Re: [PATCH master] of: fdt: fix overflows when parsing sizes

[Sascha Hauer <s.hauer@pengutronix.de>]

On Mon, Jul 22, 2024 at 07:24:10PM +0200, Ahmad Fatoum wrote:
> The function dt_struct_advance() is used to advance a pointer to the next
> offset within the structure block, while checking that the result is in
> bounds.
> 
> Unfortunately, the function used a signed size argument. This had the
> effect that a too-large size in the FDT wrapped around and caused the
> pointer to move backwards.
> 
> This issue was found by libfuzzer which generated an FDT that
> always triggered an out-of-memory condition: One struct indicated a size
> that caused the pointer to move backwards.
> 
> The resulting loop allocated memory on every iteration and eventually
> ran out.
> 
> Fix this by using unsigned sizes and treating wrap around as an
> error case.
> 
> Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
> ---
>  drivers/of/fdt.c | 11 ++++++-----
>  1 file changed, 6 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/of/fdt.c b/drivers/of/fdt.c
> index 8dca41990c87..237468cd8164 100644
> --- a/drivers/of/fdt.c
> +++ b/drivers/of/fdt.c
> @@ -32,12 +32,13 @@ static inline bool __dt_ptr_ok(const struct fdt_header *fdt, const void *p,
>  }
>  #define dt_ptr_ok(fdt, p) __dt_ptr_ok(fdt, p, sizeof(*(p)), __alignof__(*(p)))
>  
> -static inline uint32_t dt_struct_advance(struct fdt_header *f, uint32_t dt, int size)
> +static inline uint32_t dt_struct_advance(struct fdt_header *f, uint32_t dt, uint32_t size)
>  {
> -	dt += size;
> -	dt = ALIGN(dt, 4);
> +	if (check_add_overflow(dt, size, &dt))
> +		return 0;
>  
> -	if (dt > f->off_dt_struct + f->size_dt_struct)
> +	dt = ALIGN(dt, 4);
> +	if ((!dt && size) || dt > f->off_dt_struct + f->size_dt_struct)
>  		return 0;

I am not sure I fully understand the newly added (!dt && size).

I think it's for the case when the initial addition results in something
like 0xfffffffe and the ALIGN(dt, 4) makes dt become 0, right?

I think dt being zero is a an error anyway, so what is the && size good
for?

>  
>  	return dt;

When dt is zero it is returned here which will be considered an error by
the caller anyway, so it seems the (!dt && size) check doesn't add
anything.

Note we have dt_struct_advance() twice in the tree. Care to fix the
other place as well?

Sascha

-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |