2 Commits

Author SHA1 Message Date
romanett 9b034f8291 Enable nullable reference types centrally; migrate all unset and annotations-only projects (#4700)
# Description

Nullable reference types are now enabled once, in `common.props`, and
every project in the repository builds warning-clean with them.

**Configuration**
- `<Nullable>enable</Nullable>` lives in `common.props`. All per-project
`<Nullable>` settings are removed: 178 project files (enable,
annotations, and the `disable` in the three Aot test projects).
- 419 per-file `#nullable enable`/`restore` directives and the 4
`#nullable disable` directives are removed.
- Kept on purpose:
- The 5 `<auto-generated>` schema files (`UANodeSet.cs`, `UA Type
Dictionary.cs`, `OPCBinarySchema.cs`, `SecuredApplication.cs`,
`PInvokeHelper.cs`) keep `#nullable enable`. The compiler ignores the
project setting in generated code, so the directive is required there.
- `tests/Opc.Ua.Interop.LegacyPeer` keeps `disable`. It builds against
the 1.5.378 packages and does not import the repository props.
- `docs/DeveloperGuide.md` and `.github/copilot-instructions.md` now
describe the central setting.

**Migration** (the 43 unset and 13 annotations-only projects, roughly
10k diagnostics)
- **Production code: annotated at the declarations.** Covers
`tools/Opc.Ua.SourceGeneration.Core`, `src/Opc.Ua.Lds.Server` and the
LDS sample.
- Members, parameters and returns are nullable where null actually
flows.
- Null-check helpers carry `[NotNullWhen]` (`IsNull`, `IsValidXmlName`,
`IsMethodTypeDesign`).
- `CoreUtils.Format`, `Utils.Format`, `ITemplateWriter.WriteLine` take
`params object?[]`. `Template.AddReplacement` takes `string?` (rendering
already skips null replacements).
- `IRegisteredServerStore.RegisterAsync` returns
`Task<RegistrationEntry?>`: it returns null when a registration is
removed.
- Dereferences backed by an invariant (for example, the validated model)
use `!`, so runtime behaviour is unchanged.
- **Behavioural changes in production code** (small, deliberate):
- `NodeSetToModelDesign` throws `InvalidOperationException` when the
file is not a NodeSet. Before, it threw a NullReferenceException on the
next line.
- `ModelDesignValidator` starts its lookup tables empty instead of null.
A redundant `if (m_nodes == null) return;` guard is gone.
- **Test code: minimal annotations.**
  - `?` on values that can be null, `!` where the test asserts non-null.
  - `= null!` for fixture members assigned in SetUp.
- `TryGetStructure<T>(out T? x)` where the inferred `T?` broke the
`IEncodeable` constraint.
- A few redundant `== null` checks on structs (`NodeId`, `ArrayOf<T>`)
folded into the adjacent `IsNull`/`Count` checks.

**Notes for reviewers**
- The diff is large (~1,500 files) but almost entirely mechanical: `?`,
`!` and directive removal. The non-annotation changes are listed above.
- Known follow-up, not done here:
`IModelDesign.Namespaces/TargetNamespace/NamespaceUris` stay
non-nullable even though the validator returns null before it loads a
dictionary. Making them nullable would ripple through every generator
call site.

## Related Issues

- Fixes #4439

## Checklist

- [ ] I have signed the
[CLA](https://opcfoundation.org/license/cla/ContributorLicenseAgreementv1.0.pdf)
and read the
[CONTRIBUTING](https://github.com/OPCFoundation/UA-.NETStandard/blob/master/CONTRIBUTING.md)
doc.
- [ ] I have added tests that prove my fix is effective or that my
feature works and increased code coverage.
- [x] I have added all necessary documentation.
- [x] I have verified that my changes do not introduce (new) build or
analyzer warnings.
- [ ] I ran **all** tests locally using the **UA.slnx** solution against
at least .net **framework** and .net **10**, and all passed.
- [ ] I fixed **all** failing and flaky tests in the CI pipelines and
**all** CodeQL warnings.
- [ ] I have addressed **all** PR feedback received.

