diff mbox series

crypto: ecc - Protect ecc_digits_from_bytes from reading too many bytes

Message ID 20240426225553.3038070-1-stefanb@linux.ibm.com
State New
Headers show
Series crypto: ecc - Protect ecc_digits_from_bytes from reading too many bytes | expand

Commit Message

Stefan Berger April 26, 2024, 10:55 p.m. UTC
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 <stefanb@linux.ibm.com>
---
 include/crypto/internal/ecc.h | 7 +++++++
 1 file changed, 7 insertions(+)

Comments

Jarkko Sakkinen April 28, 2024, 10:12 p.m. UTC | #1
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 <stefanb@linux.ibm.com>
> ---
>  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;

BR, Jarkko
Lukas Wunner April 29, 2024, 3:30 a.m. UTC | #2
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 <stefanb@linux.ibm.com>
> > ---
> >  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.
Jarkko Sakkinen April 29, 2024, 10:14 a.m. UTC | #3
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 <stefanb@linux.ibm.com>
> > > ---
> > >  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
Stefan Berger April 29, 2024, 11:11 a.m. UTC | #4
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 <stefanb@linux.ibm.com>
>>>> ---
>>>>   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
Jarkko Sakkinen April 29, 2024, 1:12 p.m. UTC | #5
On Mon Apr 29, 2024 at 2:11 PM EEST, Stefan Berger wrote:
>
>
> 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 <stefanb@linux.ibm.com>
> >>>> ---
> >>>>   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?

Hmm... so as far as I'm concerned you pick what works for you. Just
was pointing out at it would make to simplify the original a bit :-)

BR, Jarkko
Herbert Xu May 3, 2024, 10:30 a.m. UTC | #6
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.

Cheers,
Jarkko Sakkinen May 3, 2024, 11:49 p.m. UTC | #7
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
Jarkko Sakkinen May 3, 2024, 11:51 p.m. UTC | #8
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
diff mbox series

Patch

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);