20 files changed, 545 insertions, 43 deletions

diff --git a/src/common/WorkQueue.cc b/src/common/WorkQueue.cc
index a40200a68bd..6b648a78021 100644
--- a/src/common/WorkQueue.cc
+++ b/src/common/WorkQueue.cc
@@ -49,7 +49,13 @@ ThreadPool::ThreadPool(CephContext *cct_, string nm, int n, const char *option)
   }
 }
 
-void ThreadPool::TPHandle::reset_tp_timeout() {
+void ThreadPool::TPHandle::suspend_tp_timeout()
+{
+  cct->get_heartbeat_map()->clear_timeout(hb);
+}
+
+void ThreadPool::TPHandle::reset_tp_timeout()
+{
   cct->get_heartbeat_map()->reset_timeout(
     hb, grace, suicide_grace);
 }
diff --git a/src/common/WorkQueue.h b/src/common/WorkQueue.h
index d936d77abef..b2742accdce 100644
--- a/src/common/WorkQueue.h
+++ b/src/common/WorkQueue.h
@@ -49,6 +49,7 @@ public:
       : cct(cct), hb(hb), grace(grace), suicide_grace(suicide_grace) {}
   public:
     void reset_tp_timeout();
+    void suspend_tp_timeout();
   };
 private:
 
diff --git a/src/common/config_opts.h b/src/common/config_opts.h
index 3b9d025393f..1c7a917602a 100644
--- a/src/common/config_opts.h
+++ b/src/common/config_opts.h
@@ -80,7 +80,7 @@ SUBSYS(journal, 1, 3)
 SUBSYS(ms, 0, 5)
 SUBSYS(mon, 1, 5)
 SUBSYS(monc, 0, 10)
-SUBSYS(paxos, 0, 5)
+SUBSYS(paxos, 1, 5)
 SUBSYS(tp, 0, 5)
 SUBSYS(auth, 1, 5)
 SUBSYS(crypto, 1, 5)
diff --git a/src/mon/Monitor.cc b/src/mon/Monitor.cc
index 119ef740aa8..2c21e6eac69 100644
--- a/src/mon/Monitor.cc
+++ b/src/mon/Monitor.cc
@@ -619,7 +619,7 @@ void Monitor::bootstrap()
 {
   dout(10) << "bootstrap" << dendl;
 
-  sync_reset();
+  sync_reset_requester();
   unregister_cluster_logger();
   cancel_probe_timeout();
 
@@ -806,23 +806,27 @@ void Monitor::sync_obtain_latest_monmap(bufferlist &bl)
   latest_monmap.encode(bl, CEPH_FEATURES_ALL);
 }
 
-void Monitor::sync_reset()
+void Monitor::sync_reset_requester()
 {
+  dout(10) << __func__ << dendl;
+
   if (sync_timeout_event) {
     timer.cancel_event(sync_timeout_event);
     sync_timeout_event = NULL;
   }
 
-  // leader state
-  sync_providers.clear();
-
-  // requester state
   sync_provider = entity_inst_t();
   sync_cookie = 0;
   sync_full = false;
   sync_start_version = 0;
 }
 
+void Monitor::sync_reset_provider()
+{
+  dout(10) << __func__ << dendl;
+  sync_providers.clear();
+}
+
 void Monitor::sync_start(entity_inst_t &other, bool full)
 {
   dout(10) << __func__ << " " << other << (full ? " full" : " recent") << dendl;
@@ -832,7 +836,7 @@ void Monitor::sync_start(entity_inst_t &other, bool full)
   state = STATE_SYNCHRONIZING;
 
   // make sure are not a provider for anyone!
-  sync_reset();
+  sync_reset_provider();
 
   sync_full = full;
 
@@ -923,8 +927,6 @@ void Monitor::sync_finish(version_t last_committed)
   t.erase("mon_sync", "last_committed_floor");
   store->apply_transaction(t);
 
-  sync_reset();
-
   assert(g_conf->mon_sync_requester_kill_at != 9);
 
   init_paxos();
@@ -1173,7 +1175,6 @@ void Monitor::handle_sync_chunk(MMonSync *m)
 void Monitor::handle_sync_no_cookie(MMonSync *m)
 {
   dout(10) << __func__ << dendl;
-  sync_reset();
   bootstrap();
 }
 
@@ -1763,7 +1764,7 @@ void Monitor::get_health(string& status, bufferlist *detailbl, Formatter *f)
       }
 
       if (f) {
-        f->open_object_section(name.c_str());
+        f->open_object_section("mon");
         f->dump_string("name", name.c_str());
         f->dump_float("skew", skew);
         f->dump_float("latency", latency);
diff --git a/src/mon/Monitor.h b/src/mon/Monitor.h
index bed48ecee34..69dfefe144a 100644
--- a/src/mon/Monitor.h
+++ b/src/mon/Monitor.h
@@ -168,6 +168,7 @@ public:
     case STATE_ELECTING: return "electing";
     case STATE_LEADER: return "leader";
     case STATE_PEON: return "peon";
+    case STATE_SHUTDOWN: return "shutdown";
     default: return "???";
     }
   }
@@ -301,10 +302,14 @@ private:
   set<string> get_sync_targets_names();
 
