ipc: check for PID liveness via libc:kill

Signed-off-by: NotAShelf <raf@notashelf.dev>
Change-Id: I961c138cf8341bad6a27c450706de11d6a6a6964

src/ipc.rs

@@ -53,6 +53,25 @@ impl From<serde_json::Error> for IpcError {
   }
 }
 
+/// Check whether a process with the given PID is still alive.
+#[cfg(unix)]
+fn is_process_alive(pid: u32) -> bool {
+  unsafe {
+    if libc::kill(pid as i32, 0) == 0 {
+      return true;
+    }
+    let errno = *libc::__errno_location();
+    // EPERM means the process exists but we can't signal it.
+    errno == libc::EPERM
+  }
+}
+
+#[cfg(not(unix))]
+fn is_process_alive(_pid: u32) -> bool {
+  // Cannot verify on non-Unix; assume alive to stay safe.
+  true
+}
+
 /// Represents an ongoing operation tracked in IPC
 #[derive(Clone, Debug, Serialize, Deserialize)]
 pub struct OngoingOperation {
@@ -279,7 +298,7 @@ impl IpcCoordinator {
     let mut operations = self.load_operations()?;
     log::debug!("Loaded {} existing operations", operations.len());
 
-    // Clean up stale operations (older than 10 minutes)
+    // Clean up stale operations (older than 10 minutes or with dead PIDs)
     let now = SystemTime::now()
       .duration_since(SystemTime::UNIX_EPOCH)
       .unwrap_or(Duration::ZERO)
@@ -289,18 +308,21 @@ impl IpcCoordinator {
       .iter()
       .filter(|op| {
         op.status == OperationStatus::Running
-          && now.saturating_sub(op.started_at) > 600
+          && (now.saturating_sub(op.started_at) > 600
+            || !is_process_alive(op.pid))
       })
       .count();
 
     operations.retain(|op| {
-      if op.status == OperationStatus::Running
-        && now.saturating_sub(op.started_at) > 600
-      {
-        false // Remove stale operations
-      } else {
-        true
+      if op.status == OperationStatus::Running {
+        if !is_process_alive(op.pid) {
+          return false; // Remove dead operations
+        }
+        if now.saturating_sub(op.started_at) > 600 {
+          return false; // Remove old operations
+        }
       }
+      true
     });
 
     if stale_count > 0 {
@@ -311,7 +333,9 @@ impl IpcCoordinator {
     let conflicting: Vec<_> = operations
       .iter()
       .filter(|op| {
-        op.r#type == operation_type && op.status == OperationStatus::Running
+        op.r#type == operation_type
+          && op.status == OperationStatus::Running
+          && is_process_alive(op.pid)
       })
       .collect();
 
@@ -403,7 +427,9 @@ impl IpcCoordinator {
       let conflicts: Vec<_> = operations
         .iter()
         .filter(|op| {
-          op.r#type == operation_type && op.status == OperationStatus::Running
+          op.r#type == operation_type
+            && op.status == OperationStatus::Running
+            && is_process_alive(op.pid)
         })
         .collect();
 
@@ -426,7 +452,9 @@ impl IpcCoordinator {
   pub fn has_running_operation(&self, operation_type: OperationType) -> bool {
     let operations = self.load_operations().unwrap_or_default();
     operations.iter().any(|op| {
-      op.r#type == operation_type && op.status == OperationStatus::Running
+      op.r#type == operation_type
+        && op.status == OperationStatus::Running
+        && is_process_alive(op.pid)
     })
   }
 
@@ -439,7 +467,9 @@ impl IpcCoordinator {
     operations
       .into_iter()
       .filter(|op| {
-        op.r#type == operation_type && op.status == OperationStatus::Running
+        op.r#type == operation_type
+          && op.status == OperationStatus::Running
+          && is_process_alive(op.pid)
       })
       .collect()
   }