Merge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2

* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2: (28 commits)
  ocfs2: Teach ocfs2_drop_lock() to use ->set_lvb() callback
  ocfs2: Remove ->unblock lockres operation
  ocfs2: move downconvert worker to lockres ops
  ocfs2: Remove unused dlmglue functions
  ocfs2: Have the metadata lock use generic dlmglue functions
  ocfs2: Add ->set_lvb callback in dlmglue
  ocfs2: Add ->check_downconvert callback in dlmglue
  ocfs2: Check for refreshing locks in generic unblock function
  ocfs2: don't unconditionally pass LVB flags
  ocfs2: combine inode and generic blocking AST functions
  ocfs2: Add ->get_osb() dlmglue locking operation
  ocfs2: remove ->unlock_ast() callback from ocfs2_lock_res_ops
  ocfs2: combine inode and generic AST functions
  ocfs2: Clean up lock resource refresh flags
  ocfs2: Remove i_generation from inode lock names
  ocfs2: Encode i_generation in the meta data lvb
  ocfs2: Free up some space in the lvb
  ocfs2: Remove special casing for inode creation in ocfs2_dentry_attach_lock()
  ocfs2: manually d_move() during ocfs2_rename()
  [PATCH] Allow file systems to manually d_move() inside of ->rename()
  ...

fs/namei.c

@@ -2370,7 +2370,8 @@ static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
 		dput(new_dentry);
 	}
 	if (!error)
-		d_move(old_dentry,new_dentry);
+		if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
+			d_move(old_dentry,new_dentry);
 	return error;
 }
 
@@ -2393,8 +2394,7 @@ static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
 	else
 		error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
 	if (!error) {
-		/* The following d_move() should become unconditional */
-		if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
+		if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
 			d_move(old_dentry, new_dentry);
 	}
 	if (target)

fs/nfs/dir.c

@@ -1669,8 +1669,7 @@ out:
 	if (rehash)
 		d_rehash(rehash);
 	if (!error) {
-		if (!S_ISDIR(old_inode->i_mode))
-			d_move(old_dentry, new_dentry);
+		d_move(old_dentry, new_dentry);
 		nfs_renew_times(new_dentry);
 		nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir));
 	}

fs/nfs/super.c

@@ -71,7 +71,7 @@ static struct file_system_type nfs_fs_type = {
 	.name		= "nfs",
 	.get_sb		= nfs_get_sb,
 	.kill_sb	= nfs_kill_super,
-	.fs_flags	= FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
+	.fs_flags	= FS_RENAME_DOES_D_MOVE|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
 };
 
 struct file_system_type nfs_xdev_fs_type = {
@@ -79,7 +79,7 @@ struct file_system_type nfs_xdev_fs_type = {
 	.name		= "nfs",
 	.get_sb		= nfs_xdev_get_sb,
 	.kill_sb	= nfs_kill_super,
-	.fs_flags	= FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
+	.fs_flags	= FS_RENAME_DOES_D_MOVE|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
 };
 
 static struct super_operations nfs_sops = {
@@ -107,7 +107,7 @@ static struct file_system_type nfs4_fs_type = {
 	.name		= "nfs4",
 	.get_sb		= nfs4_get_sb,
 	.kill_sb	= nfs4_kill_super,
-	.fs_flags	= FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
+	.fs_flags	= FS_RENAME_DOES_D_MOVE|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
 };
 
 struct file_system_type nfs4_xdev_fs_type = {
@@ -115,7 +115,7 @@ struct file_system_type nfs4_xdev_fs_type = {
 	.name		= "nfs4",
 	.get_sb		= nfs4_xdev_get_sb,
 	.kill_sb	= nfs4_kill_super,
-	.fs_flags	= FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
+	.fs_flags	= FS_RENAME_DOES_D_MOVE|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
 };
 
 struct file_system_type nfs4_referral_fs_type = {
@@ -123,7 +123,7 @@ struct file_system_type nfs4_referral_fs_type = {
 	.name		= "nfs4",
 	.get_sb		= nfs4_referral_get_sb,
 	.kill_sb	= nfs4_kill_super,
-	.fs_flags	= FS_ODD_RENAME|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
+	.fs_flags	= FS_RENAME_DOES_D_MOVE|FS_REVAL_DOT|FS_BINARY_MOUNTDATA,
 };
 
 static struct super_operations nfs4_sops = {

fs/ocfs2/cluster/tcp_internal.h

@@ -44,11 +44,17 @@
  * locking semantics of the file system using the protocol.  It should 
  * be somewhere else, I'm sure, but right now it isn't.
  *
+ * New in version 4:
+ * 	- Remove i_generation from lock names for better stat performance.
+ *
+ * New in version 3:
+ * 	- Replace dentry votes with a cluster lock
+ *
  * New in version 2:
  * 	- full 64 bit i_size in the metadata lock lvbs
  * 	- introduction of "rw" lock and pushing meta/data locking down
  */
-#define O2NET_PROTOCOL_VERSION 2ULL
+#define O2NET_PROTOCOL_VERSION 4ULL
 struct o2net_handshake {
 	__be64	protocol_version;
 	__be64	connector_id;

fs/ocfs2/dcache.h

@@ -28,4 +28,31 @@
 
 extern struct dentry_operations ocfs2_dentry_ops;
 
+struct ocfs2_dentry_lock {
+	unsigned int		dl_count;
+	u64			dl_parent_blkno;
+
+	/*
+	 * The ocfs2_dentry_lock keeps an inode reference until
+	 * dl_lockres has been destroyed. This is usually done in
+	 * ->d_iput() anyway, so there should be minimal impact.
+	 */
+	struct inode		*dl_inode;
+	struct ocfs2_lock_res	dl_lockres;
+};
+
+int ocfs2_dentry_attach_lock(struct dentry *dentry, struct inode *inode,
+			     u64 parent_blkno);
+
+void ocfs2_dentry_lock_put(struct ocfs2_super *osb,
+			   struct ocfs2_dentry_lock *dl);
+
+struct dentry *ocfs2_find_local_alias(struct inode *inode, u64 parent_blkno,
+				      int skip_unhashed);
+
+void ocfs2_dentry_move(struct dentry *dentry, struct dentry *target,
+		       struct inode *old_dir, struct inode *new_dir);
+
+extern spinlock_t dentry_attach_lock;
+
 #endif /* OCFS2_DCACHE_H */

fs/ocfs2/dlm/dlmapi.h

@@ -182,6 +182,7 @@ enum dlm_status dlmlock(struct dlm_ctxt *dlm,
 			struct dlm_lockstatus *lksb,
 			int flags,
 			const char *name,
+			int namelen,
 			dlm_astlockfunc_t *ast,
 			void *data,
 			dlm_bastlockfunc_t *bast);

fs/ocfs2/dlm/dlmast.c

@@ -320,8 +320,8 @@ int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data)
 
 	res = dlm_lookup_lockres(dlm, name, locklen);
 	if (!res) {
-		mlog(ML_ERROR, "got %sast for unknown lockres! "
-			       "cookie=%u:%llu, name=%.*s, namelen=%u\n",
+		mlog(0, "got %sast for unknown lockres! "
+		     "cookie=%u:%llu, name=%.*s, namelen=%u\n",
 		     past->type == DLM_AST ? "" : "b",
 		     dlm_get_lock_cookie_node(cookie),
 		     dlm_get_lock_cookie_seq(cookie),
@@ -462,7 +462,7 @@ int dlm_send_proxy_ast_msg(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
 			mlog(ML_ERROR, "sent AST to node %u, it returned "
 			     "DLM_MIGRATING!\n", lock->ml.node);
 			BUG();
-		} else if (status != DLM_NORMAL) {
+		} else if (status != DLM_NORMAL && status != DLM_IVLOCKID) {
 			mlog(ML_ERROR, "AST to node %u returned %d!\n",
 			     lock->ml.node, status);
 			/* ignore it */

fs/ocfs2/dlm/dlmcommon.h

@@ -747,6 +747,7 @@ void dlm_change_lockres_owner(struct dlm_ctxt *dlm,
 			      u8 owner);
 struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm,
 						 const char *lockid,
+						 int namelen,
 						 int flags);
 struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm,
 					  const char *name,

fs/ocfs2/dlm/dlmlock.c

@@ -540,8 +540,8 @@ static inline void dlm_get_next_cookie(u8 node_num, u64 *cookie)
 
 enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode,
 			struct dlm_lockstatus *lksb, int flags,
-			const char *name, dlm_astlockfunc_t *ast, void *data,
-			dlm_bastlockfunc_t *bast)
+			const char *name, int namelen, dlm_astlockfunc_t *ast,
+			void *data, dlm_bastlockfunc_t *bast)
 {
 	enum dlm_status status;
 	struct dlm_lock_resource *res = NULL;
@@ -571,7 +571,7 @@ enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode,
 	recovery = (flags & LKM_RECOVERY);
 
 	if (recovery &&
-	    (!dlm_is_recovery_lock(name, strlen(name)) || convert) ) {
+	    (!dlm_is_recovery_lock(name, namelen) || convert) ) {
 		dlm_error(status);
 		goto error;
 	}
@@ -643,7 +643,7 @@ retry_convert:
 		}
 
 		status = DLM_IVBUFLEN;
-		if (strlen(name) > DLM_LOCKID_NAME_MAX || strlen(name) < 1) {
+		if (namelen > DLM_LOCKID_NAME_MAX || namelen < 1) {
 			dlm_error(status);
 			goto error;
 		}
@@ -659,7 +659,7 @@ retry_convert:
 			dlm_wait_for_recovery(dlm);
 
 		/* find or create the lock resource */
-		res = dlm_get_lock_resource(dlm, name, flags);
+		res = dlm_get_lock_resource(dlm, name, namelen, flags);
 		if (!res) {
 			status = DLM_IVLOCKID;
 			dlm_error(status);

fs/ocfs2/dlm/dlmmaster.c

@@ -740,6 +740,7 @@ struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm,
  */
 struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm,
 					  const char *lockid,
+					  int namelen,
 					  int flags)
 {
 	struct dlm_lock_resource *tmpres=NULL, *res=NULL;
@@ -748,13 +749,12 @@ struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm,
 	int blocked = 0;
 	int ret, nodenum;
 	struct dlm_node_iter iter;
-	unsigned int namelen, hash;
+	unsigned int hash;
 	int tries = 0;
 	int bit, wait_on_recovery = 0;
 
 	BUG_ON(!lockid);
 
-	namelen = strlen(lockid);
 	hash = dlm_lockid_hash(lockid, namelen);
 
 	mlog(0, "get lockres %s (len %d)\n", lockid, namelen);

fs/ocfs2/dlm/dlmrecovery.c

@@ -2285,7 +2285,8 @@ again:
 	memset(&lksb, 0, sizeof(lksb));
 
 	ret = dlmlock(dlm, LKM_EXMODE, &lksb, LKM_NOQUEUE|LKM_RECOVERY,
-		      DLM_RECOVERY_LOCK_NAME, dlm_reco_ast, dlm, dlm_reco_bast);
+		      DLM_RECOVERY_LOCK_NAME, DLM_RECOVERY_LOCK_NAME_LEN,
+		      dlm_reco_ast, dlm, dlm_reco_bast);
 
 	mlog(0, "%s: dlmlock($RECOVERY) returned %d, lksb=%d\n",
 	     dlm->name, ret, lksb.status);

fs/ocfs2/dlm/userdlm.c

@@ -102,10 +102,10 @@ static inline void user_recover_from_dlm_error(struct user_lock_res *lockres)
 	spin_unlock(&lockres->l_lock);
 }
 
-#define user_log_dlm_error(_func, _stat, _lockres) do {		\
-	mlog(ML_ERROR, "Dlm error \"%s\" while calling %s on "	\
-		"resource %s: %s\n", dlm_errname(_stat), _func,	\
-		_lockres->l_name, dlm_errmsg(_stat));		\
+#define user_log_dlm_error(_func, _stat, _lockres) do {			\
+	mlog(ML_ERROR, "Dlm error \"%s\" while calling %s on "		\
+		"resource %.*s: %s\n", dlm_errname(_stat), _func,	\
+		_lockres->l_namelen, _lockres->l_name, dlm_errmsg(_stat)); \
 } while (0)
 
 /* WARNING: This function lives in a world where the only three lock
@@ -127,21 +127,22 @@ static void user_ast(void *opaque)
 	struct user_lock_res *lockres = opaque;
 	struct dlm_lockstatus *lksb;
 
-	mlog(0, "AST fired for lockres %s\n", lockres->l_name);
+	mlog(0, "AST fired for lockres %.*s\n", lockres->l_namelen,
+	     lockres->l_name);
 
 	spin_lock(&lockres->l_lock);
 
 	lksb = &(lockres->l_lksb);
 	if (lksb->status != DLM_NORMAL) {
-		mlog(ML_ERROR, "lksb status value of %u on lockres %s\n",
-		     lksb->status, lockres->l_name);
+		mlog(ML_ERROR, "lksb status value of %u on lockres %.*s\n",
+		     lksb->status, lockres->l_namelen, lockres->l_name);
 		spin_unlock(&lockres->l_lock);
 		return;
 	}
 
 	mlog_bug_on_msg(lockres->l_requested == LKM_IVMODE,
-			"Lockres %s, requested ivmode. flags 0x%x\n",
-			lockres->l_name, lockres->l_flags);
+			"Lockres %.*s, requested ivmode. flags 0x%x\n",
+			lockres->l_namelen, lockres->l_name, lockres->l_flags);
 
 	/* we're downconverting. */
 	if (lockres->l_requested < lockres->l_level) {
@@ -213,8 +214,8 @@ static void user_bast(void *opaque, int level)
 {
 	struct user_lock_res *lockres = opaque;
 
-	mlog(0, "Blocking AST fired for lockres %s. Blocking level %d\n",
-		lockres->l_name, level);
+	mlog(0, "Blocking AST fired for lockres %.*s. Blocking level %d\n",
+	     lockres->l_namelen, lockres->l_name, level);
 
 	spin_lock(&lockres->l_lock);
 	lockres->l_flags |= USER_LOCK_BLOCKED;
@@ -231,7 +232,8 @@ static void user_unlock_ast(void *opaque, enum dlm_status status)
 {
 	struct user_lock_res *lockres = opaque;
 
