Re: [PATCH] comedi: zero-init data in do_insn_ioctl

[Lecomte, Arnaud]

Hi Ian,
Thanks for the feedback, I can send the revision of the patch by the end 
of the day if it is fine for you.


On 25/07/2025 10:41, Ian Abbott wrote:
> On 24/07/2025 22:04, Arnaud Lecomte wrote:
> > KMSAN reported a kernel-infoleak when copying instruction data back to
> > userspace in do_insnlist_ioctl(). The issue occurs because allocated
> > memory buffers weren't properly initialized (not
> > zero initialized)  before being copied to
> > userspace, potentially leaking kernel memory.
> >
> > Reported-by: syzbot+a5e45f768aab5892da5d@xxxxxxxxxxxxxxxxxxxxxxxxx
> > Closes: https://syzkaller.appspot.com/bug?extid=a5e45f768aab5892da5d
> > Tested-by: syzbot+a5e45f768aab5892da5d@xxxxxxxxxxxxxxxxxxxxxxxxx
> > Signed-off-by: Arnaud Lecomte <contact@xxxxxxxxxxxxxx>
> > ---
> >   drivers/comedi/comedi_fops.c | 2 +-
> >   1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/comedi/comedi_fops.c b/drivers/comedi/comedi_fops.c
> > index c83fd14dd7ad..15fee829d14c 100644
> > --- a/drivers/comedi/comedi_fops.c
> > +++ b/drivers/comedi/comedi_fops.c
> > @@ -1636,7 +1636,7 @@ static int do_insn_ioctl(struct comedi_device 
> > *dev,
> >           n_data = MAX_SAMPLES;
> >       }
> >   -    data = kmalloc_array(n_data, sizeof(unsigned int), GFP_KERNEL);
> > +    data = kcalloc(n_data, sizeof(unsigned int), GFP_KERNEL);
> >       if (!data) {
> >           ret = -ENOMEM;
> >           goto error;
>
> I thought my commit 46d8c744136c ("comedi: Fix initialization of data 
> for instructions that write to subdevice" would have fixed this as 
> long as all the driver code was playing nicely, but it seems I was 
> mistaken.
>
> I don't think it is necessary to use kcalloc().  The buffer already 
> gets initialized (partly by `copy_from_user()`) when `insn->insn & 
> INSN_MASK_WRITE` is non-zero, so it just needs to be initialized when 
> `insn->insn & INSN_MASK_WRITE` is zero too.
>
> There is nearly identical code in `do_insnlist_ioctl()` that needs 
> fixing, so it would be better to fix both at the same time.