[WSLC] Fix network policy parity between one-shot and state-aware surfaces by SohamDas2021 · Pull Request #1042 · microsoft/mxc

Soham Das (SohamDas2021) · 2026-08-25T19:34:29Z

📖 Description

I have signed the Contributor License Agreement.
This pull request is related to an issue.
If this PR changes build commands, project architecture, or key conventions, I have updated .github/copilot-instructions.md.

Summary

Three WSLc network-policy gaps where the one-shot and state-aware surfaces disagreed.

1. Redundant host lists failed open. The parser gated on needs_host_filtering(), which only inspects the list that tightens the default — so block + blockedHosts and allow + allowedHosts were neither rejected nor enforced. Replaced with has_host_lists(), which is presence-based and fail-closed.

2. portMappings was one-shot only. Added the state-aware twin on the provision phase, honoured as container port forwards and frozen post-provision (TCP only; ports must be unique and non-zero).

3. Two parser gates mislabelled state-aware errors. Both the host-list and network.proxy gates are written for one-shot, where a single config carries the whole policy. On state-aware the network mode is bound at provision, so later phases legitimately omit defaultPolicy — and a proxy-carrying exec/start/stop was rejected as malformed_request before dispatch. Both gates are now !state_aware; every state-aware phase already gates the proxy itself, so the errors surface as policy_validation and exec's cooperative proxy injection works. This last one was pre-existing on main (verified against a main build producing byte-identical output).

Validation

1,896 Rust unit tests, 277 SDK tests, all 6 codegen/version gates green
cargo fmt + clippy -D warnings (incl. -p wxc --features wslc)
Live WSLc E2E on a WSL2 host: run_wslc_all_tests.ps1 (one-shot + state-aware), 0 failures — including a real port-forward round-trip on a dynamically-allocated host port

Closes#824

GitHub Actions runs the PR validation build automatically. The ADO pipeline
(MXC-PR-Build) is the Azure version of the PR pipeline, kept in parity with the GitHub
Actions build; it runs on merge to main, and Microsoft reviewers with write access can trigger it
on a PR with /azp run. See docs/pull-requests.md.

If the dependency-feed-check check fails on a new dependency, the crate must be added to
the feed before the PR can pass. See docs/pull-requests.md
for the steps.

Microsoft Reviewers: Open in CodeFlow

azure-pipelines · 2026-08-25T19:34:37Z

Azure Pipelines: There may be pipelines that require an authorized user to comment /azp run to run.

Copilot

Pull request overview

Aligns WSLc one-shot and state-aware network-policy behavior.

Changes:

Rejects all unsupported WSLc host lists.
Adds state-aware provision-time TCP port mappings.
Improves state-aware proxy validation routing and test coverage.

Reviewed changes

Copilot reviewed 24 out of 26 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
`tests/scripts/run_wslc_state_aware_tests.ps1`	Adds port-forwarding E2E coverage.
`tests/configs/wslc_state_aware_provision_ports.json`	Adds a port-mapping fixture.
`src/core/wxc_common/src/wire.rs`	Extends the provision wire model.
`src/core/wxc_common/src/models.rs`	Adds host-list presence detection.
`src/core/wxc_common/src/config_parser.rs`	Updates WSLc policy validation.
`src/core/wxc_common/src/config_contract_adapters/dev/state_aware.rs`	Converts state-aware port mappings.
`src/core/wxc_common/src/config_contract_adapters/dev/one_shot.rs`	Reuses shared mapping conversion.
`src/core/wxc_common/src/config_contract_adapters/dev/common.rs`	Centralizes mapping conversion.
`src/core/mxc_config_contract/src/dev/state_aware/provision/wslc.rs`	Adds port mappings to the contract.
`src/core/mxc_config_contract/src/dev/state_aware/provision/mod.rs`	Boxes the larger WSLc request.
`src/core/mxc_config_contract/src/dev/request.rs`	Constructs the boxed request.
`src/backends/wslc/daemon/tests/daemon_ipc.rs`	Updates daemon test configuration.
`src/backends/wslc/daemon/src/session_manager.rs`	Applies mappings during provisioning.
`src/backends/wslc/common/src/wsl_container_runner.rs`	Rejects every host-list form.
`src/backends/wslc/common/src/state_aware.rs`	Validates and forwards mappings.
`src/backends/wslc/common/src/daemon_protocol.rs`	Extends and versions daemon IPC.
`src/backends/wslc/common/src/container_steps.rs`	Passes mappings to the SDK.
`sdk/node/tests/unit/wire-conformance-state-aware.test.ts`	Checks SDK wire parity.
`sdk/node/src/state-aware-types.ts`	Exposes provision port mappings.
`sdk/node/src/generated/wire.ts`	Regenerates rolling wire types.
`sdk/node/src/generated/v0_9_0_alpha/wire.ts`	Regenerates exact wire types.
`sdk/node/README.md`	Documents SDK behavior.
`schemas/dev/mxc-config.schema.0.9.0-dev.json`	Updates the rolling schema.
`schemas/dev/mxc-config.schema.0.9.0-alpha.json`	Updates the exact schema.
`docs/wsl/wslc-state-aware.md`	Documents state-aware forwarding.
`docs/wsl/wsl-container-getting-started.md`	Documents fail-closed host lists.

