4 files changed, 186 insertions, 143 deletions

diff --git a/fs/jbd2/checkpoint.c b/fs/jbd2/checkpoint.c
index 6a79fd0a1a3..2c62c5aae82 100644
--- a/fs/jbd2/checkpoint.c
+++ b/fs/jbd2/checkpoint.c
@@ -97,10 +97,14 @@ static int __try_to_free_cp_buf(struct journal_head *jh)
 
 	if (jh->b_jlist == BJ_None && !buffer_locked(bh) &&
 	    !buffer_dirty(bh) && !buffer_write_io_error(bh)) {
+		/*
+		 * Get our reference so that bh cannot be freed before
+		 * we unlock it
+		 */
+		get_bh(bh);
 		JBUFFER_TRACE(jh, "remove from checkpoint list");
 		ret = __jbd2_journal_remove_checkpoint(jh) + 1;
 		jbd_unlock_bh_state(bh);
-		jbd2_journal_remove_journal_head(bh);
 		BUFFER_TRACE(bh, "release");
 		__brelse(bh);
 	} else {
@@ -223,8 +227,8 @@ restart:
 			spin_lock(&journal->j_list_lock);
 			goto restart;
 		}
+		get_bh(bh);
 		if (buffer_locked(bh)) {
-			atomic_inc(&bh->b_count);
 			spin_unlock(&journal->j_list_lock);
 			jbd_unlock_bh_state(bh);
 			wait_on_buffer(bh);
@@ -243,7 +247,6 @@ restart:
 		 */
 		released = __jbd2_journal_remove_checkpoint(jh);
 		jbd_unlock_bh_state(bh);
-		jbd2_journal_remove_journal_head(bh);
 		__brelse(bh);
 	}
 
@@ -284,7 +287,7 @@ static int __process_buffer(journal_t *journal, struct journal_head *jh,
 	int ret = 0;
 
 	if (buffer_locked(bh)) {
-		atomic_inc(&bh->b_count);
+		get_bh(bh);
 		spin_unlock(&journal->j_list_lock);
 		jbd_unlock_bh_state(bh);
 		wait_on_buffer(bh);
@@ -316,12 +319,12 @@ static int __process_buffer(journal_t *journal, struct journal_head *jh,
 		ret = 1;
 		if (unlikely(buffer_write_io_error(bh)))
 			ret = -EIO;
+		get_bh(bh);
 		J_ASSERT_JH(jh, !buffer_jbddirty(bh));
 		BUFFER_TRACE(bh, "remove from checkpoint");
 		__jbd2_journal_remove_checkpoint(jh);
 		spin_unlock(&journal->j_list_lock);
 		jbd_unlock_bh_state(bh);
-		jbd2_journal_remove_journal_head(bh);
 		__brelse(bh);
 	} else {
 		/*
@@ -554,7 +557,8 @@ int jbd2_cleanup_journal_tail(journal_t *journal)
 /*
  * journal_clean_one_cp_list
  *
- * Find all the written-back checkpoint buffers in the given list and release them.
+ * Find all the written-back checkpoint buffers in the given list and
+ * release them.
  *
  * Called with the journal locked.
  * Called with j_list_lock held.
@@ -663,8 +667,8 @@ out:
  * checkpoint lists.
  *
  * The function returns 1 if it frees the transaction, 0 otherwise.
+ * The function can free jh and bh.
  *
- * This function is called with the journal locked.
  * This function is called with j_list_lock held.
  * This function is called with jbd_lock_bh_state(jh2bh(jh))
  */
@@ -684,13 +688,14 @@ int __jbd2_journal_remove_checkpoint(struct journal_head *jh)
 	}
 	journal = transaction->t_journal;
 
