Re: [PATCH v2 2/2] tty: Allow TIOCSTI to be disabled

[Jiri Slaby]

Hi,

On 15. 10. 22, 8:45, Kees Cook wrote:
> TIOCSTI continues its long history of being used in privilege escalation
> attacks[1]. Prior attempts to provide a mechanism to disable this have
> devolved into discussions around creating full-blown LSMs to provide
> arbitrary ioctl filtering, which is hugely over-engineered -- only
> TIOCSTI is being used this way. 3 years ago OpenBSD entirely removed
> TIOCSTI[2], Android has had it filtered for longer[3], and the tools that
> had historically used TIOCSTI either do not need it, are not commonly
> built with it, or have had its use removed.
>
> Provide a simple CONFIG and global sysctl to disable this for the system
> builders who have wanted this functionality for literally decades now,
> much like the ldisc_autoload CONFIG and sysctl.
>
> [1] https://lore.kernel.org/linux-hardening/Y0m9l52AKmw6Yxi1@hostpad
> [2] https://undeadly.org/cgi?action=article;sid=20170701132619
> [3] https://lore.kernel.org/lkml/CAFJ0LnFGRuEEn1tCLhoki8ZyWrKfktbF+rwwN7WzyC_kBFoQVA@xxxxxxxxxxxxxx/
>
> Cc: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
> Cc: Jiri Slaby <jirislaby@xxxxxxxxxx>
> Cc: Simon Brand <simon.brand@xxxxxxxxxxxxxxxx>
> Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
...
> --- a/drivers/tty/tty_io.c
> +++ b/drivers/tty/tty_io.c
> @@ -2275,11 +2275,15 @@ static int tty_fasync(int fd, struct file *filp, int on)
>    *  * Called functions take tty_ldiscs_lock
>    *  * current->signal->tty check is safe without locks
>    */
> +static int tty_legacy_tiocsti __read_mostly = IS_BUILTIN(CONFIG_LEGACY_TIOCSTI);

This can be bool, right? And IS_ENABLED() sounds more appropriate here.

>   static int tiocsti(struct tty_struct *tty, char __user *p)
>   {
>   	char ch, mbz = 0;
>   	struct tty_ldisc *ld;
>   
> +	if (!tty_legacy_tiocsti)
> +		return -EIO;
> +
>   	if ((current->signal->tty != tty) && !capable(CAP_SYS_ADMIN))
>   		return -EPERM;
>   	if (get_user(ch, p))
> @@ -3582,6 +3586,15 @@ void console_sysfs_notify(void)
>   }
>   
>   static struct ctl_table tty_table[] = {
> +	{
> +		.procname	= "legacy_tiocsti",
> +		.data		= &tty_legacy_tiocsti,
> +		.maxlen		= sizeof(tty_legacy_tiocsti),
> +		.mode		= 0644,
> +		.proc_handler	= proc_dointvec,
> +		.extra1		= SYSCTL_ZERO,
> +		.extra2		= SYSCTL_ONE,

Then this becomes just proc_dobool without extras.

Or we can leave it as int, allow 0, 1, and 2. 2 would log_limited the 
caller's comm before EIO. Just thinking loudly. Maybe the EIO is enough 
for users to notice. Likely…

thanks,
--
js
suse labs