...

Text file src/sigs.k8s.io/gateway-api/geps/gep-1897.md

Documentation: sigs.k8s.io/gateway-api/geps

     1# GEP-1897: BackendTLSPolicy - Explicit Backend TLS Connection Configuration
     2
     3* Issue: [#1897](https://github.com/kubernetes-sigs/gateway-api/issues/1897)
     4* Status: Experimental
     5
     6## TLDR
     7
     8This document specifically addresses the topic of conveying HTTPS from the Gateway
     9dataplane to the backend (backend TLS termination), and intends to satisfy the single
    10use case “As a client implementation of Gateway API, I need to know how to connect to
    11a backend pod that has its own certificate”. TLS configuration can be a nebulous topic,
    12so in order to drive resolution this GEP focuses only on this single piece of functionality.
    13
    14Furthermore, for Gateway API to handle the case where the service or backend owner is doing their own TLS, _and_
    15the service or backend owner wants to validate the clients connecting to it, two things need to happen:
    16
    17- The service or backend owner has to provide a method for the Gateway owner to retrieve a certificate.
    18- Gateway API has to provide a way for the Gateway to configure and apply the validation options.
    19
    20## Immediate Goals
    21
    221. The solution must satisfy the following use case: the backend pod has its own
    23certificate and the gateway implementation client needs to know how to connect to the
    24backend pod. (Use case #4 in [Gateway API TLS Use Cases](#references))
    252. In terms of the Gateway API personas, only the application developer persona applies in this
    26solution. The application developer should control the gateway to backend TLS settings,
    27not the cluster operator, as requiring a cluster operator to manage certificate renewals
    28and revocations would be extremely cumbersome.
    293. The solution should consider client certificate settings used in the TLS handshake **from
    30Gateway to backend**, such as server name indication, trusted certificates,
    31and CA certificates.
    32
    33## Longer Term Goals
    34
    35These are worthy goals, but deserve a different GEP for proper attention.  This GEP is concerned entirely with the
    36controlplane, i.e. the hop between gateway and backend.
    37
    381. [TCPRoute](https://gateway-api.sigs.k8s.io/reference/spec/#gateway.networking.k8s.io/v1alpha2.TCPRoute) and
    39[GRPCRoute](https://gateway-api.sigs.k8s.io/reference/spec/#gateway.networking.k8s.io/v1alpha2.GRPCRoute) use cases
    40are not addressed here, because at this point in time these two route types are not graduated to beta.
    412. Mutual TLS (mTLS) use cases are intentionally out of scope for this GEP for two reasons.  First, the design of Gateway
    42API is backend-attached and does not currently support mutual authentication, and also because this GEP does not
    43address the case where connections to TLS are **implicitly configured** on behalf of the user, which is the norm for mTLS.
    44This GEP is about the case where an application developer needs to **explicitly express** that they expect TLS when
    45there is no automatic, implicit configuration available.
    463. Service mesh use cases are not addressed here because this GEP is specifically concerned with the connection between
    47Gateways and Backends, not Service to Service.  Service mesh use cases should ignore the design components described in
    48this proposal.
    49
    50## Non-Goals
    51
    52These are worthy goals, but will not be covered by this GEP.
    53
    541. Changes to the existing mechanisms for edge or passthrough TLS termination
    552. Providing a mechanism to decorate multiple route instances
    563. TLSRoute use cases
    574. UDPRoute use cases
    585. Controlling TLS versions or cipher suites used in TLS handshakes. (Use case #5 in [Gateway API TLS Use Cases](#references))
    596. Controlling certificates used by more than one workload (#6 in [Gateway API TLS Use Cases](#references))
    607. Client certificate settings used in TLS **from external clients to the
    61Listener** (#7 in [Gateway API TLS Use Cases](#references))
    628. Providing a mechanism for the cluster operator to override gateway to backend TLS settings.
    63
    64## Already Solved TLS Use Cases
    65
    66These are worthy goals that are already solved and thus will not be modified by the implementation.
    67
    681. Termination of TLS for HTTP routing (#1 in [Gateway API TLS Use Cases](#references))
    692. HTTPS passthrough use cases (#2 in [Gateway API TLS Use Cases](#references))
    703. Termination of TLS for non-HTTP TCP streams (#3 in [Gateway API TLS Use Cases](#references))
    71
    72## Overview - what do we want to do?
    73
    74Given that the current ingress solution specifies **edge** TLS termination (from the client to
    75the gateway), and how to handle **passthrough** TLS (from the client to the backend pod), this
    76proposed ingress solution specifies TLS origination to the **backend** (from the gateway to the
    77backend pod).  As mentioned, this solution satisfies the use case in which the backend pod
    78has its own certificate and the gateway client needs to know how to connect to the backend pod.
