[PATCH] gfs2: drain withdraw work after making filesystem read-only

[Matheus Alves de Almeida]

Currently, gfs2_put_super() drains withdraw work before making the
filesystem read-only. A journal write I/O error could enqueue more
work, which does not get drained before releasing filesystem data.

Flush any pending withdraw work after gfs2_make_fs_ro() in
gfs2_put_super().

Reported-by: syzbot+88ffc513fdcbbb39cdf7@xxxxxxxxxxxxxxxxxxxxxxxxx
Closes: https://syzkaller.appspot.com/bug?extid=88ffc513fdcbbb39cdf7
Fixes: f458aafc5c21 ("gfs2: wait for withdraw earlier during unmount")
Signed-off-by: Matheus Alves de Almeida <matheus.aalmeida@xxxxxxxxxxxx>
---
 fs/gfs2/super.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/fs/gfs2/super.c b/fs/gfs2/super.c
index 06302c293..45260280a 100644
--- a/fs/gfs2/super.c
+++ b/fs/gfs2/super.c
@@ -599,9 +599,14 @@ static void gfs2_put_super(struct super_block *sb)
 	/* Wait for withdraw to complete */
 	flush_work(&sdp->sd_withdraw_work);
 
-	if (!sb_rdonly(sb))
+	if (!sb_rdonly(sb)) {
 		gfs2_make_fs_ro(sdp);
-	else {
+		/*
+		 * A journal I/O error during gfs2_make_fs_ro() may queue
+		 * withdraw work.
+		 */
+		flush_work(&sdp->sd_withdraw_work);
+	} else {
 		if (gfs2_withdrawn(sdp))
 			gfs2_destroy_threads(sdp);
 
-- 
2.43.0