From: Jason Dillaman <dillaman@redhat.com>
Date: Fri, 2 Mar 2018 14:41:06 +0000 (-0500)
Subject: librbd: avoid potential for lock inversion in journal IO replay
X-Git-Tag: v13.0.2~31^2~2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=59802c5255c17ff5715a53c7f57a41ea7413db4c;p=ceph.git

librbd: avoid potential for lock inversion in journal IO replay

If an IO replay immediately fails due to a image shrink applied after
the IO, the lock order will be incorrect.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
---

diff --git a/src/librbd/journal/Replay.cc b/src/librbd/journal/Replay.cc
index da91f28bc0d..2f33ab5e915 100644
--- a/src/librbd/journal/Replay.cc
+++ b/src/librbd/journal/Replay.cc
@@ -884,7 +884,7 @@ void Replay<I>::handle_aio_modify_complete(Context *on_ready, Context *on_safe,
 
   if (r < 0) {
     lderr(cct) << ": AIO modify op failed: " << cpp_strerror(r) << dendl;
-    on_safe->complete(r);
+    m_image_ctx.op_work_queue->queue(on_safe, r);
     return;
   }