The "no new warnings" box covers net10.0 and net48: full non-incremental
builds of `UA.slnx`, `tools/SourceGeneration.slnx`,
`tools/MigrationAnalyzer.slnx` and the Aot test projects finished with 0
warnings and 0 errors under the repository's warnings-as-errors
settings. net9.0 and net8.0 also had no nullable diagnostics, but they
were not rebuilt after the final fixes. Tests have not been run locally
yet.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-08 08:48:13 +02:00
romanett a0d569d756 Fix Server and GDS audit findings, review findings and spec-conformance gaps (#4571)
## Summary

Fixes the findings of a bug audit of `Opc.Ua.Server`,
`Opc.Ua.Gds.Server` and `Opc.Ua.Gds.Client`. Rebased on master after
#4516.

What went in:
- **Audit:** 93 findings, each checked against the OPC UA specification.
- **Code review of those fixes:** 35 further findings.
- **Spec grounding of every fix commit:** 18 conformance gaps, now
closed.

Commits are one per finding, named `Fix <ID>`, `Fix review <ID>` or `Fix
spec gap <name>`. The IDs refer to the audit report (S = Server areas,
G1 = GDS server, G2 = GDS client).

Validation:
- Each finding was traced in code and checked against OPC 10000 (parts
3, 4, 5, 9, 11, 12, 13, 18, 20).
- Every fix commit was checked against the spec text again afterwards: 0
non-conformant, remaining gaps fixed.
- Regression tests were added for most fixes.
- Suites run on net10.0 before the rebase: Server, GDS, History,
Configuration, Features, Core.Security, DI, Redundancy, LDS,
InformationModel, PubSub.Server, Client, Sessions and the three
Subscriptions suites, all green. The same suites on the rebased branch
are still running; results will be posted as a comment.

## Behaviour and user-facing changes

### Secure defaults
- **`Historize()` / `HistorianBuilder`** grant HistoryRead only. History
writes need an explicit `historyAccessLevel: HistoryRead |
HistoryWrite`.
- **File-system mount:**
- Anonymous sessions can no longer create, delete, move or open files
for writing (`Bad_UserAccessDenied`).
  - UserWritable and UserExecutable now reflect the user.
- `AllowAnonymousWrite` restores the old behaviour; the ReferenceServer
sample turns it on.
- **Default PKI root** for the hosted server, hosted GDS, hosted LDS and
`OpcUaApplicationConfigurationProvider`:
- Moved from the shared temp directory to `{LocalApplicationData}/OPC
Foundation/{ApplicationName}/pki`, and startup fails if no per-user
directory exists.
- When certificate stores of an earlier version still exist under temp,
a warning names both paths. Existing deployments that relied on the
default get a new application certificate (and the hosted GDS a new CA)
unless `PkiRoot` is set to the old path.
- **GDS:**
  - `GdsServerOptions.AutoApprove` defaults to `false`.
- The default KeyCredential store grants no roles unless
`AuthorizeRoles` allows them.
- Certificate requests (StartSigningRequest, StartNewKeyPairRequest,
FinishRequest) need an encrypted channel.
- **Hosted server anonymous logins:**
- Anonymous is rejected (`Bad_IdentityTokenRejected`) when the default
authenticator options disable it.
- A custom `AddIdentityAuthenticator<T>()` on its own no longer disables
anonymous.
- With anonymous disabled and no configured token policies, the endpoint
advertises the authenticators' token types instead of Anonymous.
- **Runtime-added roles:** their properties and methods need
SecurityAdmin over an encrypted channel, like the well-known roles.
- **Disabled users:** users disabled through Part 18 UserManagement can
no longer log in through `RegisterDefaultAuthenticators(IUserDatabase)`.

