From be9d26f57443a52649c5c3d8f9034d9117ff36c5 Mon Sep 17 00:00:00 2001
From: Jason Dillaman <dillaman@redhat.com>
Date: Tue, 14 Jul 2015 09:06:59 -0400
Subject: [PATCH] librbd: prevent concurrent image refreshes

With multiple concurrent AIO requests, it's possible for multiple
image refreshes to be scheduled.

Signed-off-by: Jason Dillaman <dillaman@redhat.com>
---
 src/librbd/ImageCtx.cc |  2 +-
 src/librbd/ImageCtx.h  |  3 +++
 src/librbd/internal.cc | 32 ++++++++++++++++++++------------
 3 files changed, 24 insertions(+), 13 deletions(-)

diff --git a/src/librbd/ImageCtx.cc b/src/librbd/ImageCtx.cc
index f371658d75722..d93e149b24a82 100644
--- a/src/librbd/ImageCtx.cc
+++ b/src/librbd/ImageCtx.cc
@@ -88,7 +88,7 @@ public:
       readahead(),
       total_bytes_read(0), copyup_finisher(NULL),
       object_map(*this), aio_work_queue(NULL), op_work_queue(NULL),
-      asok_hook(new LibrbdAdminSocketHook(this))
+      refresh_in_progress(false), asok_hook(new LibrbdAdminSocketHook(this))
   {
     md_ctx.dup(p);
     data_ctx.dup(p);
diff --git a/src/librbd/ImageCtx.h b/src/librbd/ImageCtx.h
index 4fd1112e112d9..9e21e45493c32 100644
--- a/src/librbd/ImageCtx.h
+++ b/src/librbd/ImageCtx.h
@@ -136,6 +136,9 @@ namespace librbd {
     AioImageRequestWQ *aio_work_queue;
     ContextWQ *op_work_queue;
 
+    Cond refresh_cond;
+    bool refresh_in_progress;
+
     // Configuration
     static const string METADATA_CONF_PREFIX;
     bool non_blocking_aio;
diff --git a/src/librbd/internal.cc b/src/librbd/internal.cc
index a0c76070737da..763cee170dad0 100644
--- a/src/librbd/internal.cc
+++ b/src/librbd/internal.cc
@@ -2463,17 +2463,34 @@ reprotect_and_return_err:
     CephContext *cct = ictx->cct;
     ldout(cct, 20) << "ictx_check " << ictx << dendl;
 
-    ictx->refresh_lock.Lock();
-    bool needs_refresh = ictx->last_refresh != ictx->refresh_seq;
-    ictx->refresh_lock.Unlock();
+    bool needs_refresh = false;
+    int refresh_seq;
+    {
+      Mutex::Locker refresh_locker(ictx->refresh_lock);
+      while (ictx->refresh_in_progress) {
+        ictx->refresh_cond.Wait(ictx->refresh_lock);
+      }
+
+      if (ictx->last_refresh != ictx->refresh_seq) {
+        ictx->refresh_in_progress = true;
+        needs_refresh = true;
+        refresh_seq = ictx->refresh_seq;
+      }
+    }
 
     if (needs_refresh) {
       int r = ictx_refresh(ictx);
+
+      Mutex::Locker refresh_locker(ictx->refresh_lock);
+      ictx->refresh_in_progress = false;
+      ictx->refresh_cond.Signal();
+
       if (r < 0) {
 	lderr(cct) << "Error re-reading rbd header: " << cpp_strerror(-r)
 		   << dendl;
         return r;
       }
+      ictx->last_refresh = refresh_seq;
     }
     return 0;
   }
@@ -2524,10 +2541,6 @@ reprotect_and_return_err:
 
     ldout(cct, 20) << "ictx_refresh " << ictx << dendl;
 
-    ictx->refresh_lock.Lock();
-    int refresh_seq = ictx->refresh_seq;
-    ictx->refresh_lock.Unlock();
-
     ::SnapContext new_snapc;
     bool new_snap = false;
     vector<string> snap_names;
@@ -2701,11 +2714,6 @@ reprotect_and_return_err:
     if (new_snap) {
       _flush(ictx);
     }
-
-    ictx->refresh_lock.Lock();
-    ictx->last_refresh = refresh_seq;
-    ictx->refresh_lock.Unlock();
-
     return 0;
   }
 
-- 
2.39.5