[v3,0/3] Add support for x509 certs with NIST p256 and p192 keys

Message

Stefan Berger Jan. 27, 2021, 12:33 p.m. UTC

From: Stefan Berger <[e-mail]>

This series of patches adds support for x509 certificates signed by a CA
that uses NIST p256 or p192 keys for signing. It also adds support for
certificates where the public key is a NIST p256 or p192 key. The math
for ECDSA signature verification is also added.

Since self-signed certificates are verified upon loading, the following
script can be used for testing:

k=$(keyctrl newring test @u)

while :; do
	for hash in sha1 sha224 sha256 sha384 sha512; do
		openssl req \
			-x509 \
			-${hash} \
			-newkey ec \
			-pkeyopt ec_paramgen_curve:prime256v1 \
			-keyout key.pem \
			-days 365 \
			-subj '/CN=test' \
			-nodes \
			-outform der \
			-out cert.der
		keyctl padd asymmetric testkey $k < cert.der
		if [ $? -ne 0 ]; then
			echo "ERROR"
			exit 1
		fi
	done
done

It also works with restricted keyrings where an RSA key is used to sign
a NIST P256/P192 key. Scripts for testing are here:

https://github.com/stefanberger/eckey-testing

The ECDSA signature verification will be used by IMA Appraisal where ECDSA
file signatures stored in RPM packages will use substantially less space
than if RSA signatures were to be used.

   Stefan

v2->v3:
  - patch 2 now includes linux/scatterlist.h

v1->v2:
  - using faster vli_sub rather than newly added vli_mod_fast to 'reduce'
    result
  - rearranged switch statements to follow after RSA
  - 3rd patch from 1st posting is now 1st patch

Stefan Berger (3):
  x509: Detect sm2 keys by their parameters OID
  x509: Add support for parsing x509 certs with NIST p256 keys
  x509: Add support for NIST p192 keys in certificates and akcipher

 crypto/Makefile                           |   9 +-
 crypto/asymmetric_keys/public_key.c       |  19 ++
 crypto/asymmetric_keys/x509_cert_parser.c |  45 ++-
 crypto/ecc.c                              | 318 ++++++++++++++++++++++
 crypto/ecc.h                              |   2 +
 crypto/ecc_curve_defs.h                   |   4 +
 crypto/eccsignature.asn1                  |   4 +
 include/linux/oid_registry.h              |   6 +
 8 files changed, 404 insertions(+), 3 deletions(-)
 create mode 100644 crypto/eccsignature.asn1

Comments

David Howells Jan. 27, 2021, 1:05 p.m. UTC | #1

Stefan Berger <[e-mail]> wrote:

> k=$(keyctrl newring test @u)

keyctl - but I can fix that.

David

David Howells Jan. 27, 2021, 2:22 p.m. UTC | #2

Stefan Berger <[e-mail]> wrote:

> This series of patches adds support for x509 certificates signed by a CA
> that uses NIST p256 or p192 keys for signing. It also adds support for
> certificates where the public key is a NIST p256 or p192 key. The math
> for ECDSA signature verification is also added.
> 
> Since self-signed certificates are verified upon loading, the following
> script can be used for testing:
> 
> k=$(keyctrl newring test @u)
> 
> while :; do
> 	for hash in sha1 sha224 sha256 sha384 sha512; do
> 		openssl req \
> 			-x509 \
> 			-${hash} \
> 			-newkey ec \
> 			-pkeyopt ec_paramgen_curve:prime256v1 \
> 			-keyout key.pem \
> 			-days 365 \
> 			-subj '/CN=test' \
> 			-nodes \
> 			-outform der \
> 			-out cert.der
> 		keyctl padd asymmetric testkey $k < cert.der
> 		if [ $? -ne 0 ]; then
> 			echo "ERROR"
> 			exit 1
> 		fi
> 	done
> done
> 
> It also works with restricted keyrings where an RSA key is used to sign
> a NIST P256/P192 key. Scripts for testing are here:
> 
> https://github.com/stefanberger/eckey-testing
> 
> The ECDSA signature verification will be used by IMA Appraisal where ECDSA
> file signatures stored in RPM packages will use substantially less space
> than if RSA signatures were to be used.

I've pulled this into my keys-next branch.

David

Nym Seddon Jan. 27, 2021, 4:12 p.m. UTC | #3

Hi Stefan,

In the recommendations from SafeCurves (https://safecurves.cr.yp.to/twist.html) there are a number of attacks against ECC twists. Two of those attacks are relevant against NIST P192: invalid-curve attacks and invalid-curve attacks against ladders.

Both attacks can be mitigated by checking the supplied public key is on the correct curve, before performing curve operations.

