to the Client, then the SecurityPolicy shall be the same as the one specified in the request. In the Server, the SecurityPolicy shall be the same as the one used to originally create the SecureChannel. The receiver shall check that the Certificate is trusted first and return Bad_CertificateUntrusted on error. The receiver shall then verify the SenderCertificate using the rules defined in OPC 10000-4. The receiver shall report the appropriate error if Certificate validation fails. The receiver shall verify the ReceiverCertificateThumbprint and report a Bad_CertificateUnknown error if it does not recognize it.

If the Message is secured with symmetric algorithms, then a Bad_SecureChannel TokenUnknown e rror shall be reported if the TokenId refers to a SecurityToken that has expired or is not recognized.

If decryption or signature validation fails, then a Bad_SecurityChecksFailed error is reported. If an implementation allows multiple SecurityModes to be used the receiver shall also verify that the Message was secured properly as required by the SecurityMode specified in the OpenSecureChannel request.

After the security validation is complete the receiver shall verify the RequestId and the SequenceNumber. If these checks fail a Bad_SecurityChecksFailed error is reported. The RequestId only needs to be verified by the Client since only the Client knows if it is valid or not.

At this point the SecureChannel knows it is dealing with an authenticated Message that was not tampered with or resent. This means the SecureChannel can return secured error responses if any further problems are encountered.

Stacks that implement UASC shall have a mechanism to log errors when invalid Messages are discarded. This mechanism is intended for developers, systems integrators and administrators to debug network system configuration issues and to detect attacks on the network.