osdc/Objecter: possible race condition with connection reset

If the connection quickly fails before the private session reference
can be associated with the connection, the connection will remain
closed and any OSD ops against the session will remain stuck.

Fixes: http://tracker.ceph.com/issues/36183
Signed-off-by: Jason Dillaman <dillaman@redhat.com>
(cherry picked from commit 74ca33cb49d2c258324447b1ca366ed4e604202a)

diff --git a/src/osdc/Objecter.cc b/src/osdc/Objecter.cc
index 98bd84eb6cf6b1c9261e768f5a5939cd3df0c489..3ad0bc516e9f1f0929b150db886e74f7e0bec891 100644 (file)
--- a/src/osdc/Objecter.cc
+++ b/src/osdc/Objecter.cc
@@ -1834,6 +1834,7 @@ void Objecter::get_session(Objecter::OSDSession *s)
 
 void Objecter::_reopen_session(OSDSession *s)
 {
+  // rwlock is locked unique
   // s->lock is locked
 
   entity_inst_t inst = osdmap->get_inst(s->osd);
@@ -4417,12 +4418,13 @@ bool Objecter::ms_handle_reset(Connection *con)
   if (!initialized)
     return false;
   if (con->get_peer_type() == CEPH_ENTITY_TYPE_OSD) {
+    unique_lock wl(rwlock);
+
     auto priv = con->get_priv();
     auto session = static_cast<OSDSession*>(priv.get());
     if (session) {
       ldout(cct, 1) << "ms_handle_reset " << con << " session " << session
                    << " osd." << session->osd << dendl;
-      unique_lock wl(rwlock);
       if (!initialized) {
        wl.unlock();
        return false;