-	mlog(0, "UNLOCK AST called on lock %s\n", lockres->l_name);
+	mlog(0, "UNLOCK AST called on lock %.*s\n", lockres->l_namelen,
+	     lockres->l_name);
 
 	if (status != DLM_NORMAL && status != DLM_CANCELGRANT)
 		mlog(ML_ERROR, "Dlm returns status %d\n", status);
@@ -244,8 +246,6 @@ static void user_unlock_ast(void *opaque, enum dlm_status status)
 	    && !(lockres->l_flags & USER_LOCK_IN_CANCEL)) {
 		lockres->l_level = LKM_IVMODE;
 	} else if (status == DLM_CANCELGRANT) {
-		mlog(0, "Lock %s, cancel fails, flags 0x%x\n",
-		     lockres->l_name, lockres->l_flags);
 		/* We tried to cancel a convert request, but it was
 		 * already granted. Don't clear the busy flag - the
 		 * ast should've done this already. */
@@ -255,8 +255,6 @@ static void user_unlock_ast(void *opaque, enum dlm_status status)
 	} else {
 		BUG_ON(!(lockres->l_flags & USER_LOCK_IN_CANCEL));
 		/* Cancel succeeded, we want to re-queue */
-		mlog(0, "Lock %s, cancel succeeds, flags 0x%x\n",
-		     lockres->l_name, lockres->l_flags);
 		lockres->l_requested = LKM_IVMODE; /* cancel an
 						    * upconvert
 						    * request. */
@@ -287,13 +285,14 @@ static void user_dlm_unblock_lock(void *opaque)
 	struct user_lock_res *lockres = (struct user_lock_res *) opaque;
 	struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);
 
-	mlog(0, "processing lockres %s\n", lockres->l_name);
+	mlog(0, "processing lockres %.*s\n", lockres->l_namelen,
+	     lockres->l_name);
 
 	spin_lock(&lockres->l_lock);
 
 	mlog_bug_on_msg(!(lockres->l_flags & USER_LOCK_QUEUED),
-			"Lockres %s, flags 0x%x\n",
-			lockres->l_name, lockres->l_flags);
+			"Lockres %.*s, flags 0x%x\n",
+			lockres->l_namelen, lockres->l_name, lockres->l_flags);
 
 	/* notice that we don't clear USER_LOCK_BLOCKED here. If it's
 	 * set, we want user_ast clear it. */
@@ -305,22 +304,16 @@ static void user_dlm_unblock_lock(void *opaque)
 	 * flag, and finally we might get another bast which re-queues
 	 * us before our ast for the downconvert is called. */
 	if (!(lockres->l_flags & USER_LOCK_BLOCKED)) {
-		mlog(0, "Lockres %s, flags 0x%x: queued but not blocking\n",
-			lockres->l_name, lockres->l_flags);
 		spin_unlock(&lockres->l_lock);
 		goto drop_ref;
 	}
 
 	if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
-		mlog(0, "lock is in teardown so we do nothing\n");
 		spin_unlock(&lockres->l_lock);
 		goto drop_ref;
 	}
 
 	if (lockres->l_flags & USER_LOCK_BUSY) {
-		mlog(0, "Cancel lock %s, flags 0x%x\n",
-		     lockres->l_name, lockres->l_flags);
-
 		if (lockres->l_flags & USER_LOCK_IN_CANCEL) {
 			spin_unlock(&lockres->l_lock);
 			goto drop_ref;
@@ -372,6 +365,7 @@ static void user_dlm_unblock_lock(void *opaque)
 			 &lockres->l_lksb,
 			 LKM_CONVERT|LKM_VALBLK,
 			 lockres->l_name,
+			 lockres->l_namelen,
 			 user_ast,
 			 lockres,
 			 user_bast);
@@ -420,16 +414,16 @@ int user_dlm_cluster_lock(struct user_lock_res *lockres,
 
 	if (level != LKM_EXMODE &&
 	    level != LKM_PRMODE) {
-		mlog(ML_ERROR, "lockres %s: invalid request!\n",
-		     lockres->l_name);
+		mlog(ML_ERROR, "lockres %.*s: invalid request!\n",
+		     lockres->l_namelen, lockres->l_name);
 		status = -EINVAL;
 		goto bail;
 	}
 
-	mlog(0, "lockres %s: asking for %s lock, passed flags = 0x%x\n",
-		lockres->l_name,
-		(level == LKM_EXMODE) ? "LKM_EXMODE" : "LKM_PRMODE",
-		lkm_flags);
+	mlog(0, "lockres %.*s: asking for %s lock, passed flags = 0x%x\n",
+	     lockres->l_namelen, lockres->l_name,
+	     (level == LKM_EXMODE) ? "LKM_EXMODE" : "LKM_PRMODE",
+	     lkm_flags);
 
 again:
 	if (signal_pending(current)) {
@@ -474,15 +468,13 @@ again:
 		BUG_ON(level == LKM_IVMODE);
 		BUG_ON(level == LKM_NLMODE);
 
-		mlog(0, "lock %s, get lock from %d to level = %d\n",
-			lockres->l_name, lockres->l_level, level);
-
 		/* call dlm_lock to upgrade lock now */
 		status = dlmlock(dlm,
 				 level,
 				 &lockres->l_lksb,
 				 local_flags,
 				 lockres->l_name,
+				 lockres->l_namelen,
 				 user_ast,
 				 lockres,
 				 user_bast);
@@ -498,9 +490,6 @@ again:
 			goto bail;
 		}
 
-		mlog(0, "lock %s, successfull return from dlmlock\n",
-			lockres->l_name);
-
 		user_wait_on_busy_lock(lockres);
 		goto again;
 	}
@@ -508,9 +497,6 @@ again:
 	user_dlm_inc_holders(lockres, level);
 	spin_unlock(&lockres->l_lock);
 
-	mlog(0, "lockres %s: Got %s lock!\n", lockres->l_name,
-		(level == LKM_EXMODE) ? "LKM_EXMODE" : "LKM_PRMODE");
-
 	status = 0;
 bail:
 	return status;
@@ -538,13 +524,11 @@ void user_dlm_cluster_unlock(struct user_lock_res *lockres,
 {
 	if (level != LKM_EXMODE &&
 	    level != LKM_PRMODE) {
-		mlog(ML_ERROR, "lockres %s: invalid request!\n", lockres->l_name);
+		mlog(ML_ERROR, "lockres %.*s: invalid request!\n",
+		     lockres->l_namelen, lockres->l_name);
 		return;
 	}
 
-	mlog(0, "lockres %s: dropping %s lock\n", lockres->l_name,
-		(level == LKM_EXMODE) ? "LKM_EXMODE" : "LKM_PRMODE");
-
 	spin_lock(&lockres->l_lock);
 	user_dlm_dec_holders(lockres, level);
 	__user_dlm_cond_queue_lockres(lockres);
@@ -602,6 +586,7 @@ void user_dlm_lock_res_init(struct user_lock_res *lockres,
 	memcpy(lockres->l_name,
 	       dentry->d_name.name,
 	       dentry->d_name.len);
+	lockres->l_namelen = dentry->d_name.len;
 }
 
 int user_dlm_destroy_lock(struct user_lock_res *lockres)
@@ -609,11 +594,10 @@ int user_dlm_destroy_lock(struct user_lock_res *lockres)
 	int status = -EBUSY;
 	struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);
 
