From: Samuel Just <sam.just@inktank.com>
Date: Wed, 25 Sep 2013 05:47:36 +0000 (-0700)
Subject: ReplicatedPG: take obc read lock prior to recovering an object to replicas
X-Git-Tag: v0.71~68^2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=a1f6b14e7fcd680f74a33e92845e54cbde69a5a3;p=ceph.git

ReplicatedPG: take obc read lock prior to recovering an object to replicas

Signed-off-by: Samuel Just <sam.just@inktank.com>
---

diff --git a/src/osd/ReplicatedPG.cc b/src/osd/ReplicatedPG.cc
index fe22b22e2cbb..191e7deb77a1 100644
--- a/src/osd/ReplicatedPG.cc
+++ b/src/osd/ReplicatedPG.cc
@@ -7821,11 +7821,19 @@ int ReplicatedPG::prep_object_replica_pushes(
   start_recovery_op(soid);
   assert(!recovering.count(soid));
   recovering.insert(soid);
+
+  /* We need this in case there is an in progress write on the object.  In fact,
+   * the only possible write is an update to the xattr due to a lost_revert --
+   * a client write would be blocked since the object is degraded.
+   * In almost all cases, therefore, this lock should be uncontended.
+   */
+  obc->ondisk_read_lock();
   pgbackend->recover_object(
     soid,
     ObjectContextRef(),
     obc, // has snapset context
     h);
+  obc->ondisk_read_unlock();
   return 1;
 }