[
https://issues.apache.org/jira/browse/TOMEE-4648?focusedWorklogId=1032739&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-1032739
]
ASF GitHub Bot logged work on TOMEE-4648:
-----------------------------------------
Author: ASF GitHub Bot
Created on: 29/Jul/26 08:24
Start Date: 29/Jul/26 08:24
Worklog Time Spent: 10m
Work Description: jungm commented on PR #2851:
URL: https://github.com/apache/tomee/pull/2851#issuecomment-5115023205
Ran a full build with tests locally on JDK 17 (matching the CI workflow's
JDK), `clean install -Pstyle,rat -DfailIfNoTests=false` — i.e. the CI command
minus the `-DskipTests`. **Result: BUILD SUCCESS, zero test failures across the
reactor.**
Worth flagging why this needed running by hand:
`main-pull-request-build.yml` runs with `-DskipTests`, so the green `build`
check on this PR is compile + `style`/`rat` sanity only and never exercised the
change. Nothing in PR CI would have caught a regression here.
Two caveats, both pre-existing and unrelated to this PR:
- `server/openejb-multicast` hangs indefinitely on this machine in
`MulticastDiscoveryAgent$Broadcaster.heartbeat` → `DatagramSocket.send` (local
network doesn't do multicast). I confirmed the log stopped advancing for ~23
min, killed it, and resumed the reactor from the next module. It fails the same
way on an unpatched tree, so it isn't related to this change — but it does mean
that one module is unverified locally and I'd rely on CI/another environment
for it.
- `maven/tomee-maven-plugin` logs `Error occurred during initialization of
VM` from a missing third-party artifact
(`org.apache.sirona:sirona-javaagent:0.2-incubating:jar:shaded`). The test
still passes and the module builds; also pre-existing.
`tomee-security` specifically: **108 tests, 0 failures**, with `rat` and
`style` clean on the new files.
If you'd rather see this on real CI infrastructure than my laptop, I can't
trigger an ASF Actions run myself — the workflow is `pull_request`-scoped and
already reported. If it's useful I could add a temporary commit enabling tests
in the PR workflow so CI runs them, but that's a workflow change I'd rather not
sneak into this PR without your say-so.
_🤖 Addressed by [Claude Code](https://claude.com/claude-code)_
Issue Time Tracking
-------------------
Worklog Id: (was: 1032739)
Time Spent: 40m (was: 0.5h)
> Jakarta Security: BASIC mechanism rejects valid credentials with an
> application IdentityStore
> ---------------------------------------------------------------------------------------------
>
> Key: TOMEE-4648
> URL: https://issues.apache.org/jira/browse/TOMEE-4648
> Project: TomEE
> Issue Type: Bug
> Reporter: Markus Jung
> Assignee: Markus Jung
> Priority: Major
> Time Spent: 40m
> Remaining Estimate: 0h
>
> When the BASIC authentication mechanism validates credentials against an
> application-supplied {{IdentityStore}}, TomEE answers 401 even when the
> caller sends valid credentials. This shows up in the plain BASIC test and in
> the decorated and custom-handler variants
> ({{AppCustomAuthenticationMechanismHandler2IT}}), so the problem is not about
> wrapping the mechanism.
> h2. Root cause
> The BASIC mechanism passed a {{BasicAuthenticationCredential}} to
> {{IdentityStoreHandler.validate(Credential)}}. The spec's default
> {{IdentityStore.validate(Credential)}} dispatches to a {{validate(...)}}
> overload only on an _exact_ parameter-type match (its javadoc explicitly
> states it does not look for the most specific overload). Because
> {{BasicAuthenticationCredential extends UsernamePasswordCredential}}, a store
> declaring the idiomatic {{validate(UsernamePasswordCredential)}} overload —
> exactly what the TCK's {{TestIdentityStore}} does — was never invoked and
> returned {{NOT_VALIDATED}}, producing a 401 for correct credentials. Built-in
> stores (e.g. Tomcat users) were unaffected, which is why it only surfaced
> with an application store.
> Fix: the BASIC mechanism now hands the identity store a plain
> {{UsernamePasswordCredential}} while still parsing the header via
> {{BasicAuthenticationCredential}}. This matches the Soteria RI, which
> constructs {{new UsernamePasswordCredential(...)}} for the same reason.
> h2. Behaviour change (release note)
> This is _not_ behaviour-preserving in one direction. An application
> {{IdentityStore}} that declares {{validate(BasicAuthenticationCredential)}}
> was previously dispatched to by the exact-match rule and will no longer be
> invoked on the BASIC path; the default {{validate(Credential)}} then applies
> and the store returns {{NOT_VALIDATED}}, so the caller gets a 401. Such
> stores must declare {{validate(UsernamePasswordCredential)}} (or
> {{validate(Credential)}}) instead. This aligns TomEE with the Jakarta
> Security RI, but it is a breaking change for any application relying on the
> previous TomEE-specific behaviour.
> To make this class of failure diagnosable,
> {{TomEEIdentityStoreHandler.validate}} now logs at debug when no store
> validated the credential — previously the caller simply received a 401 with
> nothing logged anywhere, which is what made this issue hard to pin down.
> h2. Steps to reproduce / TCK reference
> Run the Jakarta Security 4.0 TCK reactor against TomEE Plus (Java 21) through
> the {{security}} runner in {{runner-standalone}}. The following tests fail
> and are excluded in {{runner-standalone/exclusions/security.txt}} in the
> apache/tomee-tck harness repo:
> * {{AppCustomAuthenticationMechanismHandler2IT}} (3 failing methods)
> * {{AppMemBasicDecorateIT#testAuthenticated}}
> * {{AppMemBasicIT#testAuthenticated}}
> Remove the matching lines from {{security.txt}} once fixed, then re-run the
> {{security}} runner to confirm all three test classes pass.
> h2. Note on the OpenID modules (not a TomEE bug)
> {{OpenId2DefaultIT}} and {{OpenId3DefaultIT}} were previously listed here as
> token-validation failures. Investigation showed this was a test-harness
> environment problem, not a TomEE defect: these modules start their bundled
> OpenID provider through Tomcat's {{startup.sh}}, which requires
> {{JAVA_HOME}}/{{JRE_HOME}} and ignores {{PATH}}. With those unset the
> provider never started, the client's {{.well-known}} discovery fetch was
> refused, and the tests failed downstream in a way that looked like a token
> check. With {{JAVA_HOME}} set, both OpenID modules pass unchanged. The runner
> has been fixed in apache/tomee-tck to derive {{JAVA_HOME}} when unset; the
> two OpenID entries can be dropped from {{security.txt}} once a corrected CI
> run confirms them.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
