diff --git a/Sources/NIO/BlockingIOThreadPool.swift b/Sources/NIO/BlockingIOThreadPool.swift
index 2d0ec74b2b..9335186961 100644
--- a/Sources/NIO/BlockingIOThreadPool.swift
+++ b/Sources/NIO/BlockingIOThreadPool.swift
@@ -68,7 +68,9 @@ public final class BlockingIOThreadPool {
         self.lock.withLock {
             switch self.state {
             case .running(let items):
-                items.forEach { $0(.cancelled) }
+                queue.async {
+                    items.forEach { $0(.cancelled) }
+                }
                 self.state = .shuttingDown(Array(repeating: true, count: numberOfThreads))
                 (0..<numberOfThreads).forEach { _ in
                     self.semaphore.signal()
@@ -162,7 +164,7 @@ public final class BlockingIOThreadPool {
         self.queues.enumerated().forEach { idAndQueue in
             let id = idAndQueue.0
             let q = idAndQueue.1
-            q.async { [unowned self] in
+            q.async {
                 self.process(identifier: id)
             }
         }
diff --git a/Tests/NIOTests/BlockingIOThreadPoolTest+XCTest.swift b/Tests/NIOTests/BlockingIOThreadPoolTest+XCTest.swift
index c0a3538d51..eee7e22303 100644
--- a/Tests/NIOTests/BlockingIOThreadPoolTest+XCTest.swift
+++ b/Tests/NIOTests/BlockingIOThreadPoolTest+XCTest.swift
@@ -29,6 +29,9 @@ extension BlockingIOThreadPoolTest {
                 ("testDoubleShutdownWorks", testDoubleShutdownWorks),
                 ("testStateCancelled", testStateCancelled),
                 ("testStateActive", testStateActive),
+                ("testLoseLastReferenceAndShutdownWhileTaskStillRunning", testLoseLastReferenceAndShutdownWhileTaskStillRunning),
+                ("testDeadLockIfCalledOutWithLockHeld", testDeadLockIfCalledOutWithLockHeld),
+                ("testPoolDoesGetReleasedWhenStoppedAndReferencedDropped", testPoolDoesGetReleasedWhenStoppedAndReferencedDropped),
            ]
    }
 }
diff --git a/Tests/NIOTests/BlockingIOThreadPoolTest.swift b/Tests/NIOTests/BlockingIOThreadPoolTest.swift
index a5fb0b8577..a7fcedf97f 100644
--- a/Tests/NIOTests/BlockingIOThreadPoolTest.swift
+++ b/Tests/NIOTests/BlockingIOThreadPoolTest.swift
@@ -15,6 +15,7 @@
 import XCTest
 import NIO
 import Dispatch
+import Foundation
 
 class BlockingIOThreadPoolTest: XCTestCase {
     func testDoubleShutdownWorks() throws {
@@ -48,4 +49,80 @@ class BlockingIOThreadPoolTest: XCTestCase {
         group.wait()
         try threadPool.syncShutdownGracefully()
     }
+
+    func testLoseLastReferenceAndShutdownWhileTaskStillRunning() throws {
+        let blockThreadSem = DispatchSemaphore(value: 0)
+        let allDoneSem = DispatchSemaphore(value: 0)
+
+        ({
+            let threadPool = BlockingIOThreadPool(numberOfThreads: 2)
+            threadPool.start()
+            threadPool.submit { _ in
+                Foundation.Thread.sleep(forTimeInterval: 0.1)
+            }
+            threadPool.submit { _ in
+                blockThreadSem.wait()
+            }
+            threadPool.shutdownGracefully { error in
+                XCTAssertNil(error)
+                allDoneSem.signal()
+            }
+        })()
+        blockThreadSem.signal()
+        allDoneSem.wait()
+    }
+
+    func testDeadLockIfCalledOutWithLockHeld() throws {
+        let blockRunningSem = DispatchSemaphore(value: 0)
+        let blockOneThreadSem = DispatchSemaphore(value: 0)
+        let threadPool = BlockingIOThreadPool(numberOfThreads: 1)
+        let allDone = DispatchSemaphore(value: 0)
+        threadPool.start()
+        // enqueue one that'll block the whole pool (1 thread only)
+        threadPool.submit { state in
+            XCTAssertEqual(state, .active)
+            blockRunningSem.signal()
+            blockOneThreadSem.wait()
+        }
+        blockRunningSem.wait()
+        // enqueue one that will be cancelled and then calls shutdown again which needs the lock
+        threadPool.submit { state in
+            XCTAssertEqual(state, .cancelled)
+            threadPool.shutdownGracefully { error in
+                XCTAssertNil(error)
+            }
+        }
+        threadPool.shutdownGracefully { error in
+            XCTAssertNil(error)
+            allDone.signal()
+        }
+        blockOneThreadSem.signal() // that'll unblock the thread in the pool
+        allDone.wait()
+    }
+
+    func testPoolDoesGetReleasedWhenStoppedAndReferencedDropped() throws {
+        let taskRunningSem = DispatchSemaphore(value: 0)
+        let doneSem = DispatchSemaphore(value: 0)
+        let shutdownDoneSem = DispatchSemaphore(value: 0)
+        weak var weakThreadPool: BlockingIOThreadPool? = nil
+        ({
+            let threadPool = BlockingIOThreadPool(numberOfThreads: 1)
+            weakThreadPool = threadPool
+            threadPool.start()
+            threadPool.submit { state in
+                XCTAssertEqual(.active, state)
+                taskRunningSem.signal()
+                doneSem.wait()
+            }
+            taskRunningSem.wait()
+            threadPool.shutdownGracefully { error in
+                XCTAssertNil(error)
+                shutdownDoneSem.signal()
+            }
+        })()
+        XCTAssertNotNil(weakThreadPool)
+        doneSem.signal()
+        shutdownDoneSem.wait()
+        assert(weakThreadPool == nil, within: .seconds(1))
+    }
 }