rgw: meta sync thread crash at RGWMetaSyncShardCR

Fixes: http://tracker.ceph.com/issues/20251
Signed-off-by: fang yuxiang fang.yuxiang@eisoo.com

diff --git a/src/rgw/rgw_data_sync.cc b/src/rgw/rgw_data_sync.cc
index 7769c8dc9de2d22a4040ce7a70a555ceecb42168..da4db761daa9719f585fdeffe9168bb445cbafed 100644 (file)
--- a/src/rgw/rgw_data_sync.cc
+++ b/src/rgw/rgw_data_sync.cc
@@ -1084,8 +1084,8 @@ class RGWDataSyncShardCR : public RGWCoroutine {
 
   set<string> spawned_keys;
 
-  RGWContinuousLeaseCR *lease_cr;
-  RGWCoroutinesStack *lease_stack;
+  boost::intrusive_ptr<RGWContinuousLeaseCR> lease_cr;
+  boost::intrusive_ptr<RGWCoroutinesStack> lease_stack;
   string status_oid;
 
 
@@ -1126,7 +1126,6 @@ public:
     delete marker_tracker;
     if (lease_cr) {
       lease_cr->abort();
-      lease_cr->put();
     }
     if (error_repo) {
       error_repo->put();
@@ -1174,14 +1173,12 @@ public:
     string lock_name = "sync_lock";
     if (lease_cr) {
       lease_cr->abort();
-      lease_cr->put();
     }
     RGWRados *store = sync_env->store;
-    lease_cr = new RGWContinuousLeaseCR(sync_env->async_rados, store,
-                                        rgw_raw_obj(store->get_zone_params().log_pool, status_oid),
-                                        lock_name, lock_duration, this);
-    lease_cr->get();
-    lease_stack = spawn(lease_cr, false);
+    lease_cr.reset(new RGWContinuousLeaseCR(sync_env->async_rados, store,
+                                            rgw_raw_obj(store->get_zone_params().log_pool, status_oid),
+                                            lock_name, lock_duration, this));
+    lease_stack.reset(spawn(lease_cr.get(), false));
   }
 
   int full_sync() {
@@ -1358,7 +1355,7 @@ public:
             set_status() << "num_spawned() > spawn_window";
             yield wait_for_child();
             int ret;
-            while (collect(&ret, lease_stack)) {
+            while (collect(&ret, lease_stack.get())) {
               if (ret < 0) {
                 ldout(sync_env->cct, 0) << "ERROR: a sync operation returned error" << dendl;
                 /* we have reported this error */
@@ -2701,12 +2698,12 @@ int RGWRunBucketSyncCoroutine::operate()
     yield {
       set_status("acquiring sync lock");
       auto store = sync_env->store;
-      lease_cr = new RGWContinuousLeaseCR(sync_env->async_rados, store,
-                                          rgw_raw_obj(store->get_zone_params().log_pool, status_oid),
-                                          "sync_lock",
-                                          cct->_conf->rgw_sync_lease_period,
-                                          this);
-      lease_stack = spawn(lease_cr.get(), false);
+      lease_cr.reset(new RGWContinuousLeaseCR(sync_env->async_rados, store,
+                                              rgw_raw_obj(store->get_zone_params().log_pool, status_oid),
+                                              "sync_lock",
+                                              cct->_conf->rgw_sync_lease_period,
+                                              this));
+      lease_stack.reset(spawn(lease_cr.get(), false));
     }
     while (!lease_cr->is_locked()) {
       if (lease_cr->is_done()) {

diff --git a/src/rgw/rgw_sync.cc b/src/rgw/rgw_sync.cc
index 7a8a8bec2f12f533cb4eaf79426a72004fb2b321..99927dbaf0aa57a3048ee703f121e1be3be5b386 100644 (file)
--- a/src/rgw/rgw_sync.cc
+++ b/src/rgw/rgw_sync.cc
@@ -593,8 +593,8 @@ class RGWInitSyncStatusCoroutine : public RGWCoroutine {
 
   rgw_meta_sync_info status;
   vector<RGWMetadataLogInfo> shards_info;
-  RGWContinuousLeaseCR *lease_cr;
-  RGWCoroutinesStack *lease_stack;
+  boost::intrusive_ptr<RGWContinuousLeaseCR> lease_cr;
+  boost::intrusive_ptr<RGWCoroutinesStack> lease_stack;
 public:
   RGWInitSyncStatusCoroutine(RGWMetaSyncEnv *_sync_env,
                              const rgw_meta_sync_info &status)
@@ -605,7 +605,6 @@ public:
   ~RGWInitSyncStatusCoroutine() override {
     if (lease_cr) {
       lease_cr->abort();
-      lease_cr->put();
     }
   }
 
@@ -617,11 +616,10 @@ public:
        uint32_t lock_duration = cct->_conf->rgw_sync_lease_period;
         string lock_name = "sync_lock";
         RGWRados *store = sync_env->store;
-       lease_cr = new RGWContinuousLeaseCR(sync_env->async_rados, store,
-                                            rgw_raw_obj(store->get_zone_params().log_pool, sync_env->status_oid()),
-                                            lock_name, lock_duration, this);
-        lease_cr->get();
-        lease_stack = spawn(lease_cr, false);
+        lease_cr.reset(new RGWContinuousLeaseCR(sync_env->async_rados, store,
+                                                rgw_raw_obj(store->get_zone_params().log_pool, sync_env->status_oid()),
+                                                lock_name, lock_duration, this));
+        lease_stack.reset(spawn(lease_cr.get(), false));
       }
       while (!lease_cr->is_locked()) {
         if (lease_cr->is_done()) {
@@ -655,7 +653,7 @@ public:
        }
       }
 
