xfs: launder in-memory btree buffers before transaction commit

As we've noted in various places, all current users of in-memory btrees
are online fsck.  Online fsck only stages a btree long enough to rebuild
an ondisk data structure, which means that the in-memory btree is
ephemeral.  Furthermore, if we encounter /any/ errors while updating an
in-memory btree, all we do is tear down all the staged data and return
an errno to userspace.  In-memory btrees need not be transactional, so
their buffers should not be committed to the ondisk log, nor should they
be checkpointed by the AIL.  That's just as well since the ephemeral
nature of the btree means that the buftarg and the buffers may disappear
quickly anyway.

Therefore, we need a way to launder the btree buffers that get attached
to the transaction by the generic btree code.  Because the buffers are
directly mapped to backing file pages, there's no need to bwrite them
back to the tmpfs file.  All we need to do is clean enough of the buffer
log item state so that the bli can be detached from the buffer, remove
the bli from the transaction's log item list, and reset the transaction
dirty state as if the laundered items had never been there.

For simplicity, create xfbtree transaction commit and cancel helpers
that launder the in-memory btree buffers for callers.  Once laundered,
call the write verifier on non-stale buffers to avoid integrity issues,
or punch a hole in the backing file for stale buffers.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>

2 files changed, 122 insertions, 0 deletions

diff --git a/fs/xfs/libxfs/xfs_btree_mem.c b/fs/xfs/libxfs/xfs_btree_mem.c
index cb156e9363a5..036061fe32cc 100644
--- a/fs/xfs/libxfs/xfs_btree_mem.c
+++ b/fs/xfs/libxfs/xfs_btree_mem.c
@@ -226,3 +226,122 @@ xfbtree_get_maxrecs(
 
 	return xfbt->maxrecs[level != 0];
 }
+
+/* If this log item is a buffer item that came from the xfbtree, return it. */
+static inline struct xfs_buf *
+xfbtree_buf_match(
+	struct xfbtree			*xfbt,
+	const struct xfs_log_item	*lip)
+{
+	const struct xfs_buf_log_item	*bli;
+	struct xfs_buf			*bp;
+
+	if (lip->li_type != XFS_LI_BUF)
+		return NULL;
+
+	bli = container_of(lip, struct xfs_buf_log_item, bli_item);
+	bp = bli->bli_buf;
+	if (bp->b_target != xfbt->target)
+		return NULL;
+
+	return bp;
+}
+
+/*
+ * Commit changes to the incore btree immediately by writing all dirty xfbtree
+ * buffers to the backing xfile.  This detaches all xfbtree buffers from the
+ * transaction, even on failure.  The buffer locks are dropped between the
+ * delwri queue and submit, so the caller must synchronize btree access.
+ *
+ * Normally we'd let the buffers commit with the transaction and get written to
+ * the xfile via the log, but online repair stages ephemeral btrees in memory
+ * and uses the btree_staging functions to write new btrees to disk atomically.
+ * The in-memory btree (and its backing store) are discarded at the end of the
+ * repair phase, which means that xfbtree buffers cannot commit with the rest
+ * of a transaction.
+ *
+ * In other words, online repair only needs the transaction to collect buffer
+ * pointers and to avoid buffer deadlocks, not to guarantee consistency of
+ * updates.
+ */
+int
+xfbtree_trans_commit(
+	struct xfbtree		*xfbt,
+	struct xfs_trans	*tp)
+{
+	struct xfs_log_item	*lip, *n;
+	bool			tp_dirty = false;
+	int			error = 0;
+
+	/*
+	 * For each xfbtree buffer attached to the transaction, write the dirty
+	 * buffers to the xfile and release them.
+	 */
+	list_for_each_entry_safe(lip, n, &tp->t_items, li_trans) {
+		struct xfs_buf	*bp = xfbtree_buf_match(xfbt, lip);
+
+		if (!bp) {
+			if (test_bit(XFS_LI_DIRTY, &lip->li_flags))
+				tp_dirty |= true;
+			continue;
+		}
+
+		trace_xfbtree_trans_commit_buf(xfbt, bp);
+
+		xmbuf_trans_bdetach(tp, bp);
+
+		/*
+		 * If the buffer fails verification, note the failure but
+		 * continue walking the transaction items so that we remove all
+		 * ephemeral btree buffers.
+		 */
+		if (!error)
+			error = xmbuf_finalize(bp);
+
+		xfs_buf_relse(bp);
+	}
+
+	/*
+	 * Reset the transaction's dirty flag to reflect the dirty state of the
+	 * log items that are still attached.
+	 */
+	tp->t_flags = (tp->t_flags & ~XFS_TRANS_DIRTY) |
+			(tp_dirty ? XFS_TRANS_DIRTY : 0);
+
+	return error;
+}
+
+/*
+ * Cancel changes to the incore btree by detaching all the xfbtree buffers.
+ * Changes are not undone, so callers must not access the btree ever again.
+ */
+void
+xfbtree_trans_cancel(
+	struct xfbtree		*xfbt,
+	struct xfs_trans	*tp)
+{
+	struct xfs_log_item	*lip, *n;
+	bool			tp_dirty = false;
+
+	list_for_each_entry_safe(lip, n, &tp->t_items, li_trans) {
+		struct xfs_buf	*bp = xfbtree_buf_match(xfbt, lip);
+
+		if (!bp) {
+			if (test_bit(XFS_LI_DIRTY, &lip->li_flags))
+				tp_dirty |= true;
+			continue;
+		}
+
+		trace_xfbtree_trans_cancel_buf(xfbt, bp);
+
+		xmbuf_trans_bdetach(tp, bp);
+		xfs_buf_relse(bp);
+	}
+
+	/*
+	 * Reset the transaction's dirty flag to reflect the dirty state of the
+	 * log items that are still attached.
+	 */
+	tp->t_flags = (tp->t_flags & ~XFS_TRANS_DIRTY) |
+			(tp_dirty ? XFS_TRANS_DIRTY : 0);
+}
diff --git a/fs/xfs/libxfs/xfs_btree_mem.h b/fs/xfs/libxfs/xfs_btree_mem.h
index ecc2ceac3ed4..1c3825786ec8 100644
--- a/fs/xfs/libxfs/xfs_btree_mem.h
+++ b/fs/xfs/libxfs/xfs_btree_mem.h
@@ -65,6 +65,9 @@ int xfbtree_free_block(struct xfs_btree_cur *cur, struct xfs_buf *bp);
 int xfbtree_init(struct xfs_mount *mp, struct xfbtree *xfbt,
 		struct xfs_buftarg *btp, const struct xfs_btree_ops *ops);
 void xfbtree_destroy(struct xfbtree *xfbt);
+
+int xfbtree_trans_commit(struct xfbtree *xfbt, struct xfs_trans *tp);
+void xfbtree_trans_cancel(struct xfbtree *xfbt, struct xfs_trans *tp);
 #else
 # define xfbtree_verify_bno(...)	(false)
 #endif /* CONFIG_XFS_BTREE_IN_MEM */