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:   ebiederm@xmission.com (Eric W. Biederman)
To:     Oleg Nesterov <oleg@redhat.com>
Cc:     qianli zhao <zhaoqianligood@gmail.com>, christian@brauner.io,
        axboe@kernel.dk, Thomas Gleixner <tglx@linutronix.de>,
        Peter Collingbourne <pcc@google.com>,
        linux-kernel@vger.kernel.org, Qianli Zhao <zhaoqianli@xiaomi.com>
References: <1615296712-175334-1-git-send-email-zhaoqianligood@gmail.com>
        <20210309182657.GA1408@redhat.com>
        <CAPx_LQEQto2fget=kCnfjSCdAp9XCwc2AWgzwxpbxdyCgNs5Mg@mail.gmail.com>
        <m1lfavufep.fsf@fess.ebiederm.org> <20210310173236.GB8973@redhat.com>
Date:   Wed, 10 Mar 2021 16:13:06 -0600
In-Reply-To: <20210310173236.GB8973@redhat.com> (Oleg Nesterov's message of
        "Wed, 10 Mar 2021 18:32:37 +0100")
Message-ID: <m1zgzamzd9.fsf@fess.ebiederm.org>
User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/26.1 (gnu/linux)
MIME-Version: 1.0
Content-Type: text/plain
Subject: Re: [PATCH] exit: trigger panic when init process is set to SIGNAL_GROUP_EXIT
Precedence: bulk

Oleg Nesterov <oleg@redhat.com> writes:

> On 03/10, Eric W. Biederman wrote:
>>
>> 	/* If global init has exited,
>>          * panic immediately to get a useable coredump.
>>          */
>> 	if (unlikely(is_global_init(tsk) &&
>> 	    (thread_group_empty(tsk) ||
>>             (tsk->signal->flags & SIGNAL_GROUP_EXIT)))) {
>> 		panic("Attempted to kill init!	exitcode=0x%08x\n",
>>                 	tsk->signal->group_exit_code ?: (int)code);
>> 	}
>>
>> The thread_group_empty test is needed to handle single threaded
>> inits.
>
> But we can't rely on thread_group_empty(). Just suppose that the main
> thread exit first, then the 2nd (last) thread exits too.

It took me a minute.  I think you are pointing out that there is a case
where we do not set SIGNAL_GROUP_EXIT and that init actually exits.

The case where all of the threads do pthread_exit() aka do_exit().

I think that implies that to have a comprehensive test would
need to do:

	group_dead = atomic_dec_and_test(&tsk->signal->live);

	/* If global init has exited,
	 * panic immediately to get a useable coredump.
	*/
	if (unlikely(is_global_init(tsk) &&
	    (group_dead || thread_group_empty(tsk) ||
             (tsk->signal->flags & SIGNAL_GROUP_EXIT)))) {
		panic("Attempted to kill init!	exitcode=0x%08x\n",
		tsk->signal->group_exit_code ?: (int)code);
	}

Leaving the test where it is.  Yes.  I think that should work.


Eric