HTTP Message Signatures

HTTP permits and sometimes requires intermediaries to transform messages in a variety of ways. This may result in a recipient receiving a message that is not bitwise equivalent to the message that was originally sent. In such a case, the recipient will be unable to verify a signature over the raw bytes of the sender's HTTP message, as verifying digital signatures or MACs requires both signer and verifier to have the exact same signature base. Since the exact raw bytes of the message cannot be relied upon as a reliable source for a signature base, the signer and verifier must independently create the signature base from their respective versions of the message, via a mechanism that is resilient to safe changes that do not alter the meaning of the message.¶

For a variety of reasons, it is impractical to strictly define what constitutes a safe change versus an unsafe one. Applications use HTTP in a wide variety of ways, and may disagree on whether a particular piece of information in a message (e.g., the body, or the Date header field) is relevant. Thus a general purpose solution must provide signers with some degree of control over which message components are signed.¶

HTTP applications may be running in environments that do not provide complete access to or control over HTTP messages (such as a web browser's JavaScript environment), or may be using libraries that abstract away the details of the protocol (such as the Java HTTPClient library). These applications need to be able to generate and verify signatures despite incomplete knowledge of the HTTP message.¶

• Re-ordering of header fields with different header field names (Section 3.2.2 of [MESSAGING]).
• Combination of header fields with the same field name (Section 3.2.2 of [MESSAGING]).
• Removal of header fields listed in the Connection header field (Section 6.1 of [MESSAGING]).
• Addition of header fields that indicate control options (Section 6.1 of [MESSAGING]).
• Addition or removal of a transfer coding (Section 5.7.2 of [MESSAGING]).
• Addition of header fields such as Via (Section 5.7.1 of [MESSAGING]) and Forwarded (Section 4 of [RFC7239]).

• Combination of header fields with the same field name.
• Reordering of header fields with different names.
• Conversion between different versions of the HTTP protocol (e.g., HTTP/1.x to HTTP/2, or vice-versa).
• Changes in casing (e.g., "Origin" to "origin") of any case-insensitive components such as header field names, request URI scheme, or host.
• Addition or removal of leading or trailing whitespace to a header field value.
• Addition or removal of obs-folds.
• Changes to the request-target and Host header field that when applied together do not result in a change to the message's effective request URI, as defined in Section 5.5 of [MESSAGING].

Additionally, all changes to components not covered by the signature are considered safe.¶

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶

The terms "HTTP message", "HTTP request", "HTTP response", absolute-form, absolute-path, "effective request URI", "gateway", "header field", "intermediary", request-target, "sender", and "recipient" are used as defined in [MESSAGING].¶

The term "method" is to be interpreted as defined in Section 4 of [SEMANTICS].¶

For brevity, the term "signature" on its own is used in this document to refer to both digital signatures (which use asymmetric cryptography) and keyed MACs (which use symmetric cryptography). Similarly, the verb "sign" refers to the generation of either a digital signature or keyed MAC over a given input string. The qualified term "digital signature" refers specifically to the output of an asymmetric cryptographic signing operation.¶

HTTP Message Signature:

A digital signature or keyed MAC that covers one or more portions of an HTTP message. Note that a given HTTP Message can contain multiple HTTP Message Signatures.

Signer:

The entity that is generating or has generated an HTTP Message Signature. Note that multiple entities can act as signers and apply separate HTTP Message Signatures to a given HTTP Message.

Verifier:

An entity that is verifying or has verified an HTTP Message Signature against an HTTP Message. Note that an HTTP Message Signature may be verified multiple times, potentially by different entities.

HTTP Message Component:

A portion of an HTTP message that is capable of being covered by an HTTP Message Signature.

HTTP Message Component Identifier:

A value that uniquely identifies a specific HTTP Message Component in respect to a particular HTTP Message Signature and the HTTP Message it applies to.

HTTP Message Component Value:

The value associated with a given component identifier within the context of a particular HTTP Message. Component values are derived from the HTTP Message and are usually subject to a canonicalization process.

Covered Components:

An ordered set of HTTP message component identifiers for fields (Section 2.1) and derived components (Section 2.2) that indicates the set of message components covered by the signature, never including the @signature-params identifier itself. The order of this set is preserved and communicated between the signer and verifier to facilitate reconstruction of the signature base.

Signature Base:

The sequence of bytes processed by the cryptographic algorithm to produce or verify the HTTP Message Signature. The signature base is generated by the signer and verifier using the covered components set and the HTTP Message.

HTTP Message Signature Algorithm:

A cryptographic algorithm that describes the signing and verification process for the signature, defined in terms of the HTTP_SIGN and HTTP_VERIFY primitives described in Section 3.3.

Key Material:

The key material required to create or verify the signature. The key material is often identified with an explicit key identifier, allowing the signer to indicate to the verifier which key was used.

Creation Time:

A timestamp representing the point in time that the signature was generated, as asserted by the signer.

Expiration Time:

A timestamp representing the point in time after which the signature should no longer be accepted by the verifier, as asserted by the signer.

The term "Unix time" is defined by [POSIX.1], Section 4.16.¶

This document contains non-normative examples of partial and complete HTTP messages. Some examples use a single trailing backslash '' to indicate line wrapping for long values, as per [RFC8792]. The \ character and leading spaces on wrapped lines are not part of the value.¶

• The set of component identifiers (Section 2) that are expected and required. For example, an authorization protocol could mandate that the Authorization header be covered to protect the authorization credentials and mandate the signature parameters contain a created parameter, while an API expecting HTTP message bodies could require the Digest header to be present and covered.
• A means of retrieving the key material used to verify the signature. An application will usually use the keyid parameter of the signature parameters (Section 2.2.1) and define rules for resolving a key from there, though the appropriate key could be known from other means.
• A means of determining the signature algorithm used to verify the signature is appropriate for the key material. For example, the process could use the alg parameter of the signature parameters (Section 2.2.1) to state the algorithm explicitly, derive the algorithm from the key material, or use some pre-configured algorithm agreed upon by the signer and verifier.
• A means of determining that a given key and algorithm presented in the request are appropriate for the request being made. For example, a server expecting only ECDSA signatures should know to reject any RSA signatures, or a server expecting asymmetric cryptography should know to reject any symmetric cryptography.

An application using signatures also has to ensure that the verifier will have access to all required information to re-create the signature base. For example, a server behind a reverse proxy would need to know the original request URI to make use of identifiers like @target-uri. Additionally, an application using signatures in responses would need to ensure that clients receiving signed responses have access to all the signed portions, including any portions of the request that were signed by the server.¶

The details of this kind of profiling are the purview of the application and outside the scope of this specification, however some additional considerations are discussed in Section 7.¶

The component identifier for an HTTP field is the lowercased form of its field name. While HTTP field names are case-insensitive, implementations MUST use lowercased field names (e.g., content-type, date, etag) when using them as component identifiers.¶

Unless overridden by additional parameters and rules, the HTTP field value MUST be canonicalized as a single combined value as defined in Section 5.2 of [SEMANTICS].¶

1. Create an ordered list of the field values of each instance of the field in the message, in the order that they occur (or will occur) in the message.
2. Strip leading and trailing whitespace from each item in the list. Note that since HTTP field values are not allowed to contain leading and trailing whitespace, this will be a no-op in a compliant implementation.
3. Remove any obsolete line-folding within the line and replace it with a single space (" "), as discussed in Section 5.2 of [MESSAGING]. Note that this behavior is specific to [MESSAGING] and does not apply to other versions of the HTTP specification.
4. Concatenate the list of values together with a single comma (",") and a single space (" ") between each item.

The resulting string is the canonicalized component value.¶

Note that some HTTP fields have values with multiple valid serializations that have equivalent semantics. Applications signing and processing such fields MUST consider how to handle the values of such fields to ensure that the signer and verifier can derive the same value, as discussed in Section 7.21.¶

Is serialized by the signature base generation algorithm (Section 2.3) with an empty string value following the colon and space added after the content identifier.¶

Note: these are shown here using the line wrapping algorithm in [RFC8792] due to limitations in the document format that strips trailing spaces from diagrams.¶

Any HTTP field component identifiers MAY have the following parameters in specific circumstances.¶

sf

A boolean flag indicating that the field value is to be canonicalized using strict encoding of the structured field value. Section 2.1.1

key

A string parameter used to select a single member value from a dictionary structured field. Section 2.1.2

In addition to HTTP fields, there are a number of different components that can be derived from the control data, processing context, or other aspects of the HTTP message being signed. Such derived components can be included in the signature base by defining a component identifier and the derivation method for its component value.¶

Derived component identifiers MUST start with the "at" @ character. This differentiates derived component identifiers from HTTP field names, which cannot contain the @ character as per Section 5.1 of [SEMANTICS]. Processors of HTTP Message Signatures MUST treat derived component identifiers separately from field names, as discussed in Section 7.18.¶

@signature-params

The signature metadata parameters for this signature. (Section 2.2.1)

@method

The method used for a request. (Section 2.2.2)

@target-uri

The full target URI for a request. (Section 2.2.3)

@authority

The authority of the target URI for a request. (Section 2.2.4)

@scheme

The scheme of the target URI for a request. (Section 2.2.5)

@request-target

The request target. (Section 2.2.6)

@path

The absolute path portion of the target URI for a request. (Section 2.2.7)

@query

The query portion of the target URI for a request. (Section 2.2.8)

@query-params

The parsed query parameters of the target URI for a request. (Section 2.2.9)

@status

The status code for a response. (Section 2.2.10).

@request-response

A signature from a request message that resulted in this response message. (Section 2.2.11)

Additional derived component identifiers MAY be defined and registered in the HTTP Signatures Derived Component Identifier Registry. (Section 6.3)¶

request:

Values derived from and results applied to an HTTP request message as described in {{Section 3.4 of SEMANTICS.

response:

Values derived from and results applied to an HTTP response message as described in Section 3.4 of [SEMANTICS].

related-response:

Values derived from an HTTP request message and results applied to the HTTP response message that is responding to that specific request.

A component identifier definition MUST define all targets to which it can be applied.¶

1. Let the output be an empty string.
2. For each message component item in the covered components set (in order):
   1. Append the component identifier for the covered component serialized according to the component-identifier rule. Note that this serialization places the component identifier in double quotes and appends any parameters outside of the quotes.
   2. Append a single colon :
   3. Append a single space " "
   4. Determine the component value for the component identifier.
      • If the component identifier starts with an "at" character (@), derive the component's value from the message according to the specific rules defined for the derived component identifier, as in Section 2.2. If the derived component identifier is unknown or the value cannot be derived, produce an error.
      • If the component identifier does not start with an "at" character (@), canonicalize the HTTP field value as described in Section 2.1. If the value cannot be calculated, produce an error.
   5. Append the covered component's canonicalized component value.
   6. Append a single newline \n
3. Append the signature parameters component (Section 2.2.1) as follows:
   1. Append the component identifier for the signature parameters serialized according to the component-identifier rule, i.e. "@signature-params"
   2. Append a single colon :
   3. Append a single space " "
   4. Append the signature parameters' canonicalized component value as defined in Section 2.2.1
4. Return the output string.

• The signer or verifier does not understand the derived component identifier.
• The component identifier identifies a field that is not present in the message or whose value is malformed.
• The component identifier indicates that a structured field serialization is used (via the sf parameter), but the field in question is known to not be a structured field or the type of structured field is not known to the implementation.
• The component identifier is a dictionary member identifier that references a field that is not present in the message, is not a Dictionary Structured Field, or whose value is malformed.
• The component identifier is a dictionary member identifier or a named query parameter identifier that references a member that is not present in the component value, or whose value is malformed. E.g., the identifier is "example-dict";key="c" and the value of the Example-Dict header field is a=1, b=2, which does not have the c value.

