mesh_remote: add external connection support for ALPC meshes by jstarks · Pull Request #3125 · microsoft/openvmm

jstarks · 2026-03-24T23:40:30Z

Enable already-running processes to join an ALPC mesh at runtime, which is needed for modify/attach scenarios where a control client connects to a running VMM process.

Enable already-running processes to join an ALPC mesh at runtime, which is needed for openvmm-neo modify/attach scenarios where a control client connects to a running VMM process. The core problem is that ALPC mesh invitations today are only delivered via inherited handles during fork/exec. There is no mechanism for an external process to discover and connect to an existing mesh. This adds two building blocks: AlpcNode::new_named() creates a mesh whose Ob directory lives at a well-known path (\BaseNamedObjects\mesh-<node-id>) instead of an anonymous directory. This lets external processes open the directory by name rather than requiring a duplicated handle. AlpcMeshListener uses a named pipe for the invitation handshake. The server listens on \\.\pipe\<name>, and when a client connects, it creates a NamedInvitation (credentials + directory path) and sends it over the pipe. The client deserializes the invitation and calls AlpcNode::join_named() to enter the mesh. The named pipe's default DACL restricts access to the current user, and the mesh secret provides ALPC connection authentication. AlpcMeshInviter is a Clone+Send handle extracted from AlpcNode that delegates invitation creation to the node's internal task, so the listener's accept loop doesn't need to hold a reference to the node. Also hardens ALPC connection authentication: each mesh now has a 256-bit secret generated from getrandom, sent as part of ALPC connection data and validated with constant-time comparison on accept. Connections with invalid secrets are rejected. PortConfig gains required_server_sid() for optional SID-based mutual authentication on the ALPC connect path.

github-actions · 2026-03-24T23:40:45Z

⚠️ Unsafe Code Detected

This PR modifies files containing unsafe Rust code. Extra scrutiny is required during review.

For more on why we check whole files, instead of just diffs, check out the Rustonomicon

Copilot

Pull request overview

Enables Windows processes that are already running to join an existing mesh_remote ALPC mesh at runtime by introducing named-directory ALPC nodes and a named-pipe-based invitation listener, removing the previous “inherited handles only” limitation.

Changes:

Add a Windows named-pipe listener (AlpcMeshListener) to distribute ALPC named invitations to external processes.
Extend Windows ALPC mesh nodes with named-directory support plus serialized invitation credentials (address + mesh secret).
Add ALPC connection mutual-auth plumbing via an optional RequiredServerSid in pal’s ALPC connect path.

Reviewed changes

Copilot reviewed 8 out of 9 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
support/pal/src/windows/alpc.rs	Adds SID validation/buffering and passes `RequiredServerSid` into `NtAlpcConnectPortEx`.
support/pal/Cargo.toml	Adds `thiserror` for Windows-only PAL error types.
support/mesh/mesh_remote/src/lib.rs	Exposes new Windows ALPC APIs (listener + new error/types) via re-exports.
support/mesh/mesh_remote/src/alpc_node.rs	Adds named-directory nodes, invitation credentials structs, mesh secret auth, and inviter handle pattern.
support/mesh/mesh_remote/src/alpc_listener.rs	New named-pipe listener/handshake implementation for runtime mesh joins.
support/mesh/mesh_remote/Cargo.toml	Adds Windows-only deps for mesh secret auth (`constant_time_eq`, `getrandom`).
support/mesh/mesh_process/src/lib.rs	Updates invitation propagation to use `AlpcInvitationCredentials` and the new `AlpcInvitation::new(...)` API.
Cargo.toml	Adds workspace dependency version for `constant_time_eq`.
Cargo.lock	Locks new crate additions (`constant_time_eq`, `getrandom` usage in mesh_remote).

support/mesh/mesh_remote/src/alpc_listener.rs

support/mesh/mesh_remote/src/alpc_node.rs

Copilot

Pull request overview

Copilot reviewed 8 out of 9 changed files in this pull request and generated 2 comments.

support/mesh/mesh_remote/src/alpc_node.rs

support/mesh/mesh_remote/src/alpc_listener.rs

The requirements for SID-based mutual auth on ALPC connections are not clear enough to justify the current API surface. Remove the expected_server_sid parameter from join_named and all internal functions in the connect path (with_id, process_connects, connect_alpc, process_one_connect). This can be re-added with a clearer design when needed.

Copilot

Pull request overview

Copilot reviewed 6 out of 7 changed files in this pull request and generated 4 comments.

