4 files changed, 62 insertions, 17 deletions

diff --git a/fs/jbd/commit.c b/fs/jbd/commit.c
index 8799207df05..f2b9a571f4c 100644
--- a/fs/jbd/commit.c
+++ b/fs/jbd/commit.c
@@ -392,6 +392,12 @@ void journal_commit_transaction(journal_t *journal)
 	jbd_debug (3, "JBD: commit phase 1\n");
 
 	/*
+	 * Clear revoked flag to reflect there is no revoked buffers
+	 * in the next transaction which is going to be started.
+	 */
+	journal_clear_buffer_revoked_flags(journal);
+
+	/*
 	 * Switch to a new revoke table.
 	 */
 	journal_switch_revoke_table(journal);
diff --git a/fs/jbd/journal.c b/fs/jbd/journal.c
index a96cff0c5f1..59c09f9541b 100644
--- a/fs/jbd/journal.c
+++ b/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);
diff --git a/fs/jbd/revoke.c b/fs/jbd/revoke.c
index 305a9076315..25c713e7071 100644
--- a/fs/jbd/revoke.c
+++ b/fs/jbd/revoke.c
@@ -47,6 +47,10 @@
  *   overwriting the new data.  We don't even need to clear the revoke
  *   bit here.
  *
+ * We cache revoke status of a buffer in the current transaction in b_states
+ * bits.  As the name says, revokevalid flag indicates that the cached revoke
+ * status of a buffer is valid and we can rely on the cached status.
+ *
  * Revoke information on buffers is a tri-state value:
  *
  * RevokeValid clear:	no cached revoke status, need to look it up
@@ -479,6 +483,36 @@ int journal_cancel_revoke(handle_t *handle, struct journal_head *jh)
 	return did_revoke;
 }
 
+/*
+ * journal_clear_revoked_flags clears revoked flag of buffers in
+ * revoke table to reflect there is no revoked buffer in the next
+ * transaction which is going to be started.
+ */
+void journal_clear_buffer_revoked_flags(journal_t *journal)
+{
+	struct jbd_revoke_table_s *revoke = journal->j_revoke;
+	int i = 0;
+
+	for (i = 0; i < revoke->hash_size; i++) {
+		struct list_head *hash_list;
+		struct list_head *list_entry;
+		hash_list = &revoke->hash_table[i];
+
+		list_for_each(list_entry, hash_list) {
+			struct jbd_revoke_record_s *record;
+			struct buffer_head *bh;
+			record = (struct jbd_revoke_record_s *)list_entry;
+			bh = __find_get_block(journal->j_fs_dev,
+					      record->blocknr,
+					      journal->j_blocksize);
+			if (bh) {
+				clear_buffer_revoked(bh);
+				__brelse(bh);
+			}
+		}
+	}
+}
+
 /* journal_switch_revoke table select j_revoke for next transaction
  * we do not want to suspend any processing until all revokes are
  * written -bzzz
diff --git a/fs/jbd/transaction.c b/fs/jbd/transaction.c
index 7e59c6e66f9..7fce94b04bc 100644
--- a/fs/jbd/transaction.c
+++ b/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 */
@@ -460,14 +477,6 @@ 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);
 }
 
 /**
@@ -475,14 +484,11 @@ void journal_lock_updates(journal_t *journal)
  * @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);