Re: [PATCH] serial: core: Fix double fetch in uart_throttle/uart_unthrottle

[Ilpo Järvinen]

On Fri, 12 Jan 2024, Gui-Dong Han wrote:

> In uart_throttle() and uart_unthrottle():
>     if (port->status & mask) {
>         port->ops->throttle/unthrottle(port);
>         mask &= ~port->status;
>     }
>     // Code segment utilizing the mask value to determine UART behavior
> 
> In uart_change_line_settings():
>     uart_port_lock_irq(uport);
>     // Code segment responsible for updating uport->status
>     uart_port_unlock_irq(uport);
> 
> In the uart_throttle() and uart_unthrottle() functions, there is a double
> fetch issue due to concurrent execution with uart_change_line_settings().
> In uart_throttle() and uart_unthrottle(), the check
> if (port->status & mask) is made, followed by mask &= ~port->status,
> where the relevant bits are cleared. However, port->status may be modified
> in uart_change_line_settings(). The current implementation does not ensure
> atomicity in the access and modification of port->status and mask. This
> can result in mask being updated based on a modified port->status value,
> leading to improper UART actions.
> 
> This possible bug is found by an experimental static analysis tool
> developed by our team, BassCheck[1]. This tool analyzes the locking APIs
> to extract function pairs that can be concurrently executed, and then
> analyzes the instructions in the paired functions to identify possible
> concurrency bugs including data races and atomicity violations. The above
> possible bug is reported when our tool analyzes the source code of
> Linux 5.17.
> 
> To resolve this double fetch, it is suggested to add a uart_port_lock pair
> in uart_throttle() and uart_unthrottle(). With this patch applied, our
> tool no longer reports the bug, with the kernel configuration allyesconfig
> for x86_64. Due to the absence of the requisite hardware, we are unable to
> conduct runtime testing of the patch. Therefore, our verification is
> solely based on code logic analysis.
> 
> [1] https://sites.google.com/view/basscheck/
> 
> Fixes: 391f93f2ec9f ("serial: core: Rework hw-assisted flow control support")
> Cc: stable@xxxxxxxxxxxxxxx
> Signed-off-by: Gui-Dong Han <2045gemini@xxxxxxxxx>
> ---
>  drivers/tty/serial/serial_core.c | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/tty/serial/serial_core.c b/drivers/tty/serial/serial_core.c
> index 80085b151b34..9d905fdf2843 100644
> --- a/drivers/tty/serial/serial_core.c
> +++ b/drivers/tty/serial/serial_core.c
> @@ -723,11 +723,13 @@ static void uart_throttle(struct tty_struct *tty)
>  		mask |= UPSTAT_AUTOXOFF;
>  	if (C_CRTSCTS(tty))
>  		mask |= UPSTAT_AUTORTS;
> -
> +
> +	uart_port_lock_irq(port);
>  	if (port->status & mask) {
>  		port->ops->throttle(port);
>  		mask &= ~port->status;
>  	}
> +	uart_port_unlock_irq(port);
>  
>  	if (mask & UPSTAT_AUTORTS)
>  		uart_clear_mctrl(port, TIOCM_RTS);
> @@ -753,10 +755,12 @@ static void uart_unthrottle(struct tty_struct *tty)
>  	if (C_CRTSCTS(tty))
>  		mask |= UPSTAT_AUTORTS;
>  
> +	uart_port_lock_irq(port);
>  	if (port->status & mask) {
>  		port->ops->unthrottle(port);
>  		mask &= ~port->status;
>  	}
> +	uart_port_unlock_irq(port);
>  
>  	if (mask & UPSTAT_AUTORTS)
>  		uart_set_mctrl(port, TIOCM_RTS);

Hi,

This is very bogus "fix". While change to the local variable gets 
"protected", uart_change_line_settings() can race after unlock and the 
value held in mask is again stale.

If, and it's a big if, this is a real problem, the patch does not fix 
anything! It proves your tool is flawed because it doesn't detect the 
race with uart_change_line_settings() issue still exists after this 
non-fix.

So NAK from me. Please provide a real fix instead if you think there is
a real issue.

Also, don't use vague wording like "leading to improper UART action" but 
describe precisely what goes wrong!

-- 
 i.