Re: [PATCH] Bluetooth: vhci: Fix race at creating hci device

[Takashi Iwai <tiwai@suse.de>]

On Wed, 20 Apr 2016 15:16:57 +0200,
Marcel Holtmann wrote:
> 
> Hi Takashi,
> 
> > hci_vhci driver creates a hci device object dynamically upon each
> > HCI_VENDOR_PKT write.  Although it checks the already created object
> > and returns an error, it's still racy and may build multiple hci_dev
> > objects concurrently when parallel writes are performed, as the device
> > tracks only a single hci_dev object.
> > 
> > This patch introduces a mutex to protect against the concurrent device
> > creations.
> > 
> > Cc: <stable@vger.kernel.org>
> > Signed-off-by: Takashi Iwai <tiwai@suse.de>
> > ---
> > drivers/bluetooth/hci_vhci.c | 22 ++++++++++++++++------
> > 1 file changed, 16 insertions(+), 6 deletions(-)
> > 
> > diff --git a/drivers/bluetooth/hci_vhci.c b/drivers/bluetooth/hci_vhci.c
> > index f67ea1c090cb..39230f30f544 100644
> > --- a/drivers/bluetooth/hci_vhci.c
> > +++ b/drivers/bluetooth/hci_vhci.c
> > @@ -50,6 +50,7 @@ struct vhci_data {
> > 	wait_queue_head_t read_wait;
> > 	struct sk_buff_head readq;
> > 
> > +	struct mutex open_mutex;
> > 	struct delayed_work open_timeout;
> > };
> > 
> > @@ -87,7 +88,7 @@ static int vhci_send_frame(struct hci_dev *hdev, struct sk_buff *skb)
> > 	return 0;
> > }
> > 
> > -static int vhci_create_device(struct vhci_data *data, __u8 opcode)
> > +static int __vhci_create_device(struct vhci_data *data, __u8 opcode)
> > {
> > 	struct hci_dev *hdev;
> > 	struct sk_buff *skb;
> > @@ -151,6 +152,19 @@ static int vhci_create_device(struct vhci_data *data, __u8 opcode)
> > 	return 0;
> > }
> > 
> > +static int vhci_create_device(struct vhci_data *data, __u8 opcode)
> > +{
> > +	int err;
> > +
> > +	mutex_lock(&data->open_mutex);
> > +	if (data->hdev)
> > +		err = -EBADFD;
> > +	else
> > +		err = __vhci_create_device(data, opcode);
> > +	mutex_unlock(&data->open_mutex);
> > +	return err;
> > +}
> > +
> > static inline ssize_t vhci_get_user(struct vhci_data *data,
> > 				    struct iov_iter *from)
> > {
> > @@ -191,11 +205,6 @@ static inline ssize_t vhci_get_user(struct vhci_data *data,
> > 	case HCI_VENDOR_PKT:
> > 		cancel_delayed_work_sync(&data->open_timeout);
> > 
> > -		if (data->hdev) {
> > -			kfree_skb(skb);
> > -			return -EBADFD;
> > -		}
> > -
> 
> why not just have the mutex around this block and the vhci_create_device in the timeout. Wouldn't that achieve exactly the same.

It's just a matter of taste :)  I prefer avoiding the duplicated
codes; instead of open-coding mutex_lock/unlock and data->hdev check
in two places, do it in the common helper.  If you prefer other way,
I'm fine with it.  Just let me know.  I'll resubmit the patch.

> Since when you actually remove this check, then you still can create another hci_dev by just writing another vendor packet. That is actually something we want to avoid.

No, it won't happen.  The removal of data->hdev in the above is merely
moving the check into the mutex protection in vhci_create_device().


thanks,

Takashi