-	mlog(0, "asked to destroy %s\n", lockres->l_name);
+	mlog(0, "asked to destroy %.*s\n", lockres->l_namelen, lockres->l_name);
 
 	spin_lock(&lockres->l_lock);
 	if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
-		mlog(0, "Lock is already torn down\n");
 		spin_unlock(&lockres->l_lock);
 		return 0;
 	}
@@ -623,8 +607,6 @@ int user_dlm_destroy_lock(struct user_lock_res *lockres)
 	while (lockres->l_flags & USER_LOCK_BUSY) {
 		spin_unlock(&lockres->l_lock);
 
-		mlog(0, "lock %s is busy\n", lockres->l_name);
-
 		user_wait_on_busy_lock(lockres);
 
 		spin_lock(&lockres->l_lock);
@@ -632,14 +614,12 @@ int user_dlm_destroy_lock(struct user_lock_res *lockres)
 
 	if (lockres->l_ro_holders || lockres->l_ex_holders) {
 		spin_unlock(&lockres->l_lock);
-		mlog(0, "lock %s has holders\n", lockres->l_name);
 		goto bail;
 	}
 
 	status = 0;
 	if (!(lockres->l_flags & USER_LOCK_ATTACHED)) {
 		spin_unlock(&lockres->l_lock);
-		mlog(0, "lock %s is not attached\n", lockres->l_name);
 		goto bail;
 	}
 
@@ -647,7 +627,6 @@ int user_dlm_destroy_lock(struct user_lock_res *lockres)
 	lockres->l_flags |= USER_LOCK_BUSY;
 	spin_unlock(&lockres->l_lock);
 
-	mlog(0, "unlocking lockres %s\n", lockres->l_name);
 	status = dlmunlock(dlm,
 			   &lockres->l_lksb,
 			   LKM_VALBLK,

fs/ocfs2/dlm/userdlm.h

@@ -53,6 +53,7 @@ struct user_lock_res {
 
 #define USER_DLM_LOCK_ID_MAX_LEN  32
 	char                     l_name[USER_DLM_LOCK_ID_MAX_LEN];
+	int                      l_namelen;
 	int                      l_level;
 	unsigned int             l_ro_holders;
 	unsigned int             l_ex_holders;

fs/ocfs2/dlmglue.h

@@ -27,10 +27,14 @@
 #ifndef DLMGLUE_H
 #define DLMGLUE_H
 
-#define OCFS2_LVB_VERSION 3
+#include "dcache.h"
+
+#define OCFS2_LVB_VERSION 4
 
 struct ocfs2_meta_lvb {
-	__be32       lvb_version;
+	__u8         lvb_version;
+	__u8         lvb_reserved0;
+	__be16       lvb_reserved1;
 	__be32       lvb_iclusters;
 	__be32       lvb_iuid;
 	__be32       lvb_igid;
@@ -41,7 +45,8 @@ struct ocfs2_meta_lvb {
 	__be16       lvb_imode;
 	__be16       lvb_inlink;
 	__be32       lvb_iattr;
-	__be32       lvb_reserved[2];
+	__be32       lvb_igeneration;
+	__be32       lvb_reserved2;
 };
 
 /* ocfs2_meta_lock_full() and ocfs2_data_lock_full() 'arg_flags' flags */
@@ -57,9 +62,14 @@ void ocfs2_dlm_shutdown(struct ocfs2_super *osb);
 void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res);
 void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
 			       enum ocfs2_lock_type type,
+			       unsigned int generation,
 			       struct inode *inode);
+void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl,
+				u64 parent, struct inode *inode);
 void ocfs2_lock_res_free(struct ocfs2_lock_res *res);
 int ocfs2_create_new_inode_locks(struct inode *inode);
+int ocfs2_create_new_lock(struct ocfs2_super *osb,
+			  struct ocfs2_lock_res *lockres, int ex, int local);
 int ocfs2_drop_inode_locks(struct inode *inode);
 int ocfs2_data_lock_full(struct inode *inode,
 			 int write,
@@ -93,7 +103,12 @@ void ocfs2_super_unlock(struct ocfs2_super *osb,
 			int ex);
 int ocfs2_rename_lock(struct ocfs2_super *osb);
 void ocfs2_rename_unlock(struct ocfs2_super *osb);
+int ocfs2_dentry_lock(struct dentry *dentry, int ex);
+void ocfs2_dentry_unlock(struct dentry *dentry, int ex);
+
 void ocfs2_mark_lockres_freeing(struct ocfs2_lock_res *lockres);
+void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
+			       struct ocfs2_lock_res *lockres);
 
 /* for the vote thread */
 void ocfs2_process_blocked_lock(struct ocfs2_super *osb,

fs/ocfs2/export.c

@@ -33,6 +33,7 @@
 
 #include "dir.h"
 #include "dlmglue.h"
+#include "dcache.h"
 #include "export.h"
 #include "inode.h"
 
@@ -57,7 +58,7 @@ static struct dentry *ocfs2_get_dentry(struct super_block *sb, void *vobjp)
 		return ERR_PTR(-ESTALE);
 	}
 
-	inode = ocfs2_iget(OCFS2_SB(sb), handle->ih_blkno);
+	inode = ocfs2_iget(OCFS2_SB(sb), handle->ih_blkno, 0);
 
 	if (IS_ERR(inode)) {
 		mlog_errno(PTR_ERR(inode));
@@ -77,6 +78,7 @@ static struct dentry *ocfs2_get_dentry(struct super_block *sb, void *vobjp)
 		mlog_errno(-ENOMEM);
 		return ERR_PTR(-ENOMEM);
 	}
+	result->d_op = &ocfs2_dentry_ops;
 
 	mlog_exit_ptr(result);
 	return result;
@@ -113,7 +115,7 @@ static struct dentry *ocfs2_get_parent(struct dentry *child)
 		goto bail_unlock;
 	}
 
-	inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno);
+	inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0);
 	if (IS_ERR(inode)) {
 		mlog(ML_ERROR, "Unable to create inode %llu\n",
 		     (unsigned long long)blkno);
@@ -127,6 +129,8 @@ static struct dentry *ocfs2_get_parent(struct dentry *child)
 		parent = ERR_PTR(-ENOMEM);
 	}
 
+	parent->d_op = &ocfs2_dentry_ops;
+
 bail_unlock:
 	ocfs2_meta_unlock(dir, 0);
 

fs/ocfs2/inode.c

@@ -54,8 +54,6 @@
 
 #include "buffer_head_io.h"
 
-#define OCFS2_FI_FLAG_NOWAIT	0x1
-#define OCFS2_FI_FLAG_DELETE	0x2
 struct ocfs2_find_inode_args
 {
 	u64		fi_blkno;
@@ -109,7 +107,7 @@ struct inode *ocfs2_ilookup_for_vote(struct ocfs2_super *osb,
 	return ilookup5(osb->sb, args.fi_ino, ocfs2_find_actor, &args);
 }
 
-struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno)
+struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, int flags)
 {
 	struct inode *inode = NULL;
 	struct super_block *sb = osb->sb;
@@ -127,7 +125,7 @@ struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno)
 	}
 
 	args.fi_blkno = blkno;
-	args.fi_flags = 0;
+	args.fi_flags = flags;
 	args.fi_ino = ino_from_blkno(sb, blkno);
 
 	inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
@@ -297,15 +295,11 @@ int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
 	OCFS2_I(inode)->ip_orphaned_slot = OCFS2_INVALID_SLOT;
 	OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
 
