fuse: write inode in fuse_release()

A race between write(2) and close(2) allows pages to be dirtied after
fuse_flush -> write_inode_now().  If these pages are not flushed from
fuse_release(), then there might not be a writable open file later.  So any
remaining dirty pages must be written back before the file is released.

This is a partial revert of the blamed commit.

Reported-by: syzbot+6e1efbd8efaaa6860e91@syzkaller.appspotmail.com
Fixes: 36ea23374d1f ("fuse: write inode in fuse_vma_close() instead of fuse_release()")
Cc: <stable@vger.kernel.org> # v5.16
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

diff --git a/fs/fuse/file.c b/fs/fuse/file.c
index 05caa2b9272e80be0c81898788a7fa3bcbaf0520..60885ff9157cb5113d30b508ff051c6c2b46fd03 100644 (file)
--- a/fs/fuse/file.c
+++ b/fs/fuse/file.c
@@ -338,6 +338,15 @@ static int fuse_open(struct inode *inode, struct file *file)
 
 static int fuse_release(struct inode *inode, struct file *file)
 {
+       struct fuse_conn *fc = get_fuse_conn(inode);
+
+       /*
+        * Dirty pages might remain despite write_inode_now() call from
+        * fuse_flush() due to writes racing with the close.
+        */
+       if (fc->writeback_cache)
+               write_inode_now(inode, 1);
+
        fuse_release_common(file, false);
 
        /* return value is ignored by VFS */