jbd2: Change j_state_lock to be a rwlock_t

Lockstat reports have shown that j_state_lock is a major source of
lock contention, especially on systems with more than 4 CPU cores.  So
change it to be a read/write spinlock.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>

fs/ext4/inode.c

@@ -5066,7 +5066,7 @@ struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
 		transaction_t *transaction;
 		tid_t tid;
 
-		spin_lock(&journal->j_state_lock);
+		read_lock(&journal->j_state_lock);
 		if (journal->j_running_transaction)
 			transaction = journal->j_running_transaction;
 		else
@@ -5075,7 +5075,7 @@ struct inode *ext4_iget(struct super_block *sb, unsigned long ino)
 			tid = transaction->t_tid;
 		else
 			tid = journal->j_commit_sequence;
-		spin_unlock(&journal->j_state_lock);
+		read_unlock(&journal->j_state_lock);
 		ei->i_sync_tid = tid;
 		ei->i_datasync_tid = tid;
 	}

fs/ext4/super.c

@@ -3232,7 +3232,7 @@ static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
 	journal->j_min_batch_time = sbi->s_min_batch_time;
 	journal->j_max_batch_time = sbi->s_max_batch_time;
 
-	spin_lock(&journal->j_state_lock);
+	write_lock(&journal->j_state_lock);
 	if (test_opt(sb, BARRIER))
 		journal->j_flags |= JBD2_BARRIER;
 	else
@@ -3241,7 +3241,7 @@ static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
 		journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
 	else
 		journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
-	spin_unlock(&journal->j_state_lock);
+	write_unlock(&journal->j_state_lock);
 }
 
 static journal_t *ext4_get_journal(struct super_block *sb,

fs/jbd2/checkpoint.c

@@ -118,13 +118,13 @@ static int __try_to_free_cp_buf(struct journal_head *jh)
 void __jbd2_log_wait_for_space(journal_t *journal)
 {
 	int nblocks, space_left;
-	assert_spin_locked(&journal->j_state_lock);
+	/* assert_spin_locked(&journal->j_state_lock); */
 
 	nblocks = jbd_space_needed(journal);
 	while (__jbd2_log_space_left(journal) < nblocks) {
 		if (journal->j_flags & JBD2_ABORT)
 			return;
-		spin_unlock(&journal->j_state_lock);
+		write_unlock(&journal->j_state_lock);
 		mutex_lock(&journal->j_checkpoint_mutex);
 
 		/*
@@ -138,7 +138,7 @@ void __jbd2_log_wait_for_space(journal_t *journal)
 		 * filesystem, so abort the journal and leave a stack
 		 * trace for forensic evidence.
 		 */
-		spin_lock(&journal->j_state_lock);
+		write_lock(&journal->j_state_lock);
 		spin_lock(&journal->j_list_lock);
 		nblocks = jbd_space_needed(journal);
 		space_left = __jbd2_log_space_left(journal);
@@ -149,7 +149,7 @@ void __jbd2_log_wait_for_space(journal_t *journal)
 			if (journal->j_committing_transaction)
 				tid = journal->j_committing_transaction->t_tid;
 			spin_unlock(&journal->j_list_lock);
-			spin_unlock(&journal->j_state_lock);
+			write_unlock(&journal->j_state_lock);
 			if (chkpt) {
 				jbd2_log_do_checkpoint(journal);
 			} else if (jbd2_cleanup_journal_tail(journal) == 0) {
@@ -167,7 +167,7 @@ void __jbd2_log_wait_for_space(journal_t *journal)
 				WARN_ON(1);
 				jbd2_journal_abort(journal, 0);
 			}
-			spin_lock(&journal->j_state_lock);
+			write_lock(&journal->j_state_lock);
 		} else {
 			spin_unlock(&journal->j_list_lock);
 		}
@@ -474,7 +474,7 @@ int jbd2_cleanup_journal_tail(journal_t *journal)
 	 * next transaction ID we will write, and where it will
 	 * start. */
 