-	if (create_ino)
-		inode->i_ino = ino_from_blkno(inode->i_sb,
-			       le64_to_cpu(fe->i_blkno));
-
-	mlog(0, "blkno = %llu, ino = %lu, create_ino = %s\n",
-	     (unsigned long long)fe->i_blkno, inode->i_ino, create_ino ? "true" : "false");
-
 	inode->i_nlink = le16_to_cpu(fe->i_links_count);
 
+	if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
+		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
+
 	if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
 		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
 		mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
@@ -343,12 +337,28 @@ int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
 		    break;
 	}
 
+	if (create_ino) {
+		inode->i_ino = ino_from_blkno(inode->i_sb,
+			       le64_to_cpu(fe->i_blkno));
+
+		/*
+		 * If we ever want to create system files from kernel,
+		 * the generation argument to
+		 * ocfs2_inode_lock_res_init() will have to change.
+		 */
+		BUG_ON(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL));
+
+		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
+					  OCFS2_LOCK_TYPE_META, 0, inode);
+	}
+
 	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
-				  OCFS2_LOCK_TYPE_RW, inode);
-	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
-				  OCFS2_LOCK_TYPE_META, inode);
+				  OCFS2_LOCK_TYPE_RW, inode->i_generation,
+				  inode);
+
 	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_data_lockres,
-				  OCFS2_LOCK_TYPE_DATA, inode);
+				  OCFS2_LOCK_TYPE_DATA, inode->i_generation,
+				  inode);
 
 	ocfs2_set_inode_flags(inode);
 	inode->i_flags |= S_NOATIME;
@@ -366,15 +376,15 @@ static int ocfs2_read_locked_inode(struct inode *inode,
 	struct ocfs2_super *osb;
 	struct ocfs2_dinode *fe;
 	struct buffer_head *bh = NULL;
-	int status;
-	int sysfile = 0;
+	int status, can_lock;
+	u32 generation = 0;
 
 	mlog_entry("(0x%p, 0x%p)\n", inode, args);
 
 	status = -EINVAL;
 	if (inode == NULL || inode->i_sb == NULL) {
 		mlog(ML_ERROR, "bad inode\n");
-		goto bail;
+		return status;
 	}
 	sb = inode->i_sb;
 	osb = OCFS2_SB(sb);
@@ -382,50 +392,110 @@ static int ocfs2_read_locked_inode(struct inode *inode,
 	if (!args) {
 		mlog(ML_ERROR, "bad inode args\n");
 		make_bad_inode(inode);
-		goto bail;
+		return status;
+	}
+
+	/*
+	 * To improve performance of cold-cache inode stats, we take
+	 * the cluster lock here if possible.
+	 *
+	 * Generally, OCFS2 never trusts the contents of an inode
+	 * unless it's holding a cluster lock, so taking it here isn't
+	 * a correctness issue as much as it is a performance
+	 * improvement.
+	 *
+	 * There are three times when taking the lock is not a good idea:
+	 *
+	 * 1) During startup, before we have initialized the DLM.
+	 *
+	 * 2) If we are reading certain system files which never get
+	 *    cluster locks (local alloc, truncate log).
+	 *
+	 * 3) If the process doing the iget() is responsible for
+	 *    orphan dir recovery. We're holding the orphan dir lock and
+	 *    can get into a deadlock with another process on another
+	 *    node in ->delete_inode().
+	 *
+	 * #1 and #2 can be simply solved by never taking the lock
+	 * here for system files (which are the only type we read
+	 * during mount). It's a heavier approach, but our main
+	 * concern is user-accesible files anyway.
+	 *
+	 * #3 works itself out because we'll eventually take the
+	 * cluster lock before trusting anything anyway.
+	 */
+	can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
+		&& !(args->fi_flags & OCFS2_FI_FLAG_NOLOCK);
+
+	/*
+	 * To maintain backwards compatibility with older versions of
+	 * ocfs2-tools, we still store the generation value for system
+	 * files. The only ones that actually matter to userspace are
+	 * the journals, but it's easier and inexpensive to just flag
+	 * all system files similarly.
+	 */
+	if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
+		generation = osb->fs_generation;
+
+	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
+				  OCFS2_LOCK_TYPE_META,
+				  generation, inode);
+
+	if (can_lock) {
+		status = ocfs2_meta_lock(inode, NULL, NULL, 0);
+		if (status) {
+			make_bad_inode(inode);
+			mlog_errno(status);
+			return status;
+		}
 	}
 
-	/* Read the FE off disk. This is safe because the kernel only
-	 * does one read_inode2 for a new inode, and if it doesn't
-	 * exist yet then nobody can be working on it! */
-	status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0, NULL);
+	status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0,
+				  can_lock ? inode : NULL);
 	if (status < 0) {
 		mlog_errno(status);
-		make_bad_inode(inode);
 		goto bail;
 	}
 
+	status = -EINVAL;
 	fe = (struct ocfs2_dinode *) bh->b_data;
 	if (!OCFS2_IS_VALID_DINODE(fe)) {
 		mlog(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n",
 		     (unsigned long long)fe->i_blkno, 7, fe->i_signature);
-		make_bad_inode(inode);
 		goto bail;
 	}
 
-	if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
-		sysfile = 1;
+	/*
+	 * This is a code bug. Right now the caller needs to
+	 * understand whether it is asking for a system file inode or
+	 * not so the proper lock names can be built.
+	 */
+	mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
+			!!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
+			"Inode %llu: system file state is ambigous\n",
+			(unsigned long long)args->fi_blkno);
 
 	if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
 	    S_ISBLK(le16_to_cpu(fe->i_mode)))
     		inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
 
-	status = -EINVAL;
 	if (ocfs2_populate_inode(inode, fe, 0) < 0) {
 		mlog(ML_ERROR, "populate failed! i_blkno=%llu, i_ino=%lu\n",
 		     (unsigned long long)fe->i_blkno, inode->i_ino);
-		make_bad_inode(inode);
 		goto bail;
 	}
 
 	BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
 
-	if (sysfile)
-	       OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
-
 	status = 0;
 
 bail:
+	if (can_lock)
+		ocfs2_meta_unlock(inode, 0);
+
+	if (status < 0)
+		make_bad_inode(inode);
+
 	if (args && bh)
 		brelse(bh);
 
@@ -898,9 +968,15 @@ void ocfs2_delete_inode(struct inode *inode)
 		goto bail_unlock_inode;
 	}
 
