fix: Backport limit spin loop in TLSActor #29922 (#31975)

johanandren · web-flow · commit 5361846b1990 · 2023-06-20T23:13:54.000+02:00
* fix: Try to limit spin loop in TLSActor #29922 Backport of fix from 2.8 release line. * Align header to make header check plugin happy
diff --git a/akka-stream/src/main/scala/akka/stream/impl/io/TLSActor.scala b/akka-stream/src/main/scala/akka/stream/impl/io/TLSActor.scala
@@ -61,6 +61,7 @@ import akka.util.ByteString
 
   import TLSActor._
 
+  private var unwrapPutBackCounter: Int = 0
   protected val outputBunch = new OutputBunch(outputCount = 2, self, this)
   outputBunch.markAllOutputs()
 
@@ -354,6 +355,7 @@ import akka.util.ByteString
 
   def flushToUser(): Unit = {
     if (tracing) log.debug("flushToUser")
+    if (unwrapPutBackCounter > 0) unwrapPutBackCounter = 0
     userOutBuffer.flip()
     if (userOutBuffer.hasRemaining) {
       val bs = ByteString(userOutBuffer)
@@ -408,7 +410,17 @@ import akka.util.ByteString
       case OK =>
         result.getHandshakeStatus match {
           case NEED_WRAP =>
-            flushToUser()
+            // https://github.com/akka/akka/issues/29922
+            // A second workaround for an infinite loop we have not been able to reproduce/isolate,
+            // if you see this, and can reproduce consistently, please report back to the Akka team
+            // with a reproducer or details about the client causing it
+            unwrapPutBackCounter += 1
+            if (unwrapPutBackCounter > 1000) {
+              throw new IllegalStateException(
+                s"Stuck in unwrap loop, bailing out, last handshake status [$lastHandshakeStatus], " +
+                s"remaining=${transportInBuffer.remaining}, out=${userOutBuffer.position()}, " +
+                "(https://github.com/akka/akka/issues/29922)")
+            }
             transportInChoppingBlock.putBack(transportInBuffer)
           case FINISHED =>
             flushToUser()
@@ -480,7 +492,6 @@ import akka.util.ByteString
     pump()
   }
 
-  // FIXME: what happens if this actor dies unexpectedly?
   override def postStop(): Unit = {
     if (tracing) log.debug("postStop")
     super.postStop()