Copilot · 2026-03-27T23:36:32Z

support/mesh/mesh_remote/src/alpc_listener.rs

+        // Read the length-prefixed invitation: [4 bytes LE length][data].
+        let mut len_buf = [0u8; 4];
+        stream
+            .read_exact(&mut len_buf)
+            .await
+            .map_err(JoinBySocketError::Connect)?;
+        let data_len = u32::from_le_bytes(len_buf) as usize;
+
+        const MAX_INVITATION_SIZE: usize = 64 * 1024;
+        if data_len > MAX_INVITATION_SIZE {
+            return Err(JoinBySocketError::InvitationTooLarge { len: data_len });
+        }
+
+        let mut data = vec![0u8; data_len];
+        stream
+            .read_exact(&mut data)
+            .await
+            .map_err(JoinBySocketError::Connect)?;


JoinBySocketError::Connect is used for read failures as well (both length and payload reads). That produces a misleading "failed to connect" error when the connection succeeded but the peer closed early or sent a truncated invitation. Consider adding a dedicated read/IO variant (or renaming Connect to something like Io) and mapping these read_exact errors to it.

support/mesh/mesh_remote/src/alpc_listener.rs

support/mesh/mesh_remote/src/alpc_node.rs

Copilot

Pull request overview

Copilot reviewed 6 out of 7 changed files in this pull request and generated 7 comments.

Copilot · 2026-03-30T19:02:50Z

support/mesh/mesh_remote/src/alpc_node.rs

+            .map_err(|_| JoinError::InvalidDirectoryPath)?;
+        let directory = open_object_directory(
+            ObjectAttributes::new().name(&path),
+            DIRECTORY_TRAVERSE | DIRECTORY_QUERY,


join_named opens the object directory with DIRECTORY_TRAVERSE | DIRECTORY_QUERY, but the joiner then creates its own ALPC port under that directory in with_id(). Opening without DIRECTORY_CREATE_OBJECT can cause CreateAlpcPort to fail with access denied; request create-object access (or reuse DIRECTORY_ALL_ACCESS).

Suggested change

DIRECTORY_TRAVERSE | DIRECTORY_QUERY,

DIRECTORY_TRAVERSE | DIRECTORY_QUERY | DIRECTORY_CREATE_OBJECT,

is this real or no?

Copilot · 2026-03-30T19:02:50Z

support/mesh/mesh_remote/src/alpc_node.rs

+        let inviter = node1.inviter();
+        let (invitation, handle) = inviter.invite(recv.into()).await.unwrap();


AlpcMeshInviter does not define an invite method, but this test calls inviter.invite(...), which will not compile. Either implement AlpcMeshInviter::invite (handle-based invitations) or update the test to use AlpcNode::invite directly.

Suggested change

let inviter = node1.inviter();

let (invitation, handle) = inviter.invite(recv.into()).await.unwrap();

let (invitation, handle) = node1.invite(recv.into()).unwrap();

Copilot · 2026-03-30T19:02:50Z

support/mesh/mesh_remote/src/alpc_listener.rs

+    pub async fn finish(mut self, port: Port) -> Result<(), FinishError> {
+        let (invitation, handle) = self
+            .inviter
+            .invite_named(port)
+            .await
+            .map_err(FinishError::Invite)?;
+
+        let data = mesh_protobuf::encode(invitation);
+
+        let len = data.len() as u32;
+        self.stream.write_all(&len.to_le_bytes()).await?;
+        self.stream.write_all(&data).await?;
+        self.stream.flush().await?;
+
+        // Wait for the client to join the mesh via the invitation.
+        handle.await;
+        Ok(())


PendingMeshConnection::finish awaits handle.await indefinitely after sending the invitation. A client can read the invitation and never join, leaving a task parked forever and keeping the invitation entry alive (resource leak/DoS). Consider adding a timeout/cancellation strategy and dropping the InvitationHandle on timeout to cancel the invitation.

Copilot · 2026-03-30T19:02:51Z

support/mesh/mesh_remote/src/alpc_listener.rs

+/// Errors from [`AlpcNode::join_by_socket`].
+#[derive(Debug, thiserror::Error)]
+#[expect(missing_docs)]
+pub enum JoinBySocketError {
+    #[error("failed to connect to mesh socket")]
+    Connect(#[source] io::Error),
+    #[error("invitation too large ({len} bytes)")]
+    InvitationTooLarge { len: usize },
+    #[error("failed to decode invitation")]
+    Decode(#[source] mesh_protobuf::Error),
+    #[error("failed to join mesh")]
+    Join(#[source] JoinError),


JoinBySocketError::Connect is used for both the initial connect and later read_exact calls, but its message says "failed to connect". This makes read/EOF errors misleading; consider a separate Io/Read variant (or rename Connect to a generic I/O error) and map the read paths accordingly.

