Protect ecc_digits_from_bytes from reading too many bytes from the input
byte array in case an insufficient number of bytes is provided to fill the
output digit array of ndigits. Therefore, initialize the most significant
digits with 0 to avoid trying to read too many bytes later on.
If too many bytes are provided on the input byte array the extra bytes
are ignored since the input variable 'ndigits' limits the number of digits
that will be filled.
Fixes: d67c96fb97b5 ("crypto: ecdsa - Convert byte arrays with key coordinates to digits")
Signed-off-by: Stefan Berger <[email protected]>
---
include/crypto/internal/ecc.h | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/include/crypto/internal/ecc.h b/include/crypto/internal/ecc.h
index 7ca1f463d1ec..56215f14ff96 100644
--- a/include/crypto/internal/ecc.h
+++ b/include/crypto/internal/ecc.h
@@ -67,9 +67,16 @@ static inline void ecc_swap_digits(const void *in, u64 *out, unsigned int ndigit
static inline void ecc_digits_from_bytes(const u8 *in, unsigned int nbytes,
u64 *out, unsigned int ndigits)
{
+ int diff = ndigits - DIV_ROUND_UP(nbytes, sizeof(u64));
unsigned int o = nbytes & 7;
__be64 msd = 0;
+ /* diff > 0: not enough input bytes: set most significant digits to 0 */
+ while (diff > 0) {
+ out[--ndigits] = 0;
+ diff--;
+ }
+
if (o) {
memcpy((u8 *)&msd + sizeof(msd) - o, in, o);
out[--ndigits] = be64_to_cpu(msd);
--
2.43.0
On Mon, Apr 29, 2024 at 01:12:00AM +0300, Jarkko Sakkinen wrote:
> On Sat Apr 27, 2024 at 1:55 AM EEST, Stefan Berger wrote:
> > Protect ecc_digits_from_bytes from reading too many bytes from the input
> > byte array in case an insufficient number of bytes is provided to fill the
> > output digit array of ndigits. Therefore, initialize the most significant
> > digits with 0 to avoid trying to read too many bytes later on.
> >
> > If too many bytes are provided on the input byte array the extra bytes
> > are ignored since the input variable 'ndigits' limits the number of digits
> > that will be filled.
> >
> > Fixes: d67c96fb97b5 ("crypto: ecdsa - Convert byte arrays with key coordinates to digits")
> > Signed-off-by: Stefan Berger <[email protected]>
> > ---
> > include/crypto/internal/ecc.h | 7 +++++++
> > 1 file changed, 7 insertions(+)
> >
> > diff --git a/include/crypto/internal/ecc.h b/include/crypto/internal/ecc.h
> > index 7ca1f463d1ec..56215f14ff96 100644
> > --- a/include/crypto/internal/ecc.h
> > +++ b/include/crypto/internal/ecc.h
> > @@ -67,9 +67,16 @@ static inline void ecc_swap_digits(const void *in, u64 *out, unsigned int ndigit
> > static inline void ecc_digits_from_bytes(const u8 *in, unsigned int nbytes,
> > u64 *out, unsigned int ndigits)
> > {
> > + int diff = ndigits - DIV_ROUND_UP(nbytes, sizeof(u64));
> > unsigned int o = nbytes & 7;
> > __be64 msd = 0;
> >
> > + /* diff > 0: not enough input bytes: set most significant digits to 0 */
> > + while (diff > 0) {
> > + out[--ndigits] = 0;
> > + diff--;
> > + }
>
> Could be just trivial for-loop:
>
> for (i = 0; i < diff; i++)
> out[--ndigits] = 0;
>
> Or also simpler while-loop could work:
>
> while (diff-- > 0)
> out[--ndigits] = 0;
Or just use memset(), which uses optimized instructions on many arches.
On Mon Apr 29, 2024 at 6:30 AM EEST, Lukas Wunner wrote:
> On Mon, Apr 29, 2024 at 01:12:00AM +0300, Jarkko Sakkinen wrote:
> > On Sat Apr 27, 2024 at 1:55 AM EEST, Stefan Berger wrote:
> > > Protect ecc_digits_from_bytes from reading too many bytes from the input
> > > byte array in case an insufficient number of bytes is provided to fill the
> > > output digit array of ndigits. Therefore, initialize the most significant
> > > digits with 0 to avoid trying to read too many bytes later on.
> > >
> > > If too many bytes are provided on the input byte array the extra bytes
> > > are ignored since the input variable 'ndigits' limits the number of digits
> > > that will be filled.
