Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1690856 > unrolled thread

Re: [PATCH V6 5/7] crypto: AES CBC multi-buffer glue code

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [PATCH V6 5/7] crypto: AES CBC multi-buffer glue code Tim Chen <tim.c.chen@linux.intel.com> - 2017-07-19 03:00 +0200
    Re: [PATCH V6 5/7] crypto: AES CBC multi-buffer glue code Megha Dey <megha.dey@intel.com> - 2017-07-19 03:10 +0200
      Re: [PATCH V6 5/7] crypto: AES CBC multi-buffer glue code Herbert Xu <herbert@gondor.apana.org.au> - 2017-07-19 09:10 +0200

#1690856 — Re: [PATCH V6 5/7] crypto: AES CBC multi-buffer glue code

On 07/17/2017 10:41 PM, Herbert Xu wrote:
> On Tue, Jun 27, 2017 at 05:26:13PM -0700, Megha Dey wrote:
>>
>> +static void completion_callback(struct mcryptd_skcipher_request_ctx *rctx,
>> +			    struct mcryptd_alg_cstate *cstate,
>> +			    int err)
>> +{
>> +	struct skcipher_request *req = cast_mcryptd_ctx_to_req(rctx);
>> +
>> +       /* remove from work list and invoke completion callback */
>> +	spin_lock(&cstate->work_lock);
>> +	list_del(&rctx->waiter);
>> +	spin_unlock(&cstate->work_lock);
>> +
>> +	if (irqs_disabled())
>> +		rctx->complete(&req->base, err);
>> +	else {
>> +		local_bh_disable();
>> +		rctx->complete(&req->base, err);
>> +		local_bh_enable();
>> +	}
>> +}
> 
> The fact that you need to do this check means that this design is
> wrong.  You should always know what context you are in.
> 

I think you are right.  The irqs_disabled check is not necessary
as we only call this function in the context of the mcryptd thread.
When I wrote the original mb algorithms I was probably unsure
and put this check in as a precaution in other mb algorithms and
Megha did the same.

>> +/*
>> + * CRYPTO_ALG_ASYNC flag is passed to indicate we have an ablk
>> + * scatter-gather walk.
>> + */
>> +static struct skcipher_alg aes_cbc_mb_alg = {
>> +	.base = {
>> +		.cra_name		= "cbc(aes)",
>> +		.cra_driver_name	= "cbc-aes-aesni-mb",
>> +		.cra_priority		= 500,
>> +		.cra_flags		= CRYPTO_ALG_INTERNAL,
>> +		.cra_blocksize		= AES_BLOCK_SIZE,
>> +		.cra_ctxsize		= CRYPTO_AES_CTX_SIZE,
>> +		.cra_module		= THIS_MODULE,
>> +	},
>> +	.min_keysize	= AES_MIN_KEY_SIZE,
>> +	.max_keysize	= AES_MAX_KEY_SIZE,
>> +	.ivsize		= AES_BLOCK_SIZE,
>> +	.setkey		= aes_set_key,
>> +	.encrypt	= mb_aes_cbc_encrypt,
>> +	.decrypt	= mb_aes_cbc_decrypt
>> +};
> 
> So this claims to be a sync algorithm.  Is this really the case?
> 
> Cheers,
> 

[toc] | [next] | [standalone]

#1690861

On Tue, 2017-07-18 at 17:52 -0700, Tim Chen wrote:
> On 07/17/2017 10:41 PM, Herbert Xu wrote:
> > On Tue, Jun 27, 2017 at 05:26:13PM -0700, Megha Dey wrote:
> >>
> >> +static void completion_callback(struct mcryptd_skcipher_request_ctx *rctx,
> >> +			    struct mcryptd_alg_cstate *cstate,
> >> +			    int err)
> >> +{
> >> +	struct skcipher_request *req = cast_mcryptd_ctx_to_req(rctx);
> >> +
> >> +       /* remove from work list and invoke completion callback */
> >> +	spin_lock(&cstate->work_lock);
> >> +	list_del(&rctx->waiter);
> >> +	spin_unlock(&cstate->work_lock);
> >> +
> >> +	if (irqs_disabled())
> >> +		rctx->complete(&req->base, err);
> >> +	else {
> >> +		local_bh_disable();
> >> +		rctx->complete(&req->base, err);
> >> +		local_bh_enable();
> >> +	}
> >> +}
> > 
> > The fact that you need to do this check means that this design is
> > wrong.  You should always know what context you are in.
> > 
> 
> I think you are right.  The irqs_disabled check is not necessary
> as we only call this function in the context of the mcryptd thread.
> When I wrote the original mb algorithms I was probably unsure
> and put this check in as a precaution in other mb algorithms and
> Megha did the same.

