Re: Re: [PATCH] usb: gadget: serial: Fixed KASAN null-ptr-deref in tty_wakeup

[Michał Mirosław]

On Wed, Apr 08, 2020 at 03:42:53PM +0800, ææ wrote:
> Cc:  Felipe Balbi <balbi@xxxxxxxxxx>,"MichaÅ MirosÅaw" <mirq-linux@xxxxxxxxxxxx>,Sergey Organov <sorganov@xxxxxxxxx>,linux-usb@xxxxxxxxxxxxxxx,linux-kernel@xxxxxxxxxxxxxxx,wenhu.wang@xxxxxxxx
> Subject: Re: [PATCH] usb: gadget: serial: Fixed KASAN null-ptr-deref in tty_wakeup>On Wed, Mar 18, 2020 at 10:56:01AM +0800, Li Tao wrote:
> >> The port->port.tty maybe reset as NULL, If gs_close() was invoked
> >> unexpectedly during running gserial_connect().
> >> 
> >> BUG: KASAN: null-ptr-deref in tty_wakeup+0x1c/0x84
> >> Call trace:
> >> [<0000000095c3c837>] dump_backtrace+0x0/0x6d4
> >> [<0000000047726bb8>] show_stack+0x18/0x20
> >> [<00000000bedb4c1e>] --dump_stack+0x20/0x28
> >> [<00000000722f2e2a>] dump_stack+0x84/0xb0
> >> [<00000000325683d4>] kasan_report_error+0x178/0x1e4
> >> [<0000000053079998>] kasan_report_error+0x0/0x1e4
> >> [<00000000b6d33afa>] --asan_load8+0x150/0x15c
> >> [<00000000188745b8>] tty_wakeup+0x1c/0x84
> >> [<0000000064f6dd21>] gs_start_io+0xd0/0x11c
> >> [<0000000063d67b6c>] gserial_connect+0x15c/0x1b0
> >> [<00000000faf7c0f9>] dm_set_alt+0xa8/0x190
> >> [<000000008deb1909>] composite_setup+0xde4/0x1edc
> >> [<00000000792ee16d>] android_setup+0x210/0x294
> >> [<00000000ab32ef30>] dwc3_ep0_delegate_req+0x48/0x68
> >> [<0000000054e26fd2>] dwc3_ep0_interrupt+0xf2c/0x16fc
> >> [<0000000050cb2262>] dwc3_interrupt+0x464/0x1f44
> >> [<00000000fdcaa6e9>] --handle_irq_event_percpu+0x184/0x398
> >> [<000000003b24ff56>] handle_irq_event_percpu+0xa0/0x134
> >> [<00000000aedda5ee>] handle_irq_event+0x60/0xa0
> >> [<000000005f51a570>] handle_fasteoi_irq+0x23c/0x31c
> >> [<000000008db2608d>] generic_handle_irq+0x70/0xa4
> >> [<00000000098683fc>] --handle_domain_irq+0x84/0xe0
> >> [<000000008ed23b46>] gic_handle_irq+0x98/0xb8
> >> 
> >> Signed-off-by: Li Tao <tao.li@xxxxxxxx>
> >> ---
> >>  drivers/usb/gadget/function/u_serial.c | 3 ++-
> >>  1 file changed, 2 insertions(+), 1 deletion(-)
> >> 
> >> diff --git a/drivers/usb/gadget/function/u_serial.c b/drivers/usb/gadget/function/u_serial.c
> >> index 8167d379e115..3c109a8f9ec4 100644
> >> --- a/drivers/usb/gadget/function/u_serial.c
> >> +++ b/drivers/usb/gadget/function/u_serial.c
> >> @@ -565,7 +565,8 @@ static int gs_start_io(struct gs_port *port)
> >>  		gs_start_tx(port);
> >>  		/* Unblock any pending writes into our circular buffer, in case
> >>  		 * we didn't in gs_start_tx() */
> >> -		tty_wakeup(port->port.tty);
> >> +		if (port->port.tty)
> >> +			tty_wakeup(port->port.tty);
> >
> >What prevents port.tty from being set to NULL between the check and the
> >call to tty_wakeup?
> >
> >Shouldn't gs_close() and gserial_connect() have some sort of locking to
> >prevent this?
> >
> I encountered this issue on Samsung EXYNOS980 platform, Maybe this is platform specific issue.
> The USB interrupt handler got port_lock on CPU 0 in gserial_connect while gs_close is waiting for 
> this lock on CPU 1.  But this port_lock will unlock in gs_start_io-->gs_start_rx temporarily, This race
> condition may cause port.tty set as NULL.

Hi,

gserial_connect() indeed accesses port[] array without taking the
portmaster's mutex. This is since commit 19b10a8828a6 ("usb: gadget:
allocate & giveback serial ports instead hard code them") introduced
dynamic allocation of the port structure.

Best Regards,
MichaÅÂMirosÅaw