[PATCH] JBD: implement journal->j_free locking

Implement the designed locking around journal->j_free.

Things get a lot better here, too.

fs/jbd/checkpoint.c

@@ -70,25 +70,33 @@ static int __try_to_free_cp_buf(struct journal_head *jh)
 }
 
 /*
- * log_wait_for_space: wait until there is space in the journal.
+ * __log_wait_for_space: wait until there is space in the journal.
  *
- * Called with the journal already locked, but it will be unlocked if we have
- * to wait for a checkpoint to free up some space in the log.
+ * Called under j-state_lock *only*.  It will be unlocked if we have to wait
+ * for a checkpoint to free up some space in the log.
  */
 
-void log_wait_for_space(journal_t *journal, int nblocks)
+void __log_wait_for_space(journal_t *journal, int nblocks)
 {
-	while (log_space_left(journal) < nblocks) {
+	assert_spin_locked(&journal->j_state_lock);
+
+	while (__log_space_left(journal) < nblocks) {
 		if (journal->j_flags & JFS_ABORT)
 			return;
 		unlock_journal(journal);
+		spin_unlock(&journal->j_state_lock);
 		down(&journal->j_checkpoint_sem);
 		lock_journal(journal);
 		
-		/* Test again, another process may have checkpointed
-		 * while we were waiting for the checkpoint lock */
-		if (log_space_left(journal) < nblocks) {
+		/*
+		 * Test again, another process may have checkpointed while we
+		 * were waiting for the checkpoint lock
+		 */
+		spin_lock(&journal->j_state_lock);
+		if (__log_space_left(journal) < nblocks) {
+			spin_unlock(&journal->j_state_lock);
 			log_do_checkpoint(journal, nblocks);
+			spin_lock(&journal->j_state_lock);
 		}
 		up(&journal->j_checkpoint_sem);
 	}
@@ -275,7 +283,7 @@ static int __flush_buffer(journal_t *journal, struct journal_head *jh,
  * Perform an actual checkpoint.  We don't write out only enough to
  * satisfy the current blocked requests: rather we submit a reasonably
  * sized chunk of the outstanding data to disk at once for
- * efficiency.  log_wait_for_space() will retry if we didn't free enough.
+ * efficiency.  __log_wait_for_space() will retry if we didn't free enough.
  * 
  * However, we _do_ take into account the amount requested so that once
  * the IO has been queued, we can return as soon as enough of it has

fs/jbd/journal.c

@@ -392,17 +392,23 @@ int journal_write_metadata_buffer(transaction_t *transaction,
  */
 
 /*
- * log_space_left: Return the number of free blocks left in the journal.
+ * __log_space_left: Return the number of free blocks left in the journal.
  *
  * Called with the journal already locked.
+ *
+ * Called under j_state_lock
  */
 
-int log_space_left (journal_t *journal)
+int __log_space_left(journal_t *journal)
 {
 	int left = journal->j_free;
 
-	/* Be pessimistic here about the number of those free blocks
-	 * which might be required for log descriptor control blocks. */
+	assert_spin_locked(&journal->j_state_lock);
+
+	/*
+	 * Be pessimistic here about the number of those free blocks which
+	 * might be required for log descriptor control blocks.
+	 */
 
 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
 

fs/jbd/transaction.c

@@ -219,12 +219,10 @@ static int start_this_handle(journal_t *journal, handle_t *handle)
 		needed += journal->j_committing_transaction->
 					t_outstanding_credits;
 	
-	if (log_space_left(journal) < needed) {
+	if (__log_space_left(journal) < needed) {
 		jbd_debug(2, "Handle %p waiting for checkpoint...\n", handle);
 		spin_unlock(&transaction->t_handle_lock);
-		spin_unlock(&journal->j_state_lock);
-		log_wait_for_space(journal, needed);
-		spin_lock(&journal->j_state_lock);
+		__log_wait_for_space(journal, needed);
 		goto repeat_locked;
 	}
 
@@ -237,7 +235,7 @@ static int start_this_handle(journal_t *journal, handle_t *handle)
 	transaction->t_handle_count++;
 	jbd_debug(4, "Handle %p given %d credits (total %d, free %d)\n",
 		  handle, nblocks, transaction->t_outstanding_credits,
-		  log_space_left(journal));
+		  __log_space_left(journal));
 	spin_unlock(&transaction->t_handle_lock);
 	spin_unlock(&journal->j_state_lock);
 	unlock_journal(journal);
@@ -330,7 +328,7 @@ int journal_extend(handle_t *handle, int nblocks)
 	int result;
 	int wanted;
 
-	lock_journal (journal);
+	lock_journal(journal);
 
 	result = -EIO;
 	if (is_handle_aborted(handle))
@@ -338,6 +336,8 @@ int journal_extend(handle_t *handle, int nblocks)
 
 	result = 1;
 
+	spin_lock(&journal->j_state_lock);
+
 	/* Don't extend a locked-down transaction! */
 	if (handle->h_transaction->t_state != T_RUNNING) {
 		jbd_debug(3, "denied handle %p %d blocks: "
@@ -345,7 +345,6 @@ int journal_extend(handle_t *handle, int nblocks)
 		goto error_out;
 	}
 
-	lock_kernel();
 	spin_lock(&transaction->t_handle_lock);
 	wanted = transaction->t_outstanding_credits + nblocks;
 	
@@ -355,7 +354,7 @@ int journal_extend(handle_t *handle, int nblocks)
 		goto unlock;
 	}
 
-	if (wanted > log_space_left(journal)) {
+	if (wanted > __log_space_left(journal)) {
 		jbd_debug(3, "denied handle %p %d blocks: "
 			  "insufficient log space\n", handle, nblocks);
 		goto unlock;
@@ -368,9 +367,9 @@ int journal_extend(handle_t *handle, int nblocks)
 	jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
 unlock:
 	spin_unlock(&transaction->t_handle_lock);
-	unlock_kernel();
 error_out:
-	unlock_journal (journal);
+	spin_unlock(&journal->j_state_lock);
+	unlock_journal(journal);
 	return result;
 }
 

include/linux/jbd.h

@@ -1000,18 +1000,19 @@ extern int	   journal_test_revoke(journal_t *, unsigned long, tid_t);
 extern void	   journal_clear_revoke(journal_t *);
 extern void	   journal_brelse_array(struct buffer_head *b[], int n);
 
-/* The log thread user interface:
+/*
+ * The log thread user interface:
  *
  * Request space in the current transaction, and force transaction commit
  * transitions on demand.
  */
 
-extern int	log_space_left (journal_t *); /* Called with journal locked */
+int __log_space_left(journal_t *); /* Called with journal locked */
 extern tid_t	log_start_commit (journal_t *, transaction_t *);
 extern int	log_wait_commit (journal_t *, tid_t);
 extern int	log_do_checkpoint (journal_t *, int);
 
-extern void	log_wait_for_space(journal_t *, int nblocks);
+void __log_wait_for_space(journal_t *, int nblocks);
 extern void	__journal_drop_transaction(journal_t *, transaction_t *);
 extern int	cleanup_journal_tail(journal_t *);