From 1ee8308c74f991e7d20c633b3e3a9f1691884be8 Mon Sep 17 00:00:00 2001
From: Sage Weil <sage@newdream.net>
Date: Fri, 20 Nov 2009 21:42:04 -0800
Subject: [PATCH] mds: do not eval subtree root during rename_prepare

The eval may lead to a scatter_writebehind, which screws up the journal
ordering.  Since the inode is also locked, we will end up evaling it
later anyway, so don't worry about it.

I'm a bit unsure about the adjust_subtree_after_rename... that is during
the rename_apply stage (after the log entry committed), so i think it's
safe to do a scatter_writebehind here.. since we're not between a predirty
and submit_entry.
---
 src/mds/MDCache.cc | 13 +++++++------
 src/mds/MDCache.h  |  6 +++---
 src/mds/Server.cc  |  4 ++--
 3 files changed, 12 insertions(+), 11 deletions(-)

diff --git a/src/mds/MDCache.cc b/src/mds/MDCache.cc
index d7e21082a41d7..621bd6375390d 100644
--- a/src/mds/MDCache.cc
+++ b/src/mds/MDCache.cc
@@ -650,7 +650,7 @@ void MDCache::list_subtrees(list<CDir*>& ls)
  * merge with parent and/or child subtrees, if is it appropriate.
  * merge can ONLY happen if both parent and child have unambiguous auth.
  */
-void MDCache::adjust_subtree_auth(CDir *dir, pair<int,int> auth)
+void MDCache::adjust_subtree_auth(CDir *dir, pair<int,int> auth, bool do_eval)
 {
   dout(7) << "adjust_subtree_auth " << dir->get_dir_auth() << " -> " << auth
 	  << " on " << *dir << dendl;
@@ -714,7 +714,8 @@ void MDCache::adjust_subtree_auth(CDir *dir, pair<int,int> auth)
       }
     }
 
-    eval_subtree_root(dir);
+    if (do_eval)
+      eval_subtree_root(dir->get_inode());
   }
 
   show_subtrees();
@@ -787,7 +788,7 @@ void MDCache::try_subtree_merge_at(CDir *dir)
       }
     }
 
-    eval_subtree_root(dir);
+    eval_subtree_root(dir->get_inode());
 
     // journal inode? 
     //  (this is a large hammer to ensure that dirfragtree updates will
@@ -831,12 +832,12 @@ void MDCache::subtree_merge_writebehind_finish(CInode *in, Mutation *mut)
   in->auth_unpin(this);
 }
 
-void MDCache::eval_subtree_root(CDir *dir)
+void MDCache::eval_subtree_root(CInode *diri)
 {
   // evaluate subtree inode filelock?
   //  (we should scatter the filelock on subtree bounds)
-  if (dir->inode->is_auth())
-    mds->locker->try_eval(dir->inode, CEPH_LOCK_IFILE | CEPH_LOCK_INEST);
+  if (diri->is_auth())
+    mds->locker->try_eval(diri, CEPH_LOCK_IFILE | CEPH_LOCK_INEST);
 }
 
 
diff --git a/src/mds/MDCache.h b/src/mds/MDCache.h
index f858ca055e5fa..ee2842d88d243 100644
--- a/src/mds/MDCache.h
+++ b/src/mds/MDCache.h
@@ -508,8 +508,8 @@ protected:
 public:
   bool is_subtrees() { return !subtrees.empty(); }
   void list_subtrees(list<CDir*>& ls);
-  void adjust_subtree_auth(CDir *root, pair<int,int> auth);
-  void adjust_subtree_auth(CDir *root, int a, int b=CDIR_AUTH_UNKNOWN) {
+  void adjust_subtree_auth(CDir *root, pair<int,int> auth, bool do_eval=true);
+  void adjust_subtree_auth(CDir *root, int a, int b=CDIR_AUTH_UNKNOWN, bool do_eval=true) {
     adjust_subtree_auth(root, pair<int,int>(a,b)); 
   }
   void adjust_bounded_subtree_auth(CDir *dir, set<CDir*>& bounds, pair<int,int> auth);
@@ -524,7 +524,7 @@ public:
   void try_subtree_merge(CDir *root);
   void try_subtree_merge_at(CDir *root);
   void subtree_merge_writebehind_finish(CInode *in, Mutation *mut);
-  void eval_subtree_root(CDir *dir);
+  void eval_subtree_root(CInode *diri);
   CDir *get_subtree_root(CDir *dir);
   bool is_leaf_subtree(CDir *dir) {
     assert(subtrees.count(dir));
diff --git a/src/mds/Server.cc b/src/mds/Server.cc
index d08f15fccff4c..4be676782034e 100644
--- a/src/mds/Server.cc
+++ b/src/mds/Server.cc
@@ -4417,7 +4417,7 @@ void Server::_rename_prepare(MDRequest *mdr,
     srcdnl->get_inode()->get_nested_dirfrags(ls);
     int auth = srcdn->authority().first;
     for (list<CDir*>::iterator p = ls.begin(); p != ls.end(); ++p) 
-      mdcache->adjust_subtree_auth(*p, auth, auth);
+      mdcache->adjust_subtree_auth(*p, auth, auth, false);
   }
 
   // do inode updates in journal, even if we aren't auth (hmm, is this necessary?)
@@ -5052,7 +5052,7 @@ void Server::do_rename_rollback(bufferlist &rbl, int master, MDRequest *mdr)
     srcdnl->get_inode()->get_nested_dirfrags(ls);
     int auth = srcdn->authority().first;
     for (list<CDir*>::iterator p = ls.begin(); p != ls.end(); ++p) 
-      mdcache->adjust_subtree_auth(*p, auth, auth);
+      mdcache->adjust_subtree_auth(*p, auth, auth, false);
   }
 
   // journal it
-- 
2.39.5