Re: [Linux-kernel-mentees] [PATCH v1] usbhid: Fix slab-out-of-bounds write in hiddev_ioctl_usage()

From: Peilin Ye
Date: Tue Jul 21 2020 - 04:39:43 EST


On Tue, Jul 21, 2020 at 10:16:37AM +0300, Dan Carpenter wrote:
> For some reason the reply-to header on your email is bogus:
>
> Reply-To: 20200720121257.GJ2571@kadam
>
> "kadam" is a system on my home network.

Ah...I thought `Reply-To` and `In-Reply-To` are the same thing...Sorry
for the beginner's mistake...

> Yeah. And in the caller it does:
>
> drivers/hid/hid-core.c
> 297 if (!parser->local.usage_index) /* Ignore padding fields */
> 298 return 0;
> 299
> 300 usages = max_t(unsigned, parser->local.usage_index,
> ^^^^^^^^^^^^^^
> 301 parser->global.report_count);
> 302
> 303 field = hid_register_field(report, usages, parser->global.report_count);
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> So ->usages is always greater or equal to ->global.report_count.
>
> 304 if (!field)
> 305 return 0;
> 306
> 307 field->physical = hid_lookup_collection(parser, HID_COLLECTION_PHYSICAL);
>
> >
> > Here, `values` equals to `global.report_count`. See how it is being
> > called:
> >
> > drivers/hid/hid-core.c:303:
> >
> > field = hid_register_field(report, usages, parser->global.report_count);
> >
> > In hid_open_report(), `global.report_count` can be set by calling
> > hid_parser_global().
> >
> > However, the syzkaller reproducer made hid_open_report() to call
> > hid_parser_main() __before__ `global.report_count` is properly set. It's
> > zero. So hid_register_field() allocated `field` with `values` equals to
> > zero - No room for value[] at all. I believe this caused the bug.
>
> I don't know if zero is valid or not. I suspect it is valid. We have
> no reason to think that it's invalid.

I see, I will stop guessing and wait for the maintainers' feedback.

Thank you,

Peilin Ye