   /**
-   * Reset the monitor's sync-related data structures and state, both
-   * for the requester- and provider-side.
+   * Reset the monitor's sync-related data structures for syncing *from* a peer
    */
-  void sync_reset();
+  void sync_reset_requester();
+
+  /**
+   * Reset sync state related to allowing others to sync from us
+   */
+  void sync_reset_provider();
 
   /**
    * Caled when a sync attempt times out (requester-side)
diff --git a/src/mon/PGMonitor.cc b/src/mon/PGMonitor.cc
index d86cbe70c19..93b0b0b3828 100644
--- a/src/mon/PGMonitor.cc
+++ b/src/mon/PGMonitor.cc
@@ -1332,12 +1332,16 @@ bool PGMonitor::preprocess_command(MMonCommand *m)
 
   // perhaps these would be better in the parsing, but it's weird
   if (prefix == "pg dump_json") {
+    vector<string> v;
+    v.push_back(string("all"));
     cmd_putval(g_ceph_context, cmdmap, "format", string("json"));
-    cmd_putval(g_ceph_context, cmdmap, "dumpcontents", string("all"));
+    cmd_putval(g_ceph_context, cmdmap, "dumpcontents", v);
     prefix = "pg dump";
   } else if (prefix == "pg dump_pools_json") {
+    vector<string> v;
+    v.push_back(string("pools"));
     cmd_putval(g_ceph_context, cmdmap, "format", string("json"));
-    cmd_putval(g_ceph_context, cmdmap, "dumpcontents", string("pool"));
+    cmd_putval(g_ceph_context, cmdmap, "dumpcontents", v);
     prefix = "pg dump";
   }
 
diff --git a/src/mon/Paxos.cc b/src/mon/Paxos.cc
index a543abed7ed..445413da13b 100644
--- a/src/mon/Paxos.cc
+++ b/src/mon/Paxos.cc
@@ -492,7 +492,7 @@ void Paxos::handle_last(MMonPaxos *last)
 
 void Paxos::collect_timeout()
 {
-  dout(5) << "collect timeout, calling fresh election" << dendl;
+  dout(1) << "collect timeout, calling fresh election" << dendl;
   collect_timeout_event = 0;
   assert(mon->is_leader());
   mon->bootstrap();
@@ -711,7 +711,7 @@ void Paxos::handle_accept(MMonPaxos *accept)
 
 void Paxos::accept_timeout()
 {
-  dout(5) << "accept timeout, calling fresh election" << dendl;
+  dout(1) << "accept timeout, calling fresh election" << dendl;
   accept_timeout_event = 0;
   assert(mon->is_leader());
   assert(is_updating() || is_updating_previous());
@@ -1004,7 +1004,7 @@ void Paxos::handle_lease_ack(MMonPaxos *ack)
 
 void Paxos::lease_ack_timeout()
 {
-  dout(5) << "lease_ack_timeout -- calling new election" << dendl;
+  dout(1) << "lease_ack_timeout -- calling new election" << dendl;
   assert(mon->is_leader());
   assert(is_active());
 
@@ -1023,7 +1023,7 @@ void Paxos::reset_lease_timeout()
 
 void Paxos::lease_timeout()
 {
-  dout(5) << "lease_timeout -- calling new election" << dendl;
+  dout(1) << "lease_timeout -- calling new election" << dendl;
   assert(mon->is_peon());
 
   lease_timeout_event = 0;
diff --git a/src/osd/OSD.cc b/src/osd/OSD.cc
index e3a7c227e15..89aa1db34eb 100644
--- a/src/osd/OSD.cc
+++ b/src/osd/OSD.cc
@@ -4957,13 +4957,22 @@ void OSD::handle_osd_map(MOSDMap *m)
   if (first > osdmap->get_epoch() + 1) {
     dout(10) << "handle_osd_map message skips epochs " << osdmap->get_epoch() + 1
 	     << ".." << (first-1) << dendl;
-    if ((m->oldest_map < first && osdmap->get_epoch() == 0) ||
-	m->oldest_map <= osdmap->get_epoch()) {
+    if (m->oldest_map <= osdmap->get_epoch() + 1) {
       monc->sub_want("osdmap", osdmap->get_epoch()+1, CEPH_SUBSCRIBE_ONETIME);
       monc->renew_subs();
       m->put();
       return;
     }
+    // always try to get the full range of maps--as many as we can.  this
+    //  1- is good to have
+    //  2- is at present the only way to ensure that we get a *full* map as
+    //     the first map!
+    if (m->oldest_map < first) {
+      monc->sub_want("osdmap", m->oldest_map - 1, CEPH_SUBSCRIBE_ONETIME);
+      monc->renew_subs();
+      m->put();
+      return;
+    }
     skip_maps = true;
   }
 