Note that the example signature base here, or anywhere else within this specification, does not include the final newline that ends the displayed example.¶

Creation of an HTTP message signature is a process that takes as its input the message and the requirements for the application. The output is a signature value and set of signature parameters that can be applied to the message.¶

1. The signer chooses an HTTP signature algorithm and key material for signing. The signer MUST choose key material that is appropriate for the signature's algorithm, and that conforms to any requirements defined by the algorithm, such as key size or format. The mechanism by which the signer chooses the algorithm and key material is out of scope for this document.
2. The signer sets the signature's creation time to the current time.
3. If applicable, the signer sets the signature's expiration time property to the time at which the signature is to expire. The expiration is a hint to the verifier, expressing the time at which the signer is no longer willing to vouch for the safety of the signature.
4. The signer creates an ordered set of component identifiers representing the message components to be covered by the signature, and attaches signature metadata parameters to this set. The serialized value of this is later used as the value of the Signature-Input field as described in Section 4.1.
   • Once an order of covered components is chosen, the order MUST NOT change for the life of the signature.
   • Each covered component identifier MUST be either an HTTP field in the message Section 2.1 or a derived component identifier listed in Section 2.2 or its associated registry.
   • Signers of a request SHOULD include some or all of the message control data in the covered components, such as the @method, @authority, @target-uri, or some combination thereof.
   • Signers SHOULD include the created signature metadata parameter to indicate when the signature was created.
   • The @signature-params derived component identifier is not explicitly listed in the list of covered component identifiers, because it is required to always be present as the last line in the signature base. This ensures that a signature always covers its own metadata.
   • Further guidance on what to include in this set and in what order is out of scope for this document.
5. The signer creates the signature base using these parameters and the signature base creation algorithm. (Section 2.3)
6. The signer uses the HTTP_SIGN primitive function to sign the signature base with the chosen signing algorithm using the key material chosen by the signer. The HTTP_SIGN primitive and several concrete applications of signing algorithms are defined in in Section 3.3.
7. The byte array output of the signature function is the HTTP message signature output value to be included in the Signature field as defined in Section 4.2.

Note that the RSA PSS algorithm in use here is non-deterministic, meaning a different signature value will be created every time the algorithm is run. The signature value provided here can be validated against the given keys, but newly-generated signature values are not expected to match the example. See Section 7.19.¶

Verification of an HTTP message signature is a process that takes as its input the message (including Signature and Signature-Input fields) and the requirements for the application. The output of the verification is either a positive verification or an error.¶

1. Parse the Signature and Signature-Input fields as described in Section 4.1 and Section 4.2, and extract the signatures to be verified.
   1. If there is more than one signature value present, determine which signature should be processed for this message based on the policy and configuration of the verifier. If an applicable signature is not found, produce an error.
   2. If the chosen Signature value does not have a corresponding Signature-Input value, produce an error.
2. Parse the values of the chosen Signature-Input field as a parameterized structured field inner list item (inner-list) to get the signature parameters for the signature to be verified.
3. Parse the value of the corresponding Signature field to get the byte array value of the signature to be verified.
4. Examine the signature parameters to confirm that the signature meets the requirements described in this document, as well as any additional requirements defined by the application such as which message components are required to be covered by the signature. (Section 3.2.1)
5. Determine the verification key material for this signature. If the key material is known through external means such as static configuration or external protocol negotiation, the verifier will use that. If the key is identified in the signature parameters, the verifier will dereference this to appropriate key material to use with the signature. The verifier has to determine the trustworthiness of the key material for the context in which the signature is presented. If a key is identified that the verifier does not know, does not trust for this request, or does not match something preconfigured, the verification MUST fail.
6. Determine the algorithm to apply for verification:
   1. If the algorithm is known through external means such as static configuration or external protocol negotiation, the verifier will use this algorithm.
   2. If the algorithm is explicitly stated in the signature parameters using a value from the HTTP Message Signatures registry, the verifier will use the referenced algorithm.
   3. If the algorithm can be determined from the keying material, such as through an algorithm field on the key value itself, the verifier will use this algorithm.
   4. If the algorithm is specified in more that one location, such as through static configuration and the algorithm signature parameter, or the algorithm signature parameter and from the key material itself, the resolved algorithms MUST be the same. If the algorithms are not the same, the verifier MUST vail the verification.
7. Use the received HTTP message and the signature's metadata to recreate the signature base, using the algorithm defined in Section 2.3. The value of the @signature-params input is the value of the Signature-Input field for this signature serialized according to the rules described in Section 2.2.1, not including the signature's label from the Signature-Input field.
8. If the key material is appropriate for the algorithm, apply the appropriate HTTP_VERIFY cryptographic verification algorithm to the signature, recalculated signature base, key material, signature value. The HTTP_VERIFY primitive and several concrete algorithms are defined in Section 3.3.
9. The results of the verification algorithm function are the final results of the cryptographic verification function.

