Message-ID: <whzoigqsbgmhxhk7qws4o63l23a5ravmpqyro4kvajomypljgf@x5pakwey5wej>
Date: Tue, 3 Sep 2024 17:05:49 +0200
From: Benjamin Tissoires <bentiss@...nel.org>
To: Peter Hutterer <peter.hutterer@...-t.net>
Cc: Jiri Kosina <jikos@...nel.org>, Vicki Pfau <vi@...rift.com>, 
	Shuah Khan <shuah@...nel.org>, Jonathan Corbet <corbet@....net>, linux-input@...r.kernel.org, 
	linux-kselftest@...r.kernel.org, linux-kernel@...r.kernel.org, bpf@...r.kernel.org, 
	linux-doc@...r.kernel.org
Subject: Re: [PATCH HID 4/7] HID: bpf: allow BPF programs to force using
 hid-generic

On Sep 03 2024, Peter Hutterer wrote:
> On Tue, Sep 03, 2024 at 01:14:34AM +0900, Benjamin Tissoires wrote:
> > The use case is when we fix a device through HID-BPF, 99% of the cases
> > we want the device to use hid-generic now instead of a dedicated device.
> 
> s/dedicated device/dedicated driver/ in the commit message
> 
> > That's because the dedicated device might also want to change the report
> > descriptor, or will be handling the device in a different way the new
> > fixed device is using.
> > 
> > In hid-core, after matching for the device (so that we only call this new
> > hook on compatible drivers), we call for `.hid_bpf_driver_probe`.
> > The function can not communicate with the device because it is not yet
> > started, but it can make educated guesses and decide to:
> > - let hid-core decide by itself
> > - force the use of this driver (by comparing the provided name with
> >   "hid-generic" for instance)
> > - force hid-core to ignore this driver for this device.
> > 
> > For API stability, we don't rely on a bitfield or a return value for
> > chosing hid-core behavior. We simply have a couple of writeable fields
> > in the new struct hid_bpf_driver, and then hid-core can make its educated
> > decision.
> > 
> > Signed-off-by: Benjamin Tissoires <bentiss@...nel.org>
> > ---
> >  Documentation/hid/hid-bpf.rst        |  2 +-
> >  drivers/hid/bpf/hid_bpf_dispatch.c   | 31 ++++++++++++++++++++++++++++
> >  drivers/hid/bpf/hid_bpf_struct_ops.c |  3 +++
> >  drivers/hid/hid-core.c               |  6 ++++++
> >  include/linux/hid_bpf.h              | 40 ++++++++++++++++++++++++++++++++++++
> >  5 files changed, 81 insertions(+), 1 deletion(-)
> > 
> > diff --git a/Documentation/hid/hid-bpf.rst b/Documentation/hid/hid-bpf.rst
> > index 5939eeafb361..05a43f11cdab 100644
> > --- a/Documentation/hid/hid-bpf.rst
> > +++ b/Documentation/hid/hid-bpf.rst
> > @@ -190,7 +190,7 @@ User API data structures available in programs:
> >  -----------------------------------------------
> >  
> >  .. kernel-doc:: include/linux/hid_bpf.h
> > -   :identifiers: hid_bpf_ctx
> > +   :identifiers: hid_bpf_ctx hid_bpf_driver
> >  
> >  Available API that can be used in all HID-BPF struct_ops programs:
> >  ------------------------------------------------------------------
> > diff --git a/drivers/hid/bpf/hid_bpf_dispatch.c b/drivers/hid/bpf/hid_bpf_dispatch.c
> > index a272a086c950..2df136d64152 100644
> > --- a/drivers/hid/bpf/hid_bpf_dispatch.c
> > +++ b/drivers/hid/bpf/hid_bpf_dispatch.c
> > @@ -189,6 +189,37 @@ u8 *call_hid_bpf_rdesc_fixup(struct hid_device *hdev, u8 *rdesc, unsigned int *s
> >  }
> >  EXPORT_SYMBOL_GPL(call_hid_bpf_rdesc_fixup);
> >  
> > +int call_hid_bpf_driver_probe(struct hid_device *hdev, struct hid_driver *hdrv,
> > +			      const struct hid_device_id *id)
> > +{
> > +	struct hid_bpf_driver drv = { 0 };
> > +	struct hid_bpf_ops *e;