### GDS server (Part 12)
- **Certificate requests:**
- FinishRequest / ReadRequest with another application's RequestId
returns `Bad_InvalidArgument`.
  - Every Start*Request returns a new RequestId.
- A CSR without the ApplicationUri is rejected
(`Bad_CertificateUriInvalid`).
- An abstract or unsupported certificate type is rejected at Start
(`Bad_InvalidArgument`).
- **Private-key password:** no longer persisted in the JSON applications
database. A password saved by an earlier version is read into memory
once and dropped on the next save.
- **CheckRevocationStatus:** validates offline against the GDS's own
trust lists and CRLs. It returns Good for good GDS-issued certificates
and `Bad_CertificateInvalid` for input that is not a certificate.
- **Certificate-group trust lists:**
- Writes need CertificateAuthorityAdmin; SecurityAdmin and application
admins are read-only.
  - Writes are validated with the Part 4 certificate validation.
- **AuthorizationService:**
- The legacy RequestAccessToken applies `AccessControl` and issues the
token for the session identity.
- StartRequestToken passes the session identity to every identity-aware
provider.
  - Unknown ResourceIds return `Bad_NotFound`.
- Expired token requests and refresh tokens are pruned, with at most 10
000 pending requests (`Bad_TooManyOperations`).
- **KeyCredential:** requests with PublicKey + SecurityPolicyUri are
rejected with `Bad_SecurityPolicyRejected` (secret encryption is not
implemented); a PublicKey without SecurityPolicyUri is rejected with
`Bad_InvalidArgument`.
- **Audit events:** access-token audit events no longer contain the
password or token data.
- **RegisterTickets / UnregisterTickets:** need the RegistrarAdmin role
when called from a session.
- **Managed application nodes:** moved from namespace 0 to their own
namespace (`ManagedApplicationsNamespaceUri`); the NodeId is the
ApplicationUri, and the mandatory child properties exist.
- **Hosted GDS:**
- New `GdsServerOptions.CertificateGroups`
(`GdsCertificateGroupOptions`), bindable from configuration.
  - The mandatory DefaultApplicationGroup is always present.
  - The default CA subject uses the resolved application name.
- **JSON applications database:**
- A corrupt file throws instead of being silently replaced by an empty
database.
- Saves are atomic and write in place on bind mounts; on net8+ builds
they also follow symlinks and keep the Unix file mode.
  - Issued certificates and trust-list links survive a reload.

### GDS client
- **Trust-list download:**
  - Reads until an empty chunk (Part 20 short reads).
- A size-limit rejection of the final end-of-file read, as sent by
released servers, counts as end of file.
- **Default certificate group:** a null group means
DefaultApplicationGroup, and vendor groups in other namespaces are no
longer mistaken for well-known groups.
- **Status codes and errors:**
  - Missing directory proxies report `Bad_ServerNotConnected`.
- Short or wrongly typed method outputs throw `BadUnexpectedError` /
`BadTypeMismatch`.
- **GdsAccessTokenProvider:** no longer caches a failed client task, and
one caller's cancellation no longer fails other callers.

### Trust lists, push and ConfigurationFile (Part 12 / Part 20)
- **AddCertificate:**
  - `IsTrustedCertificate=false` returns `Bad_CertificateInvalid`.
- Added certificates are validated (Part 4), and a certificate revoked
by a current CRL returns `Bad_CertificateRevoked`.
- **CloseAndUpdate:** validates the uploaded list against its own
certificates and CRLs, not the stores it replaces. Validation failures
are no longer recorded in the server's Rejected store.
- **RemoveCertificate:** removing a trusted or issuer CA that other
certificates still need returns `Bad_CertificateChainIncomplete`.
- **Handles:**
  - Several read handles may be open at once.
- Any open while a write handle exists returns `Bad_NotReadable` /
`Bad_NotWritable`.
  - A write open while any handle is open returns `Bad_NotWritable`.
  - Handles of closed sessions are released.
  - Idle handles close after ActivityTimeout (default 60 s).
- **Status codes:**
- Read clamps to the remaining bytes and no longer enforces the size
limit.
  - Oversized Write / AddCertificate returns `Bad_RequestTooLarge`.
  - Unsupported open modes return `Bad_NotSupported`.
- **ConfigurationFile.CloseAndUpdate:**
  - Empty Targets return `Bad_InvalidArgument`.
- The ConfigurationUpdatedAuditEvent is raised with the configuration
owner as source.
- With the new `IApplicationConfigurationFileTargetProvider`, providers
receive the Targets, a bad target returns Uncertain with per-target
results, and session-interrupting updates are applied after
RestartDelayTime.
- **Default namespace permissions:** namespaces registered after the
first permission lookup now get their NamespaceMetadata default
RolePermissions and AccessRestrictions (no stale negative cache).

