Repository navigation
fix(auth,gax): align Java mTLS and agent identity certificate handling with Python parity - #14658
Conversation
…g with Python parity - Fail closed in GAX (EndpointContext, InstantiatingGrpcChannelProvider, InstantiatingHttpJsonChannelProvider) when DefaultMtlsProviderFactory.create() throws IOException while useMtlsClientCertificate() is true. - Only poll on startup for agent identity certificates when GOOGLE_API_CERTIFICATE_CONFIG explicitly points inside the well-known workload SPIFFE directory, resolving parent directory symlinks and requiring non-empty regular files (> 0 bytes). - Return certsPresent=false immediately without polling or throwing when GOOGLE_API_CERTIFICATE_CONFIG is unset and no well-known certificates exist on disk. - Treat any non-'true' value of GOOGLE_API_USE_CLIENT_CERTIFICATE as disabled in MtlsUtils.
There was a problem hiding this comment.
Code Review
This pull request refactors mTLS configuration and certificate resolution. Key changes include optimizing the handling of the GOOGLE_API_USE_CLIENT_CERTIFICATE environment variable, ensuring that file checks verify that paths are non-empty regular files, resolving symlinks while preserving trailing missing elements, and removing startup polling when the certificate configuration is unset. Additionally, the default MtlsProvider creation is refactored to propagate IOExceptions rather than silently falling back to non-mTLS channels. Review feedback suggests optimizing useMtlsClientCertificate by avoiding redundant environment variable fetches and adding logging for unexpected exceptions caught during certificate checks in AgentIdentityUtils.
|
|
||
| AgentIdentityUtils.CertInfo info = AgentIdentityUtils.getAgentIdentityCertInfo(); | ||
| assertNotNull(info); | ||
| assertEquals(2, fakeTime.getSleepCount()); |
There was a problem hiding this comment.
This test still passes if the attrs.size() > 0 check in checkExistsOrAccessDenied is reverted. Without the check, the 0-byte files count as ready and polling returns right away, but loadAndVerifyCredentials then makes 3 read attempts with a timeService.sleep between them on the same fake clock. The callback fills in the cert and key on those two retry sleeps, so the test still ends with 2 sleeps and a valid CertInfo.
Does it make sense to fill the files in after the load retries would have run out (e.g. the cert on sleep 4 and the key on sleep 5, and assert 5 sleeps)?
There was a problem hiding this comment.
Good catch. Updated the test in a14152d5ca4 to populate the certificate on sleep 4 and the private key on sleep 5, and assert 5 total sleeps so loadAndVerifyCredentials's 3 attempts (2 sleeps) cannot mask a missing 0-byte check.
| } | ||
|
|
||
| @Test | ||
| void endpointContextBuild_whenDefaultMtlsProviderThrowsIOException_propagatesIOException() |
There was a problem hiding this comment.
nit: Could we also add a test where createDefaultMtlsProvider() throws CertificateSourceUnavailableException (here and in the two channel provider tests)? That covers the GOOGLE_API_USE_CLIENT_CERTIFICATE=true fallback when no certs exist on disk, where build() should still succeed without mTLS on the regular endpoint.
There was a problem hiding this comment.
Done in a14152d5ca4. Added unit tests covering CertificateSourceUnavailableException from createDefaultMtlsProvider() in EndpointContextTest, InstantiatingGrpcChannelProviderTest, and InstantiatingHttpJsonChannelProviderTest.
|
|
||
| /** | ||
| * Resolves symlinks along all existing prefix components of {@code path}, preserving any trailing | ||
| * non-existent path elements (matching Python's {@code os.path.realpath} behavior before startup |
There was a problem hiding this comment.
nit: this doesn't follow a symlink whose target doesn't exist yet (e.g. a config path outside the well-known dir that links to a file Cloud Run hasn't written), so it isn't quite os.path.realpath. That's fine for the /var/run -> /run case this targets, but could we reword the comment so it doesn't claim full parity?
There was a problem hiding this comment.