> > +	int idx;
>  +
> > +	if (strscpy(drv.name, hdrv->name, sizeof(drv.name)) < 0)
> > +		return 0;
> > +
> > +	idx = srcu_read_lock(&hdev->bpf.srcu);
> > +	list_for_each_entry_srcu(e, &hdev->bpf.prog_list, list,
> > +				 srcu_read_lock_held(&hdev->bpf.srcu)) {
> > +		if (!e->hid_driver_probe)
> > +			continue;
> > +
> > +		e->hid_driver_probe(hdev, &drv, id);
> > +	}
> > +
> > +	srcu_read_unlock(&hdev->bpf.srcu, idx);
> > +
> > +	if (drv.force_driver)
> > +		return 1;
> > +
> > +	if (drv.ignore_driver)
> > +		return -1;
> > +
> > +	return 0;
> > +}
> > +EXPORT_SYMBOL_GPL(call_hid_bpf_driver_probe);
> > +
> >  static int device_match_id(struct device *dev, const void *id)
> >  {
> >  	struct hid_device *hdev = to_hid_device(dev);
> > diff --git a/drivers/hid/bpf/hid_bpf_struct_ops.c b/drivers/hid/bpf/hid_bpf_struct_ops.c
> > index cd696c59ba0f..1e13a22f73a1 100644
> > --- a/drivers/hid/bpf/hid_bpf_struct_ops.c
> > +++ b/drivers/hid/bpf/hid_bpf_struct_ops.c
> > @@ -46,6 +46,7 @@ static int hid_bpf_ops_check_member(const struct btf_type *t,
> >  	case offsetof(struct hid_bpf_ops, hid_rdesc_fixup):
> >  	case offsetof(struct hid_bpf_ops, hid_hw_request):
> >  	case offsetof(struct hid_bpf_ops, hid_hw_output_report):
> > +	case offsetof(struct hid_bpf_ops, hid_driver_probe):
> >  		break;
> >  	default:
> >  		if (prog->sleepable)
> > @@ -79,6 +80,8 @@ static int hid_bpf_ops_btf_struct_access(struct bpf_verifier_log *log,
> >  		WRITE_RANGE(hid_device, name, true),
> >  		WRITE_RANGE(hid_device, uniq, true),
> >  		WRITE_RANGE(hid_device, phys, true),
> > +		WRITE_RANGE(hid_bpf_driver, force_driver, false),
> > +		WRITE_RANGE(hid_bpf_driver, ignore_driver, false),
> >  	};
> >  #undef WRITE_RANGE
> >  	const struct btf_type *state = NULL;
> > diff --git a/drivers/hid/hid-core.c b/drivers/hid/hid-core.c
> > index 988d0acbdf04..7845f0a789ec 100644
> > --- a/drivers/hid/hid-core.c
> > +++ b/drivers/hid/hid-core.c
> > @@ -2639,10 +2639,16 @@ static bool hid_check_device_match(struct hid_device *hdev,
> >  				   struct hid_driver *hdrv,
> >  				   const struct hid_device_id **id)
> >  {
> > +	int ret;
> > +
> >  	*id = hid_match_device(hdev, hdrv);
> >  	if (!*id)
> >  		return false;
> >  
> > +	ret = call_hid_bpf_driver_probe(hdev, hdrv, *id);
> > +	if (ret)
> > +		return ret > 0;
> > +
> >  	if (hdrv->match)
> >  		return hdrv->match(hdev, hid_ignore_special_drivers);
> >  
> > diff --git a/include/linux/hid_bpf.h b/include/linux/hid_bpf.h
> > index d4d063cf63b5..20693c218857 100644
> > --- a/include/linux/hid_bpf.h
> > +++ b/include/linux/hid_bpf.h
> > @@ -9,6 +9,7 @@
> >  #include <uapi/linux/hid.h>
> >  
> >  struct hid_device;
> > +struct hid_driver;
> >  
> >  /*
> >   * The following is the user facing HID BPF API.
> > @@ -80,6 +81,22 @@ struct hid_ops {
> >  
> >  extern struct hid_ops *hid_ops;
> >  
> > +/**
> > + * struct hid_bpf_driver - User accessible data for the ``hid_bpf_probe``
> > + * struct_ops
> > + *
> > + * @name: the name of the driver currently being treated
> > + * @force_driver: set this to ``true`` to force hid-core to use this driver,
> > + *		  bypassing any further decision made by this driver
> > + * @ignore_driver: set this to ``true`` to force hid-core to ignore this driver,
> > + *                bypassing any further decision made by this driver
> 
> If I set both to false or true, what happens? The two seem to be

force_driver has priority over ignore_driver.

