gfs2: fix scheduling while atomic bug in glocks

commit 20265d9 upstream.

Before this patch, in the unlikely event that gfs2_glock_dq encountered
a withdraw, it would do a wait_on_bit to wait for its journal to be
recovered, but it never released the glock's spin_lock, which caused a
scheduling-while-atomic error.

This patch unlocks the lockref spin_lock before waiting for recovery.

Fixes: 601ef0d ("gfs2: Force withdraw to replay journals and wait for it to finish")
Cc: stable@vger.kernel.org # v5.7+
Reported-by: Alexander Aring <aahringo@redhat.com>
Signed-off-by: Bob Peterson <rpeterso@redhat.com>
Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

fs/gfs2/glock.c

@@ -1457,9 +1457,11 @@ void gfs2_glock_dq(struct gfs2_holder *gh)
 	    glock_blocked_by_withdraw(gl) &&
 	    gh->gh_gl != sdp->sd_jinode_gl) {
 		sdp->sd_glock_dqs_held++;
+		spin_unlock(&gl->gl_lockref.lock);
 		might_sleep();
 		wait_on_bit(&sdp->sd_flags, SDF_WITHDRAW_RECOVERY,
 			    TASK_UNINTERRUPTIBLE);
+		spin_lock(&gl->gl_lockref.lock);
 	}
 	if (gh->gh_flags & GL_NOCACHE)
 		handle_callback(gl, LM_ST_UNLOCKED, 0, false);