Re: Subject: [PATCH v3] USB:Core: BugFix: Proper handling of Race Condition when two USB class drivers try to call init_usb_class simultaneously

Re: Subject: [PATCH v3] USB:Core: BugFix: Proper handling of Race Condition when two USB class drivers try to call init_usb_class simultaneously

[Ajay Kaher]

[-- Attachment #1: Type: text/plain, Size: 3286 bytes --]

 
On Thu, 16 Feb 2017, Alan Stern wrote: 

> On Thu, 16 Feb 2017, Ajay Kaher wrote:
> 
>> > On Thu, 14 Feb 2017, Alan Stern wrote:
>> > 
>> > I think Ajay's argument is correct and a patch is needed.  But this
>> > patch misses the race between init_usb_class() and release_usb_class().  
>> 
>> Thanks Alan for your comments, in patch v2 I have taken care for
>> release_usb_class() also. Please review again.
>> 
>> > The basic problem is that reference counting doesn't work when you try
>> > to use the same global pointer (usb_class) to refer to multiple
>> > generations of a dynamically allocated entity.  We had the same sort of
>> > problem many years ago with the usb_interface structure (and we
>> > ultimately fixed it by creating a separate usb_interface_cache
>> > structure).
>> >  
>> > The best approach here would be to forget about all the reference
>> > counting.  Get rid of usb_class entirely, and create the "usbmisc"
>> > class structure just once, when usbcore initializes.  Or, if you
>> > prefer, use a mutex to protect a routine that allocates the class
>> > structure dynamically, just once.  Either way, don't deallocate it
>> > until usbcore is unloaded.
>> 
>> usbmisc class creation should not require everytime when USB core
>> initializes. So better to keep usbmisc class creation as it is. 
>> And to prevent the race conditions just protect it with Mutex locking
>> as per patch v2.
> 
> This is not right.  What happens if usb_register_dev() runs just before
> release_usb_class() calls mutex_lock()?
 
Alan, as per my understanding I have shifted the lock from
release_usb_class() to destroy_usb_class() in patch v3. 
If it is not right, please explain in detail which race condition
I have missed and also share your suggestions.

thanks,
ajay kaher

Signed-off-by: Ajay Kaher

---

 drivers/usb/core/file.c |    6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/drivers/usb/core/file.c b/drivers/usb/core/file.c
index 822ced9..a12d184 100644
--- a/drivers/usb/core/file.c
+++ b/drivers/usb/core/file.c
@@ -27,6 +27,7 @@
 #define MAX_USB_MINORS 256
 static const struct file_operations *usb_minors[MAX_USB_MINORS];
 static DECLARE_RWSEM(minor_rwsem);
+static DEFINE_MUTEX(init_usb_class_mutex);

 static int usb_open(struct inode *inode, struct file *file)
 {
@@ -109,8 +110,10 @@ static void release_usb_class(struct kref *kref)

 static void destroy_usb_class(void)
 {
+       mutex_lock(&init_usb_class_mutex);
        if (usb_class)
                kref_put(&usb_class->kref, release_usb_class);
+       mutex_unlock(&init_usb_class_mutex);
 }

 int usb_major_init(void)
@@ -171,7 +174,10 @@ int usb_register_dev(struct usb_interface *intf,
        if (intf->minor >= 0)
                return -EADDRINUSE;

+       mutex_lock(&init_usb_class_mutex);
        retval = init_usb_class();
+       mutex_unlock(&init_usb_class_mutex);
+
        if (retval)
                return retval;







 

Re: Subject: [PATCH v3] USB:Core: BugFix: Proper handling of Race Condition when two USB class drivers try to call init_usb_class simultaneously

[Alan Stern]

On Mon, 20 Feb 2017, Ajay Kaher wrote:

> Alan, as per my understanding I have shifted the lock from
> release_usb_class() to destroy_usb_class() in patch v3. 
> If it is not right, please explain in detail which race condition
> I have missed and also share your suggestions.
> 
> thanks,
> ajay kaher
> 
> Signed-off-by: Ajay Kaher
> 
> ---
> 
>  drivers/usb/core/file.c |    6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/drivers/usb/core/file.c b/drivers/usb/core/file.c
> index 822ced9..a12d184 100644
> --- a/drivers/usb/core/file.c
> +++ b/drivers/usb/core/file.c
> @@ -27,6 +27,7 @@
>  #define MAX_USB_MINORS 256
>  static const struct file_operations *usb_minors[MAX_USB_MINORS];
>  static DECLARE_RWSEM(minor_rwsem);
> +static DEFINE_MUTEX(init_usb_class_mutex);
> 
>  static int usb_open(struct inode *inode, struct file *file)
>  {
> @@ -109,8 +110,10 @@ static void release_usb_class(struct kref *kref)
> 
>  static void destroy_usb_class(void)
>  {
> +       mutex_lock(&init_usb_class_mutex);
>         if (usb_class)
>                 kref_put(&usb_class->kref, release_usb_class);
> +       mutex_unlock(&init_usb_class_mutex);
>  }
> 
>  int usb_major_init(void)
> @@ -171,7 +174,10 @@ int usb_register_dev(struct usb_interface *intf,
>         if (intf->minor >= 0)
>                 return -EADDRINUSE;
> 
> +       mutex_lock(&init_usb_class_mutex);
>         retval = init_usb_class();
> +       mutex_unlock(&init_usb_class_mutex);
> +
>         if (retval)
>                 return retval;

Have you considered what would happen if destroy_usb_class() ran, but 
some other CPU was still holding a reference to usb_class?  And what if 
the last reference gets dropped later on, while init_usb_class() is 
running?

Maybe that's not possible here, but it is possible in general for 
refcounted objects.  So yes, this code is probably okay, but it isn't 
good form.

Alan Stern