[PATCH] btrfs: fix infinite loop with dio faulting

[Josef Bacik <josef@toxicpanda.com>]

Filipe removed the check for the case where we are constantly trying to
fault in the buffer from user space for DIO reads.  He left it for
writes, but for reads you can end up in this infinite loop trying to
fault in a page that won't fault in.  This is the patch I applied ontop
of his patch, without this I was able to get generic/475 to get stuck
infinite looping.

This patch is currently staged so we can probably just fold this into
his actual patch.

Fixes: 5ad7531dbe67 ("btrfs: fault in pages for direct io reads/writes in a more controlled way")
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
---
 fs/btrfs/file.c | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)

diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index 1876072dee9d..79f866e36368 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -3806,20 +3806,21 @@ static ssize_t btrfs_direct_read(struct kiocb *iocb, struct iov_iter *to)
 		read = ret;
 
 	if (iov_iter_count(to) > 0 && (ret == -EFAULT || ret > 0)) {
-		if (iter_is_iovec(to)) {
-			const size_t left = iov_iter_count(to);
-			const size_t size = dio_fault_in_size(iocb, to, prev_left);
+		const size_t left = iov_iter_count(to);
 
-			fault_in_iov_iter_writeable(to, size);
-			prev_left = left;
-			goto again;
-		} else {
+		if (left == prev_left) {
 			/*
 			 * fault_in_iov_iter_writeable() only works for iovecs,
 			 * return with a partial read and fallback to buffered
 			 * IO for the rest of the range.
 			 */
 			ret = read;
+		} else {
+			const size_t size = dio_fault_in_size(iocb, to, prev_left);
+
+			fault_in_iov_iter_writeable(to, size);
+			prev_left = left;
+			goto again;
 		}
 	}
 	btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
-- 
2.26.3