The ops->receive_buf() may be accessed concurrently from these two
functions. If the driver flushes data to the line discipline
receive_buf() method while tiocsti() is using the ops->receive_buf(),
the data race will happen.
For example:
tty_ioctl |tty_ldisc_receive_buf
->tioctsi | ->tty_port_default_receive_buf
| ->tty_ldisc_receive_buf
->hci_uart_tty_receive | ->hci_uart_tty_receive
->h4_recv | ->h4_recv
In this case, the h4 receive buffer will be overwritten by the
latecomer, and it cause a memory leak.
Hence, change tioctsi() function to use the exclusive lock interface
from tty_buffer to avoid the data race.
Signed-off-by: Nguyen Dinh Phi <[email protected]>
Reported-by: [email protected]
---
drivers/tty/tty_io.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/drivers/tty/tty_io.c b/drivers/tty/tty_io.c
index e8532006e960..746fe13a2054 100644
--- a/drivers/tty/tty_io.c
+++ b/drivers/tty/tty_io.c
@@ -2307,8 +2307,10 @@ static int tiocsti(struct tty_struct *tty, char __user *p)
ld = tty_ldisc_ref_wait(tty);
if (!ld)
return -EIO;
+ tty_buffer_lock_exclusive(tty->port);
if (ld->ops->receive_buf)
ld->ops->receive_buf(tty, &ch, &mbz, 1);
+ tty_buffer_unlock_exclusive(tty->port);
tty_ldisc_deref(ld);
return 0;
}
--
2.25.1
On Sun, Aug 08, 2021 at 03:05:13AM +0800, Nguyen Dinh Phi wrote:
> The ops->receive_buf() may be accessed concurrently from these two
> functions. If the driver flushes data to the line discipline
> receive_buf() method while tiocsti() is using the ops->receive_buf(),
> the data race will happen.
>
> For example:
> tty_ioctl |tty_ldisc_receive_buf
> ->tioctsi | ->tty_port_default_receive_buf
> | ->tty_ldisc_receive_buf
> ->hci_uart_tty_receive | ->hci_uart_tty_receive
> ->h4_recv | ->h4_recv
>
> In this case, the h4 receive buffer will be overwritten by the
> latecomer, and it cause a memory leak.
That looks to be a bug in the h4 code, if the receive_buf() call can not
be run at the same time, it should have a lock in it, right?
> Hence, change tioctsi() function to use the exclusive lock interface
> from tty_buffer to avoid the data race.
Where is the lock being grabbed from the other receive_buf() call path
to ensure that the lock is always needed here?
>
> Signed-off-by: Nguyen Dinh Phi <[email protected]>
> Reported-by: [email protected]
> ---
> drivers/tty/tty_io.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/drivers/tty/tty_io.c b/drivers/tty/tty_io.c
> index e8532006e960..746fe13a2054 100644
> --- a/drivers/tty/tty_io.c
> +++ b/drivers/tty/tty_io.c
> @@ -2307,8 +2307,10 @@ static int tiocsti(struct tty_struct *tty, char __user *p)
> ld = tty_ldisc_ref_wait(tty);
> if (!ld)
> return -EIO;
> + tty_buffer_lock_exclusive(tty->port);
> if (ld->ops->receive_buf)
> ld->ops->receive_buf(tty, &ch, &mbz, 1);
> + tty_buffer_unlock_exclusive(tty->port);
Did this fix the syzbot reported issue?
thanks,
greg k-h
On 8/13/2021 3:33 PM, Greg KH wrote:
>>
>> Signed-off-by: Nguyen Dinh Phi <[email protected]>
>> Reported-by: [email protected]
>> ---
>> drivers/tty/tty_io.c | 2 ++
>> 1 file changed, 2 insertions(+)
>>
>> diff --git a/drivers/tty/tty_io.c b/drivers/tty/tty_io.c
>> index e8532006e960..746fe13a2054 100644
>> --- a/drivers/tty/tty_io.c
>> +++ b/drivers/tty/tty_io.c
>> @@ -2307,8 +2307,10 @@ static int tiocsti(struct tty_struct *tty, char __user *p)
>> ld = tty_ldisc_ref_wait(tty);
>> if (!ld)
>> return -EIO;
>> + tty_buffer_lock_exclusive(tty->port);
>> if (ld->ops->receive_buf)
>> ld->ops->receive_buf(tty, &ch, &mbz, 1);
>> + tty_buffer_unlock_exclusive(tty->port);
>
> Did this fix the syzbot reported issue?
>
> thanks,
>
> greg k-h
> Yes, this fixed the syzbot reported issue.