+	JBUFFER_TRACE(jh, "removing from transaction");
 	__buffer_unlink(jh);
 	jh->b_cp_transaction = NULL;
+	jbd2_journal_put_journal_head(jh);
 
 	if (transaction->t_checkpoint_list != NULL ||
 	    transaction->t_checkpoint_io_list != NULL)
 		goto out;
-	JBUFFER_TRACE(jh, "transaction has no more buffers");
 
 	/*
 	 * There is one special case to worry about: if we have just pulled the
@@ -701,10 +706,8 @@ int __jbd2_journal_remove_checkpoint(struct journal_head *jh)
 	 * The locking here around t_state is a bit sleazy.
 	 * See the comment at the end of jbd2_journal_commit_transaction().
 	 */
-	if (transaction->t_state != T_FINISHED) {
-		JBUFFER_TRACE(jh, "belongs to running/committing transaction");
+	if (transaction->t_state != T_FINISHED)
 		goto out;
-	}
 
 	/* OK, that was the last buffer for the transaction: we can now
 	   safely remove this transaction from the log */
@@ -723,7 +726,6 @@ int __jbd2_journal_remove_checkpoint(struct journal_head *jh)
 	wake_up(&journal->j_wait_logspace);
 	ret = 1;
 out:
-	JBUFFER_TRACE(jh, "exit");
 	return ret;
 }
 
@@ -742,6 +744,8 @@ void __jbd2_journal_insert_checkpoint(struct journal_head *jh,
 	J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh)));
 	J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
 
+	/* Get reference for checkpointing transaction */
+	jbd2_journal_grab_journal_head(jh2bh(jh));
 	jh->b_cp_transaction = transaction;
 
 	if (!transaction->t_checkpoint_list) {
diff --git a/fs/jbd2/commit.c b/fs/jbd2/commit.c
index 6e28000a4b2..eef6979821a 100644
--- a/fs/jbd2/commit.c
+++ b/fs/jbd2/commit.c
@@ -219,7 +219,6 @@ static int journal_submit_data_buffers(journal_t *journal,
 			ret = err;
 		spin_lock(&journal->j_list_lock);
 		J_ASSERT(jinode->i_transaction == commit_transaction);
-		commit_transaction->t_flushed_data_blocks = 1;
 		clear_bit(__JI_COMMIT_RUNNING, &jinode->i_flags);
 		smp_mb__after_clear_bit();
 		wake_up_bit(&jinode->i_flags, __JI_COMMIT_RUNNING);
@@ -338,12 +337,6 @@ void jbd2_journal_commit_transaction(journal_t *journal)
 	 * all outstanding updates to complete.
 	 */
 
-#ifdef COMMIT_STATS
-	spin_lock(&journal->j_list_lock);
-	summarise_journal_usage(journal);
-	spin_unlock(&journal->j_list_lock);
-#endif
-
 	/* Do we need to erase the effects of a prior jbd2_journal_flush? */
 	if (journal->j_flags & JBD2_FLUSHED) {
 		jbd_debug(3, "super block updated\n");
@@ -678,12 +671,16 @@ start_journal_io:
 		err = 0;
 	}
 
+	write_lock(&journal->j_state_lock);
+	J_ASSERT(commit_transaction->t_state == T_COMMIT);
+	commit_transaction->t_state = T_COMMIT_DFLUSH;
+	write_unlock(&journal->j_state_lock);
 	/* 
 	 * If the journal is not located on the file system device,
 	 * then we must flush the file system device before we issue
 	 * the commit record
 	 */
-	if (commit_transaction->t_flushed_data_blocks &&
+	if (commit_transaction->t_need_data_flush &&
 	    (journal->j_fs_dev != journal->j_dev) &&
 	    (journal->j_flags & JBD2_BARRIER))
 		blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
@@ -760,8 +757,13 @@ wait_for_iobuf:
                    required. */
 		JBUFFER_TRACE(jh, "file as BJ_Forget");
 		jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget);
-		/* Wake up any transactions which were waiting for this
-		   IO to complete */
+		/*
+		 * Wake up any transactions which were waiting for this IO to
+		 * complete. The barrier must be here so that changes by
+		 * jbd2_journal_file_buffer() take effect before wake_up_bit()
+		 * does the waitqueue check.
+		 */
+		smp_mb();
 		wake_up_bit(&bh->b_state, BH_Unshadow);
 		JBUFFER_TRACE(jh, "brelse shadowed buffer");
 		__brelse(bh);
@@ -800,6 +802,10 @@ wait_for_iobuf:
 		jbd2_journal_abort(journal, err);
 