If any of the above steps fail or produce an error, the signature validation fails.¶

With the additional requirements that at least the method, path, authority, and cache-control be signed, and that the signature creation timestamp is recent enough at the time of verification, the verification passes.¶

HTTP Message signatures MAY use any cryptographic digital signature or MAC method that is appropriate for the key material, environment, and needs of the signer and verifier.¶

This section contains several common algorithm methods. The method to use can be communicated through the explicit algorithm signature parameter alg defined in Section 2.2.1, by reference to the key material, or through mutual agreement between the signer and verifier.¶

The Signature-Input HTTP field is a Dictionary Structured Field [RFC8941] containing the metadata for one or more message signatures generated from components within the HTTP message. Each member describes a single message signature. The member's name is an identifier that uniquely identifies the message signature within the context of the HTTP message. The member's value is the serialization of the covered components including all signature metadata parameters, using the serialization process defined in Section 2.2.1.¶

To facilitate signature validation, the Signature-Input field value MUST contain the same serialized value used in generating the signature base's @signature-params value.¶

The signer MAY include the Signature-Input field as a trailer to facilitate signing a message after its content has been processed by the signer. However, since intermediaries are allowed to drop trailers as per [SEMANTICS], it is RECOMMENDED that the Signature-Input HTTP field be included only as a header to avoid signatures being inadvertently stripped from a message.¶

Multiple Signature-Input fields MAY be included in a single HTTP message. The signature labels MUST be unique across all field values.¶

The Signature HTTP field is a Dictionary Structured field [RFC8941] containing one or more message signatures generated from components within the HTTP message. Each member's name is a signature identifier that is present as a member name in the Signature-Input Structured field within the HTTP message. Each member's value is a Byte Sequence containing the signature value for the message signature identified by the member name. Any member in the Signature HTTP field that does not have a corresponding member in the HTTP message's Signature-Input HTTP field MUST be ignored.¶

The signer MAY include the Signature field as a trailer to facilitate signing a message after its content has been processed by the signer. However, since intermediaries are allowed to drop trailers as per [SEMANTICS], it is RECOMMENDED that the Signature-Input HTTP field be included only as a header to avoid signatures being inadvertently stripped from a message.¶

Multiple Signature fields MAY be included in a single HTTP message. The signature labels MUST be unique across all field values.¶

Multiple distinct signatures MAY be included in a single message. Each distinct signature MUST have a unique label. Since Signature-Input and Signature are both defined as Dictionary Structured fields, they can be used to include multiple signatures within the same HTTP message by using distinct signature labels. These multiple signatures could be added all by the same signer or could come from several different signers. For example, a signer may include multiple signatures signing the same message components with different keys or algorithms to support verifiers with different capabilities, or a reverse proxy may include information about the client in fields when forwarding the request to a service host, including a signature over the client's original signature values.¶

The following is a non-normative example starts with a signed request from the client. The proxy takes this request validates the client's signature.¶

The proxy then alters the message before forwarding it on to the origin server, changing the target host and adding the Forwarded header defined in [RFC7239].¶

These values are added to the HTTP request message by the proxy. The original signature is included under the identifier sig1, and the reverse proxy's signature is included under the label proxy_sig. The proxy uses the key test-key-rsa to create its signature using the rsa-v1_5-sha256 signature algorithm, while the client's original signature was made using the key id of test-key-rsa-pss and an RSA PSS signature algorithm.¶

The proxy's signature and the client's original signature can be verified independently for the same message, based on the needs of the application. Since the proxy's signature covers the client signature, the backend service fronted by the proxy can trust that the proxy has validated the incoming signature.¶

The Accept-Signature HTTP header field is a Dictionary Structured field [RFC8941] containing the metadata for one or more requested message signatures to be generated from message components of the target HTTP message. Each member describes a single message signature. The member's name is an identifier that uniquely identifies the requested message signature within the context of the target HTTP message. The member's value is the serialization of the desired covered components of the target message, including any allowed signature metadata parameters, using the serialization process defined in Section 2.2.1.¶

The requested signature MAY include parameters, such as a desired algorithm or key identifier. These parameters MUST NOT include parameters that the signer is expected to generate, including the created and nonce parameters.¶

1. Parse the field value as a Dictionary
2. For each member of the dictionary:
   1. The name of the member is the label of the output signature as specified in Section 4.1
   2. Parse the value of the member to obtain the set of covered component identifiers
   3. Process the requested parameters, such as the signing algorithm and key material. If any requested parameters cannot be fulfilled, or if the requested parameters conflict with those deemed appropriate to the target message, the process fails and returns an error.
   4. Select any additional parameters necessary for completing the signature
   5. Create the Signature-Input and Signature header values and associate them with the label
3. Optionally create any additional Signature-Input and Signature values, with unique labels not found in the Accept-Signature field
4. Combine all labeled Signature-Input and Signature values and attach both headers to the target message

Note that by this process, a signature applied to a target message MUST have the same label, MUST have the same set of covered component, and MAY have additional parameters. Also note that the target message MAY include additional signatures not specified by the Accept-Signature field.¶

This document defines HTTP Signature Algorithms, for which IANA is asked to create and maintain a new registry titled "HTTP Signature Algorithms". Initial values for this registry are given in Section 6.1.2. Future assignments and modifications to existing assignment are to be made through the Expert Review registration policy [RFC8126] and shall follow the template presented in Section 6.1.1.¶

Algorithms referenced by algorithm identifiers have to be fully defined with all parameters fixed. Algorithm identifiers in this registry are to be interpreted as whole string values and not as a combination of parts. That is to say, it is expected that implementors understand rsa-pss-sha512 as referring to one specific algorithm with its hash, mask, and salt values set as defined here. Implementors do not parse out the rsa, pss, and sha512 portions of the identifier to determine parameters of the signing algorithm from the string.¶

Algorithms added to this registry MUST NOT be aliases for other entries in the registry.¶

This document defines the signature parameters structure, the values of which may have parameters containing metadata about a message signature. IANA is asked to create and maintain a new registry titled "HTTP Signature Metadata Parameters" to record and maintain the set of parameters defined for use with member values in the signature parameters structure. Initial values for this registry are given in Section 6.2.2. Future assignments and modifications to existing assignments are to be made through the Expert Review registration policy [RFC8126] and shall follow the template presented in Section 6.2.1.¶

