Re: [PATCH] usb: typec: tps6598x: Remove VLA usage

From: Kees Cook
Date: Thu Jun 21 2018 - 16:31:33 EST


On Thu, Jun 21, 2018 at 3:00 AM, Heikki Krogerus
<heikki.krogerus@xxxxxxxxxxxxxxx> wrote:
> On Wed, Jun 20, 2018 at 11:28:40AM -0700, Kees Cook wrote:
>> In the quest to remove all stack VLA usage from the kernel[1], this
>> uses the maximum buffer size and adds a sanity check.
>>
>> [1] https://lkml.kernel.org/r/CA+55aFzCG-zNmZwX4A2FQpadafLfEzK6CC=qPXydAacU1RqZWA@xxxxxxxxxxxxxx
>>
>> Signed-off-by: Kees Cook <keescook@xxxxxxxxxxxx>
>> ---
>> drivers/usb/typec/tps6598x.c | 7 ++++++-
>> 1 file changed, 6 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/usb/typec/tps6598x.c b/drivers/usb/typec/tps6598x.c
>> index 4b4c8d271b27..396193f85e6d 100644
>> --- a/drivers/usb/typec/tps6598x.c
>> +++ b/drivers/usb/typec/tps6598x.c
>> @@ -81,12 +81,17 @@ struct tps6598x {
>> struct typec_capability typec_cap;
>> };
>>
>> +#define TPS_MAX_LEN sizeof(u64)
>
> That is not big enough. The registers of this chip can be as big as 64
> bytes. The identity register alone is 25 bytes, so the above would
> make the driver fail quite fast. Can you set the maximum to 64?
>
> #define TPS_MAX_LEN 64

Oops! Yes, thanks, I missed this usage:

struct tps6598x_rx_identity_reg id;
...
ret = tps6598x_block_read(tps, TPS_REG_RX_IDENTITY_SOP,
&id, sizeof(id));

struct tps6598x_rx_identity_reg {
u8 status; /* 0 1 */
struct usb_pd_identity identity; /* 1 12 */
u32 vdo[3]; /* 13 12 */

/* size: 25, cachelines: 1, members: 3 */
/* last cacheline: 25 bytes */
};

But that's still only 25 bytes. Where is the 64? I see these:

return tps6598x_block_read(tps, reg, val, sizeof(u16));
return tps6598x_block_read(tps, reg, val, sizeof(u32));
return tps6598x_block_read(tps, reg, val, sizeof(u64));
ret = tps6598x_block_read(tps, TPS_REG_RX_IDENTITY_SOP,
&id, sizeof(id));
ret = tps6598x_block_read(tps, TPS_REG_DATA1,
out_data, out_len);
ret = tps6598x_block_read(tps, TPS_REG_DATA1, &val, sizeof(u8));

1, 2, 4, 8, 25, and "out_len" in tps6598x_exec_cmd(), which is always zero:

ret = tps6598x_exec_cmd(tps, cmd, 0, NULL, 0, NULL);
ret = tps6598x_exec_cmd(tps, cmd, 0, NULL, 0, NULL);

I clearly need a v2 of this patch, but I just want to make sure I get
the right value. :)

Thanks for the review!

-Kees

--
Kees Cook
Pixel Security