jbd2: call __jbd2_log_start_commit with j_state_lock write locked

On an SMP ARM system running ext4, I've received a report that the
first J_ASSERT in jbd2_journal_commit_transaction has been triggering:

	J_ASSERT(journal->j_running_transaction != NULL);

While investigating possible causes for this problem, I noticed that
__jbd2_log_start_commit() is getting called with j_state_lock only
read-locked, in spite of the fact that it's possible for it might
j_commit_request.  Fix this by grabbing the necessary information so
we can test to see if we need to start a new transaction before
dropping the read lock, and then calling jbd2_log_start_commit() which
will grab the write lock.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>

2 files changed, 21 insertions, 9 deletions

diff --git a/fs/jbd2/journal.c b/fs/jbd2/journal.c
index 9e46869..97e7346 100644
--- a/fs/jbd2/journal.c
+++ b/fs/jbd2/journal.c
@@ -473,7 +473,8 @@ int __jbd2_log_space_left(journal_t *journal)
 }
 
 /*
- * Called under j_state_lock.  Returns true if a transaction commit was started.
+ * Called with j_state_lock locked for writing.
+ * Returns true if a transaction commit was started.
  */
 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
 {
@@ -520,11 +521,13 @@ int jbd2_journal_force_commit_nested(journal_t *journal)
 {
 	transaction_t *transaction = NULL;
 	tid_t tid;
+	int need_to_start = 0;
 
 	read_lock(&journal->j_state_lock);
 	if (journal->j_running_transaction && !current->journal_info) {
 		transaction = journal->j_running_transaction;
-		__jbd2_log_start_commit(journal, transaction->t_tid);
+		if (!tid_geq(journal->j_commit_request, transaction->t_tid))
+			need_to_start = 1;
 	} else if (journal->j_committing_transaction)
 		transaction = journal->j_committing_transaction;
 
@@ -535,6 +538,8 @@ int jbd2_journal_force_commit_nested(journal_t *journal)
 
 	tid = transaction->t_tid;
 	read_unlock(&journal->j_state_lock);
+	if (need_to_start)
+		jbd2_log_start_commit(journal, tid);
 	jbd2_log_wait_commit(journal, tid);
 	return 1;
 }
diff --git a/fs/jbd2/transaction.c b/fs/jbd2/transaction.c
index faad2bd..1d11910 100644
--- a/fs/jbd2/transaction.c
+++ b/fs/jbd2/transaction.c
@@ -117,10 +117,10 @@ static inline void update_t_max_wait(transaction_t *transaction)
 static int start_this_handle(journal_t *journal, handle_t *handle,
 			     int gfp_mask)
 {
-	transaction_t *transaction;
-	int needed;
-	int nblocks = handle->h_buffer_credits;
-	transaction_t *new_transaction = NULL;
+	transaction_t	*transaction, *new_transaction = NULL;
+	tid_t		tid;
+	int		needed, need_to_start;
+	int		nblocks = handle->h_buffer_credits;
 
 	if (nblocks > journal->j_max_transaction_buffers) {
 		printk(KERN_ERR "JBD: %s wants too many credits (%d > %d)\n",
@@ -222,8 +222,11 @@ repeat:
 		atomic_sub(nblocks, &transaction->t_outstanding_credits);
 		prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
 				TASK_UNINTERRUPTIBLE);
-		__jbd2_log_start_commit(journal, transaction->t_tid);
+		tid = transaction->t_tid;
+		need_to_start = !tid_geq(journal->j_commit_request, tid);
 		read_unlock(&journal->j_state_lock);
+		if (need_to_start)
+			jbd2_log_start_commit(journal, tid);
 		schedule();
 		finish_wait(&journal->j_wait_transaction_locked, &wait);
 		goto repeat;
@@ -442,7 +445,8 @@ int jbd2__journal_restart(handle_t *handle, int nblocks, int gfp_mask)
 {
 	transaction_t *transaction = handle->h_transaction;
 	journal_t *journal = transaction->t_journal;
-	int ret;
+	tid_t		tid;
+	int		need_to_start, ret;
 
 	/* If we've had an abort of any type, don't even think about
 	 * actually doing the restart! */
@@ -465,8 +469,11 @@ int jbd2__journal_restart(handle_t *handle, int nblocks, int gfp_mask)
 	spin_unlock(&transaction->t_handle_lock);
 
 	jbd_debug(2, "restarting handle %p\n", handle);
-	__jbd2_log_start_commit(journal, transaction->t_tid);
+	tid = transaction->t_tid;
+	need_to_start = !tid_geq(journal->j_commit_request, tid);
 	read_unlock(&journal->j_state_lock);
+	if (need_to_start)
+		jbd2_log_start_commit(journal, tid);
 
 	lock_map_release(&handle->h_lockdep_map);
 	handle->h_buffer_credits = nblocks;