I will make this change.
> 
> >> +/*
> >> + * CRYPTO_ALG_ASYNC flag is passed to indicate we have an ablk
> >> + * scatter-gather walk.
> >> + */
> >> +static struct skcipher_alg aes_cbc_mb_alg = {
> >> +	.base = {
> >> +		.cra_name		= "cbc(aes)",
> >> +		.cra_driver_name	= "cbc-aes-aesni-mb",
> >> +		.cra_priority		= 500,
> >> +		.cra_flags		= CRYPTO_ALG_INTERNAL,
> >> +		.cra_blocksize		= AES_BLOCK_SIZE,
> >> +		.cra_ctxsize		= CRYPTO_AES_CTX_SIZE,
> >> +		.cra_module		= THIS_MODULE,
> >> +	},
> >> +	.min_keysize	= AES_MIN_KEY_SIZE,
> >> +	.max_keysize	= AES_MAX_KEY_SIZE,
> >> +	.ivsize		= AES_BLOCK_SIZE,
> >> +	.setkey		= aes_set_key,
> >> +	.encrypt	= mb_aes_cbc_encrypt,
> >> +	.decrypt	= mb_aes_cbc_decrypt
> >> +};
> > 
> > So this claims to be a sync algorithm.  Is this really the case?

yes, the inner algorithm is sync whereas the outer algorithm is async.
> > 
> > Cheers,
> > 
> 

[toc] | [prev] | [next] | [standalone]

#1691045

On Tue, Jul 18, 2017 at 06:18:59PM -0700, Megha Dey wrote:
>
> > >> +/*
> > >> + * CRYPTO_ALG_ASYNC flag is passed to indicate we have an ablk
> > >> + * scatter-gather walk.
> > >> + */
> > >> +static struct skcipher_alg aes_cbc_mb_alg = {
> > >> +	.base = {
> > >> +		.cra_name		= "cbc(aes)",
> > >> +		.cra_driver_name	= "cbc-aes-aesni-mb",
> > >> +		.cra_priority		= 500,
> > >> +		.cra_flags		= CRYPTO_ALG_INTERNAL,
> > >> +		.cra_blocksize		= AES_BLOCK_SIZE,
> > >> +		.cra_ctxsize		= CRYPTO_AES_CTX_SIZE,
> > >> +		.cra_module		= THIS_MODULE,
> > >> +	},
> > >> +	.min_keysize	= AES_MIN_KEY_SIZE,
> > >> +	.max_keysize	= AES_MAX_KEY_SIZE,
> > >> +	.ivsize		= AES_BLOCK_SIZE,
> > >> +	.setkey		= aes_set_key,
> > >> +	.encrypt	= mb_aes_cbc_encrypt,
> > >> +	.decrypt	= mb_aes_cbc_decrypt
> > >> +};
> > > 
> > > So this claims to be a sync algorithm.  Is this really the case?
> 
> yes, the inner algorithm is sync whereas the outer algorithm is async.

Are you saying that once mb_aes_cbc_encrypt returns it will never
touch the input/output again? This doesn't seem to agree with the
actual code:

+       if (!ret_rctx) {
+               /* we submitted a job, but none completed */
+               /* just notify the caller */
+               notify_callback(rctx, cstate, -EINPROGRESS);
+               return 0;
+       }

Just because an algorithm is internal doesn't mean that it can
arbitrarily violate the crypto API requirements.

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

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web