Not sure if the right place for those checks are in the signature verification code provided in these patches, or when reading public keys from the certificates. Does the kernel provide functions for checking curve points satisfy their respective curve equations?

There are also tables describing the cost of combined attacks on various curves, where NIST P224 already falls below the safe threshold. Because of that, I would recommend not implementing support for NIST P192 (since it would fair even worse).

What are your thoughts?

Best,
Nym

‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐
On Wednesday, January 27, 2021 12:33 PM, Stefan Berger <[e-mail]> wrote:

> From: Stefan Berger [e-mail]
>
> This series of patches adds support for x509 certificates signed by a CA
> that uses NIST p256 or p192 keys for signing. It also adds support for
> certificates where the public key is a NIST p256 or p192 key. The math
> for ECDSA signature verification is also added.
>
> Since self-signed certificates are verified upon loading, the following
> script can be used for testing:
>
> k=$(keyctrl newring test @u)
>
> while :; do
> for hash in sha1 sha224 sha256 sha384 sha512; do
> openssl req \
> -x509 \
> -${hash} \
> -newkey ec \
> -pkeyopt ec_paramgen_curve:prime256v1 \
> -keyout key.pem \
> -days 365 \
> -subj '/CN=test' \
> -nodes \
> -outform der \
> -out cert.der
> keyctl padd asymmetric testkey $k < cert.der
> if [ $? -ne 0 ]; then
> echo "ERROR"
> exit 1
> fi
> done
> done
>
> It also works with restricted keyrings where an RSA key is used to sign
> a NIST P256/P192 key. Scripts for testing are here:
>
> https://github.com/stefanberger/eckey-testing
>
> The ECDSA signature verification will be used by IMA Appraisal where ECDSA
> file signatures stored in RPM packages will use substantially less space
> than if RSA signatures were to be used.
>
> Stefan
>
> v2->v3:
>
> -   patch 2 now includes linux/scatterlist.h
>
>     v1->v2:
>
> -   using faster vli_sub rather than newly added vli_mod_fast to 'reduce'
>     result
>
> -   rearranged switch statements to follow after RSA
>
> -   3rd patch from 1st posting is now 1st patch
>
>     Stefan Berger (3):
>     x509: Detect sm2 keys by their parameters OID
>     x509: Add support for parsing x509 certs with NIST p256 keys
>     x509: Add support for NIST p192 keys in certificates and akcipher
>
>     crypto/Makefile | 9 +-
>     crypto/asymmetric_keys/public_key.c | 19 ++
>     crypto/asymmetric_keys/x509_cert_parser.c | 45 ++-
>     crypto/ecc.c | 318 ++++++++++++++++++++++
>     crypto/ecc.h | 2 +
>     crypto/ecc_curve_defs.h | 4 +
>     crypto/eccsignature.asn1 | 4 +
>     include/linux/oid_registry.h | 6 +
>     8 files changed, 404 insertions(+), 3 deletions(-)
>     create mode 100644 crypto/eccsignature.asn1
>
>     --
>     2.25.4
>

Herbert Xu Jan. 27, 2021, 7:32 p.m. UTC | #4

On Wed, Jan 27, 2021 at 02:22:08PM +0000, David Howells wrote:
>
> I've pulled this into my keys-next branch.

David, please drop them because there are issues with the Crypto API
bits.

Thanks,

David Howells Jan. 27, 2021, 9:08 p.m. UTC | #5

Herbert Xu <[e-mail]> wrote:

> > I've pulled this into my keys-next branch.
> 
> David, please drop them because there are issues with the Crypto API
> bits.

Okay, dropped.

David

Stefan Berger Jan. 27, 2021, 10:42 p.m. UTC | #6

On 1/27/21 11:12 AM, Nym Seddon wrote:
> Hi Stefan,
>
> In the recommendations from SafeCurves (https://safecurves.cr.yp.to/twist.html) there are a number of attacks against ECC twists. Two of those attacks are relevant against NIST P192: invalid-curve attacks and invalid-curve attacks against ladders.
>
> Both attacks can be mitigated by checking the supplied public key is on the correct curve, before performing curve operations.
>
> Not sure if the right place for those checks are in the signature verification code provided in these patches, or when reading public keys from the certificates. Does the kernel provide functions for checking curve points satisfy their respective curve equations?
>
> There are also tables describing the cost of combined attacks on various curves, where NIST P224 already falls below the safe threshold. Because of that, I would recommend not implementing support for NIST P192 (since it would fair even worse).
>
> What are your thoughts?


I am calling into a function performing such a test at the end of the 
function parsing the public key:

  return ecc_is_pubkey_valid_full(ctx->curve, ctx->pub_key)

https://elixir.bootlin.com/linux/latest/source/crypto/ecc.c#L1458

Is that good 'enough' ?

    Stefan