Copilot · 2026-03-30T19:02:51Z

support/mesh/mesh_remote/src/alpc_node.rs

+        // Generate a 256-bit mesh secret for ALPC connection authentication.
+        // Every connecting node must present this secret to be accepted.
+        let mut secret = [0u8; 32];
+        getrandom::fill(&mut secret).unwrap();


AlpcNode::new_mesh calls getrandom::fill(&mut secret).unwrap(), which can panic if the OS RNG fails. Since this is library code, propagate the error instead (e.g., add a NewNodeError::GetRandom variant and return it).

Suggested change

getrandom::fill(&mut secret).unwrap();

if let Err(_err) = getrandom::fill(&mut secret) {

// Propagate RNG failure as a node-creation error instead of panicking.

return Err(NewNodeError::InvalidDirectoryPath);

}

support/mesh/mesh_remote/src/alpc_node.rs

Copilot · 2026-03-30T19:02:51Z

support/mesh/mesh_remote/src/alpc_node.rs

+        let (addr, handle, init_recv) = self.invite_setup()?;
+        let dir = if dup_directory {
+            Some(self.directory.as_handle().duplicate(true, Some(0))?)
+        } else {
+            None
+        };


process_invite mutates shared state (adds the remote + inserts into invitations) before duplicating the directory handle. If handle duplication fails, the invitation entry/remote handle is left behind and can never be accepted. Either duplicate first (when dup_directory is true) or roll back the inserted invitation on error.

Suggested change

let (addr, handle, init_recv) = self.invite_setup()?;

let dir = if dup_directory {

Some(self.directory.as_handle().duplicate(true, Some(0))?)

} else {

None

};

let dir = if dup_directory {

Some(self.directory.as_handle().duplicate(true, Some(0))?)

} else {

None

};

let (addr, handle, init_recv) = self.invite_setup()?;

Copilot

Pull request overview

Copilot reviewed 6 out of 7 changed files in this pull request and generated 5 comments.

Copilot · 2026-03-30T22:03:22Z

support/mesh/mesh_remote/src/alpc_node.rs

+    /// Generate a new random mesh secret.
+    fn new() -> Self {
+        let mut secret = [0u8; 32];
+        getrandom::fill(&mut secret).unwrap();


MeshSecret::new() uses getrandom::fill(..).unwrap(), which can panic and take down the hosting process if RNG acquisition fails. Consider returning a Result from MeshSecret::new() and propagating it through NewNodeError (or at least use expect("rng failure") for a clearer crash reason).

Suggested change

getrandom::fill(&mut secret).unwrap();

getrandom::fill(&mut secret)

.expect("MeshSecret::new: RNG failure generating mesh secret");

Copilot · 2026-03-30T22:03:23Z

support/mesh/mesh_remote/src/alpc_node.rs

+    /// `new()` instead.
+    pub fn new_named(driver: impl Driver + Spawn + Clone) -> Result<Self, NewNodeError> {
+        let id = NodeId::new();
+        let path_string = format!(r"\BaseNamedObjects\mesh-{id:?}");


new_named() builds the directory path using format!("...{id:?}"), which couples the on-disk/NT namespace to Debug formatting (currently producing names like mesh-N-<uuid>). This also conflicts with the doc comment that says mesh-<random-uuid>. Prefer formatting from the underlying Uuid (e.g., id.0) or update the docs to match the actual name format.

Suggested change

let path_string = format!(r"\BaseNamedObjects\mesh-{id:?}");

let path_string = format!(r"\BaseNamedObjects\mesh-{}", id.0);

Copilot · 2026-03-30T22:03:23Z

support/mesh/mesh_remote/src/alpc_node.rs

+        let directory = open_object_directory(
+            ObjectAttributes::new().name(&path),
+            DIRECTORY_TRAVERSE | DIRECTORY_QUERY,
+        )
+        .map_err(|e| JoinError::OpenDirectory {
+            path: invitation.directory_path.clone(),
+            source: e,
+        })?;


join_named() opens the object directory with only DIRECTORY_TRAVERSE | DIRECTORY_QUERY, but the joiner then creates its own ALPC server port under that directory in with_id(). If the directory handle lacks DIRECTORY_CREATE_OBJECT, port creation can fail with access denied. Consider requesting DIRECTORY_CREATE_OBJECT (or DIRECTORY_ALL_ACCESS) when calling open_object_directory.