This document defines a method for canonicalizing HTTP message components, including components that can be derived from the context of the HTTP message outside of the HTTP fields. These components are identified by a unique string, known as the component identifier. Component identifiers for derived components always start with the "@" (at) symbol to distinguish them from HTTP header fields. IANA is asked to create and maintain a new registry typed "HTTP Signature Derived Component Identifiers" to record and maintain the set of non-field component identifiers and the methods to produce their associated component values. Initial values for this registry are given in Section 6.3.2. Future assignments and modifications to existing assignments are to be made through the Expert Review registration policy [RFC8126] and shall follow the template presented in Section 6.3.1.¶

HTTP Message Signatures only provide security if the signature is verified by the verifier. Since the message to which the signature is attached remains a valid HTTP message without the signature fields, it is possible for a verifier to ignore the output of the verification function and still process the message. Common reasons for this could be relaxed requirements in a development environment or a temporary suspension of enforcing verification during debugging an overall system. Such temporary suspensions are difficult to detect under positive-example testing since a good signature will always trigger a valid response whether or not it has been checked.¶

To detect this, verifiers should be tested using both valid and invalid signatures, ensuring that the invalid signature fails as expected.¶

The use of HTTP Message Signatures does not negate the need for TLS or its equivalent to protect information in transit. Message signatures provide message integrity over the covered message components but do not provide any confidentiality for the communication between parties.¶

TLS provides such confidentiality between the TLS endpoints. As part of this, TLS also protects the signature data itself from being captured by an attacker, which is an important step in preventing signature replay (Section 7.3).¶

When TLS is used, it needs to be deployed according to the recommendations in [BCP195].¶

Since HTTP Message Signatures allows sub-portions of the HTTP message to be signed, it is possible for two different HTTP messages to validate against the same signature. The most extreme form of this would be a signature over no message components. If such a signature were intercepted, it could be replayed at will by an attacker, attached to any HTTP message. Even with sufficient component coverage, a given signature could be applied to two similar HTTP messages, allowing a message to be replayed by an attacker with the signature intact.¶

To counteract these kinds of attacks, it's first important for the signer to cover sufficient portions of the message to differentiate it from other messages. In addition, the signature can use the nonce signature parameter to provide a per-message unique value to allow the verifier to detect replay of the signature itself if a nonce value is repeated. Furthermore, the signer can provide a timestamp for when the signature was created and a time at which the signer considers the signature to be invalid, limiting the utility of a captured signature value.¶

If a verifier wants to trigger a new signature from a signer, it can send the Accept-Signature header field with a new nonce parameter. An attacker that is simply replaying a signature would not be able to generate a new signature with the chosen nonce value.¶

Any portions of the message not covered by the signature are susceptible to modification by an attacker without affecting the signature. An attacker can take advantage of this by introducing a header field or other message component that will change the processing of the message but will not be covered by the signature. Such an altered message would still pass signature verification, but when the verifier processes the message as a whole, the unsigned content injected by the attacker would subvert the trust conveyed by the valid signature and change the outcome of processing the message.¶

To combat this, an application of this specification should require as much of the message as possible to be signed, within the limits of the application and deployment. The verifier should only trust message components that have been signed. Verifiers could also strip out any sensitive unsigned portions of the message before processing of the message continues.¶

The HTTP Message Signatures specification does not define any of its own cryptographic primitives, and instead relies on other specifications to define such elements. If the signature algorithm or key used to process the signature base is vulnerable to any attacks, the resulting signature will also be susceptible to these same attacks.¶

A common attack against signature systems is to force a signature collision, where the same signature value successfully verifies against multiple different inputs. Since this specification relies on reconstruction of the input string based on an HTTP message, and the list of components signed is fixed in the signature, it is difficult but not impossible for an attacker to effect such a collision. An attacker would need to manipulate the HTTP message and its covered message components in order to make the collision effective.¶

To counter this, only vetted keys and signature algorithms should be used to sign HTTP messages. The HTTP Message Signatures Algorithm Registry is one source of potential trusted algorithms.¶

While it is possible for an attacker to substitute the signature parameters value or the signature value separately, the signature base generation algorithm (Section 2.3) always covers the signature parameters as the final value in the input string using a deterministic serialization method. This step strongly binds the signature base with the signature value in a way that makes it much more difficult for an attacker to perform a partial substitution on the signature bases.¶

A foundational assumption of signature-based cryptographic systems is that the signing key is not compromised by an attacker. If the keys used to sign the message are exfiltrated or stolen, the attacker will be able to generate their own signatures using those keys. As a consequence, signers have to protect any signing key material from exfiltration, capture, and use by an attacker.¶

To combat this, signers can rotate keys over time to limit the amount of time stolen keys are useful. Signers can also use key escrow and storage systems to limit the attack surface against keys. Furthermore, the use of asymmetric signing algorithms exposes key material less than the use of symmetric signing algorithms (Section 7.11).¶

An attacker could effectively deny a service by modifying an otherwise benign signature parameter or signed message component. While rejecting a modified message is the desired behavior, consistently failing signatures could lead to the verifier turning off signature checking in order to make systems work again (see Section 7.1).¶

If such failures are common within an application, the signer and verifier should compare their generated signature bases with each other to determine which part of the message is being modified. However, the signer and verifier should not remove the requirement to sign the modified component when it is suspected an attacker is modifying the component.¶

The verifier needs to make sure that the signed message components match those in the message itself. This specification encourages this by requiring the verifier to derive these values from the message, but lazy cacheing or conveyance of the signature base to a processing system could lead to downstream verifiers accepting a message that does not match the presented signature.¶

Since multiple signatures can be applied to one message (Section 4.3), it is possible for an attacker to attach their own signature to a captured message without modifying existing signatures. This new signature could be completely valid based on the attacker's key, or it could be an invalid signature for any number of reasons. Each of these situations need to be accounted for.¶

A verifier processing a set of valid signatures needs to account for all of the signers, identified by the signing keys. Only signatures from expected signers should be accepted, regardless of the cryptographic validity of the signature itself.¶

