Merge branch 'next'

Conflicts:
	src/include/ceph_features.h

13 files changed, 91 insertions, 9 deletions

diff --git a/qa/workunits/misc/mkpool_layout_vxattrs.sh b/qa/workunits/misc/mkpool_layout_vxattrs.sh
new file mode 100755
index 00000000000..16b3cdfe517
--- /dev/null
+++ b/qa/workunits/misc/mkpool_layout_vxattrs.sh
@@ -0,0 +1,13 @@
+#!/bin/bash
+
+set -e
+
+touch foo.$$
+rados mkpool foo.$$
+poolid=$(ceph osd dump | grep "^pool" | awk '{print $2}' | tail -n 1)
+ceph mds add_data_pool ${poolid}
+setfattr -n ceph.file.layout.pool -v foo.$$ foo.$$
+
+# cleanup
+rados rmpool foo.$$ foo.$$ --yes-i-really-really-mean-it 
+rm foo.$$
diff --git a/src/include/ceph_features.h b/src/include/ceph_features.h
index 242e239c667..c9ff72c15f9 100644
--- a/src/include/ceph_features.h
+++ b/src/include/ceph_features.h
@@ -32,7 +32,8 @@
 #define CEPH_FEATURE_CRUSH_TUNABLES2 (1<<25)
 #define CEPH_FEATURE_CREATEPOOLID   (1<<26)
 #define CEPH_FEATURE_REPLY_CREATE_INODE   (1<<27)
-#define CEPH_FEATURE_MDSENC         (1<<28)
+#define CEPH_FEATURE_OSD_HBMSGS     (1<<28)
+#define CEPH_FEATURE_MDSENC         (1<<29)
 
 /*
  * Features supported.  Should be everything above.
@@ -66,6 +67,7 @@
 	 CEPH_FEATURE_CRUSH_TUNABLES2 |	     \
 	 CEPH_FEATURE_CREATEPOOLID |	     \
 	 CEPH_FEATURE_REPLY_CREATE_INODE |   \
+	 CEPH_FEATURE_OSD_HBMSGS | \
 	 CEPH_FEATURE_MDSENC)
 
 #define CEPH_FEATURES_SUPPORTED_DEFAULT  CEPH_FEATURES_ALL
diff --git a/src/librbd/ImageCtx.cc b/src/librbd/ImageCtx.cc
index cffc556c1d4..7909f8c620c 100644
--- a/src/librbd/ImageCtx.cc
+++ b/src/librbd/ImageCtx.cc
@@ -332,6 +332,19 @@ namespace librbd {
     return -ENOENT;
   }
 
+  int ImageCtx::is_snap_unprotected(string in_snap_name,
+				    bool *is_unprotected) const
+  {
+    assert(snap_lock.is_locked());
+    map<string, SnapInfo>::const_iterator it = snaps_by_name.find(in_snap_name);
+    if (it != snaps_by_name.end()) {
+      *is_unprotected =
+	(it->second.protection_status == RBD_PROTECTION_STATUS_UNPROTECTED);
+      return 0;
+    }
+    return -ENOENT;
+  }
+
   int ImageCtx::get_snap_size(string in_snap_name, uint64_t *out_size) const
   {
     assert(snap_lock.is_locked());
diff --git a/src/librbd/ImageCtx.h b/src/librbd/ImageCtx.h
index ed03846fcde..b24639906b2 100644
--- a/src/librbd/ImageCtx.h
+++ b/src/librbd/ImageCtx.h
@@ -104,6 +104,7 @@ namespace librbd {
     int get_parent_spec(snapid_t snap_id, parent_spec *pspec);
     int get_snap_size(std::string in_snap_name, uint64_t *out_size) const;
     int is_snap_protected(string in_snap_name, bool *is_protected) const;
+    int is_snap_unprotected(string in_snap_name, bool *is_unprotected) const;
 
     uint64_t get_current_size() const;
     uint64_t get_object_size() const;
diff --git a/src/librbd/internal.cc b/src/librbd/internal.cc
index 663556e49a3..fdadf6f6753 100644
--- a/src/librbd/internal.cc
+++ b/src/librbd/internal.cc
@@ -585,12 +585,12 @@ namespace librbd {
     if (snap_id == CEPH_NOSNAP)
       return -ENOENT;
 
-    bool is_protected;
-    r = ictx->is_snap_protected(snap_name, &is_protected);
+    bool is_unprotected;
+    r = ictx->is_snap_unprotected(snap_name, &is_unprotected);
     if (r < 0)
       return r;
 
-    if (!is_protected)
+    if (is_unprotected)
       return -EINVAL;
 
     r = cls_client::set_protection_status(&ictx->md_ctx,
@@ -662,7 +662,12 @@ reprotect_and_return_err:
       return r;
 
     Mutex::Locker l(ictx->snap_lock);
-    return ictx->is_snap_protected(snap_name, is_protected);
+    bool is_unprotected;
+    r = ictx->is_snap_unprotected(snap_name, &is_unprotected);
+    // consider both PROTECTED or UNPROTECTING to be 'protected',
+    // since in either state they can't be deleted
+    *is_protected = !is_unprotected;
+    return r;
   }
 
   int create_v1(IoCtx& io_ctx, const char *imgname, uint64_t bid,
diff --git a/src/mds/MDCache.cc b/src/mds/MDCache.cc
index 2c48d375af8..da5ac0663ed 100644
--- a/src/mds/MDCache.cc
+++ b/src/mds/MDCache.cc
@@ -10581,6 +10581,7 @@ C_MDS_RetryRequest::C_MDS_RetryRequest(MDCache *c, MDRequest *r)
 
 void C_MDS_RetryRequest::finish(int r)
 {
+  mdr->retry++;
   cache->dispatch_request(mdr);
   mdr->put();
 }
diff --git a/src/mds/MDS.cc b/src/mds/MDS.cc
index 80afa241356..75dcf9e6dba 100644
--- a/src/mds/MDS.cc
+++ b/src/mds/MDS.cc
@@ -690,6 +690,10 @@ void MDS::handle_mds_beacon(MMDSBeacon *m)
   m->put();
 }
 
+void MDS::request_osdmap(Context *c) {
+  objecter->wait_for_new_map(c, osdmap->get_epoch());
+}
+
 /* This function DOES put the passed message before returning*/
 void MDS::handle_command(MMonCommand *m)
 {
diff --git a/src/mds/MDS.h b/src/mds/MDS.h
index f61ad8ddac5..50555e831d9 100644
--- a/src/mds/MDS.h
+++ b/src/mds/MDS.h
@@ -394,6 +394,8 @@ class MDS : public Dispatcher {
   void beacon_send();
   void handle_mds_beacon(MMDSBeacon *m);
 
+  void request_osdmap(Context *c);
+
   // messages
   bool _dispatch(Message *m);
 
diff --git a/src/mds/Mutation.h b/src/mds/Mutation.h
index bb5f1f6febc..313edef30d8 100644
--- a/src/mds/Mutation.h
+++ b/src/mds/Mutation.h
@@ -188,6 +188,12 @@ struct MDRequest : public Mutation {
   // -- i am an internal op
   int internal_op;
 
