Re: [PATCH V4 7/9] aacraid: Fix AIF triggered IOP_RESET

[Tomas Henzl <thenzl@redhat.com>]

On 29.1.2016 23:45, Raghava Aditya Renukunta wrote:
> From: Raghava Aditya Renukunta <raghavaaditya.renukunta@pmcs.com>
>
> while driver removal is in progress or PCI shutdown is invoked, driver
> kills AIF aacraid thread, but IOCTL requests from the management tools
> re-start AIF thread leading to IOP_RESET.
>
> Fixed by setting adapter_shutdown flag when PCI shutdown is invoked.
>
> Changes in V2:
> Set adapter_shutdown flag before shutdown command is sent to \
> controller
>
> Changes in V3:
> Call aac_send_shut_shutdown first thing in __aac_shutdown
> Convert adapter_shutdown to atomic_t variable to prevent \
> SMP coherency issues(race conditions)
>
> Changes in V4:
> Used mutex to protect ioctl path and adapter_shutdown to prevent \
> race conditions.
>
> Signed-off-by: Raghava Aditya Renukunta <raghavaaditya.renukunta@pmcs.com>
> Reviewed-by: Shane Seymour <shane.seymour@hpe.com>
> Reviewed-by: Johannes Thumshirn <jthumshirn@suse.de>
> ---
>  drivers/scsi/aacraid/aacraid.h  |  2 +-
>  drivers/scsi/aacraid/commctrl.c |  3 ++
>  drivers/scsi/aacraid/comminit.c |  6 ++--
>  drivers/scsi/aacraid/linit.c    | 63 +++++++++++++++++++++++++++++++----------
>  4 files changed, 56 insertions(+), 18 deletions(-)
>
> diff --git a/drivers/scsi/aacraid/aacraid.h b/drivers/scsi/aacraid/aacraid.h
> index 2916288..6c55749 100644
> --- a/drivers/scsi/aacraid/aacraid.h
> +++ b/drivers/scsi/aacraid/aacraid.h
> @@ -1123,7 +1123,7 @@ struct aac_dev
>  
>  	struct fib		*free_fib;
>  	spinlock_t		fib_lock;
> -
> +	struct mutex		ioctl_mutex;
>  	struct aac_queue_block *queues;
>  	/*
>  	 *	The user API will use an IOCTL to register itself to receive
> diff --git a/drivers/scsi/aacraid/commctrl.c b/drivers/scsi/aacraid/commctrl.c
> index 54195a1..8d3438c 100644
> --- a/drivers/scsi/aacraid/commctrl.c
> +++ b/drivers/scsi/aacraid/commctrl.c
> @@ -855,6 +855,9 @@ int aac_do_ioctl(struct aac_dev * dev, int cmd, void __user *arg)
>  {
>  	int status;
>  
> +	if (dev->adapter_shutdown)
> +		return -EACCES;
> +
>  	/*
>  	 *	HBA gets first crack
>  	 */
> diff --git a/drivers/scsi/aacraid/comminit.c b/drivers/scsi/aacraid/comminit.c
> index 0e954e3..2b4e753 100644
> --- a/drivers/scsi/aacraid/comminit.c
> +++ b/drivers/scsi/aacraid/comminit.c
> @@ -212,8 +212,11 @@ int aac_send_shutdown(struct aac_dev * dev)
>  		return -ENOMEM;
>  	aac_fib_init(fibctx);
>  
> -	cmd = (struct aac_close *) fib_data(fibctx);
> +	mutex_lock(&dev->ioctl_mutex);
> +	dev->adapter_shutdown = 1;
> +	mutex_unlock(&dev->ioctl_mutex);
>  
> +	cmd = (struct aac_close *) fib_data(fibctx);
>  	cmd->command = cpu_to_le32(VM_CloseAll);
>  	cmd->cid = cpu_to_le32(0xfffffffe);
>  
> @@ -229,7 +232,6 @@ int aac_send_shutdown(struct aac_dev * dev)
>  	/* FIB should be freed only after getting the response from the F/W */
>  	if (status != -ERESTARTSYS)
>  		aac_fib_free(fibctx);
> -	dev->adapter_shutdown = 1;
>  	if ((dev->pdev->device == PMC_DEVICE_S7 ||
>  	     dev->pdev->device == PMC_DEVICE_S8 ||
>  	     dev->pdev->device == PMC_DEVICE_S9) &&
> diff --git a/drivers/scsi/aacraid/linit.c b/drivers/scsi/aacraid/linit.c
> index af8974e..9453e11 100644
> --- a/drivers/scsi/aacraid/linit.c
> +++ b/drivers/scsi/aacraid/linit.c
> @@ -524,10 +524,17 @@ static struct device_attribute *aac_dev_attrs[] = {
>  
>  static int aac_ioctl(struct scsi_device *sdev, int cmd, void __user * arg)
>  {
> -	struct aac_dev *dev = (struct aac_dev *)sdev->host->hostdata;
> +	int ret;
> +	struct aac_dev *aac = (struct aac_dev *)sdev->host->hostdata;
> +
>  	if (!capable(CAP_SYS_RAWIO))
>  		return -EPERM;
> -	return aac_do_ioctl(dev, cmd, arg);
> +
> +	mutex_lock(&aac->ioctl_mutex);
> +	ret = aac_do_ioctl(aac, cmd, arg);
> +	mutex_unlock(&aac->ioctl_mutex);
> +
> +	return ret;
>  }
>  
>  static int aac_eh_abort(struct scsi_cmnd* cmd)
> @@ -704,13 +711,14 @@ static long aac_cfg_ioctl(struct file *file,
>  		unsigned int cmd, unsigned long arg)
>  {
>  	int ret;
> -	struct aac_dev *aac;
> -	aac = (struct aac_dev *)file->private_data;
> -	if (!capable(CAP_SYS_RAWIO) || aac->adapter_shutdown)
> +	struct aac_dev *aac = (struct aac_dev *)file->private_data;
> +
> +	if (!capable(CAP_SYS_RAWIO))
>  		return -EPERM;
> -	mutex_lock(&aac_mutex);
> -	ret = aac_do_ioctl(file->private_data, cmd, (void __user *)arg);
> -	mutex_unlock(&aac_mutex);
> +
> +	mutex_lock(&aac->ioctl_mutex);
> +	ret = aac_do_ioctl(aac, cmd, (void __user *)arg);
> +	mutex_unlock(&aac->ioctl_mutex);
>  
>  	return ret;
>  }
> @@ -719,7 +727,10 @@ static long aac_cfg_ioctl(struct file *file,
>  static long aac_compat_do_ioctl(struct aac_dev *dev, unsigned cmd, unsigned long arg)
>  {
>  	long ret;
> -	mutex_lock(&aac_mutex);
> +
> +	if (dev->adapter_shutdown)
> +		return -EACCES;

