jbd: fix error handling for checkpoint io

When a checkpointing IO fails, current JBD code doesn't check the error
and continue journaling.  This means latest metadata can be lost from both
the journal and filesystem.

This patch leaves the failed metadata blocks in the journal space and
aborts journaling in the case of log_do_checkpoint().  To achieve this, we
need to do:

1. don't remove the failed buffer from the checkpoint list where in
   the case of __try_to_free_cp_buf() because it may be released or
   overwritten by a later transaction
2. log_do_checkpoint() is the last chance, remove the failed buffer
   from the checkpoint list and abort the journal
3. when checkpointing fails, don't update the journal super block to
   prevent the journaled contents from being cleaned.  For safety,
   don't update j_tail and j_tail_sequence either
4. when checkpointing fails, notify this error to the ext3 layer so
   that ext3 don't clear the needs_recovery flag, otherwise the
   journaled contents are ignored and cleaned in the recovery phase
5. if the recovery fails, keep the needs_recovery flag
6. prevent cleanup_journal_tail() from being called between
   __journal_drop_transaction() and journal_abort() (a race issue
   between journal_flush() and __log_wait_for_space()

Signed-off-by: Hidehiro Kawai <hidehiro.kawai.ez@hitachi.com>
Acked-by: Jan Kara <jack@suse.cz>
Cc: <linux-ext4@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

3 files changed, 64 insertions, 20 deletions

diff --git a/fs/jbd/checkpoint.c b/fs/jbd/checkpoint.c
index a5432bb..e292935 100644
--- a/fs/jbd/checkpoint.c
+++ b/fs/jbd/checkpoint.c
@@ -93,7 +93,8 @@ static int __try_to_free_cp_buf(struct journal_head *jh)
 	int ret = 0;
 	struct buffer_head *bh = jh2bh(jh);
 
-	if (jh->b_jlist == BJ_None && !buffer_locked(bh) && !buffer_dirty(bh)) {
+	if (jh->b_jlist == BJ_None && !buffer_locked(bh) &&
+	    !buffer_dirty(bh) && buffer_uptodate(bh)) {
 		JBUFFER_TRACE(jh, "remove from checkpoint list");
 		ret = __journal_remove_checkpoint(jh) + 1;
 		jbd_unlock_bh_state(bh);
@@ -160,21 +161,25 @@ static void jbd_sync_bh(journal_t *journal, struct buffer_head *bh)
  * buffers. Note that we take the buffers in the opposite ordering
  * from the one in which they were submitted for IO.
  *
+ * Return 0 on success, and return <0 if some buffers have failed
+ * to be written out.
+ *
  * Called with j_list_lock held.
  */
-static void __wait_cp_io(journal_t *journal, transaction_t *transaction)
+static int __wait_cp_io(journal_t *journal, transaction_t *transaction)
 {
 	struct journal_head *jh;
 	struct buffer_head *bh;
 	tid_t this_tid;
 	int released = 0;
+	int ret = 0;
 
 	this_tid = transaction->t_tid;
 restart:
 	/* Did somebody clean up the transaction in the meanwhile? */
 	if (journal->j_checkpoint_transactions != transaction ||
 			transaction->t_tid != this_tid)
-		return;
+		return ret;
 	while (!released && transaction->t_checkpoint_io_list) {
 		jh = transaction->t_checkpoint_io_list;
 		bh = jh2bh(jh);
@@ -194,6 +199,9 @@ restart:
 			spin_lock(&journal->j_list_lock);
 			goto restart;
 		}
+		if (unlikely(!buffer_uptodate(bh)))
+			ret = -EIO;
+
 		/*
 		 * Now in whatever state the buffer currently is, we know that
 		 * it has been written out and so we can drop it from the list
@@ -203,6 +211,8 @@ restart:
 		journal_remove_journal_head(bh);
 		__brelse(bh);
 	}
+
+	return ret;
 }
 
