seccomp: Fix tracer exit notifications during fatal signals

This fixes a ptrace vs fatal pending signals bug as manifested in
seccomp now that seccomp was reordered to happen after ptrace. The
short version is that seccomp should not attempt to call do_exit()
while fatal signals are pending under a tracer. The existing code was
trying to be as defensively paranoid as possible, but it now ends up
confusing ptrace. Instead, the syscall can just be skipped (which solves
the original concern that the do_exit() was addressing) and normal signal
handling, tracer notification, and process death can happen.

Paraphrasing from the original bug report:

If a tracee task is in a PTRACE_EVENT_SECCOMP trap, or has been resumed
after such a trap but not yet been scheduled, and another task in the
thread-group calls exit_group(), then the tracee task exits without the
ptracer receiving a PTRACE_EVENT_EXIT notification. Test case here:
https://gist.github.com/khuey/3c43ac247c72cef8c956ca73281c9be7

The bug happens because when __seccomp_filter() detects
fatal_signal_pending(), it calls do_exit() without dequeuing the fatal
signal. When do_exit() sends the PTRACE_EVENT_EXIT notification and
that task is descheduled, __schedule() notices that there is a fatal
signal pending and changes its state from TASK_TRACED to TASK_RUNNING.
That prevents the ptracer's waitpid() from returning the ptrace event.
A more detailed analysis is here:
https://github.com/mozilla/rr/issues/1762#issuecomment-237396255.
Reported-by: Robert O'Callahan <robert@ocallahan.org>
Reported-by: Kyle Huey <khuey@kylehuey.com>
Tested-by: Kyle Huey <khuey@kylehuey.com>
Fixes: 93e35efb ("x86/ptrace: run seccomp after ptrace")
Signed-off-by: Kees Cook <keescook@chromium.org>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: James Morris <james.l.morris@oracle.com>

kernel/seccomp.c

@@ -605,12 +605,16 @@ static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
 		ptrace_event(PTRACE_EVENT_SECCOMP, data);
 		/*
 		 * The delivery of a fatal signal during event
-		 * notification may silently skip tracer notification.
-		 * Terminating the task now avoids executing a system
-		 * call that may not be intended.
+		 * notification may silently skip tracer notification,
+		 * which could leave us with a potentially unmodified
+		 * syscall that the tracer would have liked to have
+		 * changed. Since the process is about to die, we just
+		 * force the syscall to be skipped and let the signal
+		 * kill the process and correctly handle any tracer exit
+		 * notifications.
 		 */
 		if (fatal_signal_pending(current))
-			do_exit(SIGSYS);
+			goto skip;
 		/* Check if the tracer forced the syscall to be skipped. */
 		this_syscall = syscall_get_nr(current, task_pt_regs(current));
 		if (this_syscall < 0)