Re: invalid read in

[Phil Sutter <phil@nwl.cc>]

Hi,

On Mon, Feb 03, 2020 at 05:14:45PM +0000, dyslexicatheist wrote:
> On Monday, February 3, 2020 4:31 PM, Phil Sutter <phil@nwl.cc> wrote:
> > On Mon, Feb 03, 2020 at 01:54:31PM +0000, dyslexicatheist wrote:
> >
> > I guess this is the typical "problem" situation in which userspace uses
> > a non-zeroed buffer to feed into sendto() and due to padding not
> > every byte was written to. So basically userspace "leaks" garbage to
> > kernel, which is something I'd consider harmless and merely a minor
> > inconvenience when analyzing with valgrind. I usually suffer from this
> > as well since libmnl()'s allocation routines don't zero the buffer
> > either.
> >
> > In your case, I'd say the error message disappears if you add
> > 'memset(&u, 0, sizeof(u))' to the beginning of nfq_set_mode().
> 
> thanks for your help Phil. I have just tried this but unfortunately it didn't change the outcome. Also tried other variations such as memset'ing both &u and the &params struct, but nada.

Maybe you need to apply the same to __build_send_cfg_msg() as well?

Cheers, Phil