@@ -7041,9 +7050,9 @@ PGRef OSD::OpWQ::_dequeue()
   return pg;
 }
 
-void OSD::OpWQ::_process(PGRef pg)
+void OSD::OpWQ::_process(PGRef pg, ThreadPool::TPHandle &handle)
 {
-  pg->lock();
+  pg->lock_suspend_timeout(handle);
   OpRequestRef op;
   {
     Mutex::Locker l(qlock);
diff --git a/src/osd/OSD.h b/src/osd/OSD.h
index 5bcff7442d7..478f766d145 100644
--- a/src/osd/OSD.h
+++ b/src/osd/OSD.h
@@ -911,7 +911,7 @@ private:
     bool _empty() {
       return pqueue.empty();
     }
-    void _process(PGRef pg);
+    void _process(PGRef pg, ThreadPool::TPHandle &handle);
   } op_wq;
 
   void enqueue_op(PG *pg, OpRequestRef op);
diff --git a/src/osd/PG.cc b/src/osd/PG.cc
index 9f957b8e054..f731441e8a4 100644
--- a/src/osd/PG.cc
+++ b/src/osd/PG.cc
@@ -193,6 +193,13 @@ PG::~PG()
 #endif
 }
 
+void PG::lock_suspend_timeout(ThreadPool::TPHandle &handle)
+{
+  handle.suspend_tp_timeout();
+  lock();
+  handle.reset_tp_timeout();
+}
+
 void PG::lock(bool no_lockdep)
 {
   _lock.Lock(no_lockdep);
diff --git a/src/osd/PG.h b/src/osd/PG.h
index 10e9a2544a9..8f572c75e19 100644
--- a/src/osd/PG.h
+++ b/src/osd/PG.h
@@ -245,6 +245,8 @@ protected:
 public:
   bool deleting;  // true while in removing or OSD is shutting down
 
+
+  void lock_suspend_timeout(ThreadPool::TPHandle &handle);
   void lock(bool no_lockdep = false);
   void unlock() {
     //generic_dout(0) << this << " " << info.pgid << " unlock" << dendl;
diff --git a/src/osd/osd_types.cc b/src/osd/osd_types.cc
index 02c1ef7b69d..fbd5cbbe9a0 100644
--- a/src/osd/osd_types.cc
+++ b/src/osd/osd_types.cc
@@ -1715,7 +1715,7 @@ bool pg_interval_t::check_new_interval(
 
     if (!i.acting.empty() &&
 	i.acting.size() >=
-	osdmap->get_pools().find(pool_id)->second.min_size) {
+	lastmap->get_pools().find(pool_id)->second.min_size) {
       if (out)
 	*out << "generate_past_intervals " << i
 	     << ": not rw,"
@@ -1730,6 +1730,7 @@ bool pg_interval_t::check_new_interval(
 	  *out << "generate_past_intervals " << i
 	       << " : primary up " << lastmap->get_up_from(i.acting[0])
 	       << "-" << lastmap->get_up_thru(i.acting[0])
+	       << " includes interval"
 	       << std::endl;
       } else if (last_epoch_clean >= i.first &&
 		 last_epoch_clean <= i.last) {
@@ -1758,7 +1759,7 @@ bool pg_interval_t::check_new_interval(
     } else {
       i.maybe_went_rw = false;
       if (out)
-	*out << "generate_past_intervals " << i << " : empty" << std::endl;
+	*out << "generate_past_intervals " << i << " : acting set is too small" << std::endl;
     }
     return true;
   } else {
diff --git a/src/osdc/Objecter.cc b/src/osdc/Objecter.cc
index aefe74088e9..70d296a3ab3 100644
--- a/src/osdc/Objecter.cc
+++ b/src/osdc/Objecter.cc
@@ -2436,6 +2436,7 @@ int Objecter::recalc_command_target(CommandOp *c)
       c->session = s;
       s->command_ops.push_back(&c->session_item);
     } else {
+      c->session = NULL;
       num_homeless_ops++;
     }
     return RECALC_OP_TARGET_NEED_RESEND;
diff --git a/src/rgw/rgw_auth_s3.cc b/src/rgw/rgw_auth_s3.cc
index c93de7cd58a..f3f0c8322f0 100644
--- a/src/rgw/rgw_auth_s3.cc
+++ b/src/rgw/rgw_auth_s3.cc
@@ -6,6 +6,7 @@
 
 static const char *signed_subresources[] = {
   "acl",
+  "cors",
   "delete",
   "lifecycle",
   "location",
diff --git a/src/rgw/rgw_bucket.cc b/src/rgw/rgw_bucket.cc
index bf8da99d616..d32af5df601 100644
--- a/src/rgw/rgw_bucket.cc
+++ b/src/rgw/rgw_bucket.cc
@@ -202,7 +202,11 @@ int rgw_bucket_set_attrs(RGWRados *store, rgw_bucket& bucket,
   string oid;
   store->get_bucket_meta_oid(bucket, oid);
   rgw_obj obj(store->zone.domain_root, oid);
-  return store->meta_mgr->set_attrs(bucket_meta_handler, oid,
+
+  string key;
+  store->get_bucket_instance_entry(bucket, key); /* we want the bucket instance name without
+						    the oid prefix cruft */
+  return store->meta_mgr->set_attrs(bucket_instance_meta_handler, key,
                                     obj, attrs, rmattrs, objv_tracker);
 }
 
diff --git a/src/rgw/rgw_common.h b/src/rgw/rgw_common.h
index 543bdf21377..8e4126de271 100644
--- a/src/rgw/rgw_common.h
+++ b/src/rgw/rgw_common.h
@@ -542,6 +542,10 @@ struct rgw_bucket {
   std::string marker;
   std::string bucket_id;
 
