seccomp: Fix setting loaded filter count during TSYNC

commit b4d8a58 upstream.

The desired behavior is to set the caller's filter count to thread's.
This value is reported via /proc, so this fixes the inaccurate count
exposed to userspace; it is not used for reference counting, etc.

Signed-off-by: Hsuan-Chi Kuo <hsuanchikuo@gmail.com>
Link: https://lore.kernel.org/r/20210304233708.420597-1-hsuanchikuo@gmail.com
Co-developed-by: Wiktor Garbacz <wiktorg@google.com>
Signed-off-by: Wiktor Garbacz <wiktorg@google.com>
Link: https://lore.kernel.org/lkml/20210810125158.329849-1-wiktorg@google.com
Signed-off-by: Kees Cook <keescook@chromium.org>
Cc: stable@vger.kernel.org
Fixes: c818c03 ("seccomp: Report number of loaded filters in /proc/$pid/status")
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

Author: hckuo

kernel/seccomp.c

@@ -511,7 +511,7 @@ static inline void seccomp_sync_threads(unsigned long flags)
 		smp_store_release(&thread->seccomp.filter,
 				  caller->seccomp.filter);
 		atomic_set(&thread->seccomp.filter_count,
-			   atomic_read(&thread->seccomp.filter_count));
+			   atomic_read(&caller->seccomp.filter_count));
 
 		/*
 		 * Don't let an unprivileged task work around