A verifier processing a set of signatures on a message also needs to determine what to do when one or more of the signatures are not valid. If a message is accepted when at least one signature is valid, then a verifier could drop all invalid signatures from the request before processing the message further. Alternatively, if the verifier rejects a message for a single invalid signature, an attacker could use this to deny service to otherwise valid messages by injecting invalid signatures alongside the valid ones.¶

HTTP Message Signature values are identified in the Signature and Signature-Input field values by unique labels. These labels are chosen only when attaching the signature values to the message and are not accounted for in the signing process. An intermediary adding its own signature is allowed to re-label an existing signature when processing the message.¶

Therefore, applications should not rely on specific labels being present, and applications should not put semantic meaning on the labels themselves. Instead, additional signature parmeters can be used to convey whatever additional meaning is required to be attached to and covered by the signature.¶

The HTTP Message Signatures specification allows for both asymmetric and symmetric cryptography to be applied to HTTP messages. By its nature, symmetric cryptographic methods require the same key material to be known by both the signer and verifier. This effectively means that a verifier is capable of generating a valid signature, since they have access to the same key material. An attacker that is able to compromise a verifier would be able to then impersonate a signer.¶

Where possible, asymmetric methods or secure key agreement mechanisms should be used in order to avoid this type of attack. When symmetric methods are used, distribution of the key material needs to be protected by the overall system. One technique for this is the use of separate cryptographic modules that separate the verification process (and therefore the key material) from other code, minimizing the vulnerable attack surface. Another technique is the use of key derivation functions that allow the signer and verifier to agree on unique keys for each message without having to share the key values directly.¶

Additionally, if symmetric algorithms are allowed within a system, special care must be taken to avoid key downgrade attacks (Section 7.15).¶

Any ambiguity in the generation of the signature base could provide an attacker with leverage to substitute or break a signature on a message. Some message component values, particularly HTTP field values, are potentially susceptible to broken implementations that could lead to unexpected and insecure behavior. Naive implementations of this specification might implement HTTP field processing by taking the single value of a field and using it as the direct component value without processing it appropriately.¶

For example, if the handling of obs-fold field values does not remove the internal line folding and whitespace, additional newlines could be introduced into the signature base by the signer, providing a potential place for an attacker to mount a signature collision (Section 7.5) attack. Alternatively, if header fields that appear multiple times are not joined into a single string value, as is required by this specification, similar attacks can be mounted as a signed component value would show up in the input string more than once and could be substituted or otherwise attacked in this way.¶

To counter this, the entire field processing algorithm needs to be implemented by all implementations of signers and verifiers.¶

The existence of a valid signature on an HTTP message is not sufficient to prove that the message has been signed by the appropriate party. It is up to the verifier to ensure that a given key and algorithm are appropriate for the message in question. If the verifier does not perform such a step, an attacker could substitute their own signature using their own key on a message and force a verifier to accept and process it. To combat this, the verifier needs to ensure that not only does the signature validate for a message, but that the key and algorithm used are appropriate.¶

Some message components are expressed in different ways across HTTP versions. For example, the authority of the request target is sent using the Host header field in HTTP 1.1 but with the :authority pseudo-header in HTTP 2. If a signer sends an HTTP 1.1 message and signs the Host field, but the message is translated to HTTP 2 before it reaches the verifier, the signature will not validate as the Host header field could be dropped.¶

It is for this reason that HTTP Message Signatures defines a set of derived components that define a single way to get value in question, such as the @authority derived component identifier (Section 2.2.4) in lieu of the Host header field. Applications should therefore prefer derived component identifiers for such options where possible.¶

Applications of this specification need to protect against key specification downgrade attacks. For example, the same RSA key can be used for both RSA-PSS and RSA v1.5 signatures. If an application expects a key to only be used with RSA-PSS, it needs to reject signatures for that key using the weaker RSA 1.5 specification.¶

Another example of a downgrade attack occurs when an asymmetric algorithm is expected, such as RSA-PSS, but an attacker substitutes a signature using symmetric algorithm, such as HMAC. A naive verifier implementation could use the value of the public RSA key as the input to the HMAC verification function. Since the public key is known to the attacker, this would allow the attacker to create a valid HMAC signature against this known key. To prevent this, the verifier needs to ensure that both the key material and the algorithm are appropriate for the usage in question. Additionally, while this specification does allow runtime specification of the algorithm using the alg signature parameter, applications are encouraged to use other mechanisms such as static configuration or higher protocol-level algorithm specification instead.¶

Several parts of this specification rely on the parsing of structured field values [RFC8941]. In particular, normalization of HTTP structured field values (Section 2.1.1), referencing members of a dictionary structured field (Section 2.1.2), and processing the @signature-input value when verifying a signature (Section 3.2). While structured field values are designed to be relatively simple to parse, a naive or broken implementation of such a parser could lead to subtle attack surfaces being exposed in the implementation.¶

For example, if a buggy parser of the @signature-input value does not enforce proper closing of quotes around string values within the list of component identifiers, an attacker could take advantage of this and inject additional content into the signature base through manipulating the Signature-Input field value on a message.¶

To counteract this, implementations should use fully compliant and trusted parsers for all structured field processing, both on the signer and verifier side.¶

Applications of HTTP Message Signatures need to decide which message components will be covered by the signature. Depending on the application, some components could be expected to be changed by intermediaries prior to the signature's verification. If these components are covered, such changes would, by design, break the signature.¶

However, the HTTP Message Signature standard allows for flexibility in determining which components are signed precisely so that a given application can choose the appropriate portions of the message that need to be signed, avoiding problematic components. For example, a web application framework that relies on rewriting query parameters might avoid use of the @query content identifier in favor of sub-indexing the query value using @query-params content identifier instead.¶

Some components are expected to be changed by intermediaries and ought not to be signed under most circumstance. The Via and Forwarded header fields, for example, are expected to be manipulated by proxies and other middle-boxes, including replacing or entirely dropping existing values. These fields should not be covered by the signature except in very limited and tightly-coupled scenarios.¶