### Sessions, identity and roles (Part 18)
- **Session timeouts:** a failure closing one timed-out session no
longer stops the session-timeout monitor.
- **ChangePassword:** an empty old password returns
`Bad_IdentityTokenInvalid`.
- **Removing or disabling a user:** closes only that user's
UserName-token sessions.
- **Role configuration changes:** roles granted by an `ImpersonateUser`
mapping are kept.
- **X509Subject rules:** a certificate subject value containing a quote
never matches an X509Subject rule.
- **`ImpersonateUser` callback:** no longer runs under the global
session lock.
- **Session diagnostics:** LocaleIds are kept as the client sent them,
while null LocaleIds are ignored for translation.

### Subscriptions and monitored items (Part 4)
- **Acknowledgements:** a failing acknowledgement returns a per-entry
Bad status instead of failing the whole Publish.
- **`DiscardedMessageCount`:** counts messages dropped from the
retransmission queue and unsent overflow messages.
- **Durable subscriptions:** transferring a restored anonymous
subscription works (no NullReferenceException).
- **Transferred subscriptions:** sampling and permission checks use the
item's current owner session.
- **Access denied:** a monitored item whose user may no longer read the
attribute gets a `Bad_UserAccessDenied` notification instead of going
silent. Non-Value attributes need Browse permission (RolePermissions
needs ReadRolePermissions), and User* attributes are evaluated per
subscriber.
- **ModifyMonitoredItems:** a SamplingInterval of -1 uses the
subscription's publishing interval.
- **AggregateFilter priming:** needs HistoryRead / ReadHistory
permission and is bounded (100 000 values, bounded page size).
- **Resource leak:** event monitored items no longer leak a consumer
task per create/delete cycle.

### Node management services (Part 3 / Part 4)
- **Reference types:** AddNodes, AddReferences, DeleteReferences and
Browse reject ids that are not ReferenceTypes with
`Bad_ReferenceTypeIdInvalid`. AddNodes, AddReferences and
DeleteReferences also reject abstract ReferenceTypes; Browse still
accepts them (e.g. `HierarchicalReferences` with includeSubtypes).
- **Concurrent AddNodes:** duplicate BrowseNames under one parent get
`Bad_BrowseNameDuplicated`.
- **TranslateBrowsePathsToNodeIds:**
- Requires Browse permission on intermediate nodes and final targets
(RolePermissions and AccessRestrictions).
  - A long RelativePath no longer rejects the whole request.
- **SemanticChangeEvent:**
- Raised once per actual change of a semantic property, whether or not
it is monitored.
  - Also raised for properties with the SemanticChange access-level bit.
  - Reports the owning node's TypeDefinition as AffectedType.
- **Localized diagnostics:** operation-level text is translated when
OperationLocalizedText is requested.

### History and aggregates (Part 11 / Part 13)
- **DeleteRawModified:** an unspecified or reversed range returns
`Bad_InvalidTimestampArgument`, and StartTime == EndTime deletes the
value at StartTime for every provider.
- **ReadAtTime fallback:** reads only the values around each requested
time. A time whose bound is not found within the search limit returns
`Bad_BoundNotSupported` instead of failing the request.
- **ReadProcessed:** a ProcessingInterval shorter than one tick returns
`Bad_AggregateInvalidInputs`.
- **Delta / DeltaBounds:** widen to the next larger signed type when the
result does not fit the source type (e.g. a decreasing unsigned counter,
or an Int16 overflow).
- **Range / Range2:** return the source data type.

### Alarms and fluent API (Part 9)
- **Suppression:** an alarm stays suppressed while any suppression group
is active, and re-evaluation without a state change does not move
TransitionTime.
- **`MonitorVariable(variable)`:** sets SourceNode, SourceName and
InputNode to the variable and makes it a ConditionSource (HasCondition
to the alarm, HasEventSource from the object).
- **Fluent input/output arguments:** InputArguments / OutputArguments
added through the fluent API to a registered method are readable.
- **Root notifiers:** a failed root-notifier registration no longer
leaves an orphaned event worker.
- **State-machine handlers:** exceptions are logged at Error.

