staging: greybus: fix a possible data-inconsistency due to data race in get_serial_info()
Subsystems (3)
PATCH 1/1 staging: greybus: fix a possible data-inconsistency due to data race in get_serial_info()
The variables gb_tty->port.close_delay and gb_tty->port.closing_wait are ofter accessed together while holding the lock gb_tty->port.mutex. Here is an example in set_serial_info():
mutex_lock(&gb_tty->port.mutex); ... gb_tty->port.close_delay = close_delay; gb_tty->port.closing_wait = closing_wait; ... mutex_unlock(&gb_tty->port.mutex);
However, they are accessed without holding the lock gb_tty->port.mutex when are accessed in get_serial_info():
ss->close_delay = jiffies_to_msecs(gb_tty->port.close_delay) / 10; ss->closing_wait =
gb_tty->port.closing_wait == ASYNC_CLOSING_WAIT_NONE ? ASYNC_CLOSING_WAIT_NONE : jiffies_to_msecs(gb_tty->port.closing_wait) / 10;
In my opinion, this may be a harmful race, because ss->close_delay can be inconsistent with ss->closing_wait if gb_tty->port.close_delay and gb_tty->port.closing_wait are updated by another thread after the assignment to ss->close_delay.
Besides, the select operator may return wrong value if gb_tty->port.closing_wait is updated right after the condition is calculated.
To fix this possible data-inconsistency caused by data race, a lock and unlock pair is added when accessing different fields of gb_tty->port.
--- drivers/staging/greybus/uart.c | 2 ++ 1 file changed, 2 insertions(+)
drivers/staging/greybus/uart.c +2 -0
@@ -596,12 +596,14 @@ static int get_serial_info(struct tty_struct *tty, {struct gb_tty *gb_tty = tty->driver_data; +mutex_lock (&gb_tty->port.mutex); ss->line = gb_tty->minor; ss->close_delay =jiffies_to_msecs (gb_tty->port.close_delay) /10 ; ss->closing_wait = gb_tty->port.closing_wait ==ASYNC_CLOSING_WAIT_NONE ?ASYNC_CLOSING_WAIT_NONE :jiffies_to_msecs (gb_tty->port.closing_wait) /10 ; +mutex_unlock (&gb_tty->port.mutex);return 0 ; }
Signature · 1 line
2.34.1
Greg KH
maintainerQuoted 16 lines · Tuo Li
The variables
gb_tty->port.close_delayandgb_tty->port.closing_waitare ofter accessed together while holding the lockgb_tty->port.mutex. Here is an example inset_serial_info():mutex_lock(&gb_tty->port.mutex); ... gb_tty->port.close_delay = close_delay; gb_tty->port.closing_wait = closing_wait; ... mutex_unlock(&gb_tty->port.mutex);However, they are accessed without holding the lock
gb_tty->port.mutexwhen are accessed inget_serial_info():
ss->close_delay=jiffies_to_msecs(gb_tty->port.close_delay)/ 10;ss->closing_wait=
Quoted 8 lines
gb_tty->port.closing_wait == ASYNC_CLOSING_WAIT_NONE ? ASYNC_CLOSING_WAIT_NONE : jiffies_to_msecs(gb_tty->port.closing_wait) / 10;In my opinion, this may be a harmful race, because
ss->close_delaycan be inconsistent withss->closing_waitifgb_tty->port.close_delayandgb_tty->port.closing_waitare updated by another thread after the assignment toss->close_delay.
And how can that happen?
Also you have trailing whitespace in your changelog text :(
Quoted 8 lines
Besides, the select operator may return wrong value if
gb_tty->port.closing_waitis updated right after the condition is calculated.To fix this possible data-inconsistency caused by data race, a lock and unlock pair is added when accessing different fields of
gb_tty->port.Reported-by BassCheck <[email protected]>
As per the documentation for research tools like this, you need to explain how this was tested.
thanks,
greg k-h