    79
    80![image depicting TLS termination types](images/1897-TLStermtypes.png "TLS termination types")
    81
    82Gateway API is missing a mechanism for separately providing the details for the backend TLS handshake,
    83including (but not limited to):
    84
    85* intent to use TLS on the backend hop
    86* client certificate of the gateway
    87* system certificates to use in the absence of client certificates
    88
    89## Purpose - why do we want to do this?
    90
    91This proposal is _very_ tightly scoped because we have tried and failed to address this well-known
    92gap in the API specification. The lack of support for this fundamental concept is holding back
    93Gateway API adoption by users that require a solution to the use case. One of the recurring themes
    94that has held up the prior art has been interest related to service mesh, and as such this proposal
    95focuses explicitly on the ingress use case in the initial round.  Another reason for the tight scope
    96is that we have been too focused on a generic representation of everything that TLS can do, which
    97covers too much ground to address in a single GEP.
    98
    99## The history of backend TLS
   100
   101Work on this topic has spanned over three years, as documented in our repositories and other references,
   102and summarized below.
   103
   104In January 2020, in issue [TLS Termination Policy #52](https://github.com/kubernetes-sigs/gateway-api/issues/52),
   105this use case was discussed.  The discussion ended after being diverted by
   106[KEP: Adding AppProtocol to Services and Endpoints #1422](https://github.com/kubernetes/enhancements/pull/1422),
   107which was implemented and later reverted.
   108
   109In February 2020, [HTTPRoute: Add Reencrypt #81](https://github.com/kubernetes-sigs/gateway-api/pull/81)
   110added the dataplane feature as “reencrypt”, but it went stale and was closed in favor of the work done in the
   111next paragraph, which unfortunately didn’t implement the backend TLS termination feature.
   112
   113In August 2020, it resurfaced with a [comment](https://github.com/kubernetes-sigs/gateway-api/pull/256/files#r472734392)
   114on this pull request: [tls: introduce mode and sni to cert matching behavior](https://github.com/kubernetes-sigs/gateway-api/pull/256/files#top).
   115The backend TLS termination feature was deferred at that time.  Other TLS discussion was documented in
   116[[SIG-NETWORK] TLS config in service-apis](https://docs.google.com/document/d/15fkzMrhN_7tA-i2mHKwZpqcjN1o2Pe9Am9Qt828x1lo/edit#heading=h.wym7wehwll44)
   117, a list of TLS features that had been collected in June 2020, itself based on spreadsheet
   118[Service API: TLS related issues](https://docs.google.com/spreadsheets/d/18KE61Y6InCmoQHZcbrYYRZS5Cnt7n33s5dTxUlhHgIA/edit#gid=0).
   119
   120In December 2021, this was discussed as a beta blocker in issue
   121[Docs mentions Reencrypt for HTTPRoute and TLSRoute is available #968](https://github.com/kubernetes-sigs/gateway-api/issues/968).
   122
   123A March 2022 issue documents another request for it: [Provide a way to configure TLS from a Gateway to Backends #1067](https://github.com/kubernetes-sigs/gateway-api/issues/1067)
   124
   125A June 2022 issue documents a documentation issue related to it:
   126[Unclear how to specify upstream (webserver) HTTP protocol #1244](https://github.com/kubernetes-sigs/gateway-api/discussions/1244)
   127
   128A July 2022 discussion [Specify Re-encrypt TLS Termination (i.e., Upstream TLS) #1285](https://github.com/kubernetes-sigs/gateway-api/discussions/1285)
   129collected most of the historical context preceding the backend TLS termination feature, with the intention of
   130collecting evidence that this feature is still unresolved.  This was followed by
   131[GEP: Describe Backend Properties #1282](https://github.com/kubernetes-sigs/gateway-api/issues/1282).
   132
   133In August 2022, [Add Provisional GEP-1282 document #1333](https://github.com/kubernetes-sigs/gateway-api/pull/1333)
   134was created, and in October 2022, a GEP update with proposed implementation
   135[GEP-1282 Backend Properties - Update implementation #1430](https://github.com/kubernetes-sigs/gateway-api/pull/1430)
   136was followed by intense discussion and closed in favor of a downsize in scope.
   137
   138In January 2023 we closed GEP-1282 and began a new discussion on enumerating TLS use cases in
   139[Gateway API TLS Use Cases](#references), for the purposes of a clear definition and separation of concerns.
   140This GEP is the outcome of the TLS use case #4 in
   141[Gateway API TLS Use Cases](#references) as mentioned in the Immediate Goals section above.
