---

yaml
---
r: 281456
b: refs/heads/master
c: 0048278
h: refs/heads/master
v: v3

[refs]

@@ -1,2 +1,2 @@
 ---
-refs/heads/master: a9e36da655e54545c3289b2a0700b5c443de0edd
+refs/heads/master: 0048278552e9752fd578c3d8deee756987c10873

trunk/fs/jbd/journal.c

@@ -721,7 +721,6 @@ static journal_t * journal_init_common (void)
 	init_waitqueue_head(&journal->j_wait_checkpoint);
 	init_waitqueue_head(&journal->j_wait_commit);
 	init_waitqueue_head(&journal->j_wait_updates);
-	mutex_init(&journal->j_barrier);
 	mutex_init(&journal->j_checkpoint_mutex);
 	spin_lock_init(&journal->j_revoke_lock);
 	spin_lock_init(&journal->j_list_lock);

trunk/fs/jbd/transaction.c

@@ -426,17 +426,34 @@ int journal_restart(handle_t *handle, int nblocks)
  * void journal_lock_updates () - establish a transaction barrier.
  * @journal:  Journal to establish a barrier on.
  *
- * This locks out any further updates from being started, and blocks
- * until all existing updates have completed, returning only once the
- * journal is in a quiescent state with no updates running.
- *
- * The journal lock should not be held on entry.
+ * This locks out any further updates from being started, and blocks until all
+ * existing updates have completed, returning only once the journal is in a
+ * quiescent state with no updates running.
+ *
+ * We do not use simple mutex for synchronization as there are syscalls which
+ * want to return with filesystem locked and that trips up lockdep. Also
+ * hibernate needs to lock filesystem but locked mutex then blocks hibernation.
+ * Since locking filesystem is rare operation, we use simple counter and
+ * waitqueue for locking.
  */
 void journal_lock_updates(journal_t *journal)
 {
 	DEFINE_WAIT(wait);
 
+wait:
+	/* Wait for previous locked operation to finish */
+	wait_event(journal->j_wait_transaction_locked,
+		   journal->j_barrier_count == 0);
+
 	spin_lock(&journal->j_state_lock);
+	/*
+	 * Check reliably under the lock whether we are the ones winning the race
+	 * and locking the journal
+	 */
+	if (journal->j_barrier_count > 0) {
+		spin_unlock(&journal->j_state_lock);
+		goto wait;
+	}
 	++journal->j_barrier_count;
 
 	/* Wait until there are no running updates */
@@ -460,29 +477,18 @@ void journal_lock_updates(journal_t *journal)
 		spin_lock(&journal->j_state_lock);
 	}
 	spin_unlock(&journal->j_state_lock);
-
-	/*
-	 * We have now established a barrier against other normal updates, but
-	 * we also need to barrier against other journal_lock_updates() calls
-	 * to make sure that we serialise special journal-locked operations
-	 * too.
-	 */
-	mutex_lock(&journal->j_barrier);
 }
 
 /**
  * void journal_unlock_updates (journal_t* journal) - release barrier
  * @journal:  Journal to release the barrier on.
  *
  * Release a transaction barrier obtained with journal_lock_updates().
- *
- * Should be called without the journal lock held.
  */
 void journal_unlock_updates (journal_t *journal)
 {
 	J_ASSERT(journal->j_barrier_count != 0);
 
-	mutex_unlock(&journal->j_barrier);
 	spin_lock(&journal->j_state_lock);
 	--journal->j_barrier_count;
 	spin_unlock(&journal->j_state_lock);

trunk/include/linux/jbd.h

@@ -497,7 +497,6 @@ struct transaction_s
  * @j_format_version: Version of the superblock format
  * @j_state_lock: Protect the various scalars in the journal
  * @j_barrier_count:  Number of processes waiting to create a barrier lock
- * @j_barrier: The barrier lock itself
  * @j_running_transaction: The current running transaction..
  * @j_committing_transaction: the transaction we are pushing to disk
  * @j_checkpoint_transactions: a linked circular list of all transactions
@@ -580,9 +579,6 @@ struct journal_s
 	 */
 	int			j_barrier_count;
 
-	/* The barrier lock itself */
-	struct mutex		j_barrier;
-
 	/*
 	 * Transactions: The current running transaction...
 	 * [j_state_lock] [caller holding open handle]