+  std::string oid; /*
+                    * runtime in-memory only info. If not empty, points to the bucket instance object
+                    */
+
   rgw_bucket() { }
   rgw_bucket(const char *n) : name(n) {
     assert(*n == '.'); // only rgw private buckets should be initialized without pool
diff --git a/src/rgw/rgw_op.cc b/src/rgw/rgw_op.cc
index 7760a2f5c52..e672de154ab 100644
--- a/src/rgw/rgw_op.cc
+++ b/src/rgw/rgw_op.cc
@@ -241,8 +241,10 @@ static int get_policy_from_attr(CephContext *cct, RGWRados *store, void *ctx,
   }
 
   if (obj.object.empty()) {
+    rgw_obj instance_obj;
+    store->get_bucket_instance_obj(bucket_info.bucket, instance_obj);
     return get_bucket_policy_from_attr(cct, store, ctx, bucket_info, bucket_attrs,
-                                       policy, obj);
+                                       policy, instance_obj);
   }
   return get_obj_policy_from_attr(cct, store, ctx, bucket_info, bucket_attrs,
                                   policy, obj);
@@ -1898,9 +1900,8 @@ void RGWPutCORS::execute()
 
   RGWObjVersionTracker *ptracker = (s->object ? NULL : &s->bucket_info.objv_tracker);
 
-  string no_obj;
   cors_config->encode(bl);
-  obj.init(s->bucket, no_obj);
+  store->get_bucket_instance_obj(s->bucket, obj);
   store->set_atomic(s->obj_ctx, obj);
   ret = store->set_attr(s->obj_ctx, obj, RGW_ATTR_CORS, bl, ptracker);
 }
@@ -1917,13 +1918,12 @@ void RGWDeleteCORS::execute()
 {
   bufferlist bl;
   rgw_obj obj;
-  string no_obj;
   if (!s->bucket_cors) {
     dout(2) << "No CORS configuration set yet for this bucket" << dendl;
     ret = -ENOENT;
     return;
   }
-  obj.init(s->bucket, no_obj);
+  store->get_bucket_instance_obj(s->bucket, obj);
   store->set_atomic(s->obj_ctx, obj);
   map<string, bufferlist> orig_attrs, attrs, rmattrs;
   map<string, bufferlist>::iterator iter;
@@ -2516,10 +2516,10 @@ int RGWHandler::read_cors_config(void)
   bufferlist bl;
  
   dout(10) << "Going to read cors from attrs" << dendl;
-  string no_object;
-  rgw_obj no_obj(s->bucket, no_object);
-  if (no_obj.bucket.name.size()) {
-    ret = store->get_attr(s->obj_ctx, no_obj, RGW_ATTR_CORS, bl);
+  rgw_obj obj;
+  store->get_bucket_instance_obj(s->bucket, obj);
+  if (obj.bucket.name.size()) {
+    ret = store->get_attr(s->obj_ctx, obj, RGW_ATTR_CORS, bl);
     if (ret >= 0) {
       bufferlist::iterator iter = bl.begin();
       s->bucket_cors = new RGWCORSConfiguration();
diff --git a/src/rgw/rgw_rados.cc b/src/rgw/rgw_rados.cc
index 8af03b03a8f..aba5cdf0ee2 100644
--- a/src/rgw/rgw_rados.cc
+++ b/src/rgw/rgw_rados.cc
@@ -831,6 +831,11 @@ void RGWRados::finalize()
     RGWRESTConn *conn = iter->second;
     delete conn;
   }
+
+  for (iter = region_conn_map.begin(); iter != region_conn_map.end(); ++iter) {
+    RGWRESTConn *conn = iter->second;
+    delete conn;
+  }
 }
 
 /** 
@@ -896,6 +901,12 @@ int RGWRados::init_complete()
     }
     RGWRegion& region = iter->second;
     rest_master_conn = new RGWRESTConn(cct, this, region.endpoints);
+
+    for (iter = region_map.regions.begin(); iter != region_map.regions.end(); ++iter) {
+      RGWRegion& region = iter->second;
+
+      region_conn_map[region.name] = new RGWRESTConn(cct, this, region.endpoints);
+    }
   }
 
   map<string, RGWZone>::iterator ziter;
@@ -2535,7 +2546,17 @@ int RGWRados::copy_obj(void *ctx,
 
     RGWRESTConn *conn;
     if (source_zone.empty()) {
-      conn = rest_master_conn;
+      if (dest_bucket_info.region.empty()) {
+        /* source is in the master region */
+        conn = rest_master_conn;
+      } else {
+        map<string, RGWRESTConn *>::iterator iter = region_conn_map.find(src_bucket_info.region);
+        if (iter == zone_conn_map.end()) {
+          ldout(cct, 0) << "could not find region connection to region: " << source_zone << dendl;
+          return -ENOENT;
+        }
+        conn = iter->second;
+      }
     } else {
       map<string, RGWRESTConn *>::iterator iter = zone_conn_map.find(source_zone);
       if (iter == zone_conn_map.end()) {
@@ -2886,7 +2907,7 @@ int RGWRados::set_bucket_owner(rgw_bucket& bucket, ACLOwner& owner)
 {
   RGWBucketInfo info;
   map<string, bufferlist> attrs;
-  int r = get_bucket_instance_info(NULL, bucket, info, NULL, &attrs);
+  int r = get_bucket_info(NULL, bucket.name, info, NULL, &attrs);
   if (r < 0) {
     ldout(cct, 0) << "NOTICE: get_bucket_info on bucket=" << bucket.name << " returned err=" << r << dendl;
     return r;
@@ -2919,7 +2940,7 @@ int RGWRados::set_buckets_enabled(vector<rgw_bucket>& buckets, bool enabled)
 
     RGWBucketInfo info;
     map<string, bufferlist> attrs;
-    int r = get_bucket_instance_info(NULL, bucket, info, NULL, &attrs);
+    int r = get_bucket_info(NULL, bucket.name, info, NULL, &attrs);
     if (r < 0) {
       ldout(cct, 0) << "NOTICE: get_bucket_info on bucket=" << bucket.name << " returned err=" << r << ", skipping bucket" << dendl;
       ret = r;
@@ -4546,6 +4567,17 @@ void RGWRados::get_bucket_meta_oid(rgw_bucket& bucket, string& oid)
   oid = RGW_BUCKET_INSTANCE_MD_PREFIX + entry;
 }
 
+void RGWRados::get_bucket_instance_obj(rgw_bucket& bucket, rgw_obj& obj)
+{
+  if (!bucket.oid.empty()) {
+    obj.init(zone.domain_root, bucket.oid);
+  } else {
+    string oid;
+    get_bucket_meta_oid(bucket, oid);
+    obj.init(zone.domain_root, oid);
+  }
+}
+
 int RGWRados::get_bucket_instance_info(void *ctx, const string& meta_key, RGWBucketInfo& info,
                                        time_t *pmtime, map<string, bufferlist> *pattrs)
 {
@@ -4562,7 +4594,11 @@ int RGWRados::get_bucket_instance_info(void *ctx, rgw_bucket& bucket, RGWBucketI
                                        time_t *pmtime, map<string, bufferlist> *pattrs)
 {
   string oid;
-  get_bucket_meta_oid(bucket, oid);
+  if (!bucket.oid.empty()) {
+    get_bucket_meta_oid(bucket, oid);
+  } else {
+    oid = bucket.oid;
+  }
 
   return get_bucket_instance_from_oid(ctx, oid, info, pmtime, pattrs);
 }
@@ -4586,6 +4622,7 @@ int RGWRados::get_bucket_instance_from_oid(void *ctx, string& oid, RGWBucketInfo
     ldout(cct, 0) << "ERROR: could not decode buffer info, caught buffer::error" << dendl;
     return -EIO;
   }
+  info.bucket.oid = oid;
   return 0;
 }
 
@@ -4628,6 +4665,7 @@ int RGWRados::get_bucket_info(void *ctx, string& bucket_name, RGWBucketInfo& inf
 
   if (entry_point.has_bucket_info) {
     info = entry_point.old_bucket_info;
+    info.bucket.oid = bucket_name;
     info.ep_objv = ot.read_version;
     ldout(cct, 20) << "rgw_get_bucket_info: old bucket info, bucket=" << info.bucket << " owner " << info.owner << dendl;
     return 0;
diff --git a/src/rgw/rgw_rados.h b/src/rgw/rgw_rados.h
index bcc40900299..d01f76ec224 100644
--- a/src/rgw/rgw_rados.h
+++ b/src/rgw/rgw_rados.h
@@ -888,6 +888,7 @@ public:
   RGWRegionMap region_map;
   RGWRESTConn *rest_master_conn;
   map<string, RGWRESTConn *> zone_conn_map;
+  map<string, RGWRESTConn *> region_conn_map;
 
   RGWMetadataManager *meta_mgr;
 
@@ -1285,6 +1286,7 @@ public:
   int decode_policy(bufferlist& bl, ACLOwner *owner);
   int get_bucket_stats(rgw_bucket& bucket, uint64_t *bucket_ver, uint64_t *master_ver, map<RGWObjCategory, RGWBucketStats>& stats,
                        string *max_marker);
+  void get_bucket_instance_obj(rgw_bucket& bucket, rgw_obj& obj);
   void get_bucket_instance_entry(rgw_bucket& bucket, string& entry);
   void get_bucket_meta_oid(rgw_bucket& bucket, string& oid);
 
diff --git a/src/test/test_osd_types.cc b/src/test/test_osd_types.cc
index 7a43b7b892a..fa4ae6163ac 100644
--- a/src/test/test_osd_types.cc
+++ b/src/test/test_osd_types.cc
@@ -17,6 +17,7 @@
 
 #include "include/types.h"
 #include "osd/osd_types.h"
+#include "osd/OSDMap.h"
 #include "gtest/gtest.h"
 #include "common/Thread.h"
 
@@ -117,6 +118,421 @@ TEST(hobject, prefixes5)
   ASSERT_EQ(prefixes_out, prefixes_correct);
 }
 
