Merge branch 'testing' into unstable

Conflicts:
	configure.ac
	src/mds/MDCache.cc

debian/changelog

@@ -1,3 +1,22 @@
+ceph (0.21.2-1) unstable; urgency=low
+
+  * osd: less log noise
+  * osd: mark down old heartbeat peers
+  * filejournal: clean up init sequence, fewer confusing errors on startup
+  * msgr: fix throttler leak when closing connections (fixes deadlock)
+  * osdmaptool: don't crash on corrupt input
+  * crush: error out instead of crashing on ambiguous hierarchy
+  * mds: error to client on invalid op
+  * mds: fix ENOTEMPTY checks on rmdir
+  * osd: fix race between reads and cloned objects
+  * auth: fix keyring load search when $HOME not defined
+  * client: fix xattr cap writeback
+  * client: fix snap vs metadata writeback
+  * osd: fix journal, btrfs throttling
+  * msgr: fix memory leak on closed connections
+
+ -- Sage Weil <sage@newdream.net>  Thu, 26 Aug 2010 15:22:42 -0700
+
 ceph (0.21.1-1) unstable; urgency=high
 
   * ceph.spec: many fixes

src/mds/CInode.h

@@ -701,7 +701,12 @@ public:
 	lock->set_state(LOCK_EXCL);
       else if (issued & CEPH_CAP_GWR)
 	lock->set_state(LOCK_MIX);
-      else
+      else if (lock->is_dirty()) {
+	if (is_replicated())
+	  lock->set_state(LOCK_MIX);
+	else
+	  lock->set_state(LOCK_LOCK);
+      } else
 	lock->set_state(LOCK_SYNC);
     } else {
       if (lock->is_xlocked())
@@ -716,6 +721,8 @@ public:
 	choose_ideal_loner() >= 0)
       try_set_loner();
     choose_lock_state(&filelock, issued);
+    choose_lock_state(&nestlock, issued);
+    choose_lock_state(&dirfragtreelock, issued);
     choose_lock_state(&authlock, issued);
     choose_lock_state(&xattrlock, issued);
     choose_lock_state(&linklock, issued);

src/mds/MDCache.cc

@@ -3935,7 +3935,7 @@ void MDCache::rejoin_gather_finish()
   }
   
   process_imported_caps();
-  process_reconnected_caps();
+  choose_lock_states_and_reconnect_caps();
 
   vector<CInode*> recover_q, check_q;
   identify_files_to_recover(rejoin_recover_q, rejoin_check_q);
@@ -4005,18 +4005,17 @@ void MDCache::check_realm_past_parents(SnapRealm *realm)
 /*
  * choose lock states based on reconnected caps
  */
-void MDCache::process_reconnected_caps()
+void MDCache::choose_lock_states_and_reconnect_caps()
 {
-  dout(10) << "process_reconnected_caps" << dendl;
+  dout(10) << "choose_lock_states_and_reconnect_caps" << dendl;
 
   map<client_t,MClientSnap*> splits;
 
-  // adjust lock states appropriately
-  for (map<CInode*,map<client_t,inodeno_t> >::iterator p = reconnected_caps.begin();
-       p != reconnected_caps.end();
-       p++) {
-    CInode *in = p->first;
-
+  for (hash_map<vinodeno_t,CInode*>::iterator i = inode_map.begin();
+       i != inode_map.end();
+       ++i) {
+    CInode *in = i->second;
+ 
     in->choose_lock_states();
     dout(15) << " chose lock states on " << *in << dendl;
 
@@ -4024,21 +4023,25 @@ void MDCache::process_reconnected_caps()
 
     check_realm_past_parents(realm);
 