   142
   143## API
   144
   145To allow the gateway client to know how to connect to the backend pod, when the backend pod has its own
   146certificate, we implement a metaresource named `BackendTLSPolicy`, that was previously introduced with the name
   147`TLSConnectionPolicy` as a hypothetical Direct Policy Attachment example in
   148[GEP-713: Metaresources and PolicyAttachment](https://gateway-api.sigs.k8s.io/geps/gep-713/).
   149Because naming is hard, a new name may be
   150substituted without blocking acceptance of the content of the API change.
   151
   152The selection of the applicable Gateway API persona is important in the design of BackendTLSPolicy, because it provides
   153a way to explicitly describe the _expectations_ of the connection to the application.  BackendTLSPolicy is configured
   154by the application developer Gateway API persona to signal what the application developer _expects_ in connections to
   155the application, from a TLS perspective.  Only the application developer can know what the application expects, so it is
   156important that this configuration be managed by that persona.
   157
   158During the course of discussion of this proposal, we did consider allowing the cluster operator persona to have some access
   159to Gateway cert validation, but as mentioned, BackendTLSPolicy is used primarily to signal what the application
   160developer expects in the connection.  Granting this expectation to any other role would blur the lines between role
   161responsibilities, which compromises the role-oriented design principle of Gateway API. As mentioned in Non-goal #8,
   162providing a mechanism for the cluster operator gateway role to override gateway to backend TLS settings is not covered
   163by this proposal, but should be addressed in a future update.  One idea is to use two types: ApplicationBackendTLSPolicy,
   164and GatewayBackendTLSPolicy, where the application developer is responsible for the former, the cluster operator is
   165responsible for the latter, and the cluster operator may configure whether certain settings may be overridden by
   166application developers.
   167
   168The BackendTLSPolicy must contain these configuration items to allow the Gateway to operate successfully
   169as a TLS Client:
   170
   171- An explicit signal that TLS should be used by this connection.
   172- A hostname the Gateway should use to connect to the backend.
   173- A reference to one or more certificates to use in the TLS handshake, signed by a CA or self-signed.
   174- An indication that system certificates may be used.
   175
   176BackendTLSPolicy is defined as a Direct Policy Attachment without defaults or overrides, applied to a Service that
   177accesses the backend in question, where the BackendTLSPolicy resides in the same namespace as the Service it is
   178applied to.  The BackendTLSPolicy and the Service must reside in the same namespace in order to prevent the
   179complications involved with sharing trust across namespace boundaries.  We chose the Service resource as a target,
   180rather than the Route resource, so that we can reuse the same BackendTLSPolicy for all the different Routes that
   181might point to this Service.
   182For the use case where certificates are stored in their own namespace, users may create Secrets and use ReferenceGrants
   183for a BackendTLSPolicy-to-Secret binding.  Implementations must respect a ReferenceGrant for cross-namespace Secret
   184sharing to BackendTLSPolicy, even if they don't for other cross-namespace sharing.
   185
   186One of the areas of concern for this API is that we need to indicate how and when the API implementations should use the
   187backend destination certificate authority.  This solution proposes, as introduced in
   188[GEP-713](https://gateway-api.sigs.k8s.io/geps/gep-713/), that the implementation
   189should watch the connections to a specified TargetRef (Service), and if the Service matches a BackendTLSPolicy, then
   190assume the connection is TLS, and verify that the TargetRef’s certificate can be validated by the client (Gateway) using
   191the provided certificates and hostname before the connection is made. On the question of how to signal
   192that there was a failure in the certificate validation, this is left up to the implementation to return a response error
   193that is appropriate, such as one of the HTTP error codes: 400 (Bad Request), 401 (Unauthorized), 403 (Forbidden), or
   194other signal that makes the failure sufficiently clear to the requester without revealing too much about the transaction,
   195based on established security requirements.
   196
   197All policy resources must include `TargetRef` with the fields specified
   198[here](https://github.com/kubernetes-sigs/gateway-api/blob/a33a934af9ec6997b34fd9b00d2ecd13d143e48b/apis/v1alpha2/policy_types.go#L24-L41).
   199In an upcoming [extension](https://github.com/kubernetes-sigs/gateway-api/issues/2147) to TargetRef, policy resources
   200_may_ also choose to include `SectionName` and/or `Port` in the TargetRef following the same mechanics as `ParentRef`.
   201
   202BackendTLSPolicySpec contains the `TargetRef` and `TLS` fields.  The `TLS` field is a `BackendTLSPolicyConfig` and
   203contains `CertRefs`, `StandardCerts`, and `Hostname`.