Done in a14152d5ca4. Reworded the Javadoc on resolveRealPathAllowingMissing to note that it resolves symlinks on existing ancestor directories (such as /var/run to /run before startup files are created) without claiming full os.path.realpath parity.
c40c533
into
googleapis:agentic-identities-bound-token
…r agent identities (#14668) ## Summary Merges the `agentic-identities-bound-token` feature branch into `main`, adding end-to-end support for certificate-bound tokens and automatic mTLS with transparent certificate rotation for agent identities on Cloud Run and GKE. ### `google-auth-library-java` (`oauth2_http`) - **Default certificate-bound tokens (`#13873`, `#14595`, `#14658`):** When an agent identity SPIFFE workload certificate is present (either via `GOOGLE_API_CERTIFICATE_CONFIG` on Cloud Run or the GKE credential bundle at `/var/run/secrets/workload-spiffe-credentials/x509.credential-bundle.private-key.pem`), `ComputeEngineCredentials` requests certificate-bound access tokens and ID tokens via `POST` to the Metadata Server with the leaf certificate chain. - **Opt-out controls (`#13873`, `#14559`, `#14557`):** - Process-wide: `GOOGLE_API_ENABLE_RUNTIME_BOUND_TOKEN=false` (or legacy `GOOGLE_API_PREVENT_AGENT_TOKEN_SHARING_FOR_GCP_SERVICES=false`) disables token binding. - Per-target ID tokens: `IdTokenProvider.Option.DISABLE_BOUND_ID_TOKEN` requests an unbound ID token for specific targets (such as plain HTTPS `*.run.app` URLs) while keeping other ID tokens and access tokens bound. - **Automatic mTLS & strict configuration (`#13995`, `#14595`, `#14658`):** - `MtlsUtils.useMtlsClientCertificate()` enables mTLS automatically when a workload certificate config or GKE credential bundle is present, unless `GOOGLE_API_USE_CLIENT_CERTIFICATE` is set to a non-`"true"` value. - Missing, unreadable, or malformed explicit `GOOGLE_API_CERTIFICATE_CONFIG` files fail closed with `IllegalStateException`. - **Full mTLS certificate chain (`#14621`):** `X509Provider` and `SecureConnectProvider` now populate the mTLS key store with the full certificate chain (leaf + intermediates) via `MtlsKeyStoreUtils.createMtlsKeyStore` rather than only the leaf certificate. ### `sdk-platform-java/gax-java` (`gax`, `gax-grpc`, `gax-httpjson`) - **Transparent retries on mTLS certificate rotation (`#13995`):** - `CertificateRotationTracker` checks the SHA-256 fingerprint of the workload certificate on disk after an `UNAUTHENTICATED` failure. - `ChannelPool` (gRPC) and `RefreshingHttpJsonChannel` (HTTP/JSON) rebuild transport channels with the rotated certificate without interrupting in-flight RPCs. - `AttemptCallable` (unary) and `ServerStreamingAttemptCallable` (server streaming) issue one immediate free retry on the refreshed channel when a certificate rotation is detected. - **Fail-closed default mTLS provider loading (`#14658`):** When `useMtlsClientCertificate()` is `true`, `EndpointContext`, `InstantiatingGrpcChannelProvider`, and `InstantiatingHttpJsonChannelProvider` propagate `IOException`s (while still falling back cleanly on `CertificateSourceUnavailableException` when no certificate source exists on disk). - **Conscrypt TLS 1.3 mTLS fix (`#14556`):** `InstantiatingHttpJsonChannelProvider` uses Conscrypt's own `TrustManagerFactory` and `KeyManagerFactory` for the mTLS `SSLContext`, fixing `Unknown authType: GENERIC` on TLS 1.3 and key-manager algorithm constraint mismatches on JDK 26+. ## Sub-PRs Included on `agentic-identities-bound-token` | PR | Description | | :--- | :--- | | `#13873` | `feat: Enable Bound Token for Agentic Identities` | | `#13995` | `feat(gax): support transparent retries during mTLS certificate rotations` | | `#14556` | `fix(gax-httpjson): use