-	spin_lock(&journal->j_state_lock);
+	write_lock(&journal->j_state_lock);
 	spin_lock(&journal->j_list_lock);
 	transaction = journal->j_checkpoint_transactions;
 	if (transaction) {
@@ -496,7 +496,7 @@ int jbd2_cleanup_journal_tail(journal_t *journal)
 	/* If the oldest pinned transaction is at the tail of the log
            already then there's not much we can do right now. */
 	if (journal->j_tail_sequence == first_tid) {
-		spin_unlock(&journal->j_state_lock);
+		write_unlock(&journal->j_state_lock);
 		return 1;
 	}
 
@@ -516,7 +516,7 @@ int jbd2_cleanup_journal_tail(journal_t *journal)
 	journal->j_free += freed;
 	journal->j_tail_sequence = first_tid;
 	journal->j_tail = blocknr;
-	spin_unlock(&journal->j_state_lock);
+	write_unlock(&journal->j_state_lock);
 
 	/*
 	 * If there is an external journal, we need to make sure that

fs/jbd2/commit.c

@@ -152,9 +152,9 @@ static int journal_submit_commit_record(journal_t *journal,
 		printk(KERN_WARNING
 		       "JBD2: Disabling barriers on %s, "
 		       "not supported by device\n", journal->j_devname);
-		spin_lock(&journal->j_state_lock);
+		write_lock(&journal->j_state_lock);
 		journal->j_flags &= ~JBD2_BARRIER;
-		spin_unlock(&journal->j_state_lock);
+		write_unlock(&journal->j_state_lock);
 
 		/* And try again, without the barrier */
 		lock_buffer(bh);
@@ -182,9 +182,9 @@ static int journal_wait_on_commit_record(journal_t *journal,
 		printk(KERN_WARNING
 		       "JBD2: %s: disabling barries on %s - not supported "
 		       "by device\n", __func__, journal->j_devname);
-		spin_lock(&journal->j_state_lock);
+		write_lock(&journal->j_state_lock);
 		journal->j_flags &= ~JBD2_BARRIER;
-		spin_unlock(&journal->j_state_lock);
+		write_unlock(&journal->j_state_lock);
 
 		lock_buffer(bh);
 		clear_buffer_dirty(bh);
@@ -400,7 +400,7 @@ void jbd2_journal_commit_transaction(journal_t *journal)
 	jbd_debug(1, "JBD: starting commit of transaction %d\n",
 			commit_transaction->t_tid);
 
-	spin_lock(&journal->j_state_lock);
+	write_lock(&journal->j_state_lock);
 	commit_transaction->t_state = T_LOCKED;
 
 	/*
@@ -424,9 +424,9 @@ void jbd2_journal_commit_transaction(journal_t *journal)
 					TASK_UNINTERRUPTIBLE);
 		if (atomic_read(&commit_transaction->t_updates)) {
 			spin_unlock(&commit_transaction->t_handle_lock);
-			spin_unlock(&journal->j_state_lock);
+			write_unlock(&journal->j_state_lock);
 			schedule();
-			spin_lock(&journal->j_state_lock);
+			write_lock(&journal->j_state_lock);
 			spin_lock(&commit_transaction->t_handle_lock);
 		}
 		finish_wait(&journal->j_wait_updates, &wait);
@@ -497,7 +497,7 @@ void jbd2_journal_commit_transaction(journal_t *journal)
 	start_time = ktime_get();
 	commit_transaction->t_log_start = journal->j_head;
 	wake_up(&journal->j_wait_transaction_locked);
-	spin_unlock(&journal->j_state_lock);
+	write_unlock(&journal->j_state_lock);
 
 	jbd_debug (3, "JBD: commit phase 2\n");
 
@@ -519,9 +519,9 @@ void jbd2_journal_commit_transaction(journal_t *journal)
 	 * transaction!  Now comes the tricky part: we need to write out
 	 * metadata.  Loop over the transaction's entire buffer list:
 	 */
-	spin_lock(&journal->j_state_lock);
+	write_lock(&journal->j_state_lock);
 	commit_transaction->t_state = T_COMMIT;
-	spin_unlock(&journal->j_state_lock);
+	write_unlock(&journal->j_state_lock);
 
