RE: [PATCH] fat: Use pointer to d_name[0] in put_user() for compat case

From: David Laight
Date: Mon Feb 14 2022 - 06:32:14 EST


From: Helge Deller
> Sent: 14 February 2022 11:05
>
> On 2/14/22 10:26, David Laight wrote:
> > From: David Laight
> >> Sent: 14 February 2022 09:12
> >>
> >> From: Helge Deller
> >>> Sent: 13 February 2022 22:10
> >>>
> >>> The put_user(val,ptr) macro wants a pointer in the second parameter, but in
> >>> fat_ioctl_filldir() the d_name field references a whole "array of chars".
> >>> Usually the compiler automatically converts it and uses a pointer to that
> >>> array, but it's more clean to explicitly give the real pointer to where someting
> >>> is put, which is in this case the first character of the d_name[] array.
> >>
> >> That just isn't true.
> >>
> >> In C both x->char_array and &x->char_array[0] have the same type
> >> 'char *'.
> >>
> >> The 'bug' is caused by put_user() trying to do:
> >> __typeof__(ptr) __ptr = ptr;
> >> where __typeof__ is returning char[n] not char *.
> >>
> >> I've tried a few things but can't get __typeof__ to
> >> generate a suitable type for both a simple type and array.
> >
> > Actually the issue is that put_user() writes a single variable
> > and needs a pointer to one.
> > So changing to:
> > put_user(0, &array[0]);
> > is probably fine.
>
> Ok.
>
> > But the description is all wrong.
>
> I agree it can be improved.
> Would you mind proposing a better description?

put_user() needs a pointer to a simple type.

David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)