### File system (Part 20)
- **Delete / MoveOrCopy:**
  - Only act on direct children of the directory.
- A NodeId whose kind does not match the entry returns `Bad_NotFound` (a
file NodeId can no longer delete a directory).
- Copying a directory into its own subtree returns
`Bad_InvalidArgument`.
- **Locking:**
- An open file, or a directory containing one, cannot be deleted or
moved (`Bad_InvalidState`).
  - The check is atomic with opens.
- While a Delete or Move runs, read opens get `Bad_NotReadable` and
write opens `Bad_InvalidState`.
- **Open:**
- Reserved mode bits or EraseExisting without Write return
`Bad_InvalidArgument`.
  - Read+Write returns `Bad_NotSupported`.
  - Read while a writer is open returns `Bad_NotReadable`.
- **Client `FileSystemErrors`:** maps `Bad_NotReadable` to
`IOException`.
- **RuntimeNodeSet:** well-formed NodeSets with a stray `Model` element,
or two models in one file where one requires the other, now load.

### Server core, hosting and lifecycle
- **`SessionEstablishmentQueueLimit` > 0:**
CreateSession/ActivateSession now queue for a permit. The permit is
acquired before the request is registered, and a waiting handshake
releases its worker through the request-parking policy.
- **Node manager lifecycle:**
  - Add, reload and remove no longer wait for parked Publish requests.
- A Migrate reload racing with CreateMonitoredItems no longer blocks
later lifecycle operations.
  - Late items on retired or removed managers get `Bad_NodeIdUnknown`.
- **HistoryUpdate:** a batch mixing data and event details is bounded by
both limits.
- **Server stop/start:** stopping the server disposes the configuration
watcher and certificate subscription; starting recreates them.
- **Monitored-item queue factory:** a factory supplied by the caller or
the DI container is no longer disposed by the server.
- **Audit events:** each certificate validation error raises its own
AuditCertificate event. A failed OpenSecureChannel without a service
result is audited as failed.

## New public API (additive)
- `IQueuedSessionEstablishmentLimiter`
- `StandardServer.OwnsMonitoredItemQueueFactory` (protected virtual)
- `DefaultPkiRoot.Get` (Opc.Ua.Configuration)
- `TrustList.SetCertificateValidation`
- `NamespaceMetadataRegistry.Invalidate`
- `IApplicationConfigurationFileTargetProvider` and
`ApplicationConfigurationUpdatePlan`
- `GdsServerOptions.CertificateGroups` / `GdsCertificateGroupOptions`
- `InMemoryKeyCredentialRequestStore.AuthorizeRoles`
-
`DefaultManagedApplicationsNodeManager.ManagedApplicationsNamespaceUri`
- `AuthorizationServiceManager.RequestAccessTokenAsync(...,
IUserIdentity?, ...)`
- `AllowAnonymousWrite` on the file-system node manager and its factory
- `IFileSystemHost.TryBeginMutation` / `EndMutation`, replacing
`HasOpenHandles`

## Not fixed / known limitations
- **KeyCredential:** secret encryption with the client's PublicKey
(requests needing it are rejected).
- **CSR SAN restriction:** restricting subject / SAN values copied from
a CSR (CA policy).
- **Ticket signatures:** ticket signature verification (needs a
trusted-signer configuration).
- **FileType Read+Write open:** needs a read/write stream contract in
`IFileSystemProvider`.
- **Custom `IUserDatabase` implementations:** the disabled-user check
still scans all users per login. The built-in stores do a single lookup.
- **64-bit Delta precision:** limited by the calculator's double
arithmetic.

## Tests with changed expectations
Existing tests were updated where they asserted the old behaviour:
- File-system status codes
- Trust-list handle eviction
- `DiscardedMessageCount`
- Localized diagnostics bits
- Delta on unsigned sources
- SourceNode of fluent alarms
- ReferenceServer transfer permission test (now expects
`Bad_UserAccessDenied` notifications)
- `UserDeactivationShutdownReviewTests` (the revoked session is a
UserName login)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-09-29 07:26:05 +02:00