Additional considerations for choosing signature aspects are discussed in Section 1.5.¶

The definition of HTTP field names does not allow for the use of the @ character anywhere in the name. As such, since all derived component identifiers start with the @ character, these namespaces should be completely separate. However, some HTTP implementations are not sufficiently strict about the characters accepted in HTTP headers. In such implementations, a sender (or attacker) could inject a header field starting with an @ character and have it passed through to the application code. These invalid header fields could be used to override a portion of the derived message content and substitute an arbitrary value, providing a potential place for an attacker to mount a signature collision (Section 7.5) attack.¶

To combat this, when selecting values for a message component, if the component identifier starts with the @ character, it needs to be processed as a derived component and never taken as a fields. Only if the component identifier does not start with the @ character can it be taken from the fields of the message. The algorithm discussed in Section 2.3 provides a safe order of operations.¶

Some cryptographic primitives such as RSA PSS and ECDSA have non-deterministic outputs, which include some amount of entropy within the algorithm. For such algorithms, multiple signatures generated in succession will not match. A lazy implementation of a verifier could ignore this distinction and simply check for the same value being created by re-signing the signature base. Such an implementation would work for deterministic algorithms such as HMAC and EdDSA but fail to verify valid signatures made using non-deterministic algorithms. It is therefore important that a verifier always use the correctly-defined verification function for the algorithm in question and not do a simple comparison.¶

"date"

The "date" field value represents the timestamp of the HTTP message. However, the creation time of the signature itself is encoded in the created signature parameter. These two values can be different, depending on how the signature and the HTTP message are created and serialized. Applications processing signatures for valid time windows should use the created signature parameter for such calculations. An application could also put limits on how much skew there is between the "date" field and the created signature parameter, in order to limit the application of a generated signature to different HTTP messages. See also Section 7.3 and Section 7.4.

"host"

The "host" header field is specific to HTTP 1.1, and its functionality is subsumed by the "@authority" derived component, defined in Section 2.2.4. In order to preserve the value across different HTTP versions, applications should always use the "@authority" derived component.

The signature base generation algorithm (Section 2.3) uses the value of an HTTP field as its component value. In the common case, this amounts to taking the actual bytes of the field value as the component value for both the signer and verifier. However, some field values allow for transformation of the values in semantically equivalent ways that alter the bytes used in the value itself. For example, a field definition can declare some or all of its value to be case-insensitive, or to have special handling of internal whitespace characters. Other fields have expected transformations from intermediaries, such as the removal of comments in the Via header field. In such cases, a verifier could be tripped up by using the equivalent transformed field value, which would differ from the byte value used by the signer. The verifier would have have a difficult time finding this class of errors since the value of the field is still acceptable for the application, but the actual bytes required by the signature base would not match.¶

When processing such fields, the signer and verifier have to agree how to handle such transformations, if at all. One option is to not sign problematic fields, but care must be taken to ensure that there is still sufficient signature coverage (Section 7.4) for the application. Another option is to define an application-specific canonicalization value for the field before it is added to the HTTP message, such as to always remove internal comments before signing, or to always transform values to lowercase. Since these transformations are applied prior to the field being used as input to the signature base generation algorithm, the signature base will still simply contain the byte value of the field as it appears within the message. If the transformations were to be applied after the value is extracted from the message but before it is added to the signature base, different attack surfaces such as value substitution attacks could be launched against the application. All application-specific additional rules are outside the scope of this specification, and by their very nature these transformations would harm interoperability of the implementation outside of this specific application. It is recommended that applications avoid the use of such additional rules wherever possible.¶

If a signer uses the same key with multiple verifiers, or uses the same key over time with a single verifier, the ongoing use of that key can be used to track the signer throughout the set of verifiers that messages are sent to. Since cryptographic keys are meant to be functionally unique, the use of the same key over time is a strong indicator that it is the same party signing multiple messages.¶

In many applications, this is a desirable trait, and it allows HTTP Message Signatures to be used as part of authenticating the signer to the verifier. However, unintentional tracking that a signer might not be aware of. To counter this kind of tracking, a signer can use a different key for each verifier that it is in communication with. Sometimes, a signer could also rotate their key when sending messages to a given verifier. These approaches do not negate the need for other anti-tracking techniques to be applied as necessary.¶

HTTP Message Signatures do not provide confidentiality of any of the information protected by the signature. The content of the HTTP message, including the value of all fields and the value of the signature itself, is presented in plaintext to any party with access to the message.¶

To provide confidentiality at the transport level, TLS or its equivalent can be used as discussed in Section 7.2.¶

It is important to balance the need for providing useful feedback to developers on error conditions without providing additional information to an attacker. For example, a naive but helpful server implementation might try to indicate the required key identifier needed for requesting a resource. If someone knows who controls that key, a correlation can be made between the resource's existence and the party identified by the key. Access to such information could be used by an attacker as a means to target the legitimate owner of the resource for further attacks.¶

A core design tenet of this specification is that all message components covered by the signature need to be available to the verifier in order to recreate the signature base and verify the signature. As a consequence, if an application of this specification requires that a particular field be signed, the verifier will need access to the value of that field.¶

For example, in some complex systems with intermediary processors this could cause the surprising behavior of an intermediary not being able to remove privacy-sensitive information from a message before forwarding it on for processing, for fear of breaking the signature. A possible mitigation for this specific situation would be for the intermediary to verify the signature itself, then modifying the message to remove the privacy-sensitive information. The intermediary can add its own signature at this point to signal to the next destination that the incoming signature was validated, as is shown in the example in Section 4.3.¶

9.1. Normative References

[FIPS186-4]

“Digital Signature Standard (DSS)”, 2013, <https://csrc.nist.gov/publications/detail/fips/186/4/final>.

[HTMLURL]

“URL (Living Standard)”, 2021, <https://url.spec.whatwg.org/>.