+TEST(pg_interval_t, check_new_interval)
+{
+  //
+  // Create a situation where osdmaps are the same so that
+  // each test case can diverge from it using minimal code.
+  //
+  int osd_id = 1;
+  epoch_t epoch = 40;
+  std::tr1::shared_ptr<OSDMap> osdmap(new OSDMap());
+  osdmap->set_max_osd(10);
+  osdmap->set_state(osd_id, CEPH_OSD_EXISTS);
+  osdmap->set_epoch(epoch);
+  std::tr1::shared_ptr<OSDMap> lastmap(new OSDMap());
+  lastmap->set_max_osd(10);
+  lastmap->set_state(osd_id, CEPH_OSD_EXISTS);
+  lastmap->set_epoch(epoch);
+  epoch_t same_interval_since = epoch;
+  epoch_t last_epoch_clean = same_interval_since;
+  int64_t pool_id = 200;
+  int pg_num = 4;
+  __u8 min_size = 2;
+  {
+    OSDMap::Incremental inc(epoch + 1);
+    inc.new_pools[pool_id].min_size = min_size;
+    inc.new_pools[pool_id].set_pg_num(pg_num);
+    inc.new_up_thru[osd_id] = epoch + 1;
+    osdmap->apply_incremental(inc);
+    lastmap->apply_incremental(inc);
+  }
+  vector<int> new_acting;
+  new_acting.push_back(osd_id);
+  new_acting.push_back(osd_id + 1);
+  vector<int> old_acting = new_acting;
+  vector<int> new_up;
+  new_up.push_back(osd_id);
+  vector<int> old_up = new_up;
+  pg_t pgid;
+  pgid.set_pool(pool_id);
+
+  //
+  // Do nothing if there are no modifications in
+  // acting, up or pool size and that the pool is not
+  // being split
+  //
+  {
+    map<epoch_t, pg_interval_t> past_intervals;
+
+    ASSERT_TRUE(past_intervals.empty());
+    ASSERT_FALSE(pg_interval_t::check_new_interval(old_acting,
+						   new_acting,
+						   old_up,
+						   new_up,
+						   same_interval_since,
+						   last_epoch_clean,
+						   osdmap,
+						   lastmap,
+						   pool_id,
+						   pgid,
+						   &past_intervals));
+    ASSERT_TRUE(past_intervals.empty());
+  }
+
+  //
+  // pool did not exist in the old osdmap
+  //
+  {
+    std::tr1::shared_ptr<OSDMap> lastmap(new OSDMap());
+    lastmap->set_max_osd(10);
+    lastmap->set_state(osd_id, CEPH_OSD_EXISTS);
+    lastmap->set_epoch(epoch);
+
+    map<epoch_t, pg_interval_t> past_intervals;
+
+    ASSERT_TRUE(past_intervals.empty());
+    ASSERT_TRUE(pg_interval_t::check_new_interval(old_acting,
+						  new_acting,
+						  old_up,
+						  new_up,
+						  same_interval_since,
+						  last_epoch_clean,
+						  osdmap,
+						  lastmap,
+						  pool_id,
+						  pgid,
+						  &past_intervals));
+    ASSERT_EQ((unsigned int)1, past_intervals.size());
+    ASSERT_EQ(same_interval_since, past_intervals[same_interval_since].first);
+    ASSERT_EQ(osdmap->get_epoch() - 1, past_intervals[same_interval_since].last);
+    ASSERT_EQ(osd_id, past_intervals[same_interval_since].acting[0]);
+    ASSERT_EQ(osd_id, past_intervals[same_interval_since].up[0]);
+  }
+
+  //
+  // The acting set has changed
+  //
+  {
+    vector<int> new_acting;
+    int new_primary = osd_id + 1;
+    new_acting.push_back(new_primary);
+
+    map<epoch_t, pg_interval_t> past_intervals;
+
+    ASSERT_TRUE(past_intervals.empty());
+    ASSERT_TRUE(pg_interval_t::check_new_interval(old_acting,
+						  new_acting,
+						  old_up,
+						  new_up,
+						  same_interval_since,
+						  last_epoch_clean,
+						  osdmap,
+						  lastmap,
+						  pool_id,
+						  pgid,
+						  &past_intervals));
+    ASSERT_EQ((unsigned int)1, past_intervals.size());
+    ASSERT_EQ(same_interval_since, past_intervals[same_interval_since].first);
+    ASSERT_EQ(osdmap->get_epoch() - 1, past_intervals[same_interval_since].last);
+    ASSERT_EQ(osd_id, past_intervals[same_interval_since].acting[0]);
+    ASSERT_EQ(osd_id, past_intervals[same_interval_since].up[0]);
+  }
+
+  //
+  // The up set has changed
+  //
+  {
+    vector<int> new_up;
+    int new_primary = osd_id + 1;
+    new_up.push_back(new_primary);
+
+    map<epoch_t, pg_interval_t> past_intervals;
+
+    ASSERT_TRUE(past_intervals.empty());
+    ASSERT_TRUE(pg_interval_t::check_new_interval(old_acting,
+						  new_acting,
+						  old_up,
+						  new_up,
+						  same_interval_since,
+						  last_epoch_clean,
+						  osdmap,
+						  lastmap,
+						  pool_id,
+						  pgid,
+						  &past_intervals));
+    ASSERT_EQ((unsigned int)1, past_intervals.size());
