Re: [PATCH 20/24] compat_ioctl: move HDIO ioctl handling into drivers/ide

[Arnd Bergmann]

On Thu, Dec 12, 2019 at 5:29 PM Christoph Hellwig <hch@xxxxxxxxxxxxx> wrote:
>
> > +static int put_user_long(long val, unsigned long arg)
> > +{
> > +#ifdef CONFIG_COMPAT
> > +     if (in_compat_syscall())
> > +             return put_user(val, (compat_long_t __user *)compat_ptr(arg));
> > +#endif
> > +     return put_user(val, (long __user *)arg);
> > +}
>
> We had this
>
> #ifdef CONFIG_COMPAT
>         if (in_compat_syscall())
>                 ...
>         ...
> #endif
>
> patter quite frequently.  Can we define a in_compat_syscall stub
> and make sure compat_ptr and the compat_* types are available available
> to clean this up a bit?

in_compat_syscall() already has a reasonable stub, what was missing
when I created this series is the compat_long_t definition and
the compat_ptr() helper.

With patch 01/24, we can rely on the types, but I still need to add
a generic compat_ptr() implementation for non-s390 to remove the
#ifdef here.

That is probably a good idea anyway, it just needs a bit of testing.
The last time I tried this, I ran into problems with the order of compat
header inclusions on some architectures.

I've made a prototype now, will see how it goes.

> > -     if (NULL == (void *) arg) {
> > +     if (NULL == argp) {
>
>         if (!argp) {
>
> ?

changed now.

      Arnd