[MESSAGING]

Fielding, R., Nottingham, M., and J. , “HTTP/1.1”, Internet-Draft draft-ietf-httpbis-messaging-19 (work in progress), September 2021.

[POSIX.1]

“The Open Group Base Specifications Issue 7, 2018 edition”, 2018, <https://pubs.opengroup.org/onlinepubs/9699919799/>.

[RFC2104]

Krawczyk, H., Bellare, M., and R. Canetti, “HMAC: Keyed-Hashing for Message Authentication”, RFC 2104, DOI 10.17487/RFC2104, February 1997, <https://www.rfc-editor.org/info/rfc2104>.

[RFC2119]

Bradner, S., “Key words for use in RFCs to Indicate Requirement Levels”, BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997, <https://www.rfc-editor.org/info/rfc2119>.

[RFC3986]

Berners-Lee, T., Fielding, R., and L. Masinter, “Uniform Resource Identifier (URI): Generic Syntax”, STD 66, RFC 3986, DOI 10.17487/RFC3986, January 2005, <https://www.rfc-editor.org/info/rfc3986>.

[RFC6234]

Eastlake 3rd, D. and T. Hansen, “US Secure Hash Algorithms (SHA and SHA-based HMAC and HKDF)”, RFC 6234, DOI 10.17487/RFC6234, May 2011, <https://www.rfc-editor.org/info/rfc6234>.

[RFC7517]

Jones, M., “JSON Web Key (JWK)”, RFC 7517, DOI 10.17487/RFC7517, May 2015, <https://www.rfc-editor.org/info/rfc7517>.

[RFC7518]

Jones, M., “JSON Web Algorithms (JWA)”, RFC 7518, DOI 10.17487/RFC7518, May 2015, <https://www.rfc-editor.org/info/rfc7518>.

[RFC8017]

Moriarty, K., Ed., Kaliski, B., Jonsson, J., and A. Rusch, “PKCS #1: RSA Cryptography Specifications Version 2.2”, RFC 8017, DOI 10.17487/RFC8017, November 2016, <https://www.rfc-editor.org/info/rfc8017>.

[RFC8032]

Josefsson, S. and I. Liusvaara, “Edwards-Curve Digital Signature Algorithm (EdDSA)”, RFC 8032, DOI 10.17487/RFC8032, January 2017, <https://www.rfc-editor.org/info/rfc8032>.

[RFC8174]

Leiba, B., “Ambiguity of Uppercase vs Lowercase in RFC 2119 Key Words”, BCP 14, RFC 8174, DOI 10.17487/RFC8174, May 2017, <https://www.rfc-editor.org/info/rfc8174>.

[RFC8792]

Watsen, K., Auerswald, E., Farrel, A., and Q. Wu, “Handling Long Lines in Content of Internet-Drafts and RFCs”, RFC 8792, DOI 10.17487/RFC8792, June 2020, <https://www.rfc-editor.org/info/rfc8792>.

[RFC8941]

Nottingham, M. and P-H. Kamp, “Structured Field Values for HTTP”, RFC 8941, DOI 10.17487/RFC8941, February 2021, <https://www.rfc-editor.org/info/rfc8941>.

[SEMANTICS]

Fielding, R., Nottingham, M., and J. , “HTTP Semantics”, Internet-Draft draft-ietf-httpbis-semantics-19 (work in progress), September 2021.

9.2. Informative References

[BCP195]

Sheffer, Y., Holz, R., and P. Saint-Andre, “Recommendations for Secure Use of Transport Layer Security (TLS) and Datagram Transport Layer Security (DTLS)”, BCP 195, RFC 7525, DOI 10.17487/RFC7525, May 2015.

Moriarty, K. and S. Farrell, “Deprecating TLS 1.0 and TLS 1.1”, BCP 195, RFC 8996, DOI 10.17487/RFC8996, March 2021.

<https://www.rfc-editor.org/info/bcp195>

[I-D.ietf-httpbis-client-cert-field]

Campbell, B. and M. , “Client-Cert HTTP Header Field”, Internet-Draft draft-ietf-httpbis-client-cert-field-01 (work in progress), January 2022.

[RFC7239]

Petersson, A. and M. Nilsson, “Forwarded HTTP Extension”, RFC 7239, DOI 10.17487/RFC7239, June 2014, <https://www.rfc-editor.org/info/rfc7239>.

[RFC8126]

Cotton, M., Leiba, B., and T. Narten, “Guidelines for Writing an IANA Considerations Section in RFCs”, BCP 26, RFC 8126, DOI 10.17487/RFC8126, June 2017, <https://www.rfc-editor.org/info/rfc8126>.

[TLS]

Rescorla, E., “The Transport Layer Security (TLS) Protocol Version 1.3”, RFC 8446, DOI 10.17487/RFC8446, August 2018, <https://www.rfc-editor.org/info/rfc8446>.

This section provides cryptographic keys that are referenced in example signatures throughout this document. These keys MUST NOT be used for any purpose other than testing.¶

The key identifiers for each key are used throughout the examples in this specification. It is assumed for these examples that the signer and verifier can unambiguously dereference all key identifiers used here, and that the keys and algorithms used are appropriate for the context in which the signature is presented.¶

This command was tested with all the example keys on OpenSSL version 1.1.1m. Note that some systems cannot produce or use these keys directly, and may require additional processing.¶

In this example, there is a TLS-terminating reverse proxy sitting in front of the resource. The client does not sign the request but instead uses mutual TLS to make its call. The terminating proxy validates the TLS stream and injects a Client-Cert header according to [I-D.ietf-httpbis-client-cert-field], and then applies a signature to this field. By signing this header field, a reverse proxy can not only attest to its own validation of the initial request's TLS parameters but also authenticate itself to the backend system independently of the client's actions.¶

The internal service can validate the proxy's signature and therefore be able to trust that the client's certificate has been appropriately processed.¶