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

Darren Hoehna (dhoehna)

Just some questions.

Copilot

Pull request overview

Copilot reviewed 24 out of 26 changed files in this pull request and generated 1 comment.

Copilot

Pull request overview

Copilot reviewed 24 out of 26 changed files in this pull request and generated 1 comment.

Suppressed comments (1)

src/core/wxc_common/src/config_contract_adapters/dev/state_aware.rs:107

The exact 0.9 state-aware adapter now maps port_mappings, but its WSLC_ALL_FIELDS_REQUEST_JSON adapter/convergence test still omits this field and never asserts the mapped value. The only E2E added here uses 0.8 (the rolling parser), while backend unit tests construct WslcProvisionPhase directly, so dropping this conversion would not be detected. Add portMappings to that all-fields fixture and assert provision.port_mappings to cover the exact-parser path.

 port_mappings: port_mappings.into_option().map(|mappings| {
mappings
.into_iter()
.map(convert_wslc_port_mapping)
.collect()

Copilot

Pull request overview

Copilot reviewed 25 out of 27 changed files in this pull request and generated no new comments.

Branden Bonaby (bbonaby) · 2026-08-29T21:16:33Z

+## Port forwarding
+
+`experimental.wslc.provision.portMappings` forwards host (Windows) ports to the container, mirroring
+the one-shot `experimental.wslc.portMappings` surface. Each entry is `{ windowsPort, containerPort }`


question: (non-blocking): so WSLc has host to container networking but specifically only related to port redirection? If so I wonder if this should mean that wslc configs that want to use port mappings need to also do network.ingress.hostloopback : allow together. Up to you though if you want to keep portmappings a completely separate exception.

Yes, that is correct.
However, coupling it to network.ingress.hostLoopback isn't expressible today: WSLc passes NetworkPolicySupport::LEGACY on every phase, so the whole 0.8 directional network.ingress section is rejected. Requiring it would mean first declaring INGRESS_DEFAULT | HOST_LOOPBACK, claiming a posture WSLC can't actually enforce. They're also different directions: hostLoopback governs whether the container can reach the host's loopback, whereas portMappings is inbound NAT into the container.

Branden Bonaby (bbonaby) · 2026-08-29T21:46:33Z

@@ -710,7 +710,7 @@
 "type": "string"
 },
 "PortMapping": {


question: not related to this file but do we need to add anything for the Rust SDK in terms of the types consumers will use to author a wlsc config?. OR does it look like it's already taken care of?
Here: https://github.com/microsoft/mxc/blob/main/src/core/mxc_engine/src/policy.rs#L504

Types are already covered.
One-shot:WslcSection.port_mappings: Vec<(u16, u16)> is the authoring surface, and wire() emits protocol: "tcp".
State-aware: the Rust SDK is JSON-in/JSON-out by design here, run_state_aware_json takes the wire-format request, so there's no typed per-phase builder to extend. Callers write experimental.wslc.provision.portMappings directly, same as the other state-aware backends.

…t lists (Closes#824)

Copilot

🟢 Approval recommended

The implementation consistently threads validated mappings through all layers and includes appropriate unit, contract, and live coverage.

Review details

Files reviewed: 26/28 changed files
Comments generated: 0 new
Review effort level: Balanced

Tighten the comments this branch introduced and correct three that were wrong. The `block` + `blockedHosts` rejection was documented as a guard against failing open. That reasoning is backwards: `map_network` maps `Block` to `NetworkMode::None`, so ignoring a blocklist under that default leaves the container with no networking at all -- strictly more restricted. Only `allow` + `allowedHosts` can fail open, and the rewrite keeps that framing on the `allow` case while giving both cases the reason that actually covers them: an accepted-then-ignored list misstates what is enforced. The rejection itself was already correct; only the stated rationale changed. Two other comments claimed things the code does not do. The domain `PortMapping.protocol` doc said `"tcp" or "udp"` with `Default: "tcp"`, but `wire::TransportProtocol` declares no `udp` variant and the struct derives `Default`, so the real default is `""` -- that `"tcp"` belonged to the wire layer. Note the scope: UDP is supported in the wire model generally (`NetworkProtocol` has `Tcp/Udp/Icmp/Any`, honored by LXC, Bubblewrap, and PSEC); it is WSLc port forwarding specifically that is TCP-only. The `WslcSection` doc in `mxc_engine::policy` said per-host rules "are accepted" and "fail the run at spawn"; they are rejected when the request is built, which this branch widens to redundant lists. That last correction had no test behind it, so this adds `wslc_rejects_redundant_host_lists`. Both cases it covers pass the old `needs_host_filtering` predicate and were accepted before this branch, so the test fails against the previous behavior rather than restating it. Also expands MDE to Microsoft Defender for Endpoint on first use in the WSL getting-started doc, per review feedback. Deliberately left long: the `WslcProvisionPhase` wire doc, which records which knobs are absent on purpose and is emitted into two schema artifacts and two generated TypeScript oracles; the `One-shot only:` paragraphs justifying the `!state_aware` guards, which would otherwise read as oversights; and the `--dry-run` note explaining why a mapping function is called for its validating side effect. Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> Copilot-Session: 6b41b4f0-4bc3-4516-9a83-ab63d70b3af5