-    // also, make sure client's cap is in the correct snaprealm.
-    for (map<client_t,inodeno_t>::iterator q = p->second.begin();
-	 q != p->second.end();
-	 q++) {
-      if (q->second == realm->inode->ino()) {
-	dout(15) << "  client" << q->first << " has correct realm " << q->second << dendl;
-      } else {
-	dout(15) << "  client" << q->first << " has wrong realm " << q->second
-		 << " != " << realm->inode->ino() << dendl;
-	if (realm->have_past_parents_open()) {
-	  // ok, include in a split message _now_.
-	  prepare_realm_split(realm, q->first, in->ino(), splits);
+    map<CInode*,map<client_t,inodeno_t> >::iterator p = reconnected_caps.find(in);
+    if (p != reconnected_caps.end()) {
+
+      // also, make sure client's cap is in the correct snaprealm.
+      for (map<client_t,inodeno_t>::iterator q = p->second.begin();
+	   q != p->second.end();
+	   q++) {
+	if (q->second == realm->inode->ino()) {
+	  dout(15) << "  client" << q->first << " has correct realm " << q->second << dendl;
 	} else {
-	  // send the split later.
-	  missing_snap_parents[realm->inode][q->first].insert(in->ino());
+	  dout(15) << "  client" << q->first << " has wrong realm " << q->second
+		   << " != " << realm->inode->ino() << dendl;
+	  if (realm->have_past_parents_open()) {
+	    // ok, include in a split message _now_.
+	    prepare_realm_split(realm, q->first, in->ino(), splits);
+	  } else {
+	    // send the split later.
+	    missing_snap_parents[realm->inode][q->first].insert(in->ino());
+	  }
 	}
       }
     }

src/mds/MDCache.h

@@ -755,7 +755,7 @@ public:
     reconnected_snaprealms[ino][client] = seq;
   }
   void process_imported_caps();
-  void process_reconnected_caps();
+  void choose_lock_states_and_reconnect_caps();
   void prepare_realm_split(SnapRealm *realm, client_t client, inodeno_t ino,
 			   map<client_t,MClientSnap*>& splits);
   void do_realm_invalidate_and_update_notify(CInode *in, int snapop, bool nosend=false);

src/mds/SimpleLock.h

@@ -365,7 +365,7 @@ public:
 
 
 
-  virtual bool is_dirty() { return false; }
+  virtual bool is_dirty() const { return false; }
 
 
   // can_*

src/osd/OSD.cc

@@ -1677,7 +1677,7 @@ void OSD::send_pg_stats()
   if (osd_stat_updated || !pg_stat_queue.empty()) {
     osd_stat_updated = false;
     
-    dout(1) << "send_pg_stats - " << pg_stat_queue.size() << " pgs updated" << dendl;
+    dout(10) << "send_pg_stats - " << pg_stat_queue.size() << " pgs updated" << dendl;
     
     utime_t had_for = g_clock.now();
     had_for -= had_map_since;
@@ -4459,11 +4459,11 @@ void OSD::handle_op(MOSDOp *op)
 
   int err = -EPERM;
   if (op->may_read() && !(perm & OSD_POOL_CAP_R)) {
-    dout(0) << "no READ permission to access pool " << pg->pool->name << dendl;
+    dout(10) << "no READ permission to access pool " << pg->pool->name << dendl;
   } else if (op->may_write() && !(perm & OSD_POOL_CAP_W)) {
-    dout(0) << "no WRITE permission to access pool " << pg->pool->name << dendl;
+    dout(10) << "no WRITE permission to access pool " << pg->pool->name << dendl;
   } else if (op->require_exec_caps() && !(perm & OSD_POOL_CAP_X)) {
-    dout(0) << "no EXEC permission to access pool " << pg->pool->name << dendl;
+    dout(10) << "no EXEC permission to access pool " << pg->pool->name << dendl;
   } else {
     err = 0;
   }
@@ -4822,7 +4822,7 @@ int OSD::get_class(const string& cname, ClassVersion& version, pg_t pgid, Messag
       *pcls = cls;
       return 0;
     case ClassHandler::ClassData::CLASS_INVALID:
-      dout(0) << "class not supported" << dendl;
+      dout(1) << "class " << cname << " not supported" << dendl;
       return -EOPNOTSUPP;
     case ClassHandler::ClassData::CLASS_ERROR:
       dout(0) << "error loading class!" << dendl;
@@ -4911,7 +4911,7 @@ void OSD::init_op_flags(MOSDOp *op)
 	is_write = flags & CLS_METHOD_WR;
         is_public = flags & CLS_METHOD_PUBLIC;
 