+  // indicates how may retries of request have been made
+  int retry;
+
+  // indicator for vxattr osdmap update
+  bool waited_for_osdmap;
+
   // break rarely-used fields into a separately allocated structure 
   // to save memory for most ops
   struct More {
@@ -241,6 +247,8 @@ struct MDRequest : public Mutation {
     alloc_ino(0), used_prealloc_ino(0), snap_caps(0), did_early_reply(false),
     slave_request(0),
     internal_op(-1),
+    retry(0),
+    waited_for_osdmap(false),
     _more(0) {
     in[0] = in[1] = 0; 
   }
@@ -252,6 +260,8 @@ struct MDRequest : public Mutation {
     alloc_ino(0), used_prealloc_ino(0), snap_caps(0), did_early_reply(false),
     slave_request(0),
     internal_op(-1),
+    retry(0),
+    waited_for_osdmap(false),
     _more(0) {
     in[0] = in[1] = 0; 
   }
@@ -263,6 +273,8 @@ struct MDRequest : public Mutation {
     alloc_ino(0), used_prealloc_ino(0), snap_caps(0), did_early_reply(false),
     slave_request(0),
     internal_op(-1),
+    retry(0),
+    waited_for_osdmap(false),
     _more(0) {
     in[0] = in[1] = 0; 
   }
diff --git a/src/mds/Server.cc b/src/mds/Server.cc
index 36850307697..ac51e60d0a9 100644
--- a/src/mds/Server.cc
+++ b/src/mds/Server.cc
@@ -3531,7 +3531,7 @@ int Server::parse_layout_vxattr(string name, string value, ceph_file_layout *lay
 	int64_t pool = mds->osdmap->lookup_pg_pool_name(value);
 	if (pool < 0) {
 	  dout(10) << " unknown pool " << value << dendl;
-	  return -EINVAL;
+	  return -ENOENT;
 	}
 	layout->fl_pg_pool = pool;
       }
@@ -3589,6 +3589,17 @@ void Server::handle_set_vxattr(MDRequest *mdr, CInode *cur,
       rest = name.substr(name.find("layout"));
       int r = parse_layout_vxattr(rest, value, &dlayout->layout);
       if (r < 0) {
+	if (r == -ENOENT) {
+	  if (!mdr->waited_for_osdmap) {
+	    // send request to get latest map, but don't wait if
+	    // we don't get anything newer than what we have
+	    mdr->waited_for_osdmap = true;
+	    mds->request_osdmap(
+		  new C_MDS_RetryRequest(mdcache, mdr));
+	    return;
+	  }
+	  r = -EINVAL;
+	}
 	reply_request(mdr, r);
 	return;
       }
@@ -3608,6 +3619,17 @@ void Server::handle_set_vxattr(MDRequest *mdr, CInode *cur,
       rest = name.substr(name.find("layout"));
       int r = parse_layout_vxattr(rest, value, &layout);
       if (r < 0) {
+	if (r == -ENOENT) {
+	  if (!mdr->waited_for_osdmap) {
+	    // send request to get latest map, but don't wait if
+	    // we don't get anything newer than what we have
+	    mdr->waited_for_osdmap = true;
+	    mds->request_osdmap(
+		  new C_MDS_RetryRequest(mdcache, mdr));
+	    return;
+	  }
+	  r = -EINVAL;
+	}
 	reply_request(mdr, r);
 	return;
       }
diff --git a/src/osd/OSD.cc b/src/osd/OSD.cc
index 243eccf8048..25bc48c4550 100644
--- a/src/osd/OSD.cc
+++ b/src/osd/OSD.cc
@@ -3422,7 +3422,8 @@ bool OSD::heartbeat_dispatch(Message *m)
     break;
 
   default:
-    return false;
+    dout(0) << "dropping unexpected message " << *m << " from " << m->get_source_inst() << dendl;
+    m->put();
   }
 
   return true;
@@ -4683,8 +4684,12 @@ bool OSD::require_same_or_newer_map(OpRequestRef op, epoch_t epoch)
     int from = m->get_source().num();
     if (!osdmap->have_inst(from) ||
 	osdmap->get_cluster_addr(from) != m->get_source_inst().addr) {
-      dout(0) << "from dead osd." << from << ", dropping, sharing map" << dendl;
-      send_incremental_map(epoch, m->get_connection());
+      if (m->get_connection()->has_feature(CEPH_FEATURE_OSD_HBMSGS)) {
+	dout(10) << "from dead osd." << from << ", dropping, sharing map" << dendl;
+	send_incremental_map(epoch, m->get_connection());
+      } else {
+	dout(10) << "from dead osd." << from << ", but it lacks OSD_HBMSGS feature, not sharing map" << dendl;
+      }
       return false;
     }
   }
diff --git a/src/test/cli/ceph/help.t b/src/test/cli/ceph/help.t
index eac011dc814..9f84986f02e 100644
--- a/src/test/cli/ceph/help.t
+++ b/src/test/cli/ceph/help.t
@@ -20,6 +20,7 @@
   METADATA SERVER (MDS) COMMANDS
     ceph mds stat
     ceph mds tell <mds-id or *> injectargs '--<switch> <value> [--<switch> <value>...]'
+    ceph mds add_data_pool <pool-id>
   
   MONITOR (MON) COMMANDS
     ceph mon add <name> <ip>[:<port>]
diff --git a/src/tools/ceph.cc b/src/tools/ceph.cc
index a1f8024cfe3..d8686d5cb17 100644
--- a/src/tools/ceph.cc
+++ b/src/tools/ceph.cc
@@ -63,6 +63,7 @@ static void usage()
   cout << "METADATA SERVER (MDS) COMMANDS\n";
   cout << "  ceph mds stat\n";
   cout << "  ceph mds tell <mds-id or *> injectargs '--<switch> <value> [--<switch> <value>...]'\n";
+  cout << "  ceph mds add_data_pool <pool-id>\n";
   cout << "\n";
   cout << "MONITOR (MON) COMMANDS\n";
   cout << "  ceph mon add <name> <ip>[:<port>]\n";