Copilot · 2026-03-30T22:03:23Z

support/mesh/mesh_remote/src/alpc_listener.rs

+        // Remove stale socket file if it exists.
+        let _ = std::fs::remove_file(path);
+        let listener = UnixListener::bind(path)?;
+        let listener = PolledSocket::new(driver, listener)?;


AlpcMeshListener::create() unconditionally removes any existing file at path. This can delete an unrelated file if the path is reused or replaced. Consider only deleting the path if it is confirmed to be an AF_UNIX socket (e.g., via pal::windows::fs::is_unix_socket) before calling remove_file.

Copilot · 2026-03-30T22:03:23Z

support/mesh/mesh_remote/src/alpc_listener.rs

+impl Drop for AlpcMeshListener {
+    fn drop(&mut self) {
+        let _ = std::fs::remove_file(&self.path);
+    }


Drop for AlpcMeshListener unconditionally removes self.path. As with creation-time cleanup, it’s safer to only remove the path if it still refers to a Unix socket, to avoid deleting an unrelated file that may have been created at the same location.

chris-oo

Looks reasonable but not sure if you want another set of eyes.

…e_named

Copilot AI review requested due to automatic review settings March 24, 2026 23:40

github-actions bot added the unsafe Related to unsafe code label Mar 24, 2026

Copilot started reviewing on behalf of jstarks March 24, 2026 23:41 View session

Copilot AI reviewed Mar 24, 2026

View reviewed changes

switch to unix sockets

41ba569

jstarks requested a review from Copilot March 25, 2026 00:14

Copilot started reviewing on behalf of jstarks March 25, 2026 00:16 View session

jstarks changed the title ~~mesh_remote: add ALPC mesh listener and named-directory nodes~~ mesh_remote: add external connection support for ALPC meshes Mar 25, 2026

Copilot AI reviewed Mar 25, 2026

View reviewed changes

support/mesh/mesh_remote/src/alpc_node.rs Outdated Show resolved Hide resolved

support/mesh/mesh_remote/src/alpc_listener.rs Show resolved Hide resolved

jstarks added 3 commits March 25, 2026 15:56

cleanup

2f85a37

rustfmt

3e45bd1

Copilot AI review requested due to automatic review settings March 27, 2026 23:26

jstarks marked this pull request as ready for review March 27, 2026 23:26

jstarks requested a review from a team as a code owner March 27, 2026 23:26

Copilot started reviewing on behalf of jstarks March 27, 2026 23:29 View session

Copilot AI reviewed Mar 27, 2026

View reviewed changes

jstarks added 2 commits March 28, 2026 00:31

fix Windows compilation and test issues

f936bfc

remove dead code

668d862

Copilot AI review requested due to automatic review settings March 30, 2026 18:31

Copilot started reviewing on behalf of jstarks March 30, 2026 18:52 View session

Copilot AI reviewed Mar 30, 2026

View reviewed changes

jstarks added 3 commits March 30, 2026 19:11

fix

da63b60

feedback

b4647bb

sdlkfjsl;kfjdsf

96df603

Copilot AI review requested due to automatic review settings March 30, 2026 21:57

Copilot started reviewing on behalf of jstarks March 30, 2026 21:58 View session

Copilot AI reviewed Mar 30, 2026

View reviewed changes

chris-oo previously approved these changes Mar 30, 2026

View reviewed changes

fix(mesh_remote): fix doc link to private item AlpcMeshInviter::invit…

9f362d4

…e_named

jstarks dismissed chris-oo’s stale review via 9f362d4 March 30, 2026 22:58

	DIRECTORY_TRAVERSE \| DIRECTORY_QUERY,
	DIRECTORY_TRAVERSE \| DIRECTORY_QUERY \| DIRECTORY_CREATE_OBJECT,

		let inviter = node1.inviter();
		let (invitation, handle) = inviter.invite(recv.into()).await.unwrap();

	let inviter = node1.inviter();
	let (invitation, handle) = inviter.invite(recv.into()).await.unwrap();
	let (invitation, handle) = node1.invite(recv.into()).unwrap();

-        getrandom::fill(&mut secret).unwrap();
+        if let Err(_err) = getrandom::fill(&mut secret) {
+            // Propagate RNG failure as a node-creation error instead of panicking.
+            return Err(NewNodeError::InvalidDirectoryPath);
+        }

	getrandom::fill(&mut secret).unwrap();
	getrandom::fill(&mut secret)
	.expect("MeshSecret::new: RNG failure generating mesh secret");

	let path_string = format!(r"\BaseNamedObjects\mesh-{id:?}");
	let path_string = format!(r"\BaseNamedObjects\mesh-{}", id.0);

Conversation

jstarks commented Mar 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Mar 24, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Mar 27, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

chris-oo Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Copilot AI Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

chris-oo left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

jstarks commented Mar 24, 2026 •

edited

Loading