1 files changed, 35 insertions, 2 deletions

diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
index d96e2f3..43953e0 100644
--- a/fs/userfaultfd.c
+++ b/fs/userfaultfd.c
@@ -63,6 +63,7 @@ struct userfaultfd_wait_queue {
 	struct uffd_msg msg;
 	wait_queue_t wq;
 	struct userfaultfd_ctx *ctx;
+	bool waken;
 };
 
 struct userfaultfd_wake_range {
@@ -86,6 +87,12 @@ static int userfaultfd_wake_function(wait_queue_t *wq, unsigned mode,
 	if (len && (start > uwq->msg.arg.pagefault.address ||
 		    start + len <= uwq->msg.arg.pagefault.address))
 		goto out;
+	WRITE_ONCE(uwq->waken, true);
+	/*
+	 * The implicit smp_mb__before_spinlock in try_to_wake_up()
+	 * renders uwq->waken visible to other CPUs before the task is
+	 * waken.
+	 */
 	ret = wake_up_state(wq->private, mode);
 	if (ret)
 		/*
@@ -264,6 +271,7 @@ int handle_userfault(struct vm_fault *vmf, unsigned long reason)
 	struct userfaultfd_wait_queue uwq;
 	int ret;
 	bool must_wait, return_to_userland;
+	long blocking_state;
 
 	BUG_ON(!rwsem_is_locked(&mm->mmap_sem));
 
@@ -334,10 +342,13 @@ int handle_userfault(struct vm_fault *vmf, unsigned long reason)
 	uwq.wq.private = current;
 	uwq.msg = userfault_msg(vmf->address, vmf->flags, reason);
 	uwq.ctx = ctx;
+	uwq.waken = false;
 
 	return_to_userland =
 		(vmf->flags & (FAULT_FLAG_USER|FAULT_FLAG_KILLABLE)) ==
 		(FAULT_FLAG_USER|FAULT_FLAG_KILLABLE);
+	blocking_state = return_to_userland ? TASK_INTERRUPTIBLE :
+			 TASK_KILLABLE;
 
 	spin_lock(&ctx->fault_pending_wqh.lock);
 	/*
@@ -350,8 +361,7 @@ int handle_userfault(struct vm_fault *vmf, unsigned long reason)
 	 * following the spin_unlock to happen before the list_add in
 	 * __add_wait_queue.
 	 */
-	set_current_state(return_to_userland ? TASK_INTERRUPTIBLE :
-			  TASK_KILLABLE);
+	set_current_state(blocking_state);
 	spin_unlock(&ctx->fault_pending_wqh.lock);
 
 	must_wait = userfaultfd_must_wait(ctx, vmf->address, vmf->flags,
@@ -364,6 +374,29 @@ int handle_userfault(struct vm_fault *vmf, unsigned long reason)
 		wake_up_poll(&ctx->fd_wqh, POLLIN);
 		schedule();
 		ret |= VM_FAULT_MAJOR;
+
+		/*
+		 * False wakeups can orginate even from rwsem before
+		 * up_read() however userfaults will wait either for a
+		 * targeted wakeup on the specific uwq waitqueue from
+		 * wake_userfault() or for signals or for uffd
+		 * release.
+		 */
+		while (!READ_ONCE(uwq.waken)) {
+			/*
+			 * This needs the full smp_store_mb()
+			 * guarantee as the state write must be
+			 * visible to other CPUs before reading
+			 * uwq.waken from other CPUs.
+			 */
+			set_current_state(blocking_state);
+			if (READ_ONCE(uwq.waken) ||
+			    READ_ONCE(ctx->released) ||
+			    (return_to_userland ? signal_pending(current) :
+			     fatal_signal_pending(current)))
+				break;
+			schedule();
+		}
 	}
 
 	__set_current_state(TASK_RUNNING);