-      drain_all_but_stack(lease_stack); /* the lease cr still needs to run */
+      drain_all_but_stack(lease_stack.get()); /* the lease cr still needs to run */
 
       yield {
         set_status("updating sync status");
@@ -782,8 +780,8 @@ class RGWFetchAllMetaCR : public RGWCoroutine {
 
   std::unique_ptr<RGWShardedOmapCRManager> entries_index;
 
-  RGWContinuousLeaseCR *lease_cr;
-  RGWCoroutinesStack *lease_stack;
+  boost::intrusive_ptr<RGWContinuousLeaseCR> lease_cr;
+  boost::intrusive_ptr<RGWCoroutinesStack> lease_stack;
   bool lost_lock;
   bool failed;
 
@@ -798,9 +796,6 @@ public:
   }
 
   ~RGWFetchAllMetaCR() override {
-    if (lease_cr) {
-      lease_cr->put();
-    }
   }
 
   void append_section_from_set(set<string>& all_sections, const string& name) {
@@ -836,12 +831,11 @@ public:
         set_status(string("acquiring lock (") + sync_env->status_oid() + ")");
        uint32_t lock_duration = cct->_conf->rgw_sync_lease_period;
         string lock_name = "sync_lock";
-       lease_cr = new RGWContinuousLeaseCR(sync_env->async_rados,
-                                            sync_env->store,
-                                            rgw_raw_obj(sync_env->store->get_zone_params().log_pool, sync_env->status_oid()),
-                                            lock_name, lock_duration, this);
-        lease_cr->get();
-        lease_stack = spawn(lease_cr, false);
+        lease_cr.reset(new RGWContinuousLeaseCR(sync_env->async_rados,
+                                                sync_env->store,
+                                                rgw_raw_obj(sync_env->store->get_zone_params().log_pool, sync_env->status_oid()),
+                                                lock_name, lock_duration, this));
+        lease_stack = spawn(lease_cr.get(), false);
       }
       while (!lease_cr->is_locked()) {
         if (lease_cr->is_done()) {
@@ -921,7 +915,7 @@ public:
         }
       }
 
-      drain_all_but_stack(lease_stack); /* the lease cr still needs to run */
+      drain_all_but_stack(lease_stack.get()); /* the lease cr still needs to run */
 
       yield lease_cr->go_down();
 
@@ -1316,8 +1310,9 @@ class RGWMetaSyncShardCR : public RGWCoroutine {
   boost::asio::coroutine incremental_cr;
   boost::asio::coroutine full_cr;
 
-  RGWContinuousLeaseCR *lease_cr = nullptr;
-  RGWCoroutinesStack *lease_stack = nullptr;
+  boost::intrusive_ptr<RGWContinuousLeaseCR> lease_cr;
+  boost::intrusive_ptr<RGWCoroutinesStack> lease_stack;
+
   bool lost_lock = false;
 
   bool *reset_backoff;
@@ -1350,7 +1345,6 @@ public:
     delete marker_tracker;
     if (lease_cr) {
       lease_cr->abort();
-      lease_cr->put();
     }
   }
 
@@ -1442,15 +1436,11 @@ public:
       yield {
        uint32_t lock_duration = cct->_conf->rgw_sync_lease_period;
         string lock_name = "sync_lock";
-        if (lease_cr) {
-          lease_cr->put();
-        }
         RGWRados *store = sync_env->store;
-       lease_cr = new RGWContinuousLeaseCR(sync_env->async_rados, store,
-                                            rgw_raw_obj(pool, sync_env->shard_obj_name(shard_id)),
-                                            lock_name, lock_duration, this);
-        lease_cr->get();
-        lease_stack = spawn(lease_cr, false);
+        lease_cr.reset(new RGWContinuousLeaseCR(sync_env->async_rados, store,
+                                                rgw_raw_obj(pool, sync_env->shard_obj_name(shard_id)),
+                                                lock_name, lock_duration, this));
+        lease_stack.reset(spawn(lease_cr.get(), false));
         lost_lock = false;
       }
       while (!lease_cr->is_locked()) {
@@ -1547,8 +1537,7 @@ ldout(cct, 0) << __FILE__ << ":" << __LINE__ << ":" << *this << ": num_spawned()
 
       yield lease_cr->go_down();
 
-      lease_cr->put();
-      lease_cr = NULL;
+      lease_cr.reset();
 
       drain_all();
 
@@ -1580,11 +1569,10 @@ ldout(cct, 0) << __FILE__ << ":" << __LINE__ << ":" << *this << ": num_spawned()
           uint32_t lock_duration = cct->_conf->rgw_sync_lease_period;
           string lock_name = "sync_lock";
           RGWRados *store = sync_env->store;
-          lease_cr = new RGWContinuousLeaseCR(sync_env->async_rados, store,
-                                              rgw_raw_obj(pool, sync_env->shard_obj_name(shard_id)),
-                                              lock_name, lock_duration, this);
-          lease_cr->get();
-          lease_stack = spawn(lease_cr, false);
+          lease_cr.reset( new RGWContinuousLeaseCR(sync_env->async_rados, store,
+                                                   rgw_raw_obj(pool, sync_env->shard_obj_name(shard_id)),
+                                                   lock_name, lock_duration, this));
+          lease_stack.reset(spawn(lease_cr.get(), false));
           lost_lock = false;
         }
         while (!lease_cr->is_locked()) {