Re: [PATCH 2/2] serdev: use tty_port_tty_get() in ttyport_write_buf() to prevent UAF

[Markus Probst <markus.probst@posteo.de>]

[-- Attachment #1: Type: text/plain, Size: 3555 bytes --]

On Fri, 2026-07-31 at 15:22 +0200, Greg Kroah-Hartman wrote:
> On Fri, Jul 31, 2026 at 01:00:30PM +0000, Markus Probst wrote:
> > On Fri, 2026-07-31 at 10:06 +0200, Greg Kroah-Hartman wrote:
> > > From: Joshua Rogers <linux@joshua.hu>
> > > 
> > > ttyport_write_buf() snapshots serport->tty as a raw pointer without
> > > taking a reference, while ttyport_close() can concurrently release the
> > > tty via tty_release_struct(), leading to a use-after-free. Use
> > > tty_port_tty_get() to obtain a reference-counted tty pointer, matching
> > > the pattern already used by ttyport_write_wakeup().
> > > 
> > > Assisted-by: AISLE:Snapshot
> > > Cc: stable <stable@kernel.org>
> > > Signed-off-by: Joshua Rogers <linux@joshua.hu>
> > > Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> > > ---
> > >  drivers/tty/serdev/serdev-ttyport.c | 12 ++++++++++--
> > >  1 file changed, 10 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/drivers/tty/serdev/serdev-ttyport.c b/drivers/tty/serdev/serdev-ttyport.c
> > > index 48ce5b3f8308..4d37c8130dd7 100644
> > > --- a/drivers/tty/serdev/serdev-ttyport.c
> > > +++ b/drivers/tty/serdev/serdev-ttyport.c
> > > @@ -85,13 +85,21 @@ static const struct tty_port_client_operations client_ops = {
> > >  static ssize_t ttyport_write_buf(struct serdev_controller *ctrl, const u8 *data, size_t len)
> > >  {
> > >  	struct serport *serport = serdev_controller_get_drvdata(ctrl);
> > > -	struct tty_struct *tty = serport->tty;
> > > +	struct tty_struct *tty;
> > > +	ssize_t ret;
> > >  
> > >  	if (!test_bit(SERPORT_ACTIVE, &serport->flags))
> > >  		return 0;
> > >  
> > > +	tty = tty_port_tty_get(serport->port);
> > > +	if (!tty)
> > > +		return 0;
> > > +
> > >  	set_bit(TTY_DO_WRITE_WAKEUP, &tty->flags);
> > > -	return tty->ops->write(serport->tty, data, len);
> > > +	ret = tty->ops->write(tty, data, len);
> > > +	tty_kref_put(tty);
> > > +
> > > +	return ret;
> > >  }
> > >  
> > >  static void ttyport_write_flush(struct serdev_controller *ctrl)
> > 
> > It should be the responsibility of the driver to ensure
> > `serdev_device_close` and `serdev_device_write_buf` are not called
> > concurrently.
> > 
> > This also applies to
> > - serdev_device_set_baudrate
> 
> Doesn't touch a tty structure, so why is the same reference count logic
> needed here?
It calls `ttyport_set_baudrate`.

> 
> > - serdev_device_set_flow_control
It calls `ttyport_set_flow_control`.
> 
> Same here.
> 
> > - serdev_device_wait_until_sent
> 
> Or here?
It calls `ttyport_wait_until_sent`.

See `struct serdev_controller_ops ctrl_ops` in `serdev-ttyport.c`.

All of those mentioned above call into tty_ functions. Obviously the
tty_struct pointer needs to be valid for those calls, which is only the
case if the device is open.

Thanks
- Markus Probst

> 
> > and more, which currently iirc even introduce a null pointer
> > dereference if called with the serdev device closed.
> > 
> > I don't see why `serdev_device_write_buf` needs special protection
> > here, which the other functions don't need for some reason.
> 
> It's the access to the tty pointer that I think the issue is, right?
> 
> > As with the previous patch, this still makes sense for hardening.
> > With scope_guard being used,
> > 
> > Reviewed-by: Markus Probst <markus.probst@posteo.de>
> 
> Great, thanks, I'll send out a v2 next week with that change made, as
> I've done it locally and will wait a bit.
> 
> thanks,
> 
> greg k-h

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 870 bytes --]