The lock is grabbed in flush_to_ldisc() and paste_selection().
Actually, I follow the document in tty_buffer.c, where it say the
callers to receive_buff() other than flush_to_ldisc() need to exclude
the kworker from concurrent use of the line discipline.
And function tiocsti() has the following comment:
/* FIXME: may race normal receive processing */
that why I add lock in this function.
BR,
Phi.
On Sat, Aug 14, 2021 at 02:35:53AM +0800, Phi Nguyen wrote:
> On 8/13/2021 3:33 PM, Greg KH wrote:
> > >
> > > Signed-off-by: Nguyen Dinh Phi <[email protected]>
> > > Reported-by: [email protected]
> > > ---
> > > drivers/tty/tty_io.c | 2 ++
> > > 1 file changed, 2 insertions(+)
> > >
> > > diff --git a/drivers/tty/tty_io.c b/drivers/tty/tty_io.c
> > > index e8532006e960..746fe13a2054 100644
> > > --- a/drivers/tty/tty_io.c
> > > +++ b/drivers/tty/tty_io.c
> > > @@ -2307,8 +2307,10 @@ static int tiocsti(struct tty_struct *tty, char __user *p)
> > > ld = tty_ldisc_ref_wait(tty);
> > > if (!ld)
> > > return -EIO;
> > > + tty_buffer_lock_exclusive(tty->port);
> > > if (ld->ops->receive_buf)
> > > ld->ops->receive_buf(tty, &ch, &mbz, 1);
> > > + tty_buffer_unlock_exclusive(tty->port);
> >
> > Did this fix the syzbot reported issue?
> >
> > thanks,
> >
> > greg k-h
> > Yes, this fixed the syzbot reported issue.
>
> The lock is grabbed in flush_to_ldisc() and paste_selection().
> Actually, I follow the document in tty_buffer.c, where it say the callers to
> receive_buff() other than flush_to_ldisc() need to exclude the kworker from
> concurrent use of the line discipline.
>
> And function tiocsti() has the following comment:
> /* FIXME: may race normal receive processing */
> that why I add lock in this function.
As you are fixing this FIXME, please remove it in this patch, otherwise
we will not know it is resolved :)
Can you add that to the change and submit a new version?
Also, this should go to stable kernels, right? Any idea how far back?
thanks,
greg k-h
On 8/18/2021 10:03 PM, Greg KH wrote:
> On Sat, Aug 14, 2021 at 02:35:53AM +0800, Phi Nguyen wrote:
>> On 8/13/2021 3:33 PM, Greg KH wrote:
>>>>
>>>> Signed-off-by: Nguyen Dinh Phi <[email protected]>
>>>> Reported-by: [email protected]
>>>> ---
>>>> drivers/tty/tty_io.c | 2 ++
>>>> 1 file changed, 2 insertions(+)
>>>>
>>>> diff --git a/drivers/tty/tty_io.c b/drivers/tty/tty_io.c
>>>> index e8532006e960..746fe13a2054 100644
>>>> --- a/drivers/tty/tty_io.c
>>>> +++ b/drivers/tty/tty_io.c
>>>> @@ -2307,8 +2307,10 @@ static int tiocsti(struct tty_struct *tty, char __user *p)
>>>> ld = tty_ldisc_ref_wait(tty);
>>>> if (!ld)
>>>> return -EIO;
>>>> + tty_buffer_lock_exclusive(tty->port);
>>>> if (ld->ops->receive_buf)
>>>> ld->ops->receive_buf(tty, &ch, &mbz, 1);
>>>> + tty_buffer_unlock_exclusive(tty->port);
>>>
>>> Did this fix the syzbot reported issue?
>>>
>>> thanks,
>>>
>>> greg k-h
>>> Yes, this fixed the syzbot reported issue.
>>
>> The lock is grabbed in flush_to_ldisc() and paste_selection().
>> Actually, I follow the document in tty_buffer.c, where it say the callers to
>> receive_buff() other than flush_to_ldisc() need to exclude the kworker from
>> concurrent use of the line discipline.
>>
>> And function tiocsti() has the following comment:
>> /* FIXME: may race normal receive processing */
>> that why I add lock in this function.
>
> As you are fixing this FIXME, please remove it in this patch, otherwise
> we will not know it is resolved :)
>
> Can you add that to the change and submit a new version?
>
Yes, I will submit a new version.
> Also, this should go to stable kernels, right? Any idea how far back?
>
I have no idea about this question, but I see it meets specified
requirements in stable-kernel-rules.rst
BR,
Phi.