Conscrypt TrustManagerFactory for mTLS SSLContext` | | `#14559` | `feat(auth): add DISABLE_BOUND_ID_TOKEN option to opt out of certificate-bound ID tokens` | | `#14557` | `docs(auth): document certificate-bound tokens for agent identities` | | `#14595` | `feat(auth): support GKE agent identity credential bundle for bound tokens and mTLS` | | `#14626` | `fix(gax): drop jsr305 annotations left over after syncing with main` | | `#14621` | `fix(auth): present the full certificate chain for mTLS` | | `#14658` | `fix(auth,gax): align Java mTLS and agent identity certificate handling with Python parity` | ## Testing - **Automated tests:** Full unit test suites in `google-auth-library-java` (`credentials`, `oauth2_http`) and `sdk-platform-java/gax-java` (`gax`, `gax-grpc`, `gax-httpjson`) pass across JDK 8, 11, 17, 21, 25, and 26, including new unit tests for certificate discovery, bound token requests, startup polling, rotation retries, Conscrypt TLS 1.3 mTLS handshakes, and full certificate chains. - **Integrated manual testing (`45/45` scenarios passed):** Built all 9 changed artifacts (`google-auth-library-credentials`, `google-auth-library-oauth2-http`, `api-common`, `gax`, `gax-grpc`, `gax-httpjson`, `google-cloud-core`, `google-cloud-core-http`, `google-cloud-core-grpc`) together from the combined branch and ran 22 live GAPIC mTLS rotation scenarios, 14 local bound-token and GAX parity checks, 3 live Cloud Run Agent Identity job executions, and 6 live GKE Agent Identity pod scenarios. <details> <summary>Full manual test matrix (45 scenarios)</summary> ### 1. Live GAPIC mTLS Rotation (`22/22` Scenarios, `125/125` Checks) Executed real `KeyManagementServiceClient` (gRPC and HTTP/JSON) and `BigQueryReadClient` (server-streaming `ReadRows`) against live `.mtls.googleapis.com` endpoints (`cloudkms.mtls.googleapis.com`, `bigquerystorage.mtls.googleapis.com`) while rotating client certificates on disk. | Group | Scenarios | What Was Verified | Result | | :--- | :--- | :--- | :--- | | **Unary gRPC (`G1`–`G7`)** | `atomic-mv`, `inplace`, `symlink-swap`, `corrupt-recover`, `missing-recover`, `idle-gap`, `burst-25-rpcs` | `KeyManagementServiceClient` gRPC channel seamlessly picks up rotated certs across atomic rename, in-place write, Kubernetes-style symlink swap, corrupt/missing file recovery, idle gaps, and concurrent bursts. | **7/7 Passed** | | **Server-Streaming gRPC (`S1`–`S4`)** | `mid-stream`, `between-streams`, `reconnect-after-rotate`, `corrupt-mid-stream` | `BigQueryReadClient.readRowsCallable()` active streams continue uninterrupted during rotation; subsequent streams and reconnects use the new certificate. | **4/4 Passed** | | **Unary HTTP/JSON (`H1`–`H5`)** | `atomic-mv`, `inplace`, `corrupt-recover`, `missing-recover`, `burst-20-rpcs` | `KeyManagementServiceClient` HTTP/JSON transport reloads rotated certs and recovers from transient corrupt/missing files without dropping requests. | **5/5 Passed** | | **Startup & Providers (`N1a`, `N1b`, `C1`)** | `N1a`, `N1b`, `C1-conscrypt`, `C1-jdk` | Missing cert config handling at startup; full rotation works under both Conscrypt and JDK (`SunJSSE`) TLS providers. | **4/4 Passed** | | **Combined Assertions** | `G1`–`S4` assertions | Verified `125/125` individual assertions across 22 isolated JVMs. | **22/22 Passed** | ### 2. Local Bound-Token & GAX Parity (`14/14` Checks) | Check | PR / Area | What Was Verified | Result | | :--- | :--- | :--- | :--- | | **Scenarios 1–7** | `#13873` / `#14595` | Default bound token (`cnf.x5t#S256` matches leaf), `GOOGLE_API_ENABLE_RUNTIME_BOUND_TOKEN=false`, `GOOGLE_API_USE_CLIENT_CERTIFICATE=false`, cert rotation updates `x5t#S256`, and `GOOGLE_API_PREVENT_AGENT_TOKEN_SHARING_FOR_GCP_SERVICES` behavior. | **Passed** | | **Scenario 8d** | `#14658` | `GOOGLE_API_USE_CLIENT_CERTIFICATE=true` with no certs falls back immediately (`43ms` GET) with no 30s polling delay. | **Passed** | | **Scenario 8e** | `#14658` | Invalid explicit `GOOGLE_API_CERTIFICATE_CONFIG` fails closed with `IOException` instead of silently downgrading. | **Passed** | | **Scenario 10** | `#14658` | 0-byte startup cert file is polled until populated rather than failing immediately. | **Passed** | | **Scenario 11** | `#14658` | Kubernetes `..data` symlinked parent directory resolves to the real directory for file watching. | **Passed** | | **Scenario 12a** | `#14559` | `IdTokenProvider.Option.DISABLE_BOUND_ID_TOKEN` returns an unbound ID token without leaking state to subsequent default calls. | **Passed** | | **Scenario 12b** | `#14621` | `X509Provider` preserves the full 2-certificate chain (`leaf + intermediate`) in the `KeyStore`. | **Passed** | | **GAX `fail-closed`** | `#14658` | Missing explicit `GOOGLE_API_CERTIFICATE_CONFIG` throws `IllegalStateException` in both `InstantiatingGrpcChannelProvider` and `InstantiatingHttpJsonChannelProvider`. | **Passed** | | **GAX `zero-ms-fallback`** | `#14658` | `GOOGLE_API_USE_CLIENT_CERTIFICATE=true` with no certs builds plain gRPC and HTTP/JSON channels immediately (`0ms` polling delay). | **Passed** | ### 3. Live Cloud Run Agent Identity (`3/3` Scenarios) Tested on a Cloud Run Agent Identity job (`--functional-type=agent --identity-type=agent-identity`). | Scenario | Env Vars | Key Results | | :--- | :--- | :--- | | **Default (Bound)** | *(unset)* | `DefaultMtlsProviderFactory` chain length = `2` (`#14621`). Access token acquired in `414ms`. ID token `cnf.x5t#S256` matched leaf cert. `DISABLE_BOUND_ID_TOKEN` (`#14559`) returned `cnf: false` and next default call restored `cnf: true`. GAX `ProjectsClient` (gRPC & HTTP/JSON) resolved `cloudresourcemanager.mtls.googleapis.com:443` and passed mTLS + bound-token auth (`PERMISSION_DENIED` IAM check). Non-mTLS transport rejected bound token (`401`), true mTLS passed auth (`403`), and mismatched client cert was rejected (`401`). | | **Runtime Bound Opt-Out** | `GOOGLE_API_ENABLE_RUNTIME_BOUND_TOKEN=false` | Unbound access token and ID token (`cnf` absent). Non-mTLS endpoints passed auth (`403` IAM) and live Java ADK `LlmAgent` + `Gemini` call on Vertex AI succeeded (`200 OK`). | | **Client Cert Opt-Out** | `GOOGLE_API_USE_CLIENT_CERTIFICATE=false` | Unbound access token (`450ms`) and ID token (`cnf` absent). GAX `ProjectsClient` (gRPC & HTTP/JSON) resolved plain `cloudresourcemanager.googleapis.com:443` and passed auth (`PERMISSION_DENIED`). Live Java ADK `LlmAgent` + `Gemini` call on Vertex AI succeeded (`200 OK`). | ### 4. Live GKE Agent Identity (`6/6` Scenarios) Tested on a GKE Agent Identity cluster against live GCP APIs (`cloudresourcemanager`, `storage`) and an in-cluster SPIFFE mTLS receiver configured to trust **only** the SPIFFE root CA (requiring the full leaf + intermediate chain from `#14621`). | Scenario | Pod Annotations | Env Vars | Key Results | | :--- | :--- | :--- | :--- | | **`s1-default`** | `inject-podcertificates=true` | *(unset)* | `bundle exists=true`, first token fetch `1430ms`, SDK KeyStore chain length = `2`, mTLS receiver (SDK KeyStore) = `200 OK`, mTLS GCS passed auth (`404`), plain GCS rejected bound token (`401`), gRPC & HTTP/JSON resolved `.mtls.googleapis.com:443` and passed auth, `DISABLE_BOUND_ID_TOKEN` -> plain `200` / mTLS `401` / default after opt-out -> mTLS `200` / access token still bound -> plain GCS `401`. | | **`s2-runtime-bound-false`** | `inject-podcertificates=true` | `GOOGLE_API_ENABLE_RUNTIME_BOUND_TOKEN=false` | Unbound tokens (`cnf=null`), mTLS receiver rejected unbound token (`401`), plain receiver accepted (`200`), raw GCS and Storage SDK (HTTP & gRPC) passed auth (`404`), gRPC & HTTP/JSON resolved `.mtls.googleapis.com:443` and passed auth. | | **`s3-use-cert-false`** | `inject-podcertificates=true` | `GOOGLE_API_USE_CLIENT_CERTIFICATE=false` | Unbound tokens (`cnf=null`), gRPC & HTTP/JSON resolved plain `cloudresourcemanager.googleapis.com:443` and passed auth, plain receiver = `200`, plain GCS passed auth (`404`). | | **`s4-use-cert-true`** | `inject-podcertificates=true` | `GOOGLE_API_USE_CLIENT_CERTIFICATE=true` | Identical bound behavior to `s1-default`: SDK KeyStore mTLS receiver = `200`, `.mtls.googleapis.com:443` resolved and passed auth, all `DISABLE_BOUND_ID_TOKEN` checks passed. | | **`s5-no-annotation`** | *(none)* | *(unset)* | `bundle exists=false`, first token fetch `646ms` (no 30s hang), resolved plain `cloudresourcemanager.googleapis.com:443`, passed auth (`PERMISSION_DENIED` / `403`). | | **`s6-use-cert-true-no-annotation`** | *(none)* | `GOOGLE_API_USE_CLIENT_CERTIFICATE=true` | Verified `#14658` on real GKE: `bundle exists=false`, first token fetch `609ms` (no 30s polling delay), resolved plain `cloudresourcemanager.googleapis.com:443`, passed auth (`PERMISSION_DENIED` / `403`). | </details> BEGIN_COMMIT_OVERRIDE feat(auth): request certificate-bound tokens for agent identities by default. Set GOOGLE_API_ENABLE_RUNTIME_BOUND_TOKEN=false to opt out. feat(auth): add IdTokenProvider.Option.DISABLE_BOUND_ID_TOKEN to request unbound ID tokens for specific targets. feat(auth): support GKE agent identity credential bundle for bound tokens and mTLS. fix(auth): present the full certificate chain for mTLS. docs(auth): known limitation: ADK for Java doesn't support mTLS yet; agents that use it must set GOOGLE_API_ENABLE_RUNTIME_BOUND_TOKEN=false. feat(gax): enable mTLS automatically when a workload certificate is present and support transparent retries during mTLS certificate rotations. fix(gax-httpjson): use Conscrypt TrustManagerFactory and KeyManagerFactory for mTLS SSLContext. END_COMMIT_OVERRIDE --------- Co-authored-by: Kirill Logachev <kirl@google.com> Co-authored-by: Min Zhu <zhumin@google.com> Co-authored-by: Lawrence Qiu <lawrenceqiu@google.com> Co-authored-by: yangyzs <171981480+yangyzs@users.noreply.github.com> Co-authored-by: cloud-java-bot <122572305+cloud-java-bot@users.noreply.github.com> Co-authored-by: Keshav Dandeva <keshavdandeva@google.com> Co-authored-by: Mattie Fu <mattiefu@google.com> Co-authored-by: Will Howes <whowes@google.com> Co-authored-by: Jin Seop Kim <jinseop@google.com> Co-authored-by: Jiang Zhu <jiangzzhu@google.com> Co-authored-by: Knut Olav Løite <koloite@gmail.com> Co-authored-by: release-please[bot] <55107282+release-please[bot]@users.noreply.github.com> Co-authored-by: rahul2393 <irahul@google.com> Co-authored-by: Sakthivel Subramanian <179120858+sakthivelmanii@users.noreply.github.com> Co-authored-by: Gaole Meng <gaole@google.com> Co-authored-by: Pranav Iyer <pjiyer@google.com>
Why
An audit comparing the Java (
