Re: [PATCH v5 5/6] iio: core: add simple centralized mechanism for ioctl() handlers

["Ardelean, Alexandru" <alexandru.Ardelean@analog.com>]

On Sun, 2020-04-26 at 11:21 +0100, Jonathan Cameron wrote:
> [External]
> 
> On Sun, 26 Apr 2020 11:10:37 +0100
> Jonathan Cameron <jic23@kernel.org> wrote:
> 
> > On Sun, 26 Apr 2020 10:38:16 +0300
> > Alexandru Ardelean <alexandru.ardelean@analog.com> wrote:
> > 
> > > The aim of this is to reduce the organization violation of ioctl() calls
> > > in
> > > IIO core. Currently, since the chardev is split across files, event
> > > ioctl()
> > > calls need to be called in buffer ioctl() calls.
> > > 
> > > The 'industrialio-core.c' file will provide a 'iio_device_ioctl()' which
> > > will iterate over a list of ioctls registered with the IIO device. These
> > > can be event ioctl() or buffer ioctl() calls, or something else.
> > > This is needed, since there is currently one chardev per IIO device and
> > > that is used for both event handling and reading from the buffer.
> > > 
> > > Each ioctl() will have to return a IIO_IOCTL_UNHANDLED code (which is
> > > positive 1), if the ioctl() did not handle the call in any. This
> > > eliminates
> > > any potential ambiguities; if we were to have used error codes it would
> > > have been uncertain whether they were actual errors, or whether
> > > the registered ioctl() doesn't service the command.
> > > 
> > > If any ioctl() returns 0, it was considered that it was serviced
> > > successfully and the loop will exit.
> > > 
> > > One assumption for all registered ioctl() handlers is that they are
> > > statically allocated, so the iio_device_unregister() which just remove all
> > > of them from the device's ioctl() handler list.
> > > 
> > > Also, something that is a bit hard to do [at this point] and may not be
> > > worth the effort of doing, is to check whether registered ioctl()
> > > calls/commands overlap. This should be unlikely to happen, and should get
> > > caught at review time. Though, new ioctl() calls would likely not be added
> > > too often.
> > > 
> > > Signed-off-by: Alexandru Ardelean <alexandru.ardelean@analog.com>  
> > 
> > A question on locking inline. Otherwise this looks fairly clean and simple
> > to me.
> > 
> > Jonathan
> > 
> > 
> > > ---
> > >  drivers/iio/iio_core.h          | 14 ++++++++++++++
> > >  drivers/iio/industrialio-core.c | 33 +++++++++++++++++++++++++++++++++
> > >  include/linux/iio/iio.h         |  2 ++
> > >  3 files changed, 49 insertions(+)
> > > 
> > > diff --git a/drivers/iio/iio_core.h b/drivers/iio/iio_core.h
> > > index a527a66be9e5..34c3e19229d8 100644
> > > --- a/drivers/iio/iio_core.h
> > > +++ b/drivers/iio/iio_core.h
> > > @@ -17,6 +17,20 @@ struct iio_dev;
> > >  
> > >  extern struct device_type iio_device_type;
> > >  
> > > +#define IIO_IOCTL_UNHANDLED	1
> > > +struct iio_ioctl_handler {
> > > +	struct list_head entry;
> > > +	long (*ioctl)(struct iio_dev *indio_dev, struct file *filp,
> > > +		      unsigned int cmd, unsigned long arg);
> > > +};
> > > +
> > > +long iio_device_ioctl(struct iio_dev *indio_dev, struct file *filp,
> > > +		      unsigned int cmd, unsigned long arg);
> > > +
> > > +void iio_device_ioctl_handler_register(struct iio_dev *indio_dev,
> > > +				       struct iio_ioctl_handler *h);
> > > +void iio_device_ioctl_handler_unregister(struct iio_ioctl_handler *h);
> > > +
> > >  int __iio_add_chan_devattr(const char *postfix,
> > >  			   struct iio_chan_spec const *chan,
> > >  			   ssize_t (*func)(struct device *dev,
> > > diff --git a/drivers/iio/industrialio-core.c b/drivers/iio/industrialio-
> > > core.c
> > > index aec585cc8453..79e8fa8ff70b 100644
> > > --- a/drivers/iio/industrialio-core.c
> > > +++ b/drivers/iio/industrialio-core.c
> > > @@ -1531,6 +1531,7 @@ struct iio_dev *iio_device_alloc(int sizeof_priv)
> > >  	}
> > >  	dev_set_name(&dev->dev, "iio:device%d", dev->id);
> > >  	INIT_LIST_HEAD(&dev->buffer_list);
> > > +	INIT_LIST_HEAD(&dev->ioctl_handlers);
> > >  
> > >  	return dev;
> > >  }
> > > @@ -1584,6 +1585,33 @@ struct iio_dev *devm_iio_device_alloc(struct device
> > > *dev, int sizeof_priv)
> > >  }
> > >  EXPORT_SYMBOL_GPL(devm_iio_device_alloc);
> > >  
> > > +void iio_device_ioctl_handler_register(struct iio_dev *indio_dev,
> > > +				       struct iio_ioctl_handler *h)
> > > +{
> > > +	/* this assumes that all ioctl() handlers are statically allocated */
> > > +	list_add_tail(&h->entry, &indio_dev->ioctl_handlers);
> > > +}
> > > +
> > > +long iio_device_ioctl(struct iio_dev *indio_dev, struct file *filp,
> > > +		      unsigned int cmd, unsigned long arg)
> > > +{
> > > +	struct iio_ioctl_handler *h;
> > > +	int ret;
> > > +
> > > +	if (!indio_dev->info)
> > > +		return -ENODEV;
> > > +
> > > +	list_for_each_entry(h, &indio_dev->ioctl_handlers, entry) {
> > > +		ret = h->ioctl(indio_dev, filp, cmd, arg);
> > > +		if (ret == 0)
> > > +			return 0;
> > > +		if (ret != IIO_IOCTL_UNHANDLED)
> > > +			return ret;
> > > +	}
> > > +
> > > +	return -EINVAL;
> > > +}
> > > +
> > >  static int iio_check_unique_scan_index(struct iio_dev *indio_dev)
> > >  {
> > >  	int i, j;
> > > @@ -1695,6 +1723,8 @@ EXPORT_SYMBOL(__iio_device_register);
> > >   **/
> > >  void iio_device_unregister(struct iio_dev *indio_dev)
> > >  {
> > > +	struct iio_ioctl_handler *h, *t;
> > > +
> > >  	if (indio_dev->chrdev)
> > >  		cdev_device_del(indio_dev->chrdev, &indio_dev->dev);
> > >  	else
> > > @@ -1708,6 +1738,9 @@ void iio_device_unregister(struct iio_dev
> > > *indio_dev)
> > >  
> > >  	iio_disable_all_buffers(indio_dev);
> > >  
> > > +	list_for_each_entry_safe(h, t, &indio_dev->ioctl_handlers, entry)
> > > +		list_del(&h->entry);
> > > +  
> > 
> > Is there any chance anything is walking that list whilst we are deleting it?
> > I think this needs to happen under a lock as does the walk.
> 
> We may want to use the rcu methods to make the walk and delete safe whilst
> avoiding heavy weight locking on the read path
> 
> linux/rculist.h
> 
> I've not thought about this in much depth though so take that advice as
> a vague suggestion and nothing more!

