Re: [PATCH 1/1] TTY: fix PTY hangup vs close race

[Jiri Slaby <jirislaby@gmail.com>]

On 02/19/2012 10:19 PM, Jiri Slaby wrote:
> Commit d3bda5298 (TTY: get rid of BTM around devpts_*) moved
> devpts_pty_kill out of BTM, but the BTM was not protecting only
> devpts_pty_kill, but also tty->link. Hence move the function back at
> this late stage until this gets resolved properly some time later.
> 
> I was confused by tty_vhangup(tty->link) outside BTM. But inside of
> tty_vhangup, there is a check for tty == NULL. But we cannot add such
> a check here. We have to have the tty and free the devpts node...
> 
> Signed-off-by: Jiri Slaby <jslaby@suse.cz>
> Reported-by: Sasha Levin <levinsasha928@gmail.com>
> ---
> 
> Gee, I messed up Greg's address again...
> 
>  drivers/tty/pty.c |    3 ++-
>  1 files changed, 2 insertions(+), 1 deletions(-)
> 
> diff --git a/drivers/tty/pty.c b/drivers/tty/pty.c
> index fa1bd2e..95037aa 100644
> --- a/drivers/tty/pty.c
> +++ b/drivers/tty/pty.c
> @@ -54,8 +54,9 @@ static void pty_close(struct tty_struct *tty, struct file *filp)
>  	wake_up_interruptible(&tty->link->write_wait);
>  	if (tty->driver->subtype == PTY_TYPE_MASTER) {
>  		set_bit(TTY_OTHER_CLOSED, &tty->flags);
> -		tty_unlock();
> +		/* BTM protects tty->link here */
>  		devpts_pty_kill(tty->link);
> +		tty_unlock();

I'm afraid this won't help. As this is based on an assumption that
tty->link is NULL [*] and that is not just true.

Greg, please revert commit d3bda5298 completely.

[*] Your dump reveals that the code fetches tty->driver_data (mov
0x428(%rdi),%rbx) and traps at a fetch of inode->i_sbm because inode is
NULL (mov    0x28(%rbx),%rax).

Anyway I'm still interested in the tool you triggered this, because we
will need to get rid of BTM eventually.

thanks,
-- 
js