+    ASSERT_EQ(same_interval_since, past_intervals[same_interval_since].first);
+    ASSERT_EQ(osdmap->get_epoch() - 1, past_intervals[same_interval_since].last);
+    ASSERT_EQ(osd_id, past_intervals[same_interval_since].acting[0]);
+    ASSERT_EQ(osd_id, past_intervals[same_interval_since].up[0]);
+  }
+
+  //
+  // PG is splitting
+  //
+  {
+    std::tr1::shared_ptr<OSDMap> osdmap(new OSDMap());
+    osdmap->set_max_osd(10);
+    osdmap->set_state(osd_id, CEPH_OSD_EXISTS);
+    osdmap->set_epoch(epoch);
+    int new_pg_num = pg_num ^ 2;
+    OSDMap::Incremental inc(epoch + 1);
+    inc.new_pools[pool_id].min_size = min_size;
+    inc.new_pools[pool_id].set_pg_num(new_pg_num);
+    osdmap->apply_incremental(inc);
+
+    map<epoch_t, pg_interval_t> past_intervals;
+
+    ASSERT_TRUE(past_intervals.empty());
+    ASSERT_TRUE(pg_interval_t::check_new_interval(old_acting,
+						  new_acting,
+						  old_up,
+						  new_up,
+						  same_interval_since,
+						  last_epoch_clean,
+						  osdmap,
+						  lastmap,
+						  pool_id,
+						  pgid,
+						  &past_intervals));
+    ASSERT_EQ((unsigned int)1, past_intervals.size());
+    ASSERT_EQ(same_interval_since, past_intervals[same_interval_since].first);
+    ASSERT_EQ(osdmap->get_epoch() - 1, past_intervals[same_interval_since].last);
+    ASSERT_EQ(osd_id, past_intervals[same_interval_since].acting[0]);
+    ASSERT_EQ(osd_id, past_intervals[same_interval_since].up[0]);
+  }
+
+  //
+  // PG size has changed
+  //
+  {
+    std::tr1::shared_ptr<OSDMap> osdmap(new OSDMap());
+    osdmap->set_max_osd(10);
+    osdmap->set_state(osd_id, CEPH_OSD_EXISTS);
+    osdmap->set_epoch(epoch);
+    OSDMap::Incremental inc(epoch + 1);
+    __u8 new_min_size = min_size + 1;
+    inc.new_pools[pool_id].min_size = new_min_size;
+    inc.new_pools[pool_id].set_pg_num(pg_num);
+    osdmap->apply_incremental(inc);
+
+    map<epoch_t, pg_interval_t> past_intervals;
+
+    ASSERT_TRUE(past_intervals.empty());
+    ASSERT_TRUE(pg_interval_t::check_new_interval(old_acting,
+						  new_acting,
+						  old_up,
+						  new_up,
+						  same_interval_since,
+						  last_epoch_clean,
+						  osdmap,
+						  lastmap,
+						  pool_id,
+						  pgid,
+						  &past_intervals));
+    ASSERT_EQ((unsigned int)1, past_intervals.size());
+    ASSERT_EQ(same_interval_since, past_intervals[same_interval_since].first);
+    ASSERT_EQ(osdmap->get_epoch() - 1, past_intervals[same_interval_since].last);
+    ASSERT_EQ(osd_id, past_intervals[same_interval_since].acting[0]);
+    ASSERT_EQ(osd_id, past_intervals[same_interval_since].up[0]);
+  }
+  
+  //
+  // The old acting set was empty : the previous interval could not
+  // have been rw
+  //
+  {
+    vector<int> old_acting;
+
+    map<epoch_t, pg_interval_t> past_intervals;
+
+    ostringstream out;
+
+    ASSERT_TRUE(past_intervals.empty());
+    ASSERT_TRUE(pg_interval_t::check_new_interval(old_acting,
+						  new_acting,
+						  old_up,
+						  new_up,
+						  same_interval_since,
+						  last_epoch_clean,
+						  osdmap,
+						  lastmap,
+						  pool_id,
+						  pgid,
+						  &past_intervals,
+						  &out));
+    ASSERT_EQ((unsigned int)1, past_intervals.size());
+    ASSERT_FALSE(past_intervals[same_interval_since].maybe_went_rw);
+    ASSERT_NE(string::npos, out.str().find("acting set is too small"));
+  }
+
+  //
+  // The old acting set did not have enough osd : it could 
+  // not have been rw
+  //
+  {
+    vector<int> old_acting;
+    old_acting.push_back(osd_id); 
+
+    //
+    // see http://tracker.ceph.com/issues/5780
+    // the size of the old acting set should be compared 
+    // with the min_size of the old osdmap
+    //
+    // The new osdmap is created so that it triggers the
+    // bug.
+    //
+    std::tr1::shared_ptr<OSDMap> osdmap(new OSDMap());
+    osdmap->set_max_osd(10);
+    osdmap->set_state(osd_id, CEPH_OSD_EXISTS);
+    osdmap->set_epoch(epoch);
+    OSDMap::Incremental inc(epoch + 1);
+    __u8 new_min_size = old_acting.size();
+    inc.new_pools[pool_id].min_size = new_min_size;
+    inc.new_pools[pool_id].set_pg_num(pg_num);
+    osdmap->apply_incremental(inc);
+
+    ostringstream out;