 	jbd_debug(3, "JBD: commit phase 5\n");
+	write_lock(&journal->j_state_lock);
+	J_ASSERT(commit_transaction->t_state == T_COMMIT_DFLUSH);
+	commit_transaction->t_state = T_COMMIT_JFLUSH;
+	write_unlock(&journal->j_state_lock);
 
 	if (!JBD2_HAS_INCOMPAT_FEATURE(journal,
 				       JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT)) {
@@ -842,10 +848,16 @@ restart_loop:
 	while (commit_transaction->t_forget) {
 		transaction_t *cp_transaction;
 		struct buffer_head *bh;
+		int try_to_free = 0;
 
 		jh = commit_transaction->t_forget;
 		spin_unlock(&journal->j_list_lock);
 		bh = jh2bh(jh);
+		/*
+		 * Get a reference so that bh cannot be freed before we are
+		 * done with it.
+		 */
+		get_bh(bh);
 		jbd_lock_bh_state(bh);
 		J_ASSERT_JH(jh,	jh->b_transaction == commit_transaction);
 
@@ -908,28 +920,27 @@ restart_loop:
 			__jbd2_journal_insert_checkpoint(jh, commit_transaction);
 			if (is_journal_aborted(journal))
 				clear_buffer_jbddirty(bh);
-			JBUFFER_TRACE(jh, "refile for checkpoint writeback");
-			__jbd2_journal_refile_buffer(jh);
-			jbd_unlock_bh_state(bh);
 		} else {
 			J_ASSERT_BH(bh, !buffer_dirty(bh));
-			/* The buffer on BJ_Forget list and not jbddirty means
+			/*
+			 * The buffer on BJ_Forget list and not jbddirty means
 			 * it has been freed by this transaction and hence it
 			 * could not have been reallocated until this
 			 * transaction has committed. *BUT* it could be
 			 * reallocated once we have written all the data to
 			 * disk and before we process the buffer on BJ_Forget
-			 * list. */
-			JBUFFER_TRACE(jh, "refile or unfile freed buffer");
-			__jbd2_journal_refile_buffer(jh);
-			if (!jh->b_transaction) {
-				jbd_unlock_bh_state(bh);
-				 /* needs a brelse */
-				jbd2_journal_remove_journal_head(bh);
-				release_buffer_page(bh);
-			} else
-				jbd_unlock_bh_state(bh);
+			 * list.
+			 */
+			if (!jh->b_next_transaction)
+				try_to_free = 1;
 		}
+		JBUFFER_TRACE(jh, "refile or unfile buffer");
+		__jbd2_journal_refile_buffer(jh);
+		jbd_unlock_bh_state(bh);
+		if (try_to_free)
+			release_buffer_page(bh);	/* Drops bh reference */
+		else
+			__brelse(bh);
 		cond_resched_lock(&journal->j_list_lock);
 	}
 	spin_unlock(&journal->j_list_lock);
@@ -955,7 +966,7 @@ restart_loop:
 
 	jbd_debug(3, "JBD: commit phase 7\n");
 
