Re: [PATCH v5 net-next 02/12] net-shapers: implement NL get operation

[Jakub Kicinski <kuba@kernel.org>]

On Fri, 30 Aug 2024 12:55:05 +0200 Paolo Abeni wrote:
> On 8/30/24 03:20, Jakub Kicinski wrote:>> +/* Initialize the context 
> fetching the relevant device and
> >> + * acquiring a reference to it.
> >> + */
> >> +static int net_shaper_ctx_init(const struct genl_info *info, int type,
> >> +			       struct net_shaper_nl_ctx *ctx)
> >> +{
> >> +	struct net *ns = genl_info_net(info);
> >> +	struct net_device *dev;
> >> +	int ifindex;
> >> +
> >> +	memset(ctx, 0, sizeof(*ctx));
> >> +	if (GENL_REQ_ATTR_CHECK(info, type))
> >> +		return -EINVAL;
> >> +
> >> +	ifindex = nla_get_u32(info->attrs[type]);  
> > 
> > Let's limit the 'binding' thing to just driver call sites, we can
> > redo the rest easily later. This line and next pretends to take
> > "arbitrary" type but clearly wants a ifindex/netdev, right?  
> 
> There is a misunderstanding. This helper will be used in a following 
> patch (7/12) with a different 'type' argument: 
> NET_SHAPER_A_BINDING_IFINDEX. I've put a note in the commit message, but 
> was unintentionally dropped in one of the recent refactors. I'll add 
> that note back.

What I'm saying is that if you want to prep the ground for more
"binding" types you should also add:

	if (type != ...IFINDEX) {
		/* other binding types are TBD */
		return -EINVAL;
	}

> I hope you are ok with the struct net_shaper_binding * argument to most 
> helpers? does not add complexity, will help to support devlink objects 
> and swapping back and forth from/to struct net_device* can't be automated.

I am "okay" in the American sense of the word which AFAIU is "unhappy
but won't complain unless asked".

> > Maybe send a patch like this, to avoid having to allocate this space,
> > and special casing dump vs doit:
> > 
> > diff --git a/include/net/genetlink.h b/include/net/genetlink.h
> > index 9ab49bfeae78..7658f0885178 100644
> > --- a/include/net/genetlink.h
> > +++ b/include/net/genetlink.h
> > @@ -124,7 +124,8 @@ struct genl_family {
> >    * @genlhdr: generic netlink message header
> >    * @attrs: netlink attributes
> >    * @_net: network namespace
> > - * @user_ptr: user pointers
> > + * @ctx: storage space for the use by the family
> > + * @user_ptr: user pointers (deprecated, use ctx instead)
> >    * @extack: extended ACK report struct
> >    */
> >   struct genl_info {
> > @@ -135,7 +136,10 @@ struct genl_info {
> >   	struct genlmsghdr *	genlhdr;
> >   	struct nlattr **	attrs;
> >   	possible_net_t		_net;
> > -	void *			user_ptr[2];
> > +	union {
> > +		u8		ctx[48];
> > +		void *		user_ptr[2];
> > +	};
> >   	struct netlink_ext_ack *extack;
> >   };  
> 
> Makes sense. Plus likely:
> 
> #define NETLINK_CTX_SIZE 48
> 
> and use such define above and in linux/netlink.h

Aha, would be good to also have a checking macro. Maybe rename

NL_ASSERT_DUMP_CTX_FITS()

to apply more broadly? or add a new one? Weak preference for former.