+
+    map<epoch_t, pg_interval_t> past_intervals;
+
+    ASSERT_TRUE(past_intervals.empty());
+    ASSERT_TRUE(pg_interval_t::check_new_interval(old_acting,
+						  new_acting,
+						  old_up,
+						  new_up,
+						  same_interval_since,
+						  last_epoch_clean,
+						  osdmap,
+						  lastmap,
+						  pool_id,
+						  pgid,
+						  &past_intervals,
+						  &out));
+    ASSERT_EQ((unsigned int)1, past_intervals.size());
+    ASSERT_FALSE(past_intervals[same_interval_since].maybe_went_rw);
+    ASSERT_NE(string::npos, out.str().find("acting set is too small"));
+  }
+
+  //
+  // The acting set changes. The old acting set primary was up during the
+  // previous interval and may have been rw.
+  //
+  {
+    vector<int> new_acting;
+    new_acting.push_back(osd_id + 4); 
+    new_acting.push_back(osd_id + 5); 
+    
+    ostringstream out;
+
+    map<epoch_t, pg_interval_t> past_intervals;
+
+    ASSERT_TRUE(past_intervals.empty());
+    ASSERT_TRUE(pg_interval_t::check_new_interval(old_acting,
+						  new_acting,
+						  old_up,
+						  new_up,
+						  same_interval_since,
+						  last_epoch_clean,
+						  osdmap,
+						  lastmap,
+						  pool_id,
+						  pgid,
+						  &past_intervals,
+						  &out));
+    ASSERT_EQ((unsigned int)1, past_intervals.size());
+    ASSERT_TRUE(past_intervals[same_interval_since].maybe_went_rw);
+    ASSERT_NE(string::npos, out.str().find("includes interval"));
+  }
+  //
+  // The acting set changes. The old acting set primary was not up 
+  // during the old interval but last_epoch_clean is in the
+  // old interval and it may have been rw.
+  //
+  {
+    vector<int> new_acting;
+    new_acting.push_back(osd_id + 4); 
+    new_acting.push_back(osd_id + 5); 
+
+    std::tr1::shared_ptr<OSDMap> lastmap(new OSDMap());
+    lastmap->set_max_osd(10);
+    lastmap->set_state(osd_id, CEPH_OSD_EXISTS);
+    lastmap->set_epoch(epoch);
+    OSDMap::Incremental inc(epoch + 1);
+    inc.new_pools[pool_id].min_size = min_size;
+    inc.new_pools[pool_id].set_pg_num(pg_num);
+    inc.new_up_thru[osd_id] = epoch - 10;
+    lastmap->apply_incremental(inc);
+
+    ostringstream out;
+
+    map<epoch_t, pg_interval_t> past_intervals;
+
+    ASSERT_TRUE(past_intervals.empty());
+    ASSERT_TRUE(pg_interval_t::check_new_interval(old_acting,
+						  new_acting,
+						  old_up,
+						  new_up,
+						  same_interval_since,
+						  last_epoch_clean,
+						  osdmap,
+						  lastmap,
+						  pool_id,
+						  pgid,
+						  &past_intervals,
+						  &out));
+    ASSERT_EQ((unsigned int)1, past_intervals.size());
+    ASSERT_TRUE(past_intervals[same_interval_since].maybe_went_rw);
+    ASSERT_NE(string::npos, out.str().find("presumed to have been rw"));
+  }
+
+  //
+  // The acting set changes. The old acting set primary was not up 
+  // during the old interval and last_epoch_clean is before the
+  // old interval : the previous interval could not possibly have
+  // been rw.
+  //
+  {
+    vector<int> new_acting;
+    new_acting.push_back(osd_id + 4); 
+    new_acting.push_back(osd_id + 5); 
+
+    epoch_t last_epoch_clean = epoch - 10;
+
+    std::tr1::shared_ptr<OSDMap> lastmap(new OSDMap());
+    lastmap->set_max_osd(10);
+    lastmap->set_state(osd_id, CEPH_OSD_EXISTS);
+    lastmap->set_epoch(epoch);
+    OSDMap::Incremental inc(epoch + 1);
+    inc.new_pools[pool_id].min_size = min_size;
+    inc.new_pools[pool_id].set_pg_num(pg_num);
+    inc.new_up_thru[osd_id] = last_epoch_clean;
+    lastmap->apply_incremental(inc);
+
+    ostringstream out;
+
+    map<epoch_t, pg_interval_t> past_intervals;
+
+    ASSERT_TRUE(past_intervals.empty());
+    ASSERT_TRUE(pg_interval_t::check_new_interval(old_acting,
+						  new_acting,
+						  old_up,
+						  new_up,
+						  same_interval_since,
+						  last_epoch_clean,
+						  osdmap,
+						  lastmap,
+						  pool_id,
+						  pgid,
+						  &past_intervals,
+						  &out));
+    ASSERT_EQ((unsigned int)1, past_intervals.size());
+    ASSERT_FALSE(past_intervals[same_interval_since].maybe_went_rw);
+    ASSERT_NE(string::npos, out.str().find("does not include interval"));
+  }
+}
+
 TEST(pg_t, split)
 {
   pg_t pgid(0, 0, -1);