agentic-identities-bound-token) and Python (googleapis/google-cloud-python#18559) agent identity and mTLS implementations surfaced several edge-case behavioral differences:CertificateBasedAccess.useMtlsClientCertificate()istrueandDefaultMtlsProviderFactory.create()throws anIOException(for example,CertificateSourceUnavailableExceptionwhen a certificate file disappears mid-read or a configured certificate path is invalid/corrupt),EndpointContext,InstantiatingGrpcChannelProvider, andInstantiatingHttpJsonChannelProvidercaught theIOExceptionand silently degraded tomtlsProvider = null(sending requests over an unauthenticated channel or regular endpoint without client certificates).GOOGLE_API_CERTIFICATE_CONFIGis unset: InAgentIdentityUtils.getWellKnownCertificatePathWithRetry(), whenGOOGLE_API_CERTIFICATE_CONFIGwas unset andGOOGLE_API_USE_CLIENT_CERTIFICATE=truewas set on a non-agent GCE VM where no certificates exist, Java polled the well-known Cloud Run paths for ~3 seconds and threw anIOExceptionon timeout instead of falling back to unbound tokens in 0ms.AgentIdentityUtils.isPathInWellKnownDir(String)usedFiles.exists(path)before callingtoRealPath(), which skipped symlink resolution when the leaf certificate file did not yet exist on disk (e.g., on systems where/var/runis a symlink to/runandGOOGLE_API_CERTIFICATE_CONFIGuses/run/secrets/workload-spiffe-credentials/...).AgentIdentityCacheUtils.checkExistsOrAccessDenied(Path)only checked file existence viaBasicFileAttributeswithout verifyingattrs.isRegularFile() && attrs.size() > 0, which could cause startup polling to return immediately on a newly created 0-byte file before certificate contents were written.GOOGLE_API_USE_CLIENT_CERTIFICATEnon-boolean values:MtlsUtils.useMtlsClientCertificate()andMtlsUtils.getWorkloadCertPath()only checked for"true"and"false", allowing arbitrary non-boolean values (e.g.,"invalid"or"1") to fall through to auto-discovery instead of treating any non-"true"value as disabled.What changes
CertificateBasedAccess,EndpointContext,InstantiatingGrpcChannelProvider,InstantiatingHttpJsonChannelProvider):IOExceptionfromDefaultMtlsProviderFactory.create()(routed via package-privateCertificateBasedAccess.createDefaultMtlsProvider()) whencertificateBasedAccess.useMtlsClientCertificate()istrue, failing closed instead of degrading tomtlsProvider = null.useMtlsClientCertificate()isfalse,IOExceptionduring defaultMtlsProvidercreation continues to be logged atFINEand ignored.AgentIdentityUtils&AgentIdentityCacheUtils:IOExceptionthrow fromgetWellKnownCertificatePathWithRetry()whenGOOGLE_API_CERTIFICATE_CONFIGis unset; check well-known paths once (GKE_CREDENTIAL_BUNDLE_PATH,WELL_KNOWN_CREDENTIAL_BUNDLE_PATH,WELL_KNOWN_CERT_PATH+WELL_KNOWN_KEY_PATH) and immediately returncertsPresent=falseif none are present.GOOGLE_API_CERTIFICATE_CONFIGpointing inside/var/run/secrets/workload-spiffe-credentials/.resolveRealPathAllowingMissing(Path)to resolve symlinks on the nearest existing ancestor directory and re-append remaining relative path components inisPathInWellKnownDir(String).attrs.isRegularFile() && attrs.size() > 0inAgentIdentityCacheUtils.checkExistsOrAccessDenied(Path)so polling waits until certificate files are non-empty regular files (> 0bytes).MtlsUtils:GOOGLE_API_USE_CLIENT_CERTIFICATEother than"true"(case-insensitive) as disabled (false/null) in bothuseMtlsClientCertificate()andgetWorkloadCertPath().Tests
EndpointContextTest,InstantiatingGrpcChannelProviderTest,InstantiatingHttpJsonChannelProviderTest:IOExceptionfromcreateDefaultMtlsProvider()is propagated whenuseMtlsClientCertificate()istrueand ignored whenuseMtlsClientCertificate()isfalse.AgentIdentityUtilsTest:GOOGLE_API_CERTIFICATE_CONFIGtests to verify 0-sleep non-polling behavior and immediatenullreturn when no well-known certificates exist, even whenGOOGLE_API_USE_CLIENT_CERTIFICATE=true.getConfigFilePathWithRetryandgetWorkloadCertPathFromConfigWithRetry.isPathInWellKnownDirwith symlinked parent directory and non-existent leaf file.MtlsUtilsTest:GOOGLE_API_USE_CLIENT_CERTIFICATE("invalid") returnfalseinuseMtlsClientCertificate()andnullingetWorkloadCertPath()even when workload certificates exist on disk.