[PATCH] 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.
Reported-by: BassCheck bass@buaa.edu.cn Signed-off-by: Tuo Li islituo@gmail.com --- drivers/staging/greybus/uart.c | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/drivers/staging/greybus/uart.c b/drivers/staging/greybus/uart.c index 20a34599859f..b8875517ea6a 100644 --- a/drivers/staging/greybus/uart.c +++ b/drivers/staging/greybus/uart.c @@ -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; }
On Mon, Jun 26, 2023 at 04:43:39PM +0800, Tuo Li wrote:
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.
And how can that happen?
Also you have trailing whitespace in your changelog text :(
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.
Reported-by: BassCheck bass@buaa.edu.cn
As per the documentation for research tools like this, you need to explain how this was tested.
thanks,
greg k-h
Thanks for your reply!
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.
And how can that happen?
I am sorry that this data race is reported by a static analysis tool, and thus I do not know how to write a test case to trigger this data race.
Also you have trailing whitespace in your changelog text :(
I am sorry to bother you, and I will be careful in the subsequent patches.
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.
Reported-by: BassCheck bass@buaa.edu.cn
As per the documentation for research tools like this, you need to explain how this was tested.
I use a static analysis tool to detect this possible data race. It first collects all struct fields that may be protected by a lock when accessed. And then, for each of such struct fields, it calculates the proportion of the protected accesses in all accesses. If the proportion is greater than a given threshold, the tool generates warnings for those unprotected field accesses. Any suggestion on this static analysis tool will be appreciated.
Sincerely, Tuo Li
-
Greg KH -
Tuo Li