Re: [PATCH bpf-next v1 2/6] HID: bpf: allow to change the report descriptor from an eBPF program

[Benjamin Tissoires <benjamin.tissoires@redhat.com>]

On Sat, Feb 26, 2022 at 8:31 AM Song Liu <song@kernel.org> wrote:
>
> On Thu, Feb 24, 2022 at 3:09 AM Benjamin Tissoires
> <benjamin.tissoires@redhat.com> wrote:
> >
> > The report descriptor is the dictionary of the HID protocol specific
> > to the given device.
> > Changing it is a common habit in the HID world, and making that feature
> > accessible from eBPF allows to fix devices without having to install a
> > new kernel.
> >
> > Signed-off-by: Benjamin Tissoires <benjamin.tissoires@redhat.com>
>
> [...]
>
> > diff --git a/include/linux/hid.h b/include/linux/hid.h
> > index 8fd79011f461..66d949d10b78 100644
> > --- a/include/linux/hid.h
> > +++ b/include/linux/hid.h
> > @@ -1213,10 +1213,16 @@ do {                                                                    \
> >
> >  #ifdef CONFIG_BPF
> >  u8 *hid_bpf_raw_event(struct hid_device *hdev, u8 *rd, int *size);
> > +u8 *hid_bpf_report_fixup(struct hid_device *hdev, u8 *rdesc, unsigned int *size);
> >  int hid_bpf_module_init(void);
> >  void hid_bpf_module_exit(void);
> >  #else
> >  static inline u8 *hid_bpf_raw_event(struct hid_device *hdev, u8 *rd, int *size) { return rd; }
> > +static inline u8 *hid_bpf_report_fixup(struct hid_device *hdev, u8 *rdesc,
> > +                                      unsigned int *size)
> > +{
> > +       return kmemdup(rdesc, *size, GFP_KERNEL);
> > +}
> >  static inline int hid_bpf_module_init(void) { return 0; }
> >  static inline void hid_bpf_module_exit(void) {}
> >  #endif
> > diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
> > index 5978b92cacd3..a7a8d9cfcf24 100644
> > --- a/include/uapi/linux/bpf.h
> > +++ b/include/uapi/linux/bpf.h
> > @@ -999,6 +999,7 @@ enum bpf_attach_type {
> >         BPF_SK_REUSEPORT_SELECT_OR_MIGRATE,
> >         BPF_PERF_EVENT,
> >         BPF_HID_DEVICE_EVENT,
> > +       BPF_HID_RDESC_FIXUP,
> >         __MAX_BPF_ATTACH_TYPE
> >  };
> >
> > diff --git a/include/uapi/linux/bpf_hid.h b/include/uapi/linux/bpf_hid.h
> > index 243ac45a253f..c0801d7174c3 100644
> > --- a/include/uapi/linux/bpf_hid.h
> > +++ b/include/uapi/linux/bpf_hid.h
> > @@ -18,6 +18,7 @@ struct hid_device;
> >  enum hid_bpf_event {
> >         HID_BPF_UNDEF = 0,
> >         HID_BPF_DEVICE_EVENT,
> > +       HID_BPF_RDESC_FIXUP,
> >  };
> >
> >  /* type is HID_BPF_DEVICE_EVENT */
> > @@ -26,12 +27,19 @@ struct hid_bpf_ctx_device_event {
> >         unsigned long size;
> >  };
> >
> > +/* type is HID_BPF_RDESC_FIXUP */
> > +struct hid_bpf_ctx_rdesc_fixup {
> > +       __u8 data[HID_BPF_MAX_BUFFER_SIZE];
> > +       unsigned long size;
> > +};
>
> This looks same as HID_BPF_DEVICE_EVENT, do we really need to
> separate the two?

I wanted to separate them because the other types have other requirements.

However, they all need a "data" with "size" associated. So I'll add
data and size to the common definition of the struct, leaving only the
specifics in the union (which means that DEVICE_EVENT and RDESC_FIXUP
won't have a definition in the union). I'll see the look of it before
submitting v2.

Cheers,
Benjamin