Re: [PATCH 1/2] misc: c2port: core: Make copying name from userspace more secure

[Lee Jones]

On Mon, 02 Nov 2020, Rodolfo Giometti wrote:

> On 02/11/2020 14:47, Lee Jones wrote:
> > On Mon, 02 Nov 2020, gregkh@xxxxxxxxxxxxxxxxxxx wrote:
> > 
> >> On Mon, Nov 02, 2020 at 12:43:01PM +0000, Lee Jones wrote:
> >>> On Mon, 02 Nov 2020, gregkh@xxxxxxxxxxxxxxxxxxx wrote:
> >>>
> >>>> On Mon, Nov 02, 2020 at 11:49:03AM +0000, Lee Jones wrote:
> >>>>> On Mon, 02 Nov 2020, David Laight wrote:
> >>>>>
> >>>>>> From: Lee Jones
> >>>>>>> Sent: 02 November 2020 11:12
> >>>>>>>
> >>>>>>> strncpy() may not provide a NUL terminator, which means that a 1-byte
> >>>>>>> leak would be possible *if* this was ever copied to userspace.  Ensure
> >>>>>>> the buffer will always be NUL terminated by using the kernel's
> >>>>>>> strscpy() which a) uses the destination (instead of the source) size
> >>>>>>> as the bytes to copy and b) is *always* NUL terminated.
> >>>>>>>
> >>>>>>> Cc: Rodolfo Giometti <giometti@xxxxxxxxxxxx>
> >>>>>>> Cc: "Eurotech S.p.A" <info@xxxxxxxxxxx>
> >>>>>>> Reported-by: Geert Uytterhoeven <geert+renesas@xxxxxxxxx>
> >>>>>>> Acked-by: Arnd Bergmann <arnd@xxxxxxxx>
> >>>>>>> Signed-off-by: Lee Jones <lee.jones@xxxxxxxxxx>
> >>>>>>> ---
> >>>>>>>  drivers/misc/c2port/core.c | 2 +-
> >>>>>>>  1 file changed, 1 insertion(+), 1 deletion(-)
> >>>>>>>
> >>>>>>> diff --git a/drivers/misc/c2port/core.c b/drivers/misc/c2port/core.c
> >>>>>>> index 80d87e8a0bea9..b96444ec94c7e 100644
> >>>>>>> --- a/drivers/misc/c2port/core.c
> >>>>>>> +++ b/drivers/misc/c2port/core.c
> >>>>>>> @@ -923,7 +923,7 @@ struct c2port_device *c2port_device_register(char *name,
> >>>>>>>  	}
> >>>>>>>  	dev_set_drvdata(c2dev->dev, c2dev);
> >>>>>>>
> >>>>>>> -	strncpy(c2dev->name, name, C2PORT_NAME_LEN - 1);
> >>>>>>> +	strscpy(c2dev->name, name, sizeof(c2dev->name));
> >>>>>>
> >>>>>> strscpy() doesn't zero fill so if the memory isn't zeroed
> >>>>>> and a 'blind' copy to user of the structure is done
> >>>>>> then more data is leaked.
> >>>>>>
> >>>>>> strscpy() may be better, but rational isn't right.
> >>>>>
> >>>>> The original patch zeroed the data too, but I was asked to remove that
> >>>>> part [0].  In your opinion, should it be reinstated?
> >>>>>
> >>>>> [0] https://lore.kernel.org/patchwork/patch/1272290/
> >>>>
> >>>> Just keep the kzalloc() part of the patch, this portion makes no sense
> >>>> to me.
> >>>
> >>> Can do.
> >>>
> >>>> But if you REALLY want to get it correct, call dev_set_name()
> >>>> instead please, as that is what it is there for.
> >>>
> >>> The line above isn't setting the 'struct device' name.  It looks as
> >>> though 'struct c2port' has it's own member, also called 'name'.  As to
> >>> how they differ, I'm not currently aware.  Nor do I wish to mess
> >>> around with the semantics all that much.
> >>>
> >>> Going with suggestion #1.
> >>
> >> As the "device" already has a name, I suggest just getting rid of this
> >> name field anyway, no need for duplicates.
> > 
> > That definitely goes against the point I made above:
> > 
> >  "Nor do I wish to mess around with the semantics all that much."
> > 
> > It looks as though the device name 'c2port%d' varies greatly to the
> > requested name 'uc'.  I don't have enough knowledge of how user-
> > space expects to use the provided sysfs entries to be able to
> > competently merge/decide which of these should be kept and which to
> > discard.
> > 
> > Hopefully one of the authors/maintainers are reading this and can come
> > up with an acceptable solution.
> 
> User-space usage can change its behavior so, please, consider the best solution
> from the kernel space point-of-view. :)

If you're sure, I can add it to my TODO.

-- 
Lee Jones [李琼斯]
Senior Technical Lead - Developer Services
Linaro.org │ Open source software for Arm SoCs
Follow Linaro: Facebook | Twitter | Blog