-	/* Mark the inode as successfully deleted. This is important
-	 * for ocfs2_clear_inode as it will check this flag and skip
-	 * any checkpointing work */
+	/*
+	 * Mark the inode as successfully deleted.
+	 *
+	 * This is important for ocfs2_clear_inode() as it will check
+	 * this flag and skip any checkpointing work
+	 *
+	 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
+	 * the LVB for other nodes.
+	 */
 	OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
 
 bail_unlock_inode:
@@ -1025,12 +1101,10 @@ void ocfs2_drop_inode(struct inode *inode)
 	/* Testing ip_orphaned_slot here wouldn't work because we may
 	 * not have gotten a delete_inode vote from any other nodes
 	 * yet. */
-	if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED) {
-		mlog(0, "Inode was orphaned on another node, clearing nlink.\n");
-		inode->i_nlink = 0;
-	}
-
-	generic_drop_inode(inode);
+	if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
+		generic_delete_inode(inode);
+	else
+		generic_drop_inode(inode);
 
 	mlog_exit_void();
 }

fs/ocfs2/inode.h

@@ -122,7 +122,13 @@ struct buffer_head *ocfs2_bread(struct inode *inode, int block,
 void ocfs2_clear_inode(struct inode *inode);
 void ocfs2_delete_inode(struct inode *inode);
 void ocfs2_drop_inode(struct inode *inode);
-struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 feoff);
+
+/* Flags for ocfs2_iget() */
+#define OCFS2_FI_FLAG_NOWAIT	0x1
+#define OCFS2_FI_FLAG_DELETE	0x2
+#define OCFS2_FI_FLAG_SYSFILE	0x4
+#define OCFS2_FI_FLAG_NOLOCK	0x8
+struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 feoff, int flags);
 struct inode *ocfs2_ilookup_for_vote(struct ocfs2_super *osb,
 				     u64 blkno,
 				     int delete_vote);

fs/ocfs2/journal.c

@@ -1493,7 +1493,8 @@ static int ocfs2_queue_orphans(struct ocfs2_super *osb,
 			if (de->name_len == 2 && !strncmp("..", de->name, 2))
 				continue;
 
-			iter = ocfs2_iget(osb, le64_to_cpu(de->inode));
+			iter = ocfs2_iget(osb, le64_to_cpu(de->inode),
+					  OCFS2_FI_FLAG_NOLOCK);
 			if (IS_ERR(iter))
 				continue;
 

fs/ocfs2/namei.c

@@ -179,7 +179,7 @@ static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 	if (status < 0)
 		goto bail_add;
 
-	inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno);
+	inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0);
 	if (IS_ERR(inode)) {
 		mlog(ML_ERROR, "Unable to create inode %llu\n",
 		     (unsigned long long)blkno);
@@ -199,10 +199,32 @@ static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 	spin_unlock(&oi->ip_lock);
 
 bail_add:
-
 	dentry->d_op = &ocfs2_dentry_ops;
 	ret = d_splice_alias(inode, dentry);
 
+	if (inode) {
+		/*
+		 * If d_splice_alias() finds a DCACHE_DISCONNECTED
+		 * dentry, it will d_move() it on top of ourse. The
+		 * return value will indicate this however, so in
+		 * those cases, we switch them around for the locking
+		 * code.
+		 *
+		 * NOTE: This dentry already has ->d_op set from
+		 * ocfs2_get_parent() and ocfs2_get_dentry()
+		 */
+		if (ret)
+			dentry = ret;
+
+		status = ocfs2_dentry_attach_lock(dentry, inode,
+						  OCFS2_I(dir)->ip_blkno);
+		if (status) {
+			mlog_errno(status);
+			ret = ERR_PTR(status);
+			goto bail_unlock;
+		}
+	}
+
 bail_unlock:
 	/* Don't drop the cluster lock until *after* the d_add --
 	 * unlink on another node will message us to remove that
@@ -418,6 +440,13 @@ static int ocfs2_mknod(struct inode *dir,
 		goto leave;
 	}
 
+	status = ocfs2_dentry_attach_lock(dentry, inode,
+					  OCFS2_I(dir)->ip_blkno);
+	if (status) {
+		mlog_errno(status);
+		goto leave;
+	}
+
 	insert_inode_hash(inode);
 	dentry->d_op = &ocfs2_dentry_ops;
 	d_instantiate(dentry, inode);
@@ -725,6 +754,12 @@ static int ocfs2_link(struct dentry *old_dentry,
 		goto bail;
 	}
 
+	err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
+	if (err) {
+		mlog_errno(err);
+		goto bail;
+	}
+
 	atomic_inc(&inode->i_count);
 	dentry->d_op = &ocfs2_dentry_ops;
 	d_instantiate(dentry, inode);
@@ -743,6 +778,23 @@ bail:
 	return err;
 }
 
+/*
+ * Takes and drops an exclusive lock on the given dentry. This will
+ * force other nodes to drop it.
+ */
+static int ocfs2_remote_dentry_delete(struct dentry *dentry)
+{
+	int ret;
+
+	ret = ocfs2_dentry_lock(dentry, 1);
+	if (ret)
+		mlog_errno(ret);
+	else
+		ocfs2_dentry_unlock(dentry, 1);
+
+	return ret;
+}
+
 static int ocfs2_unlink(struct inode *dir,
 			struct dentry *dentry)
 {
@@ -832,8 +884,7 @@ static int ocfs2_unlink(struct inode *dir,
 	else
 		inode->i_nlink--;
 
-	status = ocfs2_request_unlink_vote(inode, dentry,
-					   (unsigned int) inode->i_nlink);
+	status = ocfs2_remote_dentry_delete(dentry);
 	if (status < 0) {
 		/* This vote should succeed under all normal
 		 * circumstances. */
@@ -1019,7 +1070,6 @@ static int ocfs2_rename(struct inode *old_dir,
 	struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
 						    // this is the 1st dirent bh
 	nlink_t old_dir_nlink = old_dir->i_nlink, new_dir_nlink = new_dir->i_nlink;
-	unsigned int links_count;
 
 	/* At some point it might be nice to break this function up a
 	 * bit. */
@@ -1093,23 +1143,26 @@ static int ocfs2_rename(struct inode *old_dir,
 		}
 	}
 
-	if (S_ISDIR(old_inode->i_mode)) {
-		/* Directories actually require metadata updates to
-		 * the directory info so we can't get away with not
-		 * doing node locking on it. */
-		status = ocfs2_meta_lock(old_inode, handle, NULL, 1);
-		if (status < 0) {
-			if (status != -ENOENT)
-				mlog_errno(status);
-			goto bail;
-		}
-
-		status = ocfs2_request_rename_vote(old_inode, old_dentry);
-		if (status < 0) {
+	/*
+	 * Though we don't require an inode meta data update if
+	 * old_inode is not a directory, we lock anyway here to ensure
+	 * the vote thread on other nodes won't have to concurrently
+	 * downconvert the inode and the dentry locks.
+	 */
+	status = ocfs2_meta_lock(old_inode, handle, NULL, 1);
+	if (status < 0) {
+		if (status != -ENOENT)
 			mlog_errno(status);
-			goto bail;
-		}
+		goto bail;
+	}
+
+	status = ocfs2_remote_dentry_delete(old_dentry);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
 
+	if (S_ISDIR(old_inode->i_mode)) {
 		status = -EIO;
 		old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
 		if (!old_inode_de_bh)
@@ -1123,14 +1176,6 @@ static int ocfs2_rename(struct inode *old_dir,
 		if (!new_inode && new_dir!=old_dir &&
 		    new_dir->i_nlink >= OCFS2_LINK_MAX)
 			goto bail;
-	} else {
-		/* Ah, the simple case - we're a file so just send a
-		 * message. */
-		status = ocfs2_request_rename_vote(old_inode, old_dentry);
-		if (status < 0) {
-			mlog_errno(status);
-			goto bail;
-		}
 	}
 
 	status = -ENOENT;
@@ -1202,13 +1247,7 @@ static int ocfs2_rename(struct inode *old_dir,
 			goto bail;
 		}
 
