From: Nitzan Mordechai <nmordech@redhat.com>
Date: Wed, 26 Nov 2025 14:36:42 +0000 (+0000)
Subject: aio_cxx: Fix mutual deadlock in PoolEIOflag test
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=9da2d94dd46adaae8926be3aeff33f3d67b17741;p=ceph.git

aio_cxx: Fix mutual deadlock in PoolEIOflag test

The LibRadosAio.PoolEIOFlag test was unstable, it was hanging due to:

 - Deadlock: The main thread held the shared mutex ('my_lock') while
   calling thread join. This created a mutual deadlock.
    * Fix: Mutex is unlocked before thread join using RAII scopes.
 - also convert to std::jthread and drop the join

Fixes: https://tracker.ceph.com/issues/73946
Signed-off-by: Nitzan Mordechai <nmordec@ibm.com>
---

diff --git a/src/test/librados/aio_cxx.cc b/src/test/librados/aio_cxx.cc
index ab26d59b2ebd..fa4e686603d8 100644
--- a/src/test/librados/aio_cxx.cc
+++ b/src/test/librados/aio_cxx.cc
@@ -34,7 +34,7 @@ class AioTestDataPP
 {
 public:
   AioTestDataPP()
-    : m_init(false),    
+    : m_init(false),
       m_oid("foo")
   {
   }
@@ -2422,25 +2422,22 @@ TEST(LibRadosAio, PoolEIOFlag) {
 
   bufferlist bl;
   bl.append("some data");
-  std::thread *t = nullptr;
+  std::unique_ptr<std::thread> t;
   std::atomic<bool> missed_eio{false};
   
   unsigned max = 1000;
   unsigned timeout = max * 10;
   unsigned long i = 1;
-  my_lock.lock();
   for (; min_failed == 0 && i <= timeout; ++i) {
     io_info *info = new io_info;
     info->i = i;
     info->c = Rados::aio_create_completion();
     info->c->set_complete_callback((void*)info, pool_io_callback);
-    inflight.insert(i);
-    my_lock.unlock();
     int r = test_data.m_ioctx.aio_write(test_data.m_oid, info->c, bl, bl.length(), 0);
-    //cout << "start " << i << " r = " << r << std::endl;
 
-    if (i == max / 2) {
-      t = new std::thread([&] {
+    // Trigger EIO after 100 ops have been submitted
+    if (i == 100) {
+      t = std::make_unique<std::thread>([&] {
         cout << "sending pool EIO time: " << ceph_clock_now() << std::endl;
         ASSERT_EQ(0, test_data.m_cluster.mon_command(
           fmt::format(R"({{
@@ -2465,8 +2462,9 @@ TEST(LibRadosAio, PoolEIOFlag) {
       break;
     }
   }
-  t->join();
-  delete t;
+  if (t && t->joinable()) {
+    t->join();
+  }
 
   // wait for ios to finish
   for (; !inflight.empty(); ++i) {
@@ -2481,8 +2479,8 @@ TEST(LibRadosAio, PoolEIOFlag) {
     GTEST_SKIP() << "eio flag missed all ios that already completed";
   }
   cout << "max_success " << max_success << ", min_failed " << min_failed << std::endl;
+  ASSERT_TRUE(min_failed > 0) << "Did not catch any EIO errors";
   ASSERT_TRUE(max_success + 1 == min_failed);
-  my_lock.unlock();
 }
 
 // This test case reproduces https://tracker.ceph.com/issues/57152