   204The names of the fields were chosen to facilitate discussion, but may be substituted without blocking acceptance of the
   205content of the API change.
   206
   207The `CertRefs` and `StandardCerts` fields are both optional, but one of them must be set for a valid TLS configuration.
   208CertRefs is a slice of
   209named config maps, each containing a single cert. We originally proposed to follow the convention established by the
   210[CertificateRefs field on Gateway](https://github.com/kubernetes-sigs/gateway-api/blob/18e79909f7310aafc625ba7c862dfcc67b385250/apis/v1beta1/gateway_types.go#L340)
   211, but the CertificateRef requires both a tls.key and tls.crt and a certificate reference only requires the tls.crt.
   212StandardCerts is an optional enum that allows users to specify whether to use the set of CA certificates trusted by the
   213Gateway (StandardCerts specified as "System"), or to use the existing CertRefs (StandardCerts specified as "").  The use
   214
   215and definition of system certificates is implementation-dependent, and the intent is that these certificates are obtained
   216from the underlying operating system. CertRefs contains one or more references to Kubernetes objects that
   217contain PEM-encoded TLS certificates, which are used to establish a TLS handshake between the gateway and backend pod.
   218References to a resource in a different namespace are invalid.
   219If CertRefs is unspecified, then StandardCerts must be set to "System" for a valid configuration.
   220If StandardCerts is unspecified, then CertRefs must be specified with at least one entry for a valid configuration.
   221If StandardCerts is set to "System" and there are no system trusted certificates or the implementation doesn't define system
   222trusted certificates, then the associated TLS connection must fail.
   223
   224The `Hostname` field is required and is to be used to configure the SNI the Gateway should use to connect to the backend.
   225Implementations must validate that at least one name in the certificate served by the backend matches this field.
   226We originally proposed using a list of allowed Subject Alternative Names, but determined that this was [not needed in
   227the first round](https://github.com/kubernetes-sigs/gateway-api/pull/2113#issuecomment-1696127092),
   228but may be added in the future.
   229
   230We originally proposed allowing the configuration of expected TLS versions, but determined that this was [not needed in
   231the first round](https://github.com/kubernetes-sigs/gateway-api/pull/2113#issuecomment-1696127092).
   232
   233Thus, the following additions would be made to the Gateway API:
   234
   235```go
   236import "sigs.k8s.io/gateway-api/apis/v1beta1"
   237
   238// BackendTLSPolicy provides a way to publish TLS configuration
   239// that enables a gateway client to connect to a backend pod.
   240type BackendTLSPolicy struct {
   241    metav1.TypeMeta   `json:",inline"`
   242    metav1.ObjectMeta `json:"metadata,omitempty"`
   243
   244    // Spec defines the desired state of BackendTLSPolicy.
   245    Spec BackendTLSPolicySpec `json:"spec"`
   246
   247    // Status defines the current state of BackendTLSPolicy.
   248    Status PolicyStatus `json:"status,omitempty"`
   249}
   250
   251// BackendTLSPolicySpec defines the desired state of
   252// BackendTLSPolicy.
   253// Note: there is no Override or Default policy configuration.
   254//
   255// Support: Core
   256type BackendTLSPolicySpec struct {
   257    // TargetRef identifies an API object to apply policy to.
   258    // Services are the only valid API target references.
   259    // Note that this config applies to the entire referenced resource
   260    // by default, but this default may change in the future to provide
   261    // a more granular application of the policy.
   262    TargetRef gatewayv1a2.PolicyTargetReference `json:"targetRef"`
   263
   264    // TLS contains backend TLS policy configuration.
   265    TLS *BackendTLSPolicyConfig `json:”tls”`
   266}
   267
   268// BackendTLSPolicyConfig contains backend TLS policy configuration.
   269// +kubebuilder:validation:XValidation:message="must not contain both CertRefs and StandardCerts",rule="(has(self.certRefs) && size(self.certRefs > 0) && has(self.standardCerts) && self.standardCerts != "")"
   270// +kubebuilder:validation:XValidation:message="must specify either CertRefs or StandardCerts",rule="!(has(self.certRefs) && size(self.certRefs > 0) || has(self.standardCerts) && self.standardCerts != "")"
   271type BackendTLSPolicyConfig struct {
   272    // CertRefs contains one or more references to
   273    // Kubernetes objects that contain PEM-encoded TLS certificates,
   274    // which are used to establish a TLS handshake between the gateway
   275    // and backend pod.
   276    //
   277    // If CertRefs is empty or unspecified, then StandardCerts must
   278    // be specified.  Only one of CertRefs or StandardCerts may be
   279    // specified, not both.