-		if (S_ISDIR(new_inode->i_mode))
-			links_count = 0;
-		else
-			links_count = (unsigned int) (new_inode->i_nlink - 1);
-
-		status = ocfs2_request_unlink_vote(new_inode, new_dentry,
-						   links_count);
+		status = ocfs2_remote_dentry_delete(new_dentry);
 		if (status < 0) {
 			mlog_errno(status);
 			goto bail;
@@ -1387,6 +1426,7 @@ static int ocfs2_rename(struct inode *old_dir,
 		}
 	}
 
+	ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
 	status = 0;
 bail:
 	if (rename_lock)
@@ -1675,6 +1715,12 @@ static int ocfs2_symlink(struct inode *dir,
 		goto bail;
 	}
 
+	status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
+	if (status) {
+		mlog_errno(status);
+		goto bail;
+	}
+
 	insert_inode_hash(inode);
 	dentry->d_op = &ocfs2_dentry_ops;
 	d_instantiate(dentry, inode);

fs/ocfs2/ocfs2_lockid.h

@@ -35,12 +35,15 @@
 #define OCFS2_LOCK_ID_MAX_LEN  32
 #define OCFS2_LOCK_ID_PAD "000000"
 
+#define OCFS2_DENTRY_LOCK_INO_START 18
+
 enum ocfs2_lock_type {
 	OCFS2_LOCK_TYPE_META = 0,
 	OCFS2_LOCK_TYPE_DATA,
 	OCFS2_LOCK_TYPE_SUPER,
 	OCFS2_LOCK_TYPE_RENAME,
 	OCFS2_LOCK_TYPE_RW,
+	OCFS2_LOCK_TYPE_DENTRY,
 	OCFS2_NUM_LOCK_TYPES
 };
 
@@ -63,6 +66,9 @@ static inline char ocfs2_lock_type_char(enum ocfs2_lock_type type)
 		case OCFS2_LOCK_TYPE_RW:
 			c = 'W';
 			break;
+		case OCFS2_LOCK_TYPE_DENTRY:
+			c = 'N';
+			break;
 		default:
 			c = '\0';
 	}
@@ -70,4 +76,23 @@ static inline char ocfs2_lock_type_char(enum ocfs2_lock_type type)
 	return c;
 }
 
+static char *ocfs2_lock_type_strings[] = {
+	[OCFS2_LOCK_TYPE_META] = "Meta",
+	[OCFS2_LOCK_TYPE_DATA] = "Data",
+	[OCFS2_LOCK_TYPE_SUPER] = "Super",
+	[OCFS2_LOCK_TYPE_RENAME] = "Rename",
+	/* Need to differntiate from [R]ename.. serializing writes is the
+	 * important job it does, anyway. */
+	[OCFS2_LOCK_TYPE_RW] = "Write/Read",
+	[OCFS2_LOCK_TYPE_DENTRY] = "Dentry",
+};
+
+static inline const char *ocfs2_lock_type_string(enum ocfs2_lock_type type)
+{
+#ifdef __KERNEL__
+	mlog_bug_on_msg(type >= OCFS2_NUM_LOCK_TYPES, "%d\n", type);
+#endif
+	return ocfs2_lock_type_strings[type];
+}
+
 #endif  /* OCFS2_LOCKID_H */

fs/ocfs2/super.c

@@ -202,7 +202,7 @@ static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 
 	mlog_entry_void();
 
-	new = ocfs2_iget(osb, osb->root_blkno);
+	new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE);
 	if (IS_ERR(new)) {
 		status = PTR_ERR(new);
 		mlog_errno(status);
@@ -210,7 +210,7 @@ static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
 	}
 	osb->root_inode = new;
 
-	new = ocfs2_iget(osb, osb->system_dir_blkno);
+	new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE);
 	if (IS_ERR(new)) {
 		status = PTR_ERR(new);
 		mlog_errno(status);
@@ -682,7 +682,7 @@ static struct file_system_type ocfs2_fs_type = {
 	.kill_sb        = kill_block_super, /* set to the generic one
 					     * right now, but do we
 					     * need to change that? */
-	.fs_flags       = FS_REQUIRES_DEV,
+	.fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
 	.next           = NULL
 };
 

fs/ocfs2/sysfile.c

@@ -28,11 +28,11 @@
 #include <linux/slab.h>
 #include <linux/highmem.h>
 
-#include "ocfs2.h"
-
 #define MLOG_MASK_PREFIX ML_INODE
 #include <cluster/masklog.h>
 
+#include "ocfs2.h"
+
 #include "alloc.h"
 #include "dir.h"
 #include "inode.h"
@@ -115,7 +115,7 @@ static struct inode * _ocfs2_get_system_file_inode(struct ocfs2_super *osb,
 		goto bail;
 	}
 
-	inode = ocfs2_iget(osb, blkno);
+	inode = ocfs2_iget(osb, blkno, OCFS2_FI_FLAG_SYSFILE);
 	if (IS_ERR(inode)) {
 		mlog_errno(PTR_ERR(inode));
 		inode = NULL;

fs/ocfs2/vote.c

@@ -74,9 +74,6 @@ struct ocfs2_vote_msg
 		__be32 v_orphaned_slot;	/* Used during delete votes */
 		__be32 v_nlink;		/* Used during unlink votes */
 	} md1;				/* Message type dependant 1 */
-	__be32 v_unlink_namelen;
-	__be64 v_unlink_parent;
-	u8  v_unlink_dirent[OCFS2_VOTE_FILENAME_LEN];
 };
 
 /* Responses are given these values to maintain backwards
@@ -100,8 +97,6 @@ struct ocfs2_vote_work {
 enum ocfs2_vote_request {
 	OCFS2_VOTE_REQ_INVALID = 0,
 	OCFS2_VOTE_REQ_DELETE,
-	OCFS2_VOTE_REQ_UNLINK,
-	OCFS2_VOTE_REQ_RENAME,
 	OCFS2_VOTE_REQ_MOUNT,
 	OCFS2_VOTE_REQ_UMOUNT,
 	OCFS2_VOTE_REQ_LAST
@@ -261,103 +256,13 @@ done:
 	return response;
 }
 
