RE: [PATCH 04/12] misc: xilinx_sdfec: Add open, close and ioctl

[Dragan Cvetic]

> -----Original Message-----
> From: Arnd Bergmann [mailto:arnd@xxxxxxxx]
> Sent: Tuesday 19 March 2019 13:18
> To: Dragan Cvetic <draganc@xxxxxxxxxx>
> Cc: gregkh <gregkh@xxxxxxxxxxxxxxxxxxx>; Michal Simek <michals@xxxxxxxxxx>; Linux ARM <linux-arm-kernel@xxxxxxxxxxxxxxxxxxx>;
> Derek Kiernan <dkiernan@xxxxxxxxxx>; Linux Kernel Mailing List <linux-kernel@xxxxxxxxxxxxxxx>
> Subject: Re: [PATCH 04/12] misc: xilinx_sdfec: Add open, close and ioctl
> 
> On Tue, Mar 19, 2019 at 1:05 PM Dragan Cvetic <dragan.cvetic@xxxxxxxxxx> wrote:
> >
> > Add char device interface per DT node present and support
> > file operations:
> > - open(), which keeps only one open per device at a time,
> > - close(), which release the open for this device,
> > - ioctl(), which provides infrastructure for a specific driver
> > control.
> 
> >  drivers/misc/xilinx_sdfec.c      | 79 ++++++++++++++++++++++++++++++++++++++++
> >  include/uapi/misc/xilinx_sdfec.h |  4 ++
> >  2 files changed, 83 insertions(+)
> >
> > diff --git a/drivers/misc/xilinx_sdfec.c b/drivers/misc/xilinx_sdfec.c
> > index a52a5c6..3407de4 100644
> > --- a/drivers/misc/xilinx_sdfec.c
> > +++ b/drivers/misc/xilinx_sdfec.c
> > @@ -81,8 +81,87 @@ struct xsdfec_dev {
> >         struct xsdfec_clks clks;
> >  };
> >
> > +static int xsdfec_dev_open(struct inode *iptr, struct file *fptr)
> > +{
> > +       struct xsdfec_dev *xsdfec;
> > +
> > +       xsdfec = container_of(iptr->i_cdev, struct xsdfec_dev, xsdfec_cdev);
> > +       if (!xsdfec)
> > +               return -EAGAIN;
> 
> The result of container_of() will not be NULL here.
> Did you mean to check i_cdev? That probably also won't
> be NULL, but that check would be more reasonable.


Will be either removed fully or changed with i_cdev check

> 
> > +       /* Only one open per device at a time */
> > +       if (!atomic_dec_and_test(&xsdfec->open_count)) {
> > +               atomic_inc(&xsdfec->open_count);
> > +               return -EBUSY;
> > +       }
> 
> What is that limitation for? Is it worse to open it twice than
> to dup() or fork()?
>
The device can be opened only once.
 
> Note that the test is not really atomic either: if three processes
> try to open the file at the same time, it gets decremented from
> 1 to -2, so only the second one sees 0 and increments it back
> to -1 afterwards...

It looks you are right. Will fix this. Thank you for the catch.

> 
> > +static long xsdfec_dev_ioctl(struct file *fptr, unsigned int cmd,
> > +                            unsigned long data)
> > +{
> > +       struct xsdfec_dev *xsdfec = fptr->private_data;
> > +       void __user *arg = NULL;
> > +       int rval = -EINVAL;
> > +       int err = 0;
> > +
> > +       if (!xsdfec)
> > +               return rval;
> > +
> > +       if (_IOC_TYPE(cmd) != XSDFEC_MAGIC) {
> > +               dev_err(xsdfec->dev, "Not a xilinx sdfec ioctl");
> > +               return -ENOTTY;
> > +       }
> 
> remove the error messages here as well.
> 
> > +       /* Access check of the argument if present */
> > +       if (_IOC_DIR(cmd) & _IOC_READ)
> > +               err = !access_ok((void *)arg, _IOC_SIZE(cmd));
> > +       else if (_IOC_DIR(cmd) & _IOC_WRITE)
> > +               err = !access_ok((void *)arg, _IOC_SIZE(cmd));
> 
> This seems odd. Why two separate checks, and why the access_ok()
> check when you do a copy_from_user() that contains the same check
> later?

Accepted, will remove it. 

> 
> If you want to get fancy here, you could just copy the data in the main
> ioctl handler based on _IOC_SIZE, and pass around normal kernel
> pointers from there.


Will not be fancy. Thank you for the advice.

> 
> >  static const struct file_operations xsdfec_fops = {
> >         .owner = THIS_MODULE,
> > +       .open = xsdfec_dev_open,
> > +       .release = xsdfec_dev_release,
> > +       .unlocked_ioctl = xsdfec_dev_ioctl,
> >  };
> 
> This lacks a .compat_ioctl pointer.

This is new for me, I have to investigate more and propose a solution.
Thank you for suggestion.

> 
>        Arnd