-	J_ASSERT(commit_transaction->t_state == T_COMMIT);
+	J_ASSERT(commit_transaction->t_state == T_COMMIT_JFLUSH);
 
 	commit_transaction->t_start = jiffies;
 	stats.run.rs_logging = jbd2_time_diff(stats.run.rs_logging,
diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c
index e0ec3db1c39..0dfa5b598e6 100644
--- a/fs/jbd2/journal.c
+++ b/fs/jbd2/journal.c
@@ -479,9 +479,12 @@ int __jbd2_log_space_left(journal_t *journal)
 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
 {
 	/*
-	 * Are we already doing a recent enough commit?
+	 * The only transaction we can possibly wait upon is the
+	 * currently running transaction (if it exists).  Otherwise,
+	 * the target tid must be an old one.
 	 */
-	if (!tid_geq(journal->j_commit_request, target)) {
+	if (journal->j_running_transaction &&
+	    journal->j_running_transaction->t_tid == target) {
 		/*
 		 * We want a new commit: OK, mark the request and wakeup the
 		 * commit thread.  We do _not_ do the commit ourselves.
@@ -493,7 +496,15 @@ int __jbd2_log_start_commit(journal_t *journal, tid_t target)
 			  journal->j_commit_sequence);
 		wake_up(&journal->j_wait_commit);
 		return 1;
-	}
+	} else if (!tid_geq(journal->j_commit_request, target))
+		/* This should never happen, but if it does, preserve
+		   the evidence before kjournald goes into a loop and
+		   increments j_commit_sequence beyond all recognition. */
+		WARN_ONCE(1, "jbd: bad log_start_commit: %u %u %u %u\n",
+			  journal->j_commit_request,
+			  journal->j_commit_sequence,
+			  target, journal->j_running_transaction ? 
+			  journal->j_running_transaction->t_tid : 0);
 	return 0;
 }
 
@@ -577,6 +588,47 @@ int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
 }
 
 /*
+ * Return 1 if a given transaction has not yet sent barrier request
+ * connected with a transaction commit. If 0 is returned, transaction
+ * may or may not have sent the barrier. Used to avoid sending barrier
+ * twice in common cases.
+ */
+int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
+{
+	int ret = 0;
+	transaction_t *commit_trans;
+
+	if (!(journal->j_flags & JBD2_BARRIER))
+		return 0;
+	read_lock(&journal->j_state_lock);
+	/* Transaction already committed? */
+	if (tid_geq(journal->j_commit_sequence, tid))
+		goto out;
+	commit_trans = journal->j_committing_transaction;
+	if (!commit_trans || commit_trans->t_tid != tid) {
+		ret = 1;
+		goto out;
+	}
+	/*
+	 * Transaction is being committed and we already proceeded to
+	 * submitting a flush to fs partition?
+	 */
+	if (journal->j_fs_dev != journal->j_dev) {
+		if (!commit_trans->t_need_data_flush ||
+		    commit_trans->t_state >= T_COMMIT_DFLUSH)
+			goto out;
+	} else {
+		if (commit_trans->t_state >= T_COMMIT_JFLUSH)
+			goto out;
+	}
+	ret = 1;
+out:
+	read_unlock(&journal->j_state_lock);
+	return ret;
+}
+EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
+
+/*
  * Wait for a specified commit to complete.
  * The caller may not hold the journal lock.
  */
@@ -2026,10 +2078,9 @@ static void journal_free_journal_head(struct journal_head *jh)
  * When a buffer has its BH_JBD bit set it is immune from being released by
  * core kernel code, mainly via ->b_count.
  *
- * A journal_head may be detached from its buffer_head when the journal_head's
- * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
- * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
- * journal_head can be dropped if needed.
+ * A journal_head is detached from its buffer_head when the journal_head's
+ * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
+ * transaction (b_cp_transaction) hold their references to b_jcount.
  *
  * Various places in the kernel want to attach a journal_head to a buffer_head
  * _before_ attaching the journal_head to a transaction.  To protect the
@@ -2042,17 +2093,16 @@ static void journal_free_journal_head(struct journal_head *jh)
  *	(Attach a journal_head if needed.  Increments b_jcount)
  *	struct journal_head *jh = jbd2_journal_add_journal_head(bh);
  *	...
+ *      (Get another reference for transaction)
+ *	jbd2_journal_grab_journal_head(bh);
  *	jh->b_transaction = xxx;
+ *	(Put original reference)
  *	jbd2_journal_put_journal_head(jh);
- *
- * Now, the journal_head's b_jcount is zero, but it is safe from being released
- * because it has a non-zero b_transaction.
  */
 