-static int ocfs2_match_dentry(struct dentry *dentry,
-			      u64 parent_blkno,
-			      unsigned int namelen,
-			      const char *name)
-{
-	struct inode *parent;
-
-	if (!dentry->d_parent) {
-		mlog(0, "Detached from parent.\n");
-		return 0;
-	}
-
-	parent = dentry->d_parent->d_inode;
-	/* Negative parent dentry? */
-	if (!parent)
-		return 0;
-
-	/* Name is in a different directory. */
-	if (OCFS2_I(parent)->ip_blkno != parent_blkno)
-		return 0;
-
-	if (dentry->d_name.len != namelen)
-		return 0;
-
-	/* comparison above guarantees this is safe. */
-	if (memcmp(dentry->d_name.name, name, namelen))
-		return 0;
-
-	return 1;
-}
-
-static void ocfs2_process_dentry_request(struct inode *inode,
-					 int rename,
-					 unsigned int new_nlink,
-					 u64 parent_blkno,
-					 unsigned int namelen,
-					 const char *name)
-{
-	struct dentry *dentry = NULL;
-	struct list_head *p;
-	struct ocfs2_inode_info *oi = OCFS2_I(inode);
-
-	mlog(0, "parent %llu, namelen = %u, name = %.*s\n",
-	     (unsigned long long)parent_blkno, namelen, namelen, name);
-
-	spin_lock(&dcache_lock);
-
-	/* Another node is removing this name from the system. It is
-	 * up to us to find the corresponding dentry and if it exists,
-	 * unhash it from the dcache. */
-	list_for_each(p, &inode->i_dentry) {
-		dentry = list_entry(p, struct dentry, d_alias);
-
-		if (ocfs2_match_dentry(dentry, parent_blkno, namelen, name)) {
-			mlog(0, "dentry found: %.*s\n",
-			     dentry->d_name.len, dentry->d_name.name);
-
-			dget_locked(dentry);
-			break;
-		}
-
-		dentry = NULL;
-	}
-
-	spin_unlock(&dcache_lock);
-
-	if (dentry) {
-		d_delete(dentry);
-		dput(dentry);
-	}
-
-	/* rename votes don't send link counts */
-	if (!rename) {
-		mlog(0, "new_nlink = %u\n", new_nlink);
-
-		/* We don't have the proper locks here to directly
-		 * change i_nlink and besides, the vote is sent
-		 * *before* the operation so it may have failed on the
-		 * other node. This passes a hint to ocfs2_drop_inode
-		 * to force ocfs2_delete_inode, who will take the
-		 * proper cluster locks to sort things out. */
-		if (new_nlink == 0) {
-			spin_lock(&oi->ip_lock);
-			oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
-			spin_unlock(&OCFS2_I(inode)->ip_lock);
-		}
-	}
-}
-
 static void ocfs2_process_vote(struct ocfs2_super *osb,
 			       struct ocfs2_vote_msg *msg)
 {
 	int net_status, vote_response;
 	int orphaned_slot = 0;
-	int rename = 0;
-	unsigned int node_num, generation, new_nlink, namelen;
-	u64 blkno, parent_blkno;
+	unsigned int node_num, generation;
+	u64 blkno;
 	enum ocfs2_vote_request request;
 	struct inode *inode = NULL;
 	struct ocfs2_msg_hdr *hdr = &msg->v_hdr;
@@ -437,18 +342,6 @@ static void ocfs2_process_vote(struct ocfs2_super *osb,
 		vote_response = ocfs2_process_delete_request(inode,
 							     &orphaned_slot);
 		break;
-	case OCFS2_VOTE_REQ_RENAME:
-		rename = 1;
-		/* fall through */
-	case OCFS2_VOTE_REQ_UNLINK:
-		parent_blkno = be64_to_cpu(msg->v_unlink_parent);
-		namelen = be32_to_cpu(msg->v_unlink_namelen);
-		/* new_nlink will be ignored in case of a rename vote */
-		new_nlink = be32_to_cpu(msg->md1.v_nlink);
-		ocfs2_process_dentry_request(inode, rename, new_nlink,
-					     parent_blkno, namelen,
-					     msg->v_unlink_dirent);
-		break;
 	default:
 		mlog(ML_ERROR, "node %u, invalid request: %u\n",
 		     node_num, request);
@@ -889,75 +782,6 @@ int ocfs2_request_delete_vote(struct inode *inode)
 	return status;
 }
 
-static void ocfs2_setup_unlink_vote(struct ocfs2_vote_msg *request,
-				    struct dentry *dentry)
-{
-	struct inode *parent = dentry->d_parent->d_inode;
-
-	/* We need some values which will uniquely identify a dentry
-	 * on the other nodes so that they can find it and run
-	 * d_delete against it. Parent directory block and full name
-	 * should suffice. */
-
-	mlog(0, "unlink/rename request: parent: %llu name: %.*s\n",
-	     (unsigned long long)OCFS2_I(parent)->ip_blkno, dentry->d_name.len,
-	     dentry->d_name.name);
-
-	request->v_unlink_parent = cpu_to_be64(OCFS2_I(parent)->ip_blkno);
-	request->v_unlink_namelen = cpu_to_be32(dentry->d_name.len);
-	memcpy(request->v_unlink_dirent, dentry->d_name.name,
-	       dentry->d_name.len);
-}
-
-int ocfs2_request_unlink_vote(struct inode *inode,
-			      struct dentry *dentry,
-			      unsigned int nlink)
-{
-	int status;
-	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
-	struct ocfs2_vote_msg *request;
-
-	if (dentry->d_name.len > OCFS2_VOTE_FILENAME_LEN)
-		return -ENAMETOOLONG;
-
-	status = -ENOMEM;
-	request = ocfs2_new_vote_request(osb, OCFS2_I(inode)->ip_blkno,
-					 inode->i_generation,
-					 OCFS2_VOTE_REQ_UNLINK, nlink);
-	if (request) {
-		ocfs2_setup_unlink_vote(request, dentry);
-
-		status = ocfs2_request_vote(inode, request, NULL);
-
-		kfree(request);
-	}
-	return status;
-}
-
-int ocfs2_request_rename_vote(struct inode *inode,
-			      struct dentry *dentry)
-{
-	int status;
-	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
-	struct ocfs2_vote_msg *request;
-
-	if (dentry->d_name.len > OCFS2_VOTE_FILENAME_LEN)
-		return -ENAMETOOLONG;
-
-	status = -ENOMEM;
-	request = ocfs2_new_vote_request(osb, OCFS2_I(inode)->ip_blkno,
-					 inode->i_generation,
-					 OCFS2_VOTE_REQ_RENAME, 0);
-	if (request) {
-		ocfs2_setup_unlink_vote(request, dentry);
-
-		status = ocfs2_request_vote(inode, request, NULL);
-
-		kfree(request);
-	}
-	return status;
-}
-
 int ocfs2_request_mount_vote(struct ocfs2_super *osb)
 {
 	int status;

fs/ocfs2/vote.h

@@ -39,11 +39,6 @@ static inline void ocfs2_kick_vote_thread(struct ocfs2_super *osb)
 }
 
 int ocfs2_request_delete_vote(struct inode *inode);
-int ocfs2_request_unlink_vote(struct inode *inode,
-			      struct dentry *dentry,
-			      unsigned int nlink);
-int ocfs2_request_rename_vote(struct inode *inode,
-			      struct dentry *dentry);
 int ocfs2_request_mount_vote(struct ocfs2_super *osb);
 int ocfs2_request_umount_vote(struct ocfs2_super *osb);
 int ocfs2_register_net_handlers(struct ocfs2_super *osb);

include/linux/fs.h

@@ -92,9 +92,10 @@ extern int dir_notify_enable;
 #define FS_REQUIRES_DEV 1 
 #define FS_BINARY_MOUNTDATA 2
 #define FS_REVAL_DOT	16384	/* Check the paths ".", ".." for staleness */
-#define FS_ODD_RENAME	32768	/* Temporary stuff; will go away as soon
-				  * as nfs_rename() will be cleaned up
-				  */
+#define FS_RENAME_DOES_D_MOVE	32768	/* FS will handle d_move()
+					 * during rename() internally.
+					 */
+
 /*
  * These are the fs-independent mount-flags: up to 32 flags are supported
  */