Re: [PATCH] Fix refcount leak in tty_port.c

[Peter Hurley <peter@hurleysoftware.com>]

On 07/12/2013 06:30 AM, Gustavo Padovan wrote:
> Hi Gianluca,
>
> * Gianluca Anzolin <gianluca@sottospazio.it> [2013-07-09 10:35:35 +0200]:
>
>> Hello,
>>
>> In linux 3.10 in the file drivers/tty/tty_port.c the function
>> tty_port_tty_hangup may leak a tty reference:
>>
>>          struct tty_struct *tty = tty_port_tty_get(port);
>>
>>          if (tty && (!check_clocal || !C_CLOCAL(tty))) {
>>                  tty_hangup(tty);
>>                  tty_kref_put(tty);
>>          }
>>
>> If tty != NULL and the second condition is false we never call tty_kref_put and
>> the reference is leaked.

Good catch.

>> Fix by nesting two if statements.
>>
>> Signed-off-by: Gianluca Anzolin <gianluca@sottospazio.it>
>
> As mentioned by Gianluca this is a regression of aa27a094 and we depend on
> this patch to go ahead with some fixes in the bluetooth subsystem.

Gustavo,

There's no direct dependency; ie., there aren't merge issues here.
We should progress with the fixes to rfcomm independent of this patch.

> Gianluca, it might help if you send a proper git inline formated patch,
> mentioning the issue and which regression you are fixing.  It makes
> maintainer's life easier.

As Gustavo points out, please inline the patch otherwise commenters
have to do it for you.

> Also add my Ack to the patch:
 >
>
> Acked-by: Gustavo Padovan <gustavo.padovan@collabora.co.uk>
>
> 	Gustavo

Copy of the Gianluca's patch with my comments

--- %< ---

Please put a proper commit message here, including that this is
a regression and the commit id that caused the regression so this patch
can eventually make its way to stable.


 > diff --git a/drivers/tty/tty_port.c b/drivers/tty/tty_port.c
 > index 121aeb9..2198f7d 100644
 > --- a/drivers/tty/tty_port.c
 > +++ b/drivers/tty/tty_port.c
 > @@ -256,8 +256,9 @@ void tty_port_tty_hangup(struct tty_port *port, bool check_clocal)
 >  {
 >  	struct tty_struct *tty = tty_port_tty_get(port);
 >
 > -	if (tty && (!check_clocal || !C_CLOCAL(tty))) {
 > -		tty_hangup(tty);
 > +	if (tty) {
 > +		if (!check_clocal || !C_CLOCAL(tty))
 > +			tty_hangup(tty);
 >  		tty_kref_put(tty);
 >  	}
 >  }


tty_kref_put() already checks for NULL tty. I would prefer:

{
   	struct tty_struct *tty = tty_port_tty_get(port);

	if (tty && (!check_clocal || !C_CLOCAL(tty)))
		tty_hangup(tty);
	tty_kref_put(tty);
}