 /*
  * Give a buffer_head a journal_head.
  *
- * Doesn't need the journal lock.
  * May sleep.
  */
 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
@@ -2116,61 +2166,29 @@ static void __journal_remove_journal_head(struct buffer_head *bh)
 	struct journal_head *jh = bh2jh(bh);
 
 	J_ASSERT_JH(jh, jh->b_jcount >= 0);
-
-	get_bh(bh);
-	if (jh->b_jcount == 0) {
-		if (jh->b_transaction == NULL &&
-				jh->b_next_transaction == NULL &&
-				jh->b_cp_transaction == NULL) {
-			J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
-			J_ASSERT_BH(bh, buffer_jbd(bh));
-			J_ASSERT_BH(bh, jh2bh(jh) == bh);
-			BUFFER_TRACE(bh, "remove journal_head");
-			if (jh->b_frozen_data) {
-				printk(KERN_WARNING "%s: freeing "
-						"b_frozen_data\n",
-						__func__);
-				jbd2_free(jh->b_frozen_data, bh->b_size);
-			}
-			if (jh->b_committed_data) {
-				printk(KERN_WARNING "%s: freeing "
-						"b_committed_data\n",
-						__func__);
-				jbd2_free(jh->b_committed_data, bh->b_size);
-			}
-			bh->b_private = NULL;
-			jh->b_bh = NULL;	/* debug, really */
-			clear_buffer_jbd(bh);
-			__brelse(bh);
-			journal_free_journal_head(jh);
-		} else {
-			BUFFER_TRACE(bh, "journal_head was locked");
-		}
+	J_ASSERT_JH(jh, jh->b_transaction == NULL);
+	J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
+	J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
+	J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
+	J_ASSERT_BH(bh, buffer_jbd(bh));
+	J_ASSERT_BH(bh, jh2bh(jh) == bh);
+	BUFFER_TRACE(bh, "remove journal_head");
+	if (jh->b_frozen_data) {
+		printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
+		jbd2_free(jh->b_frozen_data, bh->b_size);
 	}
+	if (jh->b_committed_data) {
+		printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
+		jbd2_free(jh->b_committed_data, bh->b_size);
+	}
+	bh->b_private = NULL;
+	jh->b_bh = NULL;	/* debug, really */
+	clear_buffer_jbd(bh);
+	journal_free_journal_head(jh);
 }
 
 /*
- * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
- * and has a zero b_jcount then remove and release its journal_head.   If we did
- * see that the buffer is not used by any transaction we also "logically"
- * decrement ->b_count.
- *
- * We in fact take an additional increment on ->b_count as a convenience,
- * because the caller usually wants to do additional things with the bh
- * after calling here.
- * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
- * time.  Once the caller has run __brelse(), the buffer is eligible for
- * reaping by try_to_free_buffers().
- */
-void jbd2_journal_remove_journal_head(struct buffer_head *bh)
-{
-	jbd_lock_bh_journal_head(bh);
-	__journal_remove_journal_head(bh);
-	jbd_unlock_bh_journal_head(bh);
-}
-
-/*
- * Drop a reference on the passed journal_head.  If it fell to zero then try to
+ * Drop a reference on the passed journal_head.  If it fell to zero then
  * release the journal_head from the buffer_head.
  */
 void jbd2_journal_put_journal_head(struct journal_head *jh)
@@ -2180,11 +2198,12 @@ void jbd2_journal_put_journal_head(struct journal_head *jh)
 	jbd_lock_bh_journal_head(bh);
 	J_ASSERT_JH(jh, jh->b_jcount > 0);
 	--jh->b_jcount;
-	if (!jh->b_jcount && !jh->b_transaction) {
+	if (!jh->b_jcount) {
 		__journal_remove_journal_head(bh);
+		jbd_unlock_bh_journal_head(bh);
 		__brelse(bh);
-	}
-	jbd_unlock_bh_journal_head(bh);
+	} else
+		jbd_unlock_bh_journal_head(bh);
 }
 
