1
0
Fork 0
mirror of synced 2026-06-05 14:58:19 +00:00

fix: skip credential check on output-env-credentials: false (#1778)

Closes #1554.
This commit is contained in:
Tom Keller 2026-05-14 13:39:34 -07:00 committed by GitHub
commit 58e7c47adf
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
5 changed files with 56 additions and 14 deletions

View file

@ -736,6 +736,13 @@ This example shows that you can reference the fetched credentials as outputs if
the `aws-session-token` input in a situation where session tokens are fetched the `aws-session-token` input in a situation where session tokens are fetched
and passed to this action. and passed to this action.
If you only want the credentials available as _step outputs_ and not exported to
the environment (for example, on a self-hosted runner where you do not want the
assumed-role credentials to shadow an existing EC2 instance profile), pair
`output-credentials: true` with `output-env-credentials: false`. In that mode,
the action does not run its post-credential SDK-pickup validation step, since
the credentials were never written to the environment.
### Configure multiple AWS profiles in a single workflow ### Configure multiple AWS profiles in a single workflow
```yaml ```yaml

View file

@ -4,8 +4,8 @@
"version": "6.1.1", "version": "6.1.1",
"scripts": { "scripts": {
"build": "tsc", "build": "tsc",
"lint": "biome check --error-on-warnings ./src && markdownlint -i node_modules -i CHANGELOG.md '**/*.md'", "lint": "biome check --error-on-warnings ./src ./test && markdownlint -i node_modules -i CHANGELOG.md '**/*.md'",
"lint:fix": "biome check --write ./src && markdownlint -i node_modules -i CHANGELOG.md -f '**/*.md'", "lint:fix": "biome check --write ./src ./test && markdownlint -i node_modules -i CHANGELOG.md -f '**/*.md'",
"package": "esbuild src/index.ts --bundle --platform=node --target=node24 --outfile=dist/index.js && esbuild src/cleanup/index.ts --bundle --platform=node --target=node24 --outfile=dist/cleanup/index.js && npm run license", "package": "esbuild src/index.ts --bundle --platform=node --target=node24 --outfile=dist/index.js && esbuild src/cleanup/index.ts --bundle --platform=node --target=node24 --outfile=dist/cleanup/index.js && npm run license",
"test": "npm run lint && vitest run && npm run build", "test": "npm run lint && vitest run && npm run build",
"clean": "del-cli coverage test-reports node_modules", "clean": "del-cli coverage test-reports node_modules",

View file

@ -210,11 +210,16 @@ export async function run() {
// Validate that the SDK can actually pick up credentials. // Validate that the SDK can actually pick up credentials.
// This validates cases where this action is using existing environment credentials, // This validates cases where this action is using existing environment credentials,
// and cases where the user intended to provide input credentials but the secrets inputs resolved to empty strings. // and cases where the user intended to provide input credentials but the secrets inputs resolved to empty strings.
await withRetry( // Skip when output-env-credentials is false: input IAM keys were not written to env, so
() => credentialsClient.validateCredentials(AccessKeyId, roleChaining, expectedAccountIds), // the default chain would resolve to ambient runner credentials and the access-key check
'validateCredentials', // would spuriously fail (see #1554).
); if (outputEnvCredentials) {
sourceAccountId = await withRetry(() => exportAccountId(credentialsClient, maskAccountId), 'exportAccountId'); await withRetry(
() => credentialsClient.validateCredentials(AccessKeyId, roleChaining, expectedAccountIds),
'validateCredentials',
);
sourceAccountId = await withRetry(() => exportAccountId(credentialsClient, maskAccountId), 'exportAccountId');
}
} }
if (customTags && (useGitHubOIDCProvider() || webIdentityTokenFile)) { if (customTags && (useGitHubOIDCProvider() || webIdentityTokenFile)) {
core.warning( core.warning(
@ -247,13 +252,12 @@ export async function run() {
