mds: disable early reply for dir layout and quota related requests

These requests impacts whole subtree tree, replaying them when
mds recovers may break order of requests in multimds cluster.

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>

diff --git a/src/mds/Mutation.h b/src/mds/Mutation.h
index 1d4d25728ff155ef6153f40cec92c8b74a447d6f..4738f8a5c19600b479ad98773436576e95290f55 100644 (file)
--- a/src/mds/Mutation.h
+++ b/src/mds/Mutation.h
@@ -192,11 +192,12 @@ struct MDRequestImpl : public MutationImpl {
   inodeno_t alloc_ino, used_prealloc_ino;  
   interval_set<inodeno_t> prealloc_inos;
 
-  int snap_caps;
-  int getattr_caps;       ///< caps requested by getattr
-  bool did_early_reply;
-  bool o_trunc;           ///< request is an O_TRUNC mutation
-  bool has_completed;     ///< request has already completed
+  int snap_caps = 0;
+  int getattr_caps = 0;                ///< caps requested by getattr
+  bool no_early_reply = false;
+  bool did_early_reply = false;
+  bool o_trunc = false;                ///< request is an O_TRUNC mutation
+  bool has_completed = false;  ///< request has already completed
 
   bufferlist reply_extra_bl;
 
@@ -299,8 +300,6 @@ struct MDRequestImpl : public MutationImpl {
     session(NULL), item_session_request(this),
     client_request(params.client_req), straydn(NULL), snapid(CEPH_NOSNAP),
     tracei(NULL), tracedn(NULL), alloc_ino(0), used_prealloc_ino(0),
-    snap_caps(0), getattr_caps(0),
-    did_early_reply(false), o_trunc(false), has_completed(false),
     slave_request(NULL), internal_op(params.internal_op), internal_op_finish(NULL),
     internal_op_private(NULL),
     retry(0),

diff --git a/src/mds/Server.cc b/src/mds/Server.cc
index b2eb354029364aee1095fe8cc933e0884c3327ad..cdbee50c44f1a465700b73484580e65370bf628b 100644 (file)
--- a/src/mds/Server.cc
+++ b/src/mds/Server.cc
@@ -1296,6 +1296,11 @@ void Server::early_reply(MDRequestRef& mdr, CInode *tracei, CDentry *tracedn)
   if (!g_conf->mds_early_reply)
     return;
 
+  if (mdr->no_early_reply) {
+    dout(10) << "early_reply - flag no_early_reply is set, not allowed." << dendl;
+    return;
+  }
+
   if (mdr->has_more() && mdr->more()->has_journaled_slaves) {
     dout(10) << "early_reply - there are journaled slaves, not allowed." << dendl;
     return; 
@@ -4361,6 +4366,7 @@ void Server::handle_client_setdirlayout(MDRequestRef& mdr)
   mdcache->predirty_journal_parents(mdr, &le->metablob, cur, 0, PREDIRTY_PRIMARY);
   mdcache->journal_dirty_inode(mdr.get(), &le->metablob, cur);
 
+  mdr->no_early_reply = true;
   journal_and_reply(mdr, cur, 0, le, new C_MDS_inode_update_finish(this, mdr, cur));
 }
 
@@ -4592,6 +4598,7 @@ void Server::handle_set_vxattr(MDRequestRef& mdr, CInode *cur,
 
     pi = cur->project_inode();
     pi->layout = layout;
+    mdr->no_early_reply = true;
   } else if (name.compare(0, 16, "ceph.file.layout") == 0) {
     if (!cur->is_file()) {
       respond_to_request(mdr, -EINVAL);
@@ -4637,6 +4644,7 @@ void Server::handle_set_vxattr(MDRequestRef& mdr, CInode *cur,
 
     pi = cur->project_inode();
     pi->quota = quota;
+    mdr->no_early_reply = true;
   } else if (name.find("ceph.dir.pin") == 0) {
     if (!cur->is_dir() || cur->is_root()) {
       respond_to_request(mdr, -EINVAL);
@@ -4726,6 +4734,7 @@ void Server::handle_remove_vxattr(MDRequestRef& mdr, CInode *cur,
     mdcache->predirty_journal_parents(mdr, &le->metablob, cur, 0, PREDIRTY_PRIMARY);
     mdcache->journal_dirty_inode(mdr.get(), &le->metablob, cur);
 
+    mdr->no_early_reply = true;
     journal_and_reply(mdr, cur, 0, le, new C_MDS_inode_update_finish(this, mdr, cur));
     return;
   } else if (name == "ceph.dir.layout.pool_namespace"