   280    //
   281    // If CertRefs is empty or unspecified, then system trusted
   282    // certificates should be used. If there are none, or the
   283    // implementation doesn't define system trusted certificates,
   284    // then a TLS connection must fail.
   285    //
   286    // References to a resource in a different namespace are
   287    // invalid.
   288    //
   289    // A single CertRef to a Kubernetes ConfigMap kind has "Core"
   290    // support.  Implementations MAY choose to support attaching
   291    // multiple certificates to a backend, but this behavior is
   292    // implementation-specific.  Also implementation-specific is
   293    // a CertRef of other object kinds, e.g. Secret.
   294    // 
   295    // Support: Core - An optional single reference to a Kubernetes
   296    // ConfigMap.
   297    //
   298    // Support: Implementation-specific (No reference, more than one
   299    // reference, or resource types other than ConfigMaps.
   300    // Service mesh may ignore.)
   301    //
   302    // +kubebuilder:validation:MaxItems=8
   303    // +optional
   304    CertRefs []ConfigMapObjectReference `json:”certRefs,omitempty”`
   305
   306    // StandardCerts specifies whether system CA certificates may
   307    // be used in the TLS handshake between the gateway and
   308    // backend pod.
   309    // 
   310    // If StandardCerts is unspecified or set to "", then CertRefs must
   311    // be specified with at least one entry for a valid configuration.
   312    // If StandardCerts is unspecified or set to "", then CertRefs must
   313    // be specified.  Only one of CertRefs or StandardCerts may be
   314    // specified, not both.
   315    //
   316    // StandardCerts must be set to "System" when CertRefs is unspecified.
   317    //
   318    // If StandardCerts is set to "System", then the system trusted
   319    // certificates should be used. If there are none, or the
   320    // implementation doesn't define system trusted certificates,
   321    // then a TLS connection must fail.
   322    //
   323    // Support: Core - An optional value to specify whether to use
   324    // system certificates or not.
   325    //
   326    // Support: Implementation-specific (In the absence of support
   327    // for usable system certs, may be ignored. Service mesh may ignore.)
   328    //
   329    // +optional
   330    StandardCerts *StandardCertType `json:"standardCerts,omitempty"`
   331
   332    // Hostname is the Server Name Indication that the Gateway uses to
   333    // connect to the backend.  It represents the fully qualified domain
   334    // name of a network host, as defined by RFC1123 - except that numeric
   335    // IP addresses are not allowed. Each label of the FQDN must consist
   336    // of lower case alphanumeric characters or '-', and must start and
   337    // end with an alphanumeric character.  No other punctuation is allowed.
   338    // Wildcard domain names are specifically disallowed.
   339    //
   340    // It specifies the hostname that may authenticate, and must be in the
   341    // certificate served by the matching backend.
   342    //
   343    // Support: Core - A required value used by the Gateway to connect to
   344    // the backend when a BackendTLSPolicy is specified.
   345    Hostname v1beta1.PreciseHostname `json:"hostname"`
   346}
   347
   348// StandardCertType is the type of CA certificate that will be used when
   349// the TLS.certRefs is unspecified.
   350// +kubebuilder:validation:Enum=System
   351type StandardCertType string
   352
   353const (
   354    StandardCertSystem StandardCertType = "System"
   355)
   356
   357// ConfigMapObjectReference identifies an API object including its namespace,
   358// defaulting to ConfigMap.
   359//
   360// The API object must be valid in the cluster; the Group and Kind must
   361// be registered in the cluster for this reference to be valid.
   362//
   363// References to objects with invalid Group and Kind are not valid, and must
   364// be rejected by the implementation, with appropriate Conditions set
   365// on the containing object.
   366type ConfigMapObjectReference struct {
   367    // Group is the group of the referent.  For example, "gateway.networking.k8s.io".
   368    // When unspecified or empty string, core API group is inferred.
   369    //
   370    // +optional
   371    // +kubebuilder:default=""
   372    Group *Group `json:"group"`
   373
   374    // Kind is the kind of the referent.  For example, "ConfigMap".
   375    //
   376    // +optional
   377    // +kubebuilder:default=ConfigMap
   378    Kind *Kind `json:"kind"`
   379
   380    // Name is the metadata.name of the referenced config map.
   381    // +kubebuilder:validation:Required
   382    Name ObjectName `json"name"`
   383
   384    // Namespace is the namespace of the referenced object. When unspecified, the local
   385    // namespace is inferred.
   386    //
   387    // Note that when a namespace different than the local namespace is specified,
   388    // a ReferenceGrant object is required in the referent namespace to allow that
   389    // namespace's owner to accept the reference. See the ReferenceGrant
   390    // documentation for details.