 /*
diff --git a/fs/jbd2/transaction.c b/fs/jbd2/transaction.c
index 05fa77a2371..2d7109414cd 100644
--- a/fs/jbd2/transaction.c
+++ b/fs/jbd2/transaction.c
@@ -30,6 +30,7 @@
 #include <linux/module.h>
 
 static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh);
+static void __jbd2_journal_unfile_buffer(struct journal_head *jh);
 
 /*
  * jbd2_get_transaction: obtain a new transaction_t object.
@@ -82,7 +83,7 @@ jbd2_get_transaction(journal_t *journal, transaction_t *transaction)
  */
 
 /*
- * Update transiaction's maximum wait time, if debugging is enabled.
+ * Update transaction's maximum wait time, if debugging is enabled.
  *
  * In order for t_max_wait to be reliable, it must be protected by a
  * lock.  But doing so will mean that start_this_handle() can not be
@@ -91,11 +92,10 @@ jbd2_get_transaction(journal_t *journal, transaction_t *transaction)
  * means that maximum wait time reported by the jbd2_run_stats
  * tracepoint will always be zero.
  */
-static inline void update_t_max_wait(transaction_t *transaction)
+static inline void update_t_max_wait(transaction_t *transaction,
+				     unsigned long ts)
 {
 #ifdef CONFIG_JBD2_DEBUG
-	unsigned long ts = jiffies;
-
 	if (jbd2_journal_enable_debug &&
 	    time_after(transaction->t_start, ts)) {
 		ts = jbd2_time_diff(ts, transaction->t_start);
@@ -121,6 +121,7 @@ static int start_this_handle(journal_t *journal, handle_t *handle,
 	tid_t		tid;
 	int		needed, need_to_start;
 	int		nblocks = handle->h_buffer_credits;
+	unsigned long ts = jiffies;
 
 	if (nblocks > journal->j_max_transaction_buffers) {
 		printk(KERN_ERR "JBD: %s wants too many credits (%d > %d)\n",
@@ -271,7 +272,7 @@ repeat:
 	/* OK, account for the buffers that this operation expects to
 	 * use and add the handle to the running transaction. 
 	 */
-	update_t_max_wait(transaction);
+	update_t_max_wait(transaction, ts);
 	handle->h_transaction = transaction;
 	atomic_inc(&transaction->t_updates);
 	atomic_inc(&transaction->t_handle_count);
@@ -316,7 +317,8 @@ static handle_t *new_handle(int nblocks)
  * This function is visible to journal users (like ext3fs), so is not
  * called with the journal already locked.
  *
- * Return a pointer to a newly allocated handle, or NULL on failure
+ * Return a pointer to a newly allocated handle, or an ERR_PTR() value
+ * on failure.
  */
 handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int gfp_mask)
 {
@@ -763,7 +765,6 @@ repeat:
 	if (!jh->b_transaction) {
 		JBUFFER_TRACE(jh, "no transaction");
 		J_ASSERT_JH(jh, !jh->b_next_transaction);
-		jh->b_transaction = transaction;
 		JBUFFER_TRACE(jh, "file as BJ_Reserved");
 		spin_lock(&journal->j_list_lock);
 		__jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
@@ -813,7 +814,6 @@ out:
  * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
  * @handle: transaction to add buffer modifications to
  * @bh:     bh to be used for metadata writes
- * @credits: variable that will receive credits for the buffer
  *
  * Returns an error code or 0 on success.
  *
@@ -895,8 +895,6 @@ int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
 		 * committed and so it's safe to clear the dirty bit.
 		 */
 		clear_buffer_dirty(jh2bh(jh));
-		jh->b_transaction = transaction;
-
 		/* first access by this transaction */
 		jh->b_modified = 0;
 
