Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
Date:   Fri, 29 Apr 2022 15:43:17 -0700
From:   Kees Cook <keescook@chromium.org>
To:     Sargun Dhillon <sargun@sargun.me>
Cc:     LKML <linux-kernel@vger.kernel.org>,
        Linux Containers <containers@lists.linux-foundation.org>,
        Rodrigo Campos <rodrigo@kinvolk.io>,
        Christian Brauner <christian.brauner@ubuntu.com>,
        Giuseppe Scrivano <gscrivan@redhat.com>,
        Will Drewry <wad@chromium.org>,
        Andy Lutomirski <luto@amacapital.net>,
        Alban Crequy <alban@kinvolk.io>,
        Tycho Andersen <tycho@tycho.pizza>
Subject: Re: [PATCH v3 2/2] selftests/seccomp: Add test for wait killable
 notifier
Message-ID: <202204291541.4438B18A@keescook>
References: <20220429023113.74993-1-sargun@sargun.me>
 <20220429023113.74993-3-sargun@sargun.me>
 <202204291053.E04A367@keescook>
 <20220429223557.GB1267404@ircssh-3.c.rugged-nimbus-611.internal>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20220429223557.GB1267404@ircssh-3.c.rugged-nimbus-611.internal>
Precedence: bulk

On Fri, Apr 29, 2022 at 10:35:57PM +0000, Sargun Dhillon wrote:
> On Fri, Apr 29, 2022 at 11:19:33AM -0700, Kees Cook wrote:
> > On Thu, Apr 28, 2022 at 07:31:13PM -0700, Sargun Dhillon wrote:
> > > +
> > > +	ASSERT_EQ(socketpair(PF_LOCAL, SOCK_SEQPACKET, 0, sk_pair), 0);
> > > +
> > > +	listener = user_notif_syscall(__NR_getppid,
> > > +				      SECCOMP_FILTER_FLAG_NEW_LISTENER |
> > > +				      SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV);
> > > +	ASSERT_GE(listener, 0);
> > > +
> > > +	pid = fork();
> > > +	ASSERT_GE(pid, 0);
> > > +
> > > +	if (pid == 0) {
> > > +		close(sk_pair[0]);
> > > +		handled = sk_pair[1];
> > > +
> > > +		/* Setup the sigaction without SA_RESTART */
> > > +		if (sigaction(SIGUSR1, &new_action, NULL)) {
> > > +			perror("sigaction");
> > > +			exit(1);
> > > +		}
> > > +
> > > +		/* Make sure that the syscall is completed (no EINTR) */
> > > +		ret = syscall(__NR_getppid);
> > > +		exit(ret != USER_NOTIF_MAGIC);
> > > +	}
> > > +
> > > +	while (get_proc_syscall(pid) != __NR_getppid &&
> > > +	       get_proc_stat(pid) != 'S')
> > > +		nanosleep(&delay, NULL);
> > > +
> > > +	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
> > > +	/* Kill the process to make sure it enters the wait_killable state */
> > > +	EXPECT_EQ(kill(pid, SIGUSR1), 0);
> > > +
> > > +	/* TASK_KILLABLE is considered D (Disk Sleep) state */
> > > +	while (get_proc_stat(pid) != 'D')
> > > +		nanosleep(&delay, NULL);
> > 
> > Should a NOWAIT waitpid() happen in this loop to make sure this doesn't
> > spin forever?
> > 
> > i.e. running these tests on a kernel that doesn't have the support
> > shouldn't hang -- yes it'll time out eventually but that's annoying. ;)
> > 
> Wouldn't this bail already because user_notif_syscall would assert out
> since the kernel would reject the unknown flag?

Oh yeah, duh. :P

> I might make this a little helper function, something like:
> static void wait_for_state(struct __test_metadata *_metadata, pid_t pid, char wait_for) {
> 	/* 100 ms */
> 	struct timespec delay = { .tv_nsec = 100000000 };
> 	int status;
> 
> 	while (get_proc_stat(pid) != wait_for) {
> 		ASSERT_EQ(waitpid(pid, &status, WNOHANG), 0) {
> 			if (WIFEXITED(status))
> 				TH_LOG("Process %d exited with error code %d", pid, WEXITSTATUS(status));
> 			else if (WIFSIGNALED(status))
> 				TH_LOG("Process %d exited due to signal %d", pid, WTERMSIG(status));
> 			else
> 				TH_LOG("Process %d exited due to unknown reason", pid);
> 		}
> 		nanosleep(&delay, NULL);
> 	}
> }

Yeah, though as you point out, that is likely overkill. :)

> > > +	EXPECT_EQ(ioctl(listener, SECCOMP_IOCTL_NOTIF_RECV, &req), 0);
> > > +	/* Kill the process with a fatal signal */
> > > +	EXPECT_EQ(kill(pid, SIGTERM), 0);
> > > +
> > > +	EXPECT_EQ(waitpid(pid, &status, 0), pid);
> > > +	EXPECT_EQ(true, WIFSIGNALED(status));
> > > +	EXPECT_EQ(SIGTERM, WTERMSIG(status));
> > > +}
> > 
> > Should there be a test validating the inverse of this, as in _without_
> > SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV, how should the above tests
> > behave?
> Don't we roughly get that from the user_notification_kill_in_middle
> and user_notification_signal?

Yeah, I guess that's true. Cool, cool.

> Although, I might cleanup the user_notification_signal test to disable
> SA_RESTART like these tests.

Sounds good, though maybe that can be a separate patch?

-- 
Kees Cook