> > >
> > > Fixes: d67c96fb97b5 ("crypto: ecdsa - Convert byte arrays with key coordinates to digits")
> > > Signed-off-by: Stefan Berger <[email protected]>
> > > ---
> > > include/crypto/internal/ecc.h | 7 +++++++
> > > 1 file changed, 7 insertions(+)
> > >
> > > diff --git a/include/crypto/internal/ecc.h b/include/crypto/internal/ecc.h
> > > index 7ca1f463d1ec..56215f14ff96 100644
> > > --- a/include/crypto/internal/ecc.h
> > > +++ b/include/crypto/internal/ecc.h
> > > @@ -67,9 +67,16 @@ static inline void ecc_swap_digits(const void *in, u64 *out, unsigned int ndigit
> > > static inline void ecc_digits_from_bytes(const u8 *in, unsigned int nbytes,
> > > u64 *out, unsigned int ndigits)
> > > {
> > > + int diff = ndigits - DIV_ROUND_UP(nbytes, sizeof(u64));
> > > unsigned int o = nbytes & 7;
> > > __be64 msd = 0;
> > >
> > > + /* diff > 0: not enough input bytes: set most significant digits to 0 */
> > > + while (diff > 0) {
> > > + out[--ndigits] = 0;
> > > + diff--;
> > > + }
> >
> > Could be just trivial for-loop:
> >
> > for (i = 0; i < diff; i++)
> > out[--ndigits] = 0;
> >
> > Or also simpler while-loop could work:
> >
> > while (diff-- > 0)
> > out[--ndigits] = 0;
>
> Or just use memset(), which uses optimized instructions on many arches.
Yeah, sure, that would be even better, or even memzero_explicit()?
BR, Jarkko
On 4/29/24 06:14, Jarkko Sakkinen wrote:
> On Mon Apr 29, 2024 at 6:30 AM EEST, Lukas Wunner wrote:
>> On Mon, Apr 29, 2024 at 01:12:00AM +0300, Jarkko Sakkinen wrote:
>>> On Sat Apr 27, 2024 at 1:55 AM EEST, Stefan Berger wrote:
>>>> Protect ecc_digits_from_bytes from reading too many bytes from the input
>>>> byte array in case an insufficient number of bytes is provided to fill the
>>>> output digit array of ndigits. Therefore, initialize the most significant
>>>> digits with 0 to avoid trying to read too many bytes later on.
>>>>
>>>> If too many bytes are provided on the input byte array the extra bytes
>>>> are ignored since the input variable 'ndigits' limits the number of digits
>>>> that will be filled.
>>>>
>>>> Fixes: d67c96fb97b5 ("crypto: ecdsa - Convert byte arrays with key coordinates to digits")
>>>> Signed-off-by: Stefan Berger <[email protected]>
>>>> ---
>>>> include/crypto/internal/ecc.h | 7 +++++++
>>>> 1 file changed, 7 insertions(+)
>>>>
>>>> diff --git a/include/crypto/internal/ecc.h b/include/crypto/internal/ecc.h
>>>> index 7ca1f463d1ec..56215f14ff96 100644
>>>> --- a/include/crypto/internal/ecc.h
>>>> +++ b/include/crypto/internal/ecc.h
>>>> @@ -67,9 +67,16 @@ static inline void ecc_swap_digits(const void *in, u64 *out, unsigned int ndigit
>>>> static inline void ecc_digits_from_bytes(const u8 *in, unsigned int nbytes,
>>>> u64 *out, unsigned int ndigits)
>>>> {
>>>> + int diff = ndigits - DIV_ROUND_UP(nbytes, sizeof(u64));
>>>> unsigned int o = nbytes & 7;
>>>> __be64 msd = 0;
>>>>
>>>> + /* diff > 0: not enough input bytes: set most significant digits to 0 */
>>>> + while (diff > 0) {
>>>> + out[--ndigits] = 0;
>>>> + diff--;
>>>> + }
>>>
>>> Could be just trivial for-loop:
>>>
>>> for (i = 0; i < diff; i++)
>>> out[--ndigits] = 0;
>>>
>>> Or also simpler while-loop could work:
>>>
>>> while (diff-- > 0)
>>> out[--ndigits] = 0;
>>
>> Or just use memset(), which uses optimized instructions on many arches.
>
> Yeah, sure, that would be even better, or even memzero_explicit()?
Thanks. The function isn't getting too big for an inline?
>
> BR, Jarkko
On Fri May 3, 2024 at 1:30 PM EEST, Herbert Xu wrote:
> On Mon, Apr 29, 2024 at 01:14:15PM +0300, Jarkko Sakkinen wrote:
> >
> > Yeah, sure, that would be even better, or even memzero_explicit()?
>
> memzero_explicit should only be used for stack memory.
BTW, is this in kernel documentation? It's a guideline really
and would be nice to have reminder, that's all.
BR, Jarkko
On Sat May 4, 2024 at 2:49 AM EEST, Jarkko Sakkinen wrote:
> On Fri May 3, 2024 at 1:30 PM EEST, Herbert Xu wrote:
> > On Mon, Apr 29, 2024 at 01:14:15PM +0300, Jarkko Sakkinen wrote:
> > >
> > > Yeah, sure, that would be even better, or even memzero_explicit()?
> >
> > memzero_explicit should only be used for stack memory.
>
> BTW, is this in kernel documentation? It's a guideline really
> and would be nice to have reminder, that's all.
Any case makes sense and is pretty obvious now that you said it.
BR, Jarkko