 #define NR_BATCH	64
@@ -226,7 +236,8 @@ __flush_batch(journal_t *journal, struct buffer_head **bhs, int *batch_count)
  * Try to flush one buffer from the checkpoint list to disk.
  *
  * Return 1 if something happened which requires us to abort the current
- * scan of the checkpoint list.
+ * scan of the checkpoint list.  Return <0 if the buffer has failed to
+ * be written out.
  *
  * Called with j_list_lock held and drops it if 1 is returned
  * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it
@@ -256,6 +267,9 @@ static int __process_buffer(journal_t *journal, struct journal_head *jh,
 		log_wait_commit(journal, tid);
 		ret = 1;
 	} else if (!buffer_dirty(bh)) {
+		ret = 1;
+		if (unlikely(!buffer_uptodate(bh)))
+			ret = -EIO;
 		J_ASSERT_JH(jh, !buffer_jbddirty(bh));
 		BUFFER_TRACE(bh, "remove from checkpoint");
 		__journal_remove_checkpoint(jh);
@@ -263,7 +277,6 @@ static int __process_buffer(journal_t *journal, struct journal_head *jh,
 		jbd_unlock_bh_state(bh);
 		journal_remove_journal_head(bh);
 		__brelse(bh);
-		ret = 1;
 	} else {
 		/*
 		 * Important: we are about to write the buffer, and
@@ -295,6 +308,7 @@ static int __process_buffer(journal_t *journal, struct journal_head *jh,
  * to disk. We submit larger chunks of data at once.
  *
  * The journal should be locked before calling this function.
+ * Called with j_checkpoint_mutex held.
  */
 int log_do_checkpoint(journal_t *journal)
 {
@@ -318,6 +332,7 @@ int log_do_checkpoint(journal_t *journal)
 	 * OK, we need to start writing disk blocks.  Take one transaction
 	 * and write it.
 	 */
+	result = 0;
 	spin_lock(&journal->j_list_lock);
 	if (!journal->j_checkpoint_transactions)
 		goto out;
@@ -334,7 +349,7 @@ restart:
 		int batch_count = 0;
 		struct buffer_head *bhs[NR_BATCH];
 		struct journal_head *jh;
-		int retry = 0;
+		int retry = 0, err;
 
 		while (!retry && transaction->t_checkpoint_list) {
 			struct buffer_head *bh;
@@ -347,6 +362,8 @@ restart:
 				break;
 			}
 			retry = __process_buffer(journal, jh, bhs,&batch_count);
+			if (retry < 0 && !result)
+				result = retry;
 			if (!retry && (need_resched() ||
 				spin_needbreak(&journal->j_list_lock))) {
 				spin_unlock(&journal->j_list_lock);
@@ -371,14 +388,18 @@ restart:
 		 * Now we have cleaned up the first transaction's checkpoint
 		 * list. Let's clean up the second one
 		 */
-		__wait_cp_io(journal, transaction);
+		err = __wait_cp_io(journal, transaction);
+		if (!result)
+			result = err;
 	}
 out:
 	spin_unlock(&journal->j_list_lock);
-	result = cleanup_journal_tail(journal);
 	if (result < 0)
-		return result;
-	return 0;
+		journal_abort(journal, result);
+	else
+		result = cleanup_journal_tail(journal);
+
+	return (result < 0) ? result : 0;
 }
 
 /*
@@ -394,8 +415,9 @@ out:
  * This is the only part of the journaling code which really needs to be
  * aware of transaction aborts.  Checkpointing involves writing to the
  * main filesystem area rather than to the journal, so it can proceed
- * even in abort state, but we must not update the journal superblock if
- * we have an abort error outstanding.
+ * even in abort state, but we must not update the super block if
+ * checkpointing may have failed.  Otherwise, we would lose some metadata
+ * buffers which should be written-back to the filesystem.
  */
 
 int cleanup_journal_tail(journal_t *journal)
@@ -404,6 +426,9 @@ int cleanup_journal_tail(journal_t *journal)
 	tid_t		first_tid;
 	unsigned long	blocknr, freed;
 