Branden Bonaby (bbonaby) · 2026-09-03T20:32:51Z

+ validate_wslc_port_mappings(mappings, "experimental.wslc.provision.portMappings")
+ .map_err(MxcError::policy_validation)?;


note: I guess this will conflict with your promotion PR

Branden Bonaby (bbonaby) · 2026-09-03T20:36:43Z

+ /// A WSLC provision request. Boxed: its port-mapping list makes it much
+ /// larger than its siblings.


note: probably don't need the second part of the comment.

Branden Bonaby (bbonaby) · 2026-09-03T21:19:16Z

+ fn wslc_rejects_redundant_host_lists() {
+ // A list matching the default enforces nothing, so before #824 these
+ // two were accepted and dropped. `allow` + allowedHosts is the harmful
+ // one: it reads as a restriction while leaving all egress open.


note: Jeff brought this up the other day actually. I don't think it's a WSLc issue. Its a greater config issue. I'm thinking the logic for this should go in network_parser.rs since it's a greater config violation. Thoughts?

Branden Bonaby (bbonaby) · 2026-09-03T21:20:10Z


 use crate::config_contract_adapters::dev::common::{
 convert_filesystem, convert_network, convert_process, convert_telemetry, convert_version,
+ convert_wslc_port_mapping,


note: I think this addition can be deleted right?

Branden Bonaby (bbonaby) · 2026-09-03T21:21:13Z

-fn convert_protocol(value: contract::TransportProtocol) -> wire::TransportProtocol {
- match value {
- contract::TransportProtocol::Tcp => wire::TransportProtocol::Tcp,
- }
-}
-
-fn convert_wslc_port_mapping(value: contract::PortMapping) -> wire::PortMapping {
- let contract::PortMapping {
- windows_port,
- container_port,
- protocol,
- } = value;
- wire::PortMapping {
- windows_port: windows_port.get(),
- container_port: container_port.get(),
- protocol: protocol.into_option().map(convert_protocol),
- }
-}
-


question: since this PR doesn't remove wslc from experimental, does removing these make sense?

Branden Bonaby (bbonaby) · 2026-09-03T21:24:58Z

 }

+ #[test]
+ fn wslc_rejects_redundant_block_with_blocked_hosts() {


note: see comments above, I think this isn't a wslc issue and a config issue. In general we shouldn't allow default block + block hosts, and default allow + allow hosts. So we should have the network_parser.rs handle that rather than it being backend specific.

Branden Bonaby (bbonaby) · 2026-09-03T21:26:01Z

+/// Validate + convert WSLc wire port mappings. Shared by the one-shot parser
+/// and the state-aware provision phase so the two cannot drift. `path` is the
+/// caller's JSON location; the error is a bare message each caller wraps.
+pub fn validate_wslc_port_mappings(


thought: I'm not sure model.rs is the right place to put validation code tbh. When I think about models I think of just the raw objects themselves. Is there maybe another place we can put this?

Branden Bonaby (bbonaby) · 2026-09-03T21:29:29Z

+/// one-shot-only sizing knobs (`cpuCount` / `memoryMb` / `gpu` / `storagePath`)
+/// are deliberately absent: the daemon shares a single session across sandboxes
+/// and does not apply per-sandbox sizing. `portMappings`, by contrast, is


thought: Oh this is what "sizing knobs" mean. That said, thinking about it some more
The daemon shares a single session across sandboxes and does not apply per-sandbox sizing
Would developers expect this? This seems a bit unexpected that I can do things with the one shot that I can't do with the state aware. This backend might potentially be the only one doing that. Would it be a big lift to re-align the two?

Soham Das (SohamDas2021) requested review from a team and a balanced review from Copilot August 25, 2026 19:34

Soham Das (SohamDas2021) requested a review from a team as a code owner August 25, 2026 19:34

Copilotstarted reviewing on behalf of Soham Das (SohamDas2021)August 25, 2026 19:35 View session

CopilotAI reviewed Aug 25, 2026
View reviewed changes

Comment threadsrc/core/wxc_common/src/config_parser.rs

Soham Das (SohamDas2021) mentioned this pull request Aug 25, 2026
[WSLC] Encapsulate backend errors in a typed enum for the Rust SDK #1045
Merged
3 tasks