@@ -921,8 +919,8 @@ int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
 	 */
 	JBUFFER_TRACE(jh, "cancelling revoke");
 	jbd2_journal_cancel_revoke(handle, jh);
-	jbd2_journal_put_journal_head(jh);
 out:
+	jbd2_journal_put_journal_head(jh);
 	return err;
 }
 
@@ -931,7 +929,6 @@ out:
  *     non-rewindable consequences
  * @handle: transaction
  * @bh: buffer to undo
- * @credits: store the number of taken credits here (if not NULL)
  *
  * Sometimes there is a need to distinguish between metadata which has
  * been committed to disk and that which has not.  The ext3fs code uses
@@ -1231,8 +1228,6 @@ int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh)
 			__jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
 		} else {
 			__jbd2_journal_unfile_buffer(jh);
-			jbd2_journal_remove_journal_head(bh);
-			__brelse(bh);
 			if (!buffer_jbd(bh)) {
 				spin_unlock(&journal->j_list_lock);
 				jbd_unlock_bh_state(bh);
@@ -1555,19 +1550,32 @@ void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
 		mark_buffer_dirty(bh);	/* Expose it to the VM */
 }
 
-void __jbd2_journal_unfile_buffer(struct journal_head *jh)
+/*
+ * Remove buffer from all transactions.
+ *
+ * Called with bh_state lock and j_list_lock
+ *
+ * jh and bh may be already freed when this function returns.
+ */
+static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
 {
 	__jbd2_journal_temp_unlink_buffer(jh);
 	jh->b_transaction = NULL;
+	jbd2_journal_put_journal_head(jh);
 }
 
 void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
 {
-	jbd_lock_bh_state(jh2bh(jh));
+	struct buffer_head *bh = jh2bh(jh);
+
+	/* Get reference so that buffer cannot be freed before we unlock it */
+	get_bh(bh);
+	jbd_lock_bh_state(bh);
 	spin_lock(&journal->j_list_lock);
 	__jbd2_journal_unfile_buffer(jh);
 	spin_unlock(&journal->j_list_lock);
-	jbd_unlock_bh_state(jh2bh(jh));
+	jbd_unlock_bh_state(bh);
+	__brelse(bh);
 }
 
 /*
@@ -1594,8 +1602,6 @@ __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
 		if (jh->b_jlist == BJ_None) {
 			JBUFFER_TRACE(jh, "remove from checkpoint list");
 			__jbd2_journal_remove_checkpoint(jh);
-			jbd2_journal_remove_journal_head(bh);
-			__brelse(bh);
 		}
 	}
 	spin_unlock(&journal->j_list_lock);
@@ -1658,7 +1664,6 @@ int jbd2_journal_try_to_free_buffers(journal_t *journal,
 		/*
 		 * We take our own ref against the journal_head here to avoid
 		 * having to add tons of locking around each instance of
-		 * jbd2_journal_remove_journal_head() and
 		 * jbd2_journal_put_journal_head().
 		 */
 		jh = jbd2_journal_grab_journal_head(bh);
@@ -1696,10 +1701,9 @@ static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
 	int may_free = 1;
 	struct buffer_head *bh = jh2bh(jh);
 