There is another test for the same in aac_do_ioctl, this duplicated test is needless.

> +
>  	switch (cmd) {
>  	case FSACTL_MINIPORT_REV_CHECK:
>  	case FSACTL_SENDFIB:
> @@ -753,23 +764,37 @@ static long aac_compat_do_ioctl(struct aac_dev *dev, unsigned cmd, unsigned long
>  		ret = -ENOIOCTLCMD;
>  		break;
>  	}
> -	mutex_unlock(&aac_mutex);
>  	return ret;
>  }
>  
>  static int aac_compat_ioctl(struct scsi_device *sdev, int cmd, void __user *arg)
>  {
> -	struct aac_dev *dev = (struct aac_dev *)sdev->host->hostdata;
> +	int ret;
> +	struct aac_dev *aac = (struct aac_dev *)sdev->host->hostdata;
> +
>  	if (!capable(CAP_SYS_RAWIO))
>  		return -EPERM;
> -	return aac_compat_do_ioctl(dev, cmd, (unsigned long)arg);
> +
> +	mutex_lock(&aac->ioctl_mutex);
> +	ret = aac_compat_do_ioctl(aac, cmd, (unsigned long)arg);
> +	mutex_unlock(&aac->ioctl_mutex);
> +
> +	return ret;
>  }
>  
>  static long aac_compat_cfg_ioctl(struct file *file, unsigned cmd, unsigned long arg)
>  {
> +	int ret;
> +	struct aac_dev *aac = (struct aac_dev *)file->private_data;
> +
>  	if (!capable(CAP_SYS_RAWIO))
>  		return -EPERM;
> -	return aac_compat_do_ioctl(file->private_data, cmd, arg);
> +
> +	mutex_lock(&aac->ioctl_mutex);
> +	ret =  aac_compat_do_ioctl(aac, cmd, arg);
> +	mutex_unlock(&aac->ioctl_mutex);
> +
> +	return ret;
>  }
>  #endif
>  
> @@ -1078,6 +1103,8 @@ static void __aac_shutdown(struct aac_dev * aac)
>  	int i;
>  	int cpu;
>  
> +	aac_send_shutdown(aac);
> +
>  	if (aac->aif_thread) {
>  		int i;
>  		/* Clear out events first */
> @@ -1089,7 +1116,7 @@ static void __aac_shutdown(struct aac_dev * aac)
>  		}
>  		kthread_stop(aac->thread);
>  	}
> -	aac_send_shutdown(aac);
> +
>  	aac_adapter_disable_int(aac);
>  	cpu = cpumask_first(cpu_online_mask);
>  	if (aac->pdev->device == PMC_DEVICE_S6 ||
> @@ -1193,7 +1220,7 @@ static int aac_probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
>  	if (!aac->fibs)
>  		goto out_free_host;
>  	spin_lock_init(&aac->fib_lock);
> -
> +	mutex_init(&aac->ioctl_mutex);
>  	/*
>  	 *	Map in the registers from the adapter.
>  	 */
> @@ -1474,7 +1501,10 @@ static int aac_resume(struct pci_dev *pdev)
>  	* reset this flag to unblock ioctl() as it was set at
>  	* aac_send_shutdown() to block ioctls from upperlayer
>  	*/
> +	mutex_lock(&aac->ioctl_mutex);
>  	aac->adapter_shutdown = 0;
> +	mutex_unlock(&aac->ioctl_mutex);

A mutex surrounding adapter_shutdown = 0; is needless.

> +
>  	scsi_unblock_requests(shost);
>  
>  	return 0;
> @@ -1633,7 +1663,10 @@ static void aac_pci_resume(struct pci_dev *pdev)
>  	 * reset this flag to unblock ioctl() as it was set
>  	 * at aac_send_shutdown() to block ioctls from upperlayer
>  	 */
> +	mutex_lock(&aac->ioctl_mutex);
>  	aac->adapter_shutdown = 0;
> +	mutex_unlock(&aac->ioctl_mutex);

Same here.


aac_compat_do_ioctl and  aac_do_ioctl is surrounded by mutexes - that is fine,
but by moving the mutex down to  aac_do_ioctl the code would be easier
and more readable.

With the new mutex, you have changed the functionality so that a new patch
is needed for that - before this patch there was no mutex in the main
.ioctl path.

I have found no functional issues so regardless whether you follow
my comments now, later or never -

Reviewed-by: Tomas Henzl <thenzl@redhat.com>

Cheers
Tomas









> +
>  	aac->handle_pci_error = 0;
>  
>  	shost_for_each_device(sdev, shost)