On Tue, Nov 19, 2024 at 09:07:45AM -0800, Breno Leitao wrote:
> > > #endif
> > > bool enabled;
> > > bool extended;
> >
> > > + /* Check if CPU NR should be populated, and append it to the user
> > > + * dictionary.
> > > + */
> > > + if (child_count < MAX_USERDATA_ITEMS && nt->userdata_auto & AUTO_CPU_NR)
> > > + scnprintf(&nt->userdata_complete[complete_idx],
> > > + MAX_USERDATA_ENTRY_LENGTH, " cpu=%u\n",
> > > + raw_smp_processor_id());
> >
> > I guess it may be tricky for backward compat, but shouldn't we return
> > an error rather than silently skip?
>
> yes, this should be easy to do, in fact. Probably return -E2BIG to
> userspace when trying to update the entry. I thought about something as
> the following patch, and piggy-back into it.
Back to this topic, in fact, this is not needed at all.
The configfs make item helper (userdatum_make_item()) checks for
exceeding entries, and fails if an additional entry is created.
static struct config_item *userdatum_make_item(struct config_group
*group,
const char *name)
{
....
child_count = list_count_nodes(&nt->userdata_group.cg_children);
if (child_count >= MAX_USERDATA_ITEMS)
return ERR_PTR(-ENOSPC);
I've sent an additional test for this mechanism, and make the check in
update_userdata() a warning instead of just silently dropping the entry.
https://lore.kernel.org/all/20241204-netcons_overflow_test-v1-0-a85a8d0ac...@debian.org/
