[U-Boot] Buffer overrun risk in UBI SPL for secure boot

[U-Boot] Buffer overrun risk in UBI SPL for secure boot

[Joel Peshkin]

It seems that, in the process of doing any sort of secure boot chain of
trust, anything loading a UBI volume in preparation to authenticate it,
will load a volume of unknown size into a buffer prior to checking the
signature of that volume.

Has anyone considered a solution for this?  Should all implementations just
carve out a buffer at the top of memory for ubispl_load_volume or should
the ubispl_load data structure be amended to include a size?  It would seem
appropriate to include a size, but not clear how to do that without
breaking compatibility with existing implementations.

[U-Boot] Buffer overrun risk in UBI SPL for secure boot

[Heiko Schocher]

Hello Joel,

Am 04.09.2019 um 06:57 schrieb Joel Peshkin:
> It seems that, in the process of doing any sort of secure boot chain of
> trust, anything loading a UBI volume in preparation to authenticate it,
> will load a volume of unknown size into a buffer prior to checking the
> signature of that volume.

Hmm.. it is not an unknown size, it loads the complete UBI Volume, so
it is the size of the UBI Volume ...
but yes, if an attacker changes the size of UBI Volumes ...

Could you please give us a reference to which piece of code you refer?

is it this part:

drivers/mtd/ubispl/ubispl.c
static int ipl_load(struct ubi_scan_info *ubi, const u32 vol_id, uint8_t *laddr)

yes ... no size...

> Has anyone considered a solution for this?  Should all implementations just
> carve out a buffer at the top of memory for ubispl_load_volume or should
> the ubispl_load data structure be amended to include a size?  It would seem
> appropriate to include a size, but not clear how to do that without
> breaking compatibility with existing implementations.

Hmm... yes may an option to add a maxsize to ubispl_load data struct
ubispl_load.

regarding backwards compatibility, there are not much boards yet, which
use this feature:

hs at threadripper:u-boot  [master] $ grep -lr SPL_UBI | grep defconfig
configs/am335x_igep003x_defconfig
configs/igep00x0_defconfig
hs at threadripper:u-boot  [master] $

So this should be solveable problem.

bye,
Heiko
-- 
DENX Software Engineering GmbH,      Managing Director: Wolfgang Denk
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: +49-8142-66989-52   Fax: +49-8142-66989-80   Email: hs at denx.de

[U-Boot] Buffer overrun risk in UBI SPL for secure boot

[Joel Peshkin]

Hi Heiko,

   The place where the issue comes up is in ubispl_load_volumes(), but that
calls ipl_load() internally.

   I guess there are several options....

1) Create a distinct ubispl_scan() function to do the scan without loading
anything and then a new load volume function that takes offset and limit
arguments.   This would have the advantage that a SPL that  needs to parse
one volume before loading the next would only need to scan once, then could
check sizes, then load the individual volumes as needed.  At the same time,
when we need to validate a FIT header before even deciding what we want to
load from it, we would be able to read more incrementally.

2) Add the size limit to struct ubispl_load and not worry about anyone with
non-upstreamed code forgetting to set it

3) Add the size limit to struct ubispl_load but have ubispl_load_volumes()
set the size limit to 0 (unlimited) before calling a new
ubispl_load_volumes_bounded() function.

   Which do you prefer?

   Do I presume correctly that your denx.de email address implies that an
approach you approve will be acceptable if correctly implemented?

Thanks,

Joel Peshkin


On Wed, Sep 4, 2019 at 6:01 AM Heiko Schocher <hs@denx.de> wrote:

> Hello Joel,
>
> Am 04.09.2019 um 06:57 schrieb Joel Peshkin:
> > It seems that, in the process of doing any sort of secure boot chain of
> > trust, anything loading a UBI volume in preparation to authenticate it,
> > will load a volume of unknown size into a buffer prior to checking the
> > signature of that volume.
>
> Hmm.. it is not an unknown size, it loads the complete UBI Volume, so
> it is the size of the UBI Volume ...
> but yes, if an attacker changes the size of UBI Volumes ...
>
> Could you please give us a reference to which piece of code you refer?
>
> is it this part:
>
> drivers/mtd/ubispl/ubispl.c
> static int ipl_load(struct ubi_scan_info *ubi, const u32 vol_id, uint8_t
> *laddr)
>
> yes ... no size...
>
> > Has anyone considered a solution for this?  Should all implementations
> just
> > carve out a buffer at the top of memory for ubispl_load_volume or should
> > the ubispl_load data structure be amended to include a size?  It would
> seem
> > appropriate to include a size, but not clear how to do that without
> > breaking compatibility with existing implementations.
>
> Hmm... yes may an option to add a maxsize to ubispl_load data struct
> ubispl_load.
>
> regarding backwards compatibility, there are not much boards yet, which
> use this feature:
>
> hs at threadripper:u-boot  [master] $ grep -lr SPL_UBI | grep defconfig
> configs/am335x_igep003x_defconfig
> configs/igep00x0_defconfig
> hs at threadripper:u-boot  [master] $
>
> So this should be solveable problem.
>
> bye,
> Heiko
> --
> DENX Software Engineering GmbH,      Managing Director: Wolfgang Denk
> HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
> Phone: +49-8142-66989-52   Fax: +49-8142-66989-80   Email: hs at denx.de
>

[U-Boot] Buffer overrun risk in UBI SPL for secure boot

[Joel Peshkin]

Hi Heiko,

    Adding a size limit without breaking things turns out to be much more
difficult that it would seem.  So, instead of capping the size, we have
changed the memory map we are using for uboot.  It is probably worthwhile
for others using UBISPL in a secure boot nevironment to do the same.

   Traditionally, uboot SPL or TPL loads or relocates to an address near
the top of memory and then builds its stack downwards from the top of
memory.   That means that any address we use for a volume.load_address will
eventually step on something if the volume is large enough.   So, we move
everything down by a size that is sufficient for any image that UBISPL may
need to load (32M) and place the CONFIG_SPL_LOAD_FIT_ADDRESS  Above the
stack where it can grow without hitting anything until it causes an
exception.

   I'm not sure if there is anything else to be done for this situation
except to caution people implementing secure boot environments to be aware
of their surroundings.

Regards,

Joel Peshkin