   391    //
   392    // Support: Core
   393    //
   394    // +optional
   395    Namespace *Namespace `json:"namespace,omitempty"`
   396}
   397
   398// BackendTLSPolicyConditionType is the type of a condition used
   399// as a signal by BackendTLSPolicy.  This type should be used with
   400// the BackendTLSPolicyStatus.Conditions field.
   401type BackendTLSPolicyConditionType string
   402
   403//  BackendTLSPolicyConditionReason is a reason that explains why a
   404// particular BackendTLSPolicyConditionType was generated.
   405type BackendTLSPolicyConditionReason string
   406
   407const (
   408    // This condition indicates that the BackendTLSPolicy has been
   409    // accepted as valid.
   410    // Possible reason for this condition to be True is:
   411    //
   412    // * “Accepted” 
   413    // 
   414    // Possible reasons for this condition to be False are:
   415    //
   416    // * “Invalid”
   417    // * “Pending”
   418    BackendTLSPolicyConditionAccepted BackendTLSPolicyConditionType = “Accepted”
   419
   420    // This reason is used with the “Accepted” condition when the condition is true.
   421    BackendTLSPolicyReasonAccepted BackendTLSPolicyConditionReason = “Valid”
   422
   423    // This reason is used with the “Accepted” condition when the BackendTLSPolicy is invalid,
   424    // e.g. use of a CertRef that crosses namespace boundaries.
   425    BackendTLSPolicyReasonInvalid BackendTLSPolicyConditionReason = “Invalid”
   426
   427    // This reason is used with the “Accepted” condition when the BackendTLSPolicy is pending validation.
   428    BackendTLSPolicyReasonPending BackendTLSPolicyConditionReason = “Pending”
   429)
   430```
   431
   432## How a client behaves
   433
   434This table describes the effect that a BackendTLSPolicy has on a Route.  There are only two cases where the
   435BackendTLSPolicy will signal a Route to connect to a backend using TLS, an HTTPRoute with a backend that is targeted
   436by a BackendTLSPolicy, either with or without listener TLS configured.  (There are a few other cases where it may be
   437possible, but is implementation dependent.)
   438
   439Every implementation that claims supports for BackendTLSPolicy should document for which Routes it is being implemented.
   440
   441| Route Type | Gateway Config             | Backend is targeted by a BackendTLSPolicy? | Connect to backend  with TLS? |
   442|------------|----------------------------|-----------------------------------------------|-------------------------------|
   443| HTTPRoute  | Listener tls               | Yes                                           | **Yes**                       |
   444| HTTPRoute  | No listener tls            | Yes                                           | **Yes**                       |
   445| HTTPRoute  | Listener tls               | No                                            | No                            |
   446| HTTPRoute  | No listener tls            | No                                            | No                            |
   447| TLSRoute   | Listener Mode: Passthrough | Yes                                           | No                            |
   448| TLSRoute   | Listener Mode: Terminate   | Yes                                           | Implementation-dependent      |
   449| TLSRoute   | Listener Mode: Passthrough | No                                            | No                            |
   450| TLSRoute   | Listener Mode: Terminate   | No                                            | No                            |
   451| TCPRoute   | Listener TLS               | Yes                                           | Implementation-dependent      |
   452| TCPRoute   | No listener TLS            | Yes                                           | Implementation-dependent      |
   453| TCPRoute   | Listener TLS               | No                                            | No                            |
   454| TCPRoute   | No listener TLS            | No                                            | No                            |
   455| UDPRoute   | Listener TLS               | Yes                                           | No                            |
   456| UDPRoute   | No listener TLS            | Yes                                           | No                            |
   457| UDPRoute   | Listener TLS               | No                                            | No                            |
   458| UDPRoute   | No listener TLS            | No                                            | No                            |
   459| GRPCRoute  | Listener TLS               | Yes                                           | Implementation-dependent      |
   460| GRPCRoute  | No Listener TLS            | Yes                                           | Implementation-dependent      |
   461| GRPCRoute  | Listener TLS               | No                                            | No                            |
   462| GRPCRoute  | No Listener TLS            | No                                            | No                            |
   463
   464## Request Flow
   465
   466Step 6 would be changed in the typical client/gateway API request flow for a gateway implemented using a
   467reverse proxy. This is shown as **bolded** additions in step 6 below.
   468
   4691. A client makes a request to http://foo.example.com.
   4702. DNS resolves the name to a Gateway address.
   4713. The reverse proxy receives the request on a Listener and uses the Host header to match an HTTPRoute.
   4724. Optionally, the reverse proxy can perform request header and/or path matching based on match rules of the HTTPRoute.
