[Devel] [PATCH rh7] vtty: Don't close unread master peer if slave is nonzero

[Vladimir Davydov]

On Sat, Dec 05, 2015 at 08:21:19PM +0300, Cyrill Gorcunov wrote:
> When there are several files opened on /dev/console from inside
> of a container and noone hooked on master peer, any close called
> cause master peer to be freed with TTY_CLOSING bit set. So that
> next "vzctl console $ctid $ttynum" call force kernel to allocate
> new vtty pair and in result we can't login into the container.
> 
> We've woraround master close when there is an active slave assigned
> but I miss the scenario when several fd = open(/dev/console) done and
> then one calls for close(fd).
> 
> Lets test if master peer is about to close while there are still
> active slave (and move it into vtty_release helper).
> 
> https://jira.sw.ru/browse/PSBM-41985
> https://jira.sw.ru/browse/PSBM-41672
> 
> Signed-off-by: Cyrill Gorcunov <gorcunov at virtuozzo.com>
> CC: Vladimir Davydov <vdavydov at virtuozzo.com>
> CC: Konstantin Khorenko <khorenko at virtuozzo.com>
> ---
>  drivers/tty/pty.c    |   27 +++++++++++++++++++++++++++
>  drivers/tty/tty_io.c |    7 ++-----
>  include/linux/ve.h   |    2 ++
>  3 files changed, 31 insertions(+), 5 deletions(-)
> 
> Index: linux-pcs7.git/drivers/tty/pty.c
> ===================================================================
> --- linux-pcs7.git.orig/drivers/tty/pty.c
> +++ linux-pcs7.git/drivers/tty/pty.c
> @@ -1259,6 +1259,33 @@ struct tty_driver *vtty_driver(dev_t dev
>  	return NULL;
>  }
>  
> +void vtty_release(struct tty_struct *tty, struct tty_struct *o_tty,
> +		  int *tty_closing, int *o_tty_closing)
> +{
> +	lockdep_assert_held(&tty_mutex);
> +	int pty_master;
> +
> +	if (tty->driver != vttym_driver &&
> +	    tty->driver != vttys_driver)
> +		return;
> +
> +	pty_master = (tty->driver == vttym_driver);
> +
> +	/*
> +	 * Do not close master while slave is active.
> +	 */
> +	if (!*o_tty_closing && pty_master)
> +		*tty_closing = 0;
> +
> +	/*
> +	 * Do not close master if we've closing
> +	 * not the last slave even if there is no
> +	 * readers on the master.
> +	 */
> +	if (*o_tty_closing && !*tty_closing && !pty_master)
> +		*o_tty_closing = 0;
> +}
> +
>  static void ve_vtty_fini(void *data)
>  {
>  	struct ve_struct *ve = data;
> Index: linux-pcs7.git/drivers/tty/tty_io.c
> ===================================================================
> --- linux-pcs7.git.orig/drivers/tty/tty_io.c
> +++ linux-pcs7.git/drivers/tty/tty_io.c
> @@ -1756,11 +1756,8 @@ int tty_release(struct inode *inode, str
>  			(o_tty->count <= (pty_master ? 1 : 0));
>  		do_sleep = 0;
>  
> -#ifdef CONFIG_VE
> -		if (!o_tty_closing &&
> -		    test_bit(TTY_PINNED_BY_OTHER, &tty->flags))
> -			tty_closing = 0;
> -#endif
> +		vtty_release(tty, o_tty, &tty_closing, &o_tty_closing);
> +

I suppose we should now drop TTY_PINNED_BY_OTHER. Other than that, the
patch looks OK to me:

Reviewed-by: Vladimir Davydov <vdavydov at virtuozzo.com>