-	__jbd2_journal_unfile_buffer(jh);
-
 	if (jh->b_cp_transaction) {
 		JBUFFER_TRACE(jh, "on running+cp transaction");
+		__jbd2_journal_temp_unlink_buffer(jh);
 		/*
 		 * We don't want to write the buffer anymore, clear the
 		 * bit so that we don't confuse checks in
@@ -1710,8 +1714,7 @@ static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
 		may_free = 0;
 	} else {
 		JBUFFER_TRACE(jh, "on running transaction");
-		jbd2_journal_remove_journal_head(bh);
-		__brelse(bh);
+		__jbd2_journal_unfile_buffer(jh);
 	}
 	return may_free;
 }
@@ -1989,6 +1992,8 @@ void __jbd2_journal_file_buffer(struct journal_head *jh,
 
 	if (jh->b_transaction)
 		__jbd2_journal_temp_unlink_buffer(jh);
+	else
+		jbd2_journal_grab_journal_head(bh);
 	jh->b_transaction = transaction;
 
 	switch (jlist) {
@@ -2040,9 +2045,10 @@ void jbd2_journal_file_buffer(struct journal_head *jh,
  * already started to be used by a subsequent transaction, refile the
  * buffer on that transaction's metadata list.
  *
- * Called under journal->j_list_lock
- *
+ * Called under j_list_lock
  * Called under jbd_lock_bh_state(jh2bh(jh))
+ *
+ * jh and bh may be already free when this function returns
  */
 void __jbd2_journal_refile_buffer(struct journal_head *jh)
 {
@@ -2066,6 +2072,11 @@ void __jbd2_journal_refile_buffer(struct journal_head *jh)
 
 	was_dirty = test_clear_buffer_jbddirty(bh);
 	__jbd2_journal_temp_unlink_buffer(jh);
+	/*
+	 * We set b_transaction here because b_next_transaction will inherit
+	 * our jh reference and thus __jbd2_journal_file_buffer() must not
+	 * take a new one.
+	 */
 	jh->b_transaction = jh->b_next_transaction;
 	jh->b_next_transaction = NULL;
 	if (buffer_freed(bh))
@@ -2082,30 +2093,21 @@ void __jbd2_journal_refile_buffer(struct journal_head *jh)
 }
 
 /*
- * For the unlocked version of this call, also make sure that any
- * hanging journal_head is cleaned up if necessary.
- *
- * __jbd2_journal_refile_buffer is usually called as part of a single locked
- * operation on a buffer_head, in which the caller is probably going to
- * be hooking the journal_head onto other lists.  In that case it is up
- * to the caller to remove the journal_head if necessary.  For the
- * unlocked jbd2_journal_refile_buffer call, the caller isn't going to be
- * doing anything else to the buffer so we need to do the cleanup
- * ourselves to avoid a jh leak.
- *
- * *** The journal_head may be freed by this call! ***
+ * __jbd2_journal_refile_buffer() with necessary locking added. We take our
+ * bh reference so that we can safely unlock bh.
+ *
+ * The jh and bh may be freed by this call.
  */
 void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
 {
 	struct buffer_head *bh = jh2bh(jh);
 
+	/* Get reference so that buffer cannot be freed before we unlock it */
+	get_bh(bh);
 	jbd_lock_bh_state(bh);
 	spin_lock(&journal->j_list_lock);
-
 	__jbd2_journal_refile_buffer(jh);
 	jbd_unlock_bh_state(bh);
-	jbd2_journal_remove_journal_head(bh);
-
 	spin_unlock(&journal->j_list_lock);
 	__brelse(bh);
 }
@@ -2147,6 +2149,13 @@ int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode)
 	    jinode->i_next_transaction == transaction)
 		goto done;
 
+	/*
+	 * We only ever set this variable to 1 so the test is safe. Since
+	 * t_need_data_flush is likely to be set, we do the test to save some
+	 * cacheline bouncing
+	 */
+	if (!transaction->t_need_data_flush)
+		transaction->t_need_data_flush = 1;
 	/* On some different transaction's list - should be
 	 * the committing one */
 	if (jinode->i_transaction) {