Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <e10b79cf-d6d5-ffcc-bce4-edd92b7cb6b9@molgen.mpg.de>
 <CAHmME9pktmNpcBS_DJhJ5Z+6xO9P1wroQ9_gwx8KZMBxk1FBeQ@mail.gmail.com> <5b3d6ad6-a506-7dae-0eb7-b7da080f6454@molgen.mpg.de>
In-Reply-To: <5b3d6ad6-a506-7dae-0eb7-b7da080f6454@molgen.mpg.de>
From:   "Jason A. Donenfeld" <Jason@zx2c4.com>
Date:   Tue, 15 Feb 2022 12:44:09 +0100
Message-ID: <CAHmME9rgBJx+-wy-_GXPRMYEuTNyaa7SAr+nVOPh-Am2k6A2CA@mail.gmail.com>
Subject: Re: BUG: KCSAN: data-race in add_device_randomness+0x20d/0x290
To:     Paul Menzel <pmenzel@molgen.mpg.de>
Cc:     "Theodore Y. Ts'o" <tytso@mit.edu>,
        LKML <linux-kernel@vger.kernel.org>,
        Dominik Brodowski <linux@dominikbrodowski.net>,
        Jann Horn <jannh@google.com>,
        Dmitry Vyukov <dvyukov@google.com>,
        Marco Elver <elver@google.com>,
        Nicolas Pitre <nico@fluxnic.net>,
        Rik van Riel <riel@redhat.com>,
        Oleg Nesterov <oleg@redhat.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi Paul,

Thanks for the new report. This looks like likely the same sort of
issue as before -- it's on a 1 byte read of the data that's being
passed to add_device_randomness by release_task(). It looks like there
might be a race in release_task():

void release_task(struct task_struct *p)
{
[...]
        cgroup_release(p);

        write_lock_irq(&tasklist_lock);
        ptrace_release_task(p);
        thread_pid = get_pid(p->thread_pid);
        __exit_signal(p);
[...]
}
static void __exit_signal(struct task_struct *tsk)
{
[...]
       add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
                             sizeof(unsigned long long));
[...]
}

Apparently the data that is being read by add_device_randomness() is
being modified while it's being read. This would be whatever is in
`tsk->se.sum_exec_runtime`.

I'm not sure what's happening there, if this is "normal" and the task
can be scheduled out while exiting, causing the schedule to add to
sum_exec_runtime, or what. CCing some people who seem to have touched
this code, and maybe that'll help illuminate things.

Thanks,
Jason