} while (specialCharacterWorkaround && !verifyKeys(roleCredentials.Credentials)); } while (specialCharacterWorkaround && !verifyKeys(roleCredentials.Credentials));
core.info(`Authenticated as assumedRoleId ${roleCredentials.AssumedRoleUser?.AssumedRoleId}`); core.info(`Authenticated as assumedRoleId ${roleCredentials.AssumedRoleUser?.AssumedRoleId}`);
exportCredentials(roleCredentials.Credentials, outputCredentials, outputEnvCredentials); exportCredentials(roleCredentials.Credentials, outputCredentials, outputEnvCredentials);
// We need to validate the credentials in 2 of our use-cases // Validate that the SDK can pick up the assumed-role credentials from the environment.
// First: self-hosted runners. If the GITHUB_ACTIONS environment variable // Skip when output-env-credentials is false: the credentials were never written to env,
// is set to `true` then we are NOT in a self-hosted runner. // so the default credential provider chain would resolve to ambient runner credentials
// Second: Customer provided credentials manually (IAM User keys stored in GH Secrets) // (e.g. an EC2 instance profile) and the access-key-id check would spuriously fail.
// If we are using a profile, don't validate credentials yet (since they most likely won't be in the environment). // Skip when using a profile: validation runs after the profile file is written below.
// Wait until after creds are written to the profile file to try validation. if ((!process.env.GITHUB_ACTIONS || AccessKeyId) && !awsProfile && outputEnvCredentials) {
if ((!process.env.GITHUB_ACTIONS || AccessKeyId) && !awsProfile) {
await withRetry( await withRetry(
() => () =>
credentialsClient.validateCredentials( credentialsClient.validateCredentials(

View file

@ -151,6 +151,29 @@ describe('Configure AWS Credentials', {}, () => {
}); });
}); });
// Regression test for #1554: IAM keys + role-to-assume on a self-hosted runner
// with ambient credentials (e.g. an EC2 instance profile), and output-env-credentials=false.
// The post-assume-role validation must be skipped, otherwise the SDK loads the runner's
// ambient access key (which doesn't match the assumed role's) and the action fails.
describe('AssumeRole with IAM LTC and output-env-credentials=false', {}, () => {
it('does not validate against ambient credentials', async () => {
vi.mocked(core.getInput).mockImplementation(mocks.getInput(mocks.IAM_ASSUMEROLE_NO_ENV_INPUTS));
mockedSTSClient.on(AssumeRoleCommand).resolvesOnce(mocks.outputs.STS_CREDENTIALS);
mockedSTSClient.on(GetCallerIdentityCommand).resolves({ ...mocks.outputs.GET_CALLER_IDENTITY });
// Simulate the runner's ambient instance-profile credentials.
// biome-ignore lint/suspicious/noExplicitAny: any required to mock private method
vi.spyOn(CredentialsClient.prototype as any, 'loadCredentials').mockResolvedValue({
accessKeyId: 'AMBIENTINSTANCEPROFILEID',
});
await run();
expect(core.setFailed).not.toHaveBeenCalled();
expect(core.exportVariable).not.toHaveBeenCalled();
expect(core.setOutput).toHaveBeenCalledWith('aws-access-key-id', 'STSAWSACCESSKEYID');
expect(core.setOutput).toHaveBeenCalledWith('aws-secret-access-key', 'STSAWSSECRETACCESSKEY');
expect(core.setOutput).toHaveBeenCalledWith('aws-session-token', 'STSAWSSESSIONTOKEN');
});
});
describe('AssumeRole with WebIdentityTokeFile', {}, () => { describe('AssumeRole with WebIdentityTokeFile', {}, () => {
beforeEach(() => { beforeEach(() => {
vi.mocked(core.getInput).mockImplementation(mocks.getInput(mocks.WEBIDENTITY_TOKEN_FILE_INPUTS)); vi.mocked(core.getInput).mockImplementation(mocks.getInput(mocks.WEBIDENTITY_TOKEN_FILE_INPUTS));

View file

@ -83,6 +83,14 @@ const inputs = {
'output-env-credentials': 'false', 'output-env-credentials': 'false',
'output-credentials': 'true', 'output-credentials': 'true',
}, },
IAM_ASSUMEROLE_NO_ENV_INPUTS: {
'aws-access-key-id': 'MYAWSACCESSKEYID',
'aws-secret-access-key': 'MYAWSSECRETACCESSKEY',
'role-to-assume': 'arn:aws:iam::111111111111:role/MY-ROLE',
'aws-region': 'fake-region-1',
'output-env-credentials': 'false',
'output-credentials': 'true',
},
}; };
const envs = { const envs = {