fs: simplify check in mount_setattr_commit()

In order to determine whether we need to call mnt_unhold_writers() in
mount_setattr_commit() we currently do not just check whether
MNT_WRITE_HOLD is set but also if a read-only mount was requested.

However, checking whether MNT_WRITE_HOLD is set is enough. Setting
MNT_WRITE_HOLD requires lock_mount_hash() to be held and it must be
unset before calling unlock_mount_hash(). This guarantees that if we see
MNT_WRITE_HOLD we know that we were the ones who set it earlier. We
don't need to care about why we set it. Plus, leaving this additional
read-only check in makes the code more confusing because it implies that
MNT_WRITE_HOLD could've been set by another thread when it really can't.
Remove it and update the associated comment.

Link: https://lore.kernel.org/r/20220203131411.3093040-6-brauner@kernel.org
Cc: Seth Forshee <seth.forshee@digitalocean.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: linux-fsdevel@vger.kernel.org
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Christian Brauner <brauner@kernel.org>

diff --git a/fs/namespace.c b/fs/namespace.c
index 7e5535ed155d9868c489bfe618a129b56ad0262c..ddae5c08ea8c89f6cd71bf58dc6b1d168352a20b 100644 (file)
--- a/fs/namespace.c
+++ b/fs/namespace.c
@@ -4096,13 +4096,8 @@ static void mount_setattr_commit(struct mount_kattr *kattr,
                        WRITE_ONCE(m->mnt.mnt_flags, flags);
                }
 
-               /*
-                * We either set MNT_READONLY above so make it visible
-                * before ~MNT_WRITE_HOLD or we failed to recursively
-                * apply mount options.
-                */
-               if ((kattr->attr_set & MNT_READONLY) &&
-                   (m->mnt.mnt_flags & MNT_WRITE_HOLD))
+               /* If we had to hold writers unblock them. */
+               if (m->mnt.mnt_flags & MNT_WRITE_HOLD)
                        mnt_unhold_writers(m);
 
                if (!err && kattr->propagation)