-	dout(0) << "class " << cname << " method " << mname
+	dout(10) << "class " << cname << " method " << mname
 		<< " flags=" << (is_read ? "r" : "") << (is_write ? "w" : "") << dendl;
 	if (is_read)
 	  op->rmw_flags |= CEPH_OSD_FLAG_READ;

src/osd/ReplicatedPG.cc

@@ -139,7 +139,7 @@ void ReplicatedPG::wait_for_degraded_object(const sobject_t& soid, Message *m)
 // ==========================================================
 void ReplicatedPG::do_pg_op(MOSDOp *op)
 {
-  dout(0) << "do_pg_op " << *op << dendl;
+  dout(10) << "do_pg_op " << *op << dendl;
 
   bufferlist outdata;
   int result = 0;
@@ -813,7 +813,6 @@ int ReplicatedPG::do_osd_ops(OpContext *ctx, vector<OSDOp>& ops,
 
     case CEPH_OSD_OP_READ:
       {
-	dout(0) << "CEPH_OSD_OP_READ" << dendl;
 	// read into a buffer
 	bufferlist bl;
 	int r = osd->store->read(coll_t::build_pg_coll(info.pgid), soid, op.extent.offset, op.extent.length, bl);
@@ -828,7 +827,7 @@ int ReplicatedPG::do_osd_ops(OpContext *ctx, vector<OSDOp>& ops,
 	}
 	info.stats.num_rd_kb += SHIFT_ROUND_UP(op.extent.length, 10);
 	info.stats.num_rd++;
-	dout(0) << " read got " << r << " / " << op.extent.length << " bytes from obj " << soid << dendl;
+	dout(10) << " read got " << r << " / " << op.extent.length << " bytes from obj " << soid << dendl;
 
 	__u32 seq = oi.truncate_seq;
 	// are we beyond truncate_size?
@@ -845,7 +844,7 @@ int ReplicatedPG::do_osd_ops(OpContext *ctx, vector<OSDOp>& ops,
 	  bufferlist keep;
 
 	  // keep first part of odata; trim at truncation point
-	  dout(0) << " obj " << soid << " seq " << seq
+	  dout(10) << " obj " << soid << " seq " << seq
 	           << ": trimming overlap " << from << "~" << trim << dendl;
 	  keep.substr_of(odata, 0, odata.length() - trim);
           odata.claim(keep);
@@ -932,7 +931,6 @@ int ReplicatedPG::do_osd_ops(OpContext *ctx, vector<OSDOp>& ops,
       
     case CEPH_OSD_OP_CMPXATTR:
       {
-	dout(0) << "CEPH_OSD_OP_CMPXATTR" << dendl;
 	string aname;
 	bp.copy(op.xattr.name_len, aname);
 	string name = "_" + aname;
@@ -970,16 +968,16 @@ int ReplicatedPG::do_osd_ops(OpContext *ctx, vector<OSDOp>& ops,
 	}
 
 	if (!result) {
-	  dout(0) << "comparison returned false" << dendl;
+	  dout(10) << "comparison returned false" << dendl;
 	  result = -ECANCELED;
 	  break;
 	}
 	if (result < 0) {
-	  dout(0) << "comparison returned " << result << " " << strerror(-result) << dendl;
+	  dout(10) << "comparison returned " << result << " " << strerror(-result) << dendl;
 	  break;
 	}
 
-	dout(0) << "comparison returned true" << dendl;
+	dout(10) << "comparison returned true" << dendl;
 	info.stats.num_rd++;
       }
       break;
@@ -3158,7 +3156,7 @@ void ReplicatedPG::sub_op_push(MOSDSubOp *op)
       
       dout(10) << "sub_op_push need " << data_needed << ", got " << data_subset << dendl;
       if (!data_needed.subset_of(data_subset)) {
-	dout(0) << " we did not get enough of " << soid << " object data" << dendl;
+	dout(0) << "sub_op_push we did not get enough of " << soid << " object data" << dendl;
 	op->put();
 	return;
       }