 	trace_jbd2_commit_logging(journal, commit_transaction);
 	stats.run.rs_logging = jiffies;
@@ -978,7 +978,7 @@ void jbd2_journal_commit_transaction(journal_t *journal)
 	 * __jbd2_journal_drop_transaction(). Otherwise we could race with
 	 * other checkpointing code processing the transaction...
 	 */
-	spin_lock(&journal->j_state_lock);
+	write_lock(&journal->j_state_lock);
 	spin_lock(&journal->j_list_lock);
 	/*
 	 * Now recheck if some buffers did not get attached to the transaction
@@ -986,7 +986,7 @@ void jbd2_journal_commit_transaction(journal_t *journal)
 	 */
 	if (commit_transaction->t_forget) {
 		spin_unlock(&journal->j_list_lock);
-		spin_unlock(&journal->j_state_lock);
+		write_unlock(&journal->j_state_lock);
 		goto restart_loop;
 	}
 
@@ -1038,7 +1038,7 @@ void jbd2_journal_commit_transaction(journal_t *journal)
 				journal->j_average_commit_time*3) / 4;
 	else
 		journal->j_average_commit_time = commit_time;
-	spin_unlock(&journal->j_state_lock);
+	write_unlock(&journal->j_state_lock);
 
 	if (commit_transaction->t_checkpoint_list == NULL &&
 	    commit_transaction->t_checkpoint_io_list == NULL) {

fs/jbd2/transaction.c

@@ -124,36 +124,38 @@ static int start_this_handle(journal_t *journal, handle_t *handle,
 
 	jbd_debug(3, "New handle %p going live.\n", handle);
 
-repeat:
-
 	/*
 	 * We need to hold j_state_lock until t_updates has been incremented,
 	 * for proper journal barrier handling
 	 */
-	spin_lock(&journal->j_state_lock);
-repeat_locked:
+repeat:
+	read_lock(&journal->j_state_lock);
 	if (is_journal_aborted(journal) ||
 	    (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
-		spin_unlock(&journal->j_state_lock);
+		read_unlock(&journal->j_state_lock);
 		kfree(new_transaction);
 		return -EROFS;
 	}
 
 	/* Wait on the journal's transaction barrier if necessary */
 	if (journal->j_barrier_count) {
-		spin_unlock(&journal->j_state_lock);
+		read_unlock(&journal->j_state_lock);
 		wait_event(journal->j_wait_transaction_locked,
 				journal->j_barrier_count == 0);
 		goto repeat;
 	}
 
 	if (!journal->j_running_transaction) {
-		if (!new_transaction) {
-			spin_unlock(&journal->j_state_lock);
+		read_unlock(&journal->j_state_lock);
+		if (!new_transaction)
 			goto alloc_transaction;
+		write_lock(&journal->j_state_lock);
+		if (!journal->j_running_transaction) {
+			jbd2_get_transaction(journal, new_transaction);
+			new_transaction = NULL;
 		}
-		jbd2_get_transaction(journal, new_transaction);
-		new_transaction = NULL;
+		write_unlock(&journal->j_state_lock);
+		goto repeat;
 	}
 
 	transaction = journal->j_running_transaction;
@@ -167,7 +169,7 @@ static int start_this_handle(journal_t *journal, handle_t *handle,
 
 		prepare_to_wait(&journal->j_wait_transaction_locked,
 					&wait, TASK_UNINTERRUPTIBLE);
-		spin_unlock(&journal->j_state_lock);
+		read_unlock(&journal->j_state_lock);
 		schedule();
 		finish_wait(&journal->j_wait_transaction_locked, &wait);
 		goto repeat;
@@ -194,7 +196,7 @@ static int start_this_handle(journal_t *journal, handle_t *handle,
 		prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
 				TASK_UNINTERRUPTIBLE);
 		__jbd2_log_start_commit(journal, transaction->t_tid);
-		spin_unlock(&journal->j_state_lock);
+		read_unlock(&journal->j_state_lock);
 		schedule();
 		finish_wait(&journal->j_wait_transaction_locked, &wait);
 		goto repeat;
@@ -228,8 +230,12 @@ static int start_this_handle(journal_t *journal, handle_t *handle,
 	if (__jbd2_log_space_left(journal) < jbd_space_needed(journal)) {
 		jbd_debug(2, "Handle %p waiting for checkpoint...\n", handle);
 		spin_unlock(&transaction->t_handle_lock);
-		__jbd2_log_wait_for_space(journal);
-		goto repeat_locked;
+		read_unlock(&journal->j_state_lock);
+		write_lock(&journal->j_state_lock);
+		if (__jbd2_log_space_left(journal) < jbd_space_needed(journal))
+			__jbd2_log_wait_for_space(journal);
+		write_unlock(&journal->j_state_lock);
+		goto repeat;
 	}
 