   4735. Optionally, the reverse proxy can modify the request, i.e. add/remove headers, based on filter rules of the HTTPRoute.
   4746. Lastly, the reverse proxy **optionally performs a TLS handshake** and forwards the request to one or more objects,
   475i.e. Service, in the cluster based on backendRefs rules of the HTTPRoute **and TLSTargetRef of the BackendTLSPolicy**.
   476
   477## Alternatives
   478Most alternatives are enumerated in the section on the history of backend TLS above.  A couple of additional
   479alternatives are also listed here.
   480
   4811. Expand BackendRef, which is already an expansion point.  At first, it seems logical that since listeners are handling
   482the client-gateway certs, BackendRefs could handle the gateway-backend certs.  However, when multiple Routes to target
   483the same Service, there would be unnecessary copying of the BackendRef every time the Service was targeted.  As well,
   484there could be multiple bBackendRefs with multiple rules on a rRoute, each of which might need the gateway-backend cert
   485configuration so it is not the appropriate pattern.
   4862. Extend HTTPRoute to indicate TLS backend support. Extending HTTPRoute would interfere with deployed implementations
   487too much to be a practical solution.
   4883. Add a new type of Route for backend TLS.  This is impractical because we might want to enable backend TLS on other
   489route types in the future, and because we might want to have both TLS listeners and backend TLS on a single route.
   490
   491## Prior Art
   492
   493TLS from gateway to backend for ingress exists in several implementations, and was developed independently.
   494
   495### Istio Gateway supports this with a DestinationRule:
   496
   497* A secret representing a certificate/key pair, where the certificate is valid for the route host
   498* Set Gateway spec.servers[].port.protocol: HTTPS, spec.servers[].tls.mode=SIMPLE, spec.servers[].tls.credentialName
   499* Set DestinationRule spec.trafficPolicy.tls.mode: SIMPLE
   500
   501Ref: [Istio / Understanding TLS Configuration](https://istio.io/latest/docs/ops/configuration/traffic-management/tls-configuration/#gateways)
   502and [Istio / Destination Rule](https://istio.io/latest/docs/reference/config/networking/destination-rule/#ClientTLSSettings)
   503
   504### OpenShift Route (comparable to GW API Gateway) supports this with the following route configuration items:
   505
   506* A certificate/key pair, where the certificate is valid for the route host
   507* A separate destination CA certificate enables the Ingress Controller to trust the destination’s certificate
   508* An optional, separate CA certificate that completes the certificate chain
   509
   510Ref: [Secured routes - Configuring Routes | Networking | OpenShift Container Platform 4.12](https://docs.openshift.com/container-platform/4.12/networking/routes/secured-routes.html#nw-ingress-creating-a-reencrypt-route-with-a-custom-certificate_secured-routes)
   511
   512### Contour supports this from Envoy to the backend using:
   513
   514* An Envoy client certificate
   515* A CA certificate and SubjectName which are both used to verify the backend endpoint’s identity
   516* Kubernetes Service annotation: projectcontour.io/upstream-protocol.tls
   517
   518Ref: [Upstream TLS](https://projectcontour.io/docs/v1.21.1/config/upstream-tls/)
   519
   520### GKE supports a way to encrypt traffic to the backend pods using:
   521
   522* `AppProtocol` on Service set to HTTPS
   523* Load balancer does not verify the certificate used by backend pods
   524
   525Ref: [Secure a Gateway](https://cloud.google.com/kubernetes-engine/docs/how-to/secure-gateway#load-balancer-tls)
   526
   527### Emissary supports encrypted traffic to services
   528
   529* In the `Mapping` definition, set https:// in the spec.service field
   530* A spec.tls in the `Mapping` definition, with the name of a `TLSContext`
   531* A `TLSContext` to provide a client certificate, set minimum TLS version support, SNI
   532
   533Ref: [TLS Origination](https://www.getambassador.io/docs/emissary/latest/topics/running/tls/origination)
   534
   535### NGINX implementation through CRDs (Comparable to Route or Policy of Gateway API) supports both TLS and mTLS
   536
   537* In the Upstream section of a VirtualServer or VirtualServerRoute (equivalent to HTTPRoute) there is a simple toggle to enable TLS.  This does not validate the certificate of the backend and implicitly trusts the backend in order to form the SSL tunnel.  This is not about validating the certificate but obfuscating the traffic with TLS/SSL.
