osd: verify object version during push

Fail to push if the ondisk version doesn't match the version we want to
send.

This isn't supposed to happen. If it does it means we have a bug somewhere
else.  Log something to the error log and don't push.  This is better than
the current behavior, which goes into a loop (repeatedly pulling the object
and retrying when it's not the right version).

Signed-off-by: Sage Weil <sage.weil@dreamhost.com>

diff --git a/src/osd/ReplicatedPG.cc b/src/osd/ReplicatedPG.cc
index cbe29e2257da8692179850980217a11ec003ebe1..9d142640ef91787c27d1e329f742f4b3e4584148 100644 (file)
--- a/src/osd/ReplicatedPG.cc
+++ b/src/osd/ReplicatedPG.cc
@@ -3386,7 +3386,7 @@ void ReplicatedPG::push_start(const sobject_t& soid, int peer,
 
   dout(10) << "push_start " << soid << " size " << size << " data " << data_subset
           << " cloning " << clone_subsets << dendl;    
-  send_push_op(soid, peer, size, true, complete, pi->data_subset_pushing, pi->clone_subsets);
+  send_push_op(soid, version, peer, size, true, complete, pi->data_subset_pushing, pi->clone_subsets);
 }
 
 
@@ -3394,7 +3394,7 @@ void ReplicatedPG::push_start(const sobject_t& soid, int peer,
  * push - send object to a peer
  */
 
-void ReplicatedPG::send_push_op(const sobject_t& soid, int peer, 
+void ReplicatedPG::send_push_op(const sobject_t& soid, eversion_t version, int peer, 
                                uint64_t size, bool first, bool complete,
                                interval_set<uint64_t> &data_subset,
                                map<sobject_t, interval_set<uint64_t> >& clone_subsets)
@@ -3423,6 +3423,12 @@ void ReplicatedPG::send_push_op(const sobject_t& soid, int peer,
   bv.push_back(attrset[OI_ATTR]);
   object_info_t oi(bv);
 
+  if (oi.version != version) {
+    osd->clog.error() << "push " << soid << " v " << version << " to osd" << peer
+                     << " failed because local copy is " << oi.version << "\n";
+    return;
+  }
+
   // ok
   dout(7) << "send_push_op " << soid << " v " << oi.version 
          << " size " << size
@@ -3482,7 +3488,8 @@ void ReplicatedPG::sub_op_push_reply(MOSDSubOpReply *reply)
       pi->data_subset_pushing.span_of(pi->data_subset, from, g_conf.osd_recovery_max_chunk);
       dout(10) << " pushing more, " << pi->data_subset_pushing << " of " << pi->data_subset << dendl;
       complete = pi->data_subset.range_end() == pi->data_subset_pushing.range_end();
-      send_push_op(soid, peer, pi->size, false, complete, pi->data_subset_pushing, pi->clone_subsets);
+      send_push_op(soid, pi->version, peer, pi->size, false, complete,
+                  pi->data_subset_pushing, pi->clone_subsets);
     } else {
       // done!
       peer_missing[peer].got(soid, pi->version);
@@ -3543,7 +3550,7 @@ void ReplicatedPG::sub_op_pull(MOSDSubOp *op)
     // complete==false means nothing.  we don't know because the primary may
     // not be pulling the entire object.
 
-    send_push_op(soid, op->get_source().num(), size, op->first, complete, op->data_subset, op->clone_subsets);
+    send_push_op(soid, op->version, op->get_source().num(), size, op->first, complete, op->data_subset, op->clone_subsets);
   }
   op->put();
 }

diff --git a/src/osd/ReplicatedPG.h b/src/osd/ReplicatedPG.h
index b89c5aa0c080f1bfd9d74cc3e1e2b9093d793ba6..5079f0b8ed2fb958da50f32e7650a439c3f7ad63 100644 (file)
--- a/src/osd/ReplicatedPG.h
+++ b/src/osd/ReplicatedPG.h
@@ -525,7 +525,7 @@ protected:
                  uint64_t size, eversion_t version,
                  interval_set<uint64_t> &data_subset,
                  map<sobject_t, interval_set<uint64_t> >& clone_subsets);
-  void send_push_op(const sobject_t& oid, int dest,
+  void send_push_op(const sobject_t& oid, eversion_t version, int dest,
                    uint64_t size, bool first, bool complete,
                    interval_set<uint64_t>& data_subset, 
                    map<sobject_t, interval_set<uint64_t> >& clone_subsets);