Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Kyle Huey <me@kylehuey.com>
To:     "Eric W. Biederman" <ebiederm@xmission.com>,
        Jens Axboe <axboe@kernel.dk>,
        Peter Zijlstra <peterz@infradead.org>,
        Marco Elver <elver@google.com>,
        Oleg Nesterov <oleg@redhat.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        Peter Collingbourne <pcc@google.com>,
        Alexey Gladkov <legion@kernel.org>
Cc:     Robert O'Callahan <rocallahan@gmail.com>,
        =?UTF-8?q?Marko=20M=C3=A4kel=C3=A4?= <marko.makela@mariadb.com>,
        linux-kernel@vger.kernel.org
Subject: [PATCH 2/2] signal: after notifying a ptracer of a signal, recheck for pending SIGKILLs
Date:   Sun, 31 Oct 2021 20:41:47 -0700
Message-Id: <20211101034147.6203-3-khuey@kylehuey.com>
In-Reply-To: <20211101034147.6203-1-khuey@kylehuey.com>
References: <20211101034147.6203-1-khuey@kylehuey.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Precedence: bulk

ptrace_signal will release the siglock which could result in the current
task catching a SIGKILL that was not present on our first check in
get_signal. If that happens, bail out of the current signal and go
straight to the SIGKILL. This avoids setting up a frame for a signal
that the ptracer may not be notified of but that will be visible at
PTRACE_EVENT_EXIT.

Signed-off-by: Kyle Huey <khuey@kylehuey.com>
Reported-by: Marko Mäkelä <marko.makela@mariadb.com>
---
 kernel/signal.c | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/kernel/signal.c b/kernel/signal.c
index 4d26ee5c662a..09810777c61b 100644
--- a/kernel/signal.c
+++ b/kernel/signal.c
@@ -2738,6 +2738,13 @@ bool get_signal(struct ksignal *ksig)
 			signr = ptrace_signal(signr, &ksig->info);
 			if (!signr)
 				continue;
+			/*
+			 * After calling the debugger anything could have changed.
+			 * If there's a pending SIGKILL stop processing the current
+			 * signal and skip straight to the end.
+			 */
+			if (signal_group_exit(signal))
+				goto kill;
 		}
 
 		ka = &sighand->action[signr-1];
-- 
2.33.1