+	if (is_journal_aborted(journal))
+		return 1;
+
 	/* OK, work out the oldest transaction remaining in the log, and
 	 * the log block it starts at.
 	 *
diff --git a/fs/jbd/journal.c b/fs/jbd/journal.c
index aa7143a..9e4fa52 100644
--- a/fs/jbd/journal.c
+++ b/fs/jbd/journal.c
@@ -1121,9 +1121,12 @@ recovery_error:
  *
  * Release a journal_t structure once it is no longer in use by the
  * journaled object.
+ * Return <0 if we couldn't clean up the journal.
  */
-void journal_destroy(journal_t *journal)
+int journal_destroy(journal_t *journal)
 {
+	int err = 0;
+
 	/* Wait for the commit thread to wake up and die. */
 	journal_kill_thread(journal);
 
@@ -1146,11 +1149,16 @@ void journal_destroy(journal_t *journal)
 	J_ASSERT(journal->j_checkpoint_transactions == NULL);
 	spin_unlock(&journal->j_list_lock);
 
-	/* We can now mark the journal as empty. */
-	journal->j_tail = 0;
-	journal->j_tail_sequence = ++journal->j_transaction_sequence;
 	if (journal->j_sb_buffer) {
-		journal_update_superblock(journal, 1);
+		if (!is_journal_aborted(journal)) {
+			/* We can now mark the journal as empty. */
+			journal->j_tail = 0;
+			journal->j_tail_sequence =
+				++journal->j_transaction_sequence;
+			journal_update_superblock(journal, 1);
+		} else {
+			err = -EIO;
+		}
 		brelse(journal->j_sb_buffer);
 	}
 
@@ -1160,6 +1168,8 @@ void journal_destroy(journal_t *journal)
 		journal_destroy_revoke(journal);
 	kfree(journal->j_wbuf);
 	kfree(journal);
+
+	return err;
 }
 
 
@@ -1359,10 +1369,16 @@ int journal_flush(journal_t *journal)
 	spin_lock(&journal->j_list_lock);
 	while (!err && journal->j_checkpoint_transactions != NULL) {
 		spin_unlock(&journal->j_list_lock);
+		mutex_lock(&journal->j_checkpoint_mutex);
 		err = log_do_checkpoint(journal);
+		mutex_unlock(&journal->j_checkpoint_mutex);
 		spin_lock(&journal->j_list_lock);
 	}
 	spin_unlock(&journal->j_list_lock);
+
+	if (is_journal_aborted(journal))
+		return -EIO;
+
 	cleanup_journal_tail(journal);
 
 	/* Finally, mark the journal as really needing no recovery.
@@ -1384,7 +1400,7 @@ int journal_flush(journal_t *journal)
 	J_ASSERT(journal->j_head == journal->j_tail);
 	J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
 	spin_unlock(&journal->j_state_lock);
-	return err;
+	return 0;
 }
 
 /**
diff --git a/fs/jbd/recovery.c b/fs/jbd/recovery.c
index 43bc5e5..db5e982 100644
--- a/fs/jbd/recovery.c
+++ b/fs/jbd/recovery.c
@@ -223,7 +223,7 @@ do {									\
  */
 int journal_recover(journal_t *journal)
 {
-	int			err;
+	int			err, err2;
 	journal_superblock_t *	sb;
 
 	struct recovery_info	info;
@@ -261,7 +261,10 @@ int journal_recover(journal_t *journal)
 	journal->j_transaction_sequence = ++info.end_transaction;
 
 	journal_clear_revoke(journal);
-	sync_blockdev(journal->j_fs_dev);
+	err2 = sync_blockdev(journal->j_fs_dev);
+	if (!err)
+		err = err2;
+
 	return err;
 }