I'm a bit tempted to do the simple mutex approach, but I'll take a look at the
rculist.h thingi.

The point is good about the list being walked while being deleted.

> 
> 
> > >  	indio_dev->info = NULL;
> > >  
> > >  	iio_device_wakeup_eventset(indio_dev);
> > > diff --git a/include/linux/iio/iio.h b/include/linux/iio/iio.h
> > > index 52992be44e9e..b6ca8d85629e 100644
> > > --- a/include/linux/iio/iio.h
> > > +++ b/include/linux/iio/iio.h
> > > @@ -488,6 +488,7 @@ struct iio_buffer_setup_ops {
> > >   * @currentmode:	[DRIVER] current operating mode
> > >   * @dev:		[DRIVER] device structure, should be assigned a parent
> > >   *			and owner
> > > + * @ioctl_handlers:	[INTERN] list of registered ioctl handlers
> > >   * @event_interface:	[INTERN] event chrdevs associated with interrupt
> > > lines
> > >   * @buffer:		[DRIVER] any buffer present
> > >   * @buffer_list:	[INTERN] list of all buffers currently attached
> > > @@ -529,6 +530,7 @@ struct iio_dev {
> > >  	int				modes;
> > >  	int				currentmode;
> > >  	struct device			dev;
> > > +	struct list_head		ioctl_handlers;
> > >  
> > >  	struct iio_event_interface	*event_interface;
> > >