 	/* OK, account for the buffers that this operation expects to
@@ -250,7 +256,7 @@ static int start_this_handle(journal_t *journal, handle_t *handle,
 		  atomic_read(&transaction->t_outstanding_credits),
 		  __jbd2_log_space_left(journal));
 	spin_unlock(&transaction->t_handle_lock);
-	spin_unlock(&journal->j_state_lock);
+	read_unlock(&journal->j_state_lock);
 
 	lock_map_acquire(&handle->h_lockdep_map);
 	kfree(new_transaction);
@@ -362,7 +368,7 @@ int jbd2_journal_extend(handle_t *handle, int nblocks)
 
 	result = 1;
 
-	spin_lock(&journal->j_state_lock);
+	read_lock(&journal->j_state_lock);
 
 	/* Don't extend a locked-down transaction! */
 	if (handle->h_transaction->t_state != T_RUNNING) {
@@ -394,7 +400,7 @@ int jbd2_journal_extend(handle_t *handle, int nblocks)
 unlock:
 	spin_unlock(&transaction->t_handle_lock);
 error_out:
-	spin_unlock(&journal->j_state_lock);
+	read_unlock(&journal->j_state_lock);
 out:
 	return result;
 }
@@ -432,7 +438,7 @@ int jbd2__journal_restart(handle_t *handle, int nblocks, int gfp_mask)
 	J_ASSERT(atomic_read(&transaction->t_updates) > 0);
 	J_ASSERT(journal_current_handle() == handle);
 
-	spin_lock(&journal->j_state_lock);
+	read_lock(&journal->j_state_lock);
 	spin_lock(&transaction->t_handle_lock);
 	atomic_sub(handle->h_buffer_credits,
 		   &transaction->t_outstanding_credits);
@@ -442,7 +448,7 @@ int jbd2__journal_restart(handle_t *handle, int nblocks, int gfp_mask)
 
 	jbd_debug(2, "restarting handle %p\n", handle);
 	__jbd2_log_start_commit(journal, transaction->t_tid);
-	spin_unlock(&journal->j_state_lock);
+	read_unlock(&journal->j_state_lock);
 
 	lock_map_release(&handle->h_lockdep_map);
 	handle->h_buffer_credits = nblocks;
@@ -472,7 +478,7 @@ void jbd2_journal_lock_updates(journal_t *journal)
 {
 	DEFINE_WAIT(wait);
 
-	spin_lock(&journal->j_state_lock);
+	write_lock(&journal->j_state_lock);
 	++journal->j_barrier_count;
 
 	/* Wait until there are no running updates */
@@ -490,12 +496,12 @@ void jbd2_journal_lock_updates(journal_t *journal)
 		prepare_to_wait(&journal->j_wait_updates, &wait,
 				TASK_UNINTERRUPTIBLE);
 		spin_unlock(&transaction->t_handle_lock);
-		spin_unlock(&journal->j_state_lock);
+		write_unlock(&journal->j_state_lock);
 		schedule();
 		finish_wait(&journal->j_wait_updates, &wait);
-		spin_lock(&journal->j_state_lock);
+		write_lock(&journal->j_state_lock);
 	}
-	spin_unlock(&journal->j_state_lock);
+	write_unlock(&journal->j_state_lock);
 
 	/*
 	 * We have now established a barrier against other normal updates, but
@@ -519,9 +525,9 @@ void jbd2_journal_unlock_updates (journal_t *journal)
 	J_ASSERT(journal->j_barrier_count != 0);
 
 	mutex_unlock(&journal->j_barrier);
-	spin_lock(&journal->j_state_lock);
+	write_lock(&journal->j_state_lock);
 	--journal->j_barrier_count;
-	spin_unlock(&journal->j_state_lock);
+	write_unlock(&journal->j_state_lock);
 	wake_up(&journal->j_wait_transaction_locked);
 }
 
