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;
MIME-Version: 1.0
References: <20200710002209.6757-1-apronin@chromium.org> <1594408113.4494.16.camel@HansenPartnership.com>
In-Reply-To: <1594408113.4494.16.camel@HansenPartnership.com>
From:   Andrey Pronin <apronin@chromium.org>
Date:   Fri, 10 Jul 2020 13:34:28 -0700
Message-ID: <CAP7wa8+0Bkbm-QCW=xwJ7auZ18Bj1jJev9LcMxuA=i=6ZfT9yA@mail.gmail.com>
Subject: Re: [PATCH] tpm: avoid accessing cleared ops during shutdown
To:     James Bottomley <James.Bottomley@hansenpartnership.com>
Cc:     peterhuewe@gmx.de,
        Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>,
        jgg@ziepe.ca, linux-integrity@vger.kernel.org,
        LKML <linux-kernel@vger.kernel.org>,
        Guenter Roeck <groeck@chromium.org>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Fri, Jul 10, 2020 at 12:08 PM James Bottomley
<James.Bottomley@hansenpartnership.com> wrote:
>
> On Thu, 2020-07-09 at 17:22 -0700, Andrey Pronin wrote:
> > This patch prevents NULL dereferencing when using chip->ops while
> > sending TPM2_Shutdown command if both tpm_class_shutdown handler and
> > tpm_del_char_device are called during system shutdown.
> >
> > Both these handlers set chip->ops to NULL but don't check if it's
> > already NULL when they are called before using it.
> >
> > This issue was revealed in Chrome OS after a recent set of changes
> > to the unregister order for spi controllers, such as:
> >   b4c6230bb0ba spi: Fix controller unregister order
> >   f40913d2dca1 spi: pxa2xx: Fix controller unregister order
> > and similar for other controllers.
> >
> > Signed-off-by: Andrey Pronin <apronin@chromium.org>
> > ---
> >  drivers/char/tpm/tpm-chip.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/char/tpm/tpm-chip.c b/drivers/char/tpm/tpm-
> > chip.c
> > index 8c77e88012e9..a410ca40a3c5 100644
> > --- a/drivers/char/tpm/tpm-chip.c
> > +++ b/drivers/char/tpm/tpm-chip.c
> > @@ -296,7 +296,7 @@ static int tpm_class_shutdown(struct device *dev)
> >       struct tpm_chip *chip = container_of(dev, struct tpm_chip,
> > dev);
> >
> >       down_write(&chip->ops_sem);
> > -     if (chip->flags & TPM_CHIP_FLAG_TPM2) {
> > +     if (chip->ops && (chip->flags & TPM_CHIP_FLAG_TPM2)) {
> >               if (!tpm_chip_start(chip)) {
> >                       tpm2_shutdown(chip, TPM2_SU_CLEAR);
> >                       tpm_chip_stop(chip);
> > @@ -479,7 +479,7 @@ static void tpm_del_char_device(struct tpm_chip
> > *chip)
> >
> >       /* Make the driver uncallable. */
> >       down_write(&chip->ops_sem);
> > -     if (chip->flags & TPM_CHIP_FLAG_TPM2) {
> > +     if (chip->ops && (chip->flags & TPM_CHIP_FLAG_TPM2)) {
> >               if (!tpm_chip_start(chip)) {
> >                       tpm2_shutdown(chip, TPM2_SU_CLEAR);
> >                       tpm_chip_stop(chip);
>
> I really don't think this is the right fix.  The problem is that these
> two functions are trying to open code tpm_try_get_ops/tpm_put_ops (only
> really for the tpm2 shutdown) because they want to NULL out the ops
> before final mutex unlock.  The problem with the current open coding is
> it doesn't shut down the clock if required (not really a problem for
> shutdown, but might cause issues for simple rmmod).  I think this is
> because no-one noticed the open coding when get/put was updated.
>
> This code should all be abstracted into a single function and shared
> with tpm_try_get_ops/tpm_put_ops so we can't have this happen in
> future.  Possibly there should be a chip shutdown function which is
> only active for TPM2 which does the correct try_get/shutdown/put
> operation and then a separate simple get mutex, null ops, put mutex one
> that's guaranteed to be called last.

Yes, went for a minimal patch here to stop kernel panics, didn't try to
refactor. Note that we do hold chip->ops_sem in both cases, and it's a
write-lock, not a read-lock (as tpm_try_get_ops uses) since we are
changing chip->ops. Thanks to this write-lock there, shouldn't be parallel
operations that use chip->ops (so not locking chip->tpm_mutex shouldn't
affect it).

So, if I understand the idea right, can refactor to something like:
1) extract common code between tpm_del_char_device and
tpm_class_shutdown into a shared method;
2) further extract the part between up/down(chip->ops_sem) to be
re-used between tpm_try_get_ops/tpm_put_ops and this flow;
3) still have down_write/up_write in this flow vs
get/put_device + down_read/up_read in tpm_try_get_ops case.

Please let me know if that's a bad idea.

Will be unavailable next week, but will continue after that.

>
> James
>


-- 
Andrey