Re: WARNING: CPU: 26 PID: 64391 at lib/vsprintf.c:2193 set_precision+0x84/0x90

[Petr Mladek <pmladek@suse.com>]

On Wed 2018-11-14 11:38:19, Andy Shevchenko wrote:
> On Wed, Nov 14, 2018 at 12:05:12AM +0100, Petr Mladek wrote:
> > On Tue 2018-11-13 14:23:17, Steven Rostedt wrote:
> > > On Tue, 13 Nov 2018 13:58:18 -0500
> > > Qian Cai <cai@gmx.us> wrote:
> > > 
> > > > > Care to print the len and name parameters before this line?  
> > > > len = 60612; name =
> > > 
> > > How big are pages on arm64? Because we shouldn't get to this path if
> > > the string is bigger than PAGE_SIZE. But I know that on PPC64,
> > > PAGE_SIZE can be 64K, and 60612 is less than that. Thus, if we get
> > > there, the test is against signed int:16 (16 bit signed integer) that
> > > can go up to most 32768. If the string size is bigger than that, you
> > > would get this error.
> > > 
> > > I would just say to ignore it.
> > 
> > I tend to agree.
> > 
> > > The only thing that can happen if
> > > someone does this is to trigger the warning. Unless if it is considered
> > > a form of DOS, where userspace just bombards the console by triggering
> > > this waring.
> > 
> > We are actually on the safe side because it is WARN_ONCE().
> > 
> > > But I don't see a problem with the actual design. There's
> > > no reason we should be processing string variables bigger than 32768 in
> > > vsprintf.
> > 
> > It is not even needed in this case. The string is limited also by
> > MODULE_NAME_LEN.
> 
> At least not in this code.
> 
> Are you proposing to replace strlen(name) with strnlen(name, MODULE_NAME_LEN)?

It might be a solution. Well, it looks like a wrong design when we
would need to use MODULE_NAME_LEN outside module loader code. Also
it does not handle other request_module() users that might be
affected.

On the other hand, I am not sure how a proper solution would look
like. request_module() should not limit printk format before
the arguments are substituted.

The most clean solution probably would be on the vsprintf-level.
I mean to limit the precision by the overall string length
limit. But it looks a bit weird as well.

I still tend to ignore it. The code is safe from the security point of
view. The warning would trigger only when completely misused.


Best Regards,
Petr