@@ -1314,9 +1320,9 @@ int jbd2_journal_stop(handle_t *handle)
 
 		journal->j_last_sync_writer = pid;
 
-		spin_lock(&journal->j_state_lock);
+		read_lock(&journal->j_state_lock);
 		commit_time = journal->j_average_commit_time;
-		spin_unlock(&journal->j_state_lock);
+		read_unlock(&journal->j_state_lock);
 
 		trans_time = ktime_to_ns(ktime_sub(ktime_get(),
 						   transaction->t_start_time));
@@ -1748,7 +1754,7 @@ static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh)
 		goto zap_buffer_unlocked;
 
 	/* OK, we have data buffer in journaled mode */
-	spin_lock(&journal->j_state_lock);
+	write_lock(&journal->j_state_lock);
 	jbd_lock_bh_state(bh);
 	spin_lock(&journal->j_list_lock);
 
@@ -1801,7 +1807,7 @@ static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh)
 			jbd2_journal_put_journal_head(jh);
 			spin_unlock(&journal->j_list_lock);
 			jbd_unlock_bh_state(bh);
-			spin_unlock(&journal->j_state_lock);
+			write_unlock(&journal->j_state_lock);
 			return ret;
 		} else {
 			/* There is no currently-running transaction. So the
@@ -1815,7 +1821,7 @@ static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh)
 				jbd2_journal_put_journal_head(jh);
 				spin_unlock(&journal->j_list_lock);
 				jbd_unlock_bh_state(bh);
-				spin_unlock(&journal->j_state_lock);
+				write_unlock(&journal->j_state_lock);
 				return ret;
 			} else {
 				/* The orphan record's transaction has
@@ -1839,7 +1845,7 @@ static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh)
 		jbd2_journal_put_journal_head(jh);
 		spin_unlock(&journal->j_list_lock);
 		jbd_unlock_bh_state(bh);
-		spin_unlock(&journal->j_state_lock);
+		write_unlock(&journal->j_state_lock);
 		return 0;
 	} else {
 		/* Good, the buffer belongs to the running transaction.
@@ -1858,7 +1864,7 @@ static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh)
 zap_buffer_no_jh:
 	spin_unlock(&journal->j_list_lock);
 	jbd_unlock_bh_state(bh);
-	spin_unlock(&journal->j_state_lock);
+	write_unlock(&journal->j_state_lock);
 zap_buffer_unlocked:
 	clear_buffer_dirty(bh);
 	J_ASSERT_BH(bh, !buffer_jbddirty(bh));
@@ -2165,9 +2171,9 @@ int jbd2_journal_begin_ordered_truncate(journal_t *journal,
 	/* Locks are here just to force reading of recent values, it is
 	 * enough that the transaction was not committing before we started
 	 * a transaction adding the inode to orphan list */
-	spin_lock(&journal->j_state_lock);
+	read_lock(&journal->j_state_lock);
 	commit_trans = journal->j_committing_transaction;
-	spin_unlock(&journal->j_state_lock);
+	read_unlock(&journal->j_state_lock);
 	spin_lock(&journal->j_list_lock);
 	inode_trans = jinode->i_transaction;
 	spin_unlock(&journal->j_list_lock);

fs/ocfs2/journal.c

@@ -760,13 +760,13 @@ void ocfs2_set_journal_params(struct ocfs2_super *osb)
 	if (osb->osb_commit_interval)
 		commit_interval = osb->osb_commit_interval;
 
-	spin_lock(&journal->j_state_lock);
+	write_lock(&journal->j_state_lock);
 	journal->j_commit_interval = commit_interval;
 	if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
 		journal->j_flags |= JBD2_BARRIER;
 	else
 		journal->j_flags &= ~JBD2_BARRIER;
-	spin_unlock(&journal->j_state_lock);
+	write_unlock(&journal->j_state_lock);
 }
 
 int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)

include/linux/jbd2.h

@@ -764,7 +764,7 @@ struct journal_s
 	/*
 	 * Protect the various scalars in the journal
 	 */
-	spinlock_t		j_state_lock;
+	rwlock_t		j_state_lock;
 
 	/*
 	 * Number of processes waiting to create a barrier lock [j_state_lock]