> mutually exclusive, in userspace I'd use an enum here to have a
> NOOP/FORCE_DRIVER/IGNORE_DRIVER value range (that can be extended later).
> Maybe something like that is an option?

enum also has the advantage to be exported in vmlinux.h.

FWIW, the idea behind adding new fields in a struct was to get the
backward compatibility for free. Because the verifier/relocator will see
if we are using the correct field entries.

OTOH, maybe we can make the function return the afformended enum, and
drop those two fields.

I think we should probably abort processing of any bpf sets the return
value to anything else than NOOP.

I'll work a little bit more on that.

Cheers,
Benjamin

> 
> > + */
> > +struct hid_bpf_driver {
> > +	__u8 name[64];
> > +	bool force_driver;
> > +	bool ignore_driver;
> > +};
> > +
> >  /**
> >   * struct hid_bpf_ops - A BPF struct_ops of callbacks allowing to attach HID-BPF
> >   *			programs to a HID device
> > @@ -178,6 +195,25 @@ struct hid_bpf_ops {
> >  	 */
> >  	int (*hid_hw_output_report)(struct hid_bpf_ctx *ctx, u64 source);
> >  
> > +	/**
> > +	 * @hid_driver_probe: called before the kernel ``.probe()`` function
> > +	 *
> > +	 * It has the following arguments:
> > +	 *
> > +	 * ``hdev``: The HID device kernel representation
> > +	 *
> > +	 * ``hdrv``: A BPF partially writeable representation of a HID driver
> > +	 *
> > +	 * ``id``: The device match structure found in the driver
> > +	 *
> > +	 * Note that the device has not been started yet, and thus kfuncs like
> > +	 * ``hid_hw_output_report`` will likely fail.
> 
> Just to confirm, I can access the device's report descriptor though? For
> the devices that we're looking at (e.g. the foot pedals pretending to be
> an apple keyboard) the driver name and what we can set in HID_BPF_CONFIG
> are not going to be enough, we'll have to check the rdesc too.
> 
> Cheers,
>   Peter
> 
> > +	 *
> > +	 * This function is useful to force/ignore a given supported HID driver,
> > +	 * by writing ``true`` in ``hdrv->force_driver`` or ``hdrv->ignore_driver``
> > +	 */
> > +	void (*hid_driver_probe)(struct hid_device *hdev, struct hid_bpf_driver *hdrv,
> > +				 const struct hid_device_id *id);
> >  
> >  	/* private: do not show up in the docs */
> >  	struct hid_device *hdev;
> > @@ -213,6 +249,8 @@ void hid_bpf_disconnect_device(struct hid_device *hdev);
> >  void hid_bpf_destroy_device(struct hid_device *hid);
> >  int hid_bpf_device_init(struct hid_device *hid);
> >  u8 *call_hid_bpf_rdesc_fixup(struct hid_device *hdev, u8 *rdesc, unsigned int *size);
> > +int call_hid_bpf_driver_probe(struct hid_device *hdev, struct hid_driver *hdrv,
> > +			      const struct hid_device_id *id);
> >  #else /* CONFIG_HID_BPF */
> >  static inline u8 *dispatch_hid_bpf_device_event(struct hid_device *hid, enum hid_report_type type,
> >  						u8 *data, u32 *size, int interrupt,
> > @@ -228,6 +266,8 @@ static inline int hid_bpf_connect_device(struct hid_device *hdev) { return 0; }
> >  static inline void hid_bpf_disconnect_device(struct hid_device *hdev) {}
> >  static inline void hid_bpf_destroy_device(struct hid_device *hid) {}
> >  static inline int hid_bpf_device_init(struct hid_device *hid) { return 0; }
> > +static inline int call_hid_bpf_driver_probe(struct hid_device *hdev, struct hid_driver *hdrv,
> > +					    const struct hid_device_id *id) { return 0; }
> >  /*
> >   * This specialized allocator has to be a macro for its allocations to be
> >   * accounted separately (to have a separate alloc_tag). The typecast is
> > 
> > -- 
> > 2.46.0
> > 

Powered by blists - more mailing lists