Re: [PATCH] netdev: Use flexible array for trailing private bytes

[Greg KH]

On Thu, Feb 29, 2024 at 10:59:10PM -0800, Jakub Kicinski wrote:
> On Thu, 29 Feb 2024 13:30:22 -0800 Kees Cook wrote:
> > Introduce a new struct net_device_priv that contains struct net_device
> > but also accounts for the commonly trailing bytes through the "size" and
> > "data" members.
> 
> I'm a bit unclear on the benefit. Perhaps I'm unaccustomed to "safe C".
> 
> > As many dummy struct net_device instances exist still,
> > it is non-trivial to but this flexible array inside struct net_device
> 
> put
> 
> Non-trivial, meaning what's the challenge?
> We also do somewhat silly things with netdev lifetime, because we can't
> assume netdev gets freed by netdev_free(). Cleaning up the "embedders"
> would be beneficial for multiple reasons.
> 
> > itself. But we can add a sanity check in netdev_priv() to catch any
> > attempts to access the private data of a dummy struct.
> > 
> > Adjust allocation logic to use the new full structure.
> > 
> > Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
> 
> > diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
> > index 118c40258d07..b476809d0bae 100644
> > --- a/include/linux/netdevice.h
> > +++ b/include/linux/netdevice.h
> > @@ -1815,6 +1815,8 @@ enum netdev_stat_type {
> >  	NETDEV_PCPU_STAT_DSTATS, /* struct pcpu_dstats */
> >  };
> >  
> > +#define	NETDEV_ALIGN		32
> 
> Unless someone knows what this is for it should go.
> Align priv to cacheline size.
> 
> >  /**
> >   *	struct net_device - The DEVICE structure.
> >   *
> 
> > @@ -2665,7 +2673,14 @@ void dev_net_set(struct net_device *dev, struct net *net)
> >   */
> >  static inline void *netdev_priv(const struct net_device *dev)
> >  {
> > -	return (char *)dev + ALIGN(sizeof(struct net_device), NETDEV_ALIGN);
> > +	struct net_device_priv *priv;
> > +
> > +	/* Dummy struct net_device have no trailing data. */
> > +	if (WARN_ON_ONCE(dev->reg_state == NETREG_DUMMY))
> > +		return NULL;
> 
> This is a static inline with roughly 11,000 call sites, according to 
> a quick grep. Aren't WARN_ONCE() in static inlines creating a "once"
> object in every compilation unit where they get used?

It also, if this every trips, will reboot the box for those that run
with panic-on-warn set, is that something that you all really want?

thanks,

greg k-h