   538* A Policy attachment can be provided when certification validation is required that is called egressMTLS (egress from the proxy to the upstream).  This can be tuned to perform various certificate validation tests.  It was created as a Policy becuase it implies some type of AuthN/AuthZ due to the additional checks.  This was also compatible with Open Service Mesh and NGINX Service Mesh and removed the need for a sidecar at the ingress controller.
   539* A corresponding 'IngressMTLS' policy also exists for mTLS verification of client connections to the proxy.  The Policy object is used for anything that implies AuthN/AuthZ.
   540
   541Ref: [Upstream.TLS](https://docs.nginx.com/nginx-ingress-controller/configuration/virtualserver-and-virtualserverroute-resources/#upstreamtls)
   542
   543Ref: [EgressMTLS](https://docs.nginx.com/nginx-ingress-controller/configuration/policy-resource/#egressmtls)
   544
   545Ref: [IngressMTLS](https://docs.nginx.com/nginx-ingress-controller/configuration/policy-resource/#ingressmtls)
   546
   547## Answered Questions
   548
   549Q. Bowei recommended that we mention the approach of cross-namespace referencing between Route and Service.
   550Be explicit about using the standard rules with respect to attaching policies to resources.
   551
   552A. This is mentioned in the
   553API section.
   554
   555Q. Costin recommended that Gateway SHOULD authenticate with either a JWT with audience or client cert
   556or some other means - so gateway added headers can be trusted, amongst other things.
   557
   558A. This is out of scope for this
   559proposal, which centers around application developer persona resources such as HTTPRoute and Service.
   560
   561Q. Costin mentioned we need to answer the question - is configuring the connection to a backend and TLS
   562something the route author decides - or the backend owner?
   563
   564A. This is decided by the application developer persona,
   565which would more likely, but not exclusively, be the backend owner.
   566
   567Q.Costin continued, same for SAN (Subject Alternative Name) certificates.
   568The backend owner is the application developer, and the route owner will have to collaborate with the application
   569developer to provide the appropriate configuration for TLS.  The implementation would need to take the certificate
   570provided by the application and verify that it satisfies the requirements of the route-as-client, including SAN
   571information.  Sometimes the backend owner and route owner are the same entity.
   572
   573A. This was most recently addressed by
   574adding hostname for SNI and removing allowed SANs.
   575
   576## Graduation Criteria
   577
   578This section is to record issues that were requested for discussion in the API section before this GEP graduates
   579out of `Provisional` status.
   580
   5811. Rob Scott is interested in extending the TargetRef to optionally include port, since we are targeting the entirety
   582of a Service. See the discussion in https://github.com/kubernetes-sigs/gateway-api/pull/2113/files#r1231594914,
   583and follow up issue in https://github.com/kubernetes-sigs/gateway-api/issues/2147
   5842. Michael Pleshakov asked about conflicts that could arise when multiple implementations are running in a cluster.
   585This is a gap in our policy attachment model that needs to be addressed.  See the discussion in
   586https://github.com/kubernetes-sigs/gateway-api/pull/2113/files#r1235750540. Graduating this GEP to implementable
   587requires an update to the Policy GEP to define how status can be nested to support multiple implementations. This will
   588likely look very similar to Route status.
   589See [comment](https://github.com/kubernetes-sigs/gateway-api/pull/2113#issuecomment-1696127092).
   5903. Rob Scott [wanted to note](https://github.com/kubernetes-sigs/gateway-api/pull/2113#issuecomment-1696127092) that
   591when this graduates to the standard channel, implementations of HTTPRoute may also be
   592required to watch the BackendTLSPolicy. If one of these policies is attached to a Service targeted by an HTTPRoute,
   593the implementation would be required to fully implement the policy or mark the backend invalid.
   594
   595## References
   596
   597[Gateway API TLS Use Cases](https://docs.google.com/document/d/17sctu2uMJtHmJTGtBi_awGB0YzoCLodtR6rUNmKMCs8/edit#heading=h.cxuq8vo8pcxm)
   598
   599[GEP-713: Metaresources and PolicyAttachment](https://gateway-api.sigs.k8s.io/geps/gep-713/)
   600
   601[Policy Attachment](https://gateway-api.sigs.k8s.io/reference/policy-attachment/#direct-policy-attachment)
   602
   603[Gateway API TLS](https://gateway-api.sigs.k8s.io/v1alpha2/guides/tls/)
   604
   605[SIG-NET Gateway API: TLS to the K8s.Service/Backend](https://docs.google.com/document/d/1RTYh2brg_vLX9o3pTcrWxtZSsf8Y5NQvIG52lpFcZlo)
   606
   607[SAN vs SNI](https://serverfault.com/questions/807959/what-is-the-difference-between-san-and-sni-ssl-certificates)

View as plain text