2017-09-13 20:09:15

by Mathieu Malaterre

[permalink] [raw]
Subject: [RFC PATCH] crypto: make the seed() function optional

This makes it simplier for driver author to not provide the seed() function
in case of a pseudo RNG where the seed operation is a no-op.

Document that the seed() function pointer is optional in header.

Signed-off-by: Mathieu Malaterre <[email protected]>
---
The PRNG as found on Ingenic JZ4780 is one such example. This is found on a
MIPS Creator CI20 SoC.

crypto/rng.c | 7 ++++++-
include/crypto/rng.h | 2 +-
2 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/crypto/rng.c b/crypto/rng.c
index 5e8469244960..ed08581901a9 100644
--- a/crypto/rng.c
+++ b/crypto/rng.c
@@ -35,9 +35,14 @@ static int crypto_default_rng_refcnt;

int crypto_rng_reset(struct crypto_rng *tfm, const u8 *seed, unsigned int slen)
{
+ struct rng_alg *ralg = crypto_rng_alg(tfm);
u8 *buf = NULL;
int err;

+ /* In case of PRNG, no need to seed */
+ if (!ralg->seed)
+ return 0;
+
if (!seed && slen) {
buf = kmalloc(slen, GFP_KERNEL);
if (!buf)
@@ -47,7 +52,7 @@ int crypto_rng_reset(struct crypto_rng *tfm, const u8 *seed, unsigned int slen)
seed = buf;
}

- err = crypto_rng_alg(tfm)->seed(tfm, seed, slen);
+ err = ralg->seed(tfm, seed, slen);

kzfree(buf);
return err;
diff --git a/include/crypto/rng.h b/include/crypto/rng.h
index b95ede354a66..ac5d061d0297 100644
--- a/include/crypto/rng.h
+++ b/include/crypto/rng.h
@@ -32,7 +32,7 @@ struct crypto_rng;
* random number generator requires a seed for setting
* up a new state, the seed must be provided by the
* consumer while invoking this function. The required
- * size of the seed is defined with @seedsize .
+ * size of the seed is defined with @seedsize. Optional.
* @set_ent: Set entropy that would otherwise be obtained from
* entropy source. Internal use only.
* @seedsize: The seed size required for a random number generator
--
2.11.0


2017-10-07 03:33:50

by Herbert Xu

[permalink] [raw]
Subject: Re: [RFC PATCH] crypto: make the seed() function optional

Mathieu Malaterre <[email protected]> wrote:
> This makes it simplier for driver author to not provide the seed() function
> in case of a pseudo RNG where the seed operation is a no-op.
>
> Document that the seed() function pointer is optional in header.
>
> Signed-off-by: Mathieu Malaterre <[email protected]>
> ---
> The PRNG as found on Ingenic JZ4780 is one such example. This is found on a
> MIPS Creator CI20 SoC.

So how does it seed itself? This also contradicts with the JZ4780
driver that's currently in the patch queue as it does contain a
seed function.

Cheers,
--
Email: Herbert Xu <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Subject: Re: [RFC PATCH] crypto: make the seed() function optional

Hi Herbert,

On 7 October 2017 at 09:03, Herbert Xu <[email protected]> wrote:
> Mathieu Malaterre <[email protected]> wrote:
>> This makes it simplier for driver author to not provide the seed() function
>> in case of a pseudo RNG where the seed operation is a no-op.
>>
>> Document that the seed() function pointer is optional in header.
>>
>> Signed-off-by: Mathieu Malaterre <[email protected]>
>> ---
>> The PRNG as found on Ingenic JZ4780 is one such example. This is found on a
>> MIPS Creator CI20 SoC.
>
> So how does it seed itself? This also contradicts with the JZ4780
> driver that's currently in the patch queue as it does contain a
> seed function.

The current version of JZ4780 driver in the patch queue indeed has
seed function. But when Mathieu sent this email based on v2 of the
driver. V2 did not have seed callback. Using v2 resulted in a NULL
pointer in kernel. This patch prevents that NULL pointer access.

Regardless of what JZ4780 driver has this patch makes sense.

Currently crypto framework does not mandate seed callback's presence.
If mandatory, crypto framework should error out if seed is not
implemented while registering the PRNG.

Thanks,
PrasannaKumar