From fa6e3794b45f06cced4823fc325f2639476ab6d2 Mon Sep 17 00:00:00 2001
From: Calum Murray <cmurray@redhat.com>
Date: Mon, 22 Jun 2026 14:34:03 -0400
Subject: [PATCH] fix(openshell-network-supervisor): gate proxy accept on
 symlink resolution readiness

Signed-off-by: Calum Murray <cmurray@redhat.com>
---
 .../openshell-supervisor-network/src/proxy.rs | 25 +++++++++++++++++++
 .../openshell-supervisor-network/src/run.rs   | 14 +++++++++++
 2 files changed, 39 insertions(+)

diff --git a/crates/openshell-supervisor-network/src/proxy.rs b/crates/openshell-supervisor-network/src/proxy.rs
index d467b022e..cde6b9458 100644
--- a/crates/openshell-supervisor-network/src/proxy.rs
+++ b/crates/openshell-supervisor-network/src/proxy.rs
@@ -188,6 +188,7 @@ impl ProxyHandle {
         policy_local_ctx: Option<Arc<PolicyLocalContext>>,
         denial_tx: Option<mpsc::UnboundedSender<DenialEvent>>,
         activity_tx: Option<ActivitySender>,
+        engine_ready: tokio::sync::watch::Receiver<bool>,
     ) -> Result<Self> {
         // Use override bind_addr, fall back to policy http_addr, then default
         // to loopback:3128.  The default allows the proxy to function when no
@@ -229,6 +230,30 @@ impl ProxyHandle {
         }
 
         let join = tokio::spawn(async move {
+            // Wait for the OPA engine's symlink resolution reload to complete
+            // before accepting connections. This prevents requests from
+            // observing a generation transition mid-flight, which would cause
+            // the generation guard to reject them with a 403.
+            //
+            // The TCP listener is already bound, so the OS backlog queues
+            // incoming SYN packets during this wait. Once we start accepting,
+            // queued connections drain immediately.
+            let mut engine_ready = engine_ready;
+            match tokio::time::timeout(
+                std::time::Duration::from_secs(15),
+                engine_ready.wait_for(|v| *v),
+            )
+            .await
+            {
+                Ok(_) => {}
+                Err(_) => {
+                    warn!(
+                        "Engine readiness signal not received within 15s; \
+                         proceeding with proxy accept loop"
+                    );
+                }
+            }
+
             loop {
                 match listener.accept().await {
                     Ok((stream, _addr)) => {
diff --git a/crates/openshell-supervisor-network/src/run.rs b/crates/openshell-supervisor-network/src/run.rs
index b98923051..68a287f84 100644
--- a/crates/openshell-supervisor-network/src/run.rs
+++ b/crates/openshell-supervisor-network/src/run.rs
@@ -97,6 +97,13 @@ pub async fn run_networking(
             .or_else(|| sandbox_id.map(str::to_string)),
     ));
 
+    // Readiness signal for the proxy accept loop: the proxy binds the TCP
+    // listener immediately (so the OS backlog queues early SYN packets) but
+    // defers `accept()` until symlink resolution completes. This eliminates
+    // the race where an in-flight request observes a generation transition
+    // during the OPA engine reload.
+    let (engine_ready_tx, engine_ready_rx) = tokio::sync::watch::channel(false);
+
     // Spawn a task to resolve policy binary symlinks once the workload's mount
     // namespace becomes accessible via /proc/<pid>/root/. The task starts
     // before run_process spawns the child, so first wait for the orchestrator
@@ -125,6 +132,7 @@ pub async fn run_networking(
                     "Entrypoint PID never published; binary symlink resolution skipped. \
                      Policy binary paths will be matched literally."
                 );
+                let _ = engine_ready_tx.send(true);
                 return;
             }
 
@@ -155,6 +163,7 @@ pub async fn run_networking(
                             );
                         }
                     }
+                    let _ = engine_ready_tx.send(true);
                     return;
                 }
                 debug!(
@@ -170,7 +179,11 @@ pub async fn run_networking(
                  If binaries are symlinks, use canonical paths in your policy \
                  (run 'readlink -f <path>' inside the sandbox)"
             );
+            let _ = engine_ready_tx.send(true);
         });
+    } else {
+        // No symlink resolution needed — unblock the proxy immediately.
+        let _ = engine_ready_tx.send(true);
     }
 
     // Identity cache for SHA256 TOFU when OPA is active. Only consumed by
@@ -279,6 +292,7 @@ pub async fn run_networking(
             Some(policy_local_ctx.clone()),
             denial_tx,
             activity_tx,
+            engine_ready_rx,
         )
         .await?;
         Some(proxy_handle)