[PATCH] qla2xxx: Fix possible race that could hang kthread_stop()

[PATCH] qla2xxx: Fix possible race that could hang kthread_stop()

[Bandan Das]

There is a small race window in qla2x00_do_dpc() between
checking for kthread_should_stop() and going to sleep after
setting TASK_INTERRUPTIBLE. If qla2x00_free_device() is called
in this window, kthread_stop will wait forever because there 
will be no one to wake up the process.


Signed-off-by: Bandan Das <bandan.das@stratus.com>
Signed-off-by: Nate Dailey <nate.dailey@stratus.com>
---
 drivers/scsi/qla2xxx/qla_os.c |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)

diff --git a/drivers/scsi/qla2xxx/qla_os.c b/drivers/scsi/qla2xxx/qla_os.c
index c194c23..8d14d77 100644
--- a/drivers/scsi/qla2xxx/qla_os.c
+++ b/drivers/scsi/qla2xxx/qla_os.c
@@ -3286,6 +3286,8 @@ qla2x00_do_dpc(void *data)
 		DEBUG3(printk("qla2x00: DPC handler sleeping\n"));
 
 		set_current_state(TASK_INTERRUPTIBLE);
+		if (kthread_should_stop())
+			break;
 		schedule();
 		__set_current_state(TASK_RUNNING);
 
-- 
1.7.2.3

Re: [PATCH] qla2xxx: Fix possible race that could hang kthread_stop()

[James Bottomley]

On Thu, 2011-01-27 at 15:59 -0500, Bandan Das wrote:
> There is a small race window in qla2x00_do_dpc() between
> checking for kthread_should_stop() and going to sleep after
> setting TASK_INTERRUPTIBLE. If qla2x00_free_device() is called
> in this window, kthread_stop will wait forever because there 
> will be no one to wake up the process.
> 
> 
> Signed-off-by: Bandan Das <bandan.das@stratus.com>
> Signed-off-by: Nate Dailey <nate.dailey@stratus.com>
> ---
>  drivers/scsi/qla2xxx/qla_os.c |    2 ++
>  1 files changed, 2 insertions(+), 0 deletions(-)
> 
> diff --git a/drivers/scsi/qla2xxx/qla_os.c b/drivers/scsi/qla2xxx/qla_os.c
> index c194c23..8d14d77 100644
> --- a/drivers/scsi/qla2xxx/qla_os.c
> +++ b/drivers/scsi/qla2xxx/qla_os.c
> @@ -3286,6 +3286,8 @@ qla2x00_do_dpc(void *data)
>  		DEBUG3(printk("qla2x00: DPC handler sleeping\n"));
>  
>  		set_current_state(TASK_INTERRUPTIBLE);
> +		if (kthread_should_stop())
> +			break;
>  		schedule();
>  		__set_current_state(TASK_RUNNING);

That's not really the accepted way to fix these race conditions because
of the double check of kthread_should_stop(); this is

James

---
diff --git a/drivers/scsi/qla2xxx/qla_os.c b/drivers/scsi/qla2xxx/qla_os.c
index c194c23..15ce69e 100644
--- a/drivers/scsi/qla2xxx/qla_os.c
+++ b/drivers/scsi/qla2xxx/qla_os.c
@@ -3282,10 +3282,10 @@ qla2x00_do_dpc(void *data)
 
 	set_user_nice(current, -20);
 
+	set_current_state(TASK_INTERRUPTIBLE);
 	while (!kthread_should_stop()) {
 		DEBUG3(printk("qla2x00: DPC handler sleeping\n"));
 
-		set_current_state(TASK_INTERRUPTIBLE);
 		schedule();
 		__set_current_state(TASK_RUNNING);
 
@@ -3454,7 +3454,9 @@ qla2x00_do_dpc(void *data)
 		qla2x00_do_dpc_all_vps(base_vha);
 
 		ha->dpc_active = 0;
+		set_current_state(TASK_INTERRUPTIBLE);
 	} /* End of while(1) */
+	__set_current_state(TASK_RUNNING);
 
 	DEBUG(printk("scsi(%ld): DPC handler exiting\n", base_vha->host_no));
 

Re: [PATCH] qla2xxx: Fix possible race that could hang kthread_stop()

[Bandan Das]

> That's not really the accepted way to fix these race conditions because
> of the double check of kthread_should_stop(); this is 
> James
> 
> ---
> diff --git a/drivers/scsi/qla2xxx/qla_os.c b/drivers/scsi/qla2xxx/qla_os.c
> index c194c23..15ce69e 100644
> --- a/drivers/scsi/qla2xxx/qla_os.c
> +++ b/drivers/scsi/qla2xxx/qla_os.c
> @@ -3282,10 +3282,10 @@ qla2x00_do_dpc(void *data)
>  
>  	set_user_nice(current, -20);
>  
> +	set_current_state(TASK_INTERRUPTIBLE);
>  	while (!kthread_should_stop()) {
>  		DEBUG3(printk("qla2x00: DPC handler sleeping\n"));
>  
> -		set_current_state(TASK_INTERRUPTIBLE);
>  		schedule();
>  		__set_current_state(TASK_RUNNING);
>  
> @@ -3454,7 +3454,9 @@ qla2x00_do_dpc(void *data)
>  		qla2x00_do_dpc_all_vps(base_vha);
>  
>  		ha->dpc_active = 0;
> +		set_current_state(TASK_INTERRUPTIBLE);
>  	} /* End of while(1) */
> +	__set_current_state(TASK_RUNNING);
>  
>  	DEBUG(printk("scsi(%ld): DPC handler exiting\n", base_vha->host_no));
>  
Yes, this looks more clean and will take care of the problem I am hitting.


Bandan