Re: [PATCH v4 2/4] sysctl: Move some boundary constants form sysctl.c to sysctl_vals

[Luis Chamberlain <mcgrof@kernel.org>]

On Wed, May 20, 2020 at 09:14:08AM +0800, Xiaoming Ni wrote:
> On 2020/5/19 12:44, Tetsuo Handa wrote:
> > On 2020/05/19 12:31, Xiaoming Ni wrote:
> > > Some boundary (.extra1 .extra2) constants (E.g: neg_one two) in
> > > sysctl.c are used in multiple features. Move these variables to
> > > sysctl_vals to avoid adding duplicate variables when cleaning up
> > > sysctls table.
> > > 
> > > Signed-off-by: Xiaoming Ni <nixiaoming@huawei.com>
> > > Reviewed-by: Kees Cook <keescook@chromium.org>
> > 
> > I feel that it is use of
> > 
> > 	void *extra1;
> > 	void *extra2;
> > 
> > in "struct ctl_table" that requires constant values indirection.
> > Can't we get rid of sysctl_vals using some "union" like below?
> > 
> > struct ctl_table {
> > 	const char *procname;           /* Text ID for /proc/sys, or zero */
> > 	void *data;
> > 	int maxlen;
> > 	umode_t mode;
> > 	struct ctl_table *child;        /* Deprecated */
> > 	proc_handler *proc_handler;     /* Callback for text formatting */
> > 	struct ctl_table_poll *poll;
> > 	union {
> > 		void *min_max_ptr[2];
> > 		int min_max_int[2];
> > 		long min_max_long[2];
> > 	};
> > } __randomize_layout;
> > 
> > .
> > 
> 
> net/decnet/dn_dev.c:
> static void dn_dev_sysctl_register(struct net_device *dev, struct
> dn_dev_parms *parms)
> {
> 	struct dn_dev_sysctl_table *t;
> 	int i;
> 
> 	char path[sizeof("net/decnet/conf/") + IFNAMSIZ];
> 
> 	t = kmemdup(&dn_dev_sysctl, sizeof(*t), GFP_KERNEL);
> 	if (t == NULL)
> 		return;
> 
> 	for(i = 0; i < ARRAY_SIZE(t->dn_dev_vars) - 1; i++) {
> 		long offset = (long)t->dn_dev_vars[i].data;
> 		t->dn_dev_vars[i].data = ((char *)parms) + offset;
> 	}
> 
> 	snprintf(path, sizeof(path), "net/decnet/conf/%s",
> 		dev? dev->name : parms->name);
> 
> 	t->dn_dev_vars[0].extra1 = (void *)dev;
> 
> 	t->sysctl_header = register_net_sysctl(&init_net, path, t->dn_dev_vars);
> 	if (t->sysctl_header == NULL)
> 		kfree(t);
> 	else
> 		parms->sysctl = t;
> }
> 
> A small amount of code is not used as a boundary value when using extra1.
> This scenario may not be suitable for renaming to min_max_ptr.
> 
> Should we add const to extra1 extra2 ?
> 
> --- a/include/linux/sysctl.h
> +++ b/include/linux/sysctl.h
> @@ -124,8 +124,8 @@ struct ctl_table {
>         struct ctl_table *child;        /* Deprecated */
>         proc_handler *proc_handler;     /* Callback for text formatting */
>         struct ctl_table_poll *poll;
> -       void *extra1;
> -       void *extra2;
> +       const void *extra1;
> +       const void *extra2;
>  } __randomize_layout;

Do that, compile an allyesconfig and it'll fail, but if you fix the
callers so that they use a const, then yes. That would cover only your
architecture. It is unclear if we ever used non-const for this on purpose.

  Luis