[PATCH v5 0/2] Add support for ECDSA-signed kernel modules

This series adds support for ECDSA-signed kernel modules. It also
attempts to address a kbuild issue where a developer created an ECDSA
key for signing kernel modules and then builds an older version of the
kernel, when bisecting the kernel for example, that does not support
ECDSA keys.

The first patch addresses the kbuild issue of needing to delete that
ECDSA key if it is in certs/signing_key.pem and trigger the creation
of an RSA key. However, for this to work this patch would have to be
backported to previous versions of the kernel but would also only work
for the developer if he/she used a stable version of the kernel to which
this patch was applied. So whether this patch actually achieves the
wanted effect is not always guaranteed.

The 2nd patch adds the support for the ECSDA-signed kernel modules.

This patch depends on the ECDSA support series currently queued here:
https://git.kernel.org/pub/scm/linux/kernel/git/herbert/cryptodev-2.6.git/log/?h=ecc

  Stefan

v5:
  - do not touch the key files if openssl is not installed; likely
    addresses an issue pointed out by kernel test robot

v4:
  - extending 'depends on' with MODULES to (IMA_APPRAISE_MODSIG && MODULES)
  
v3:
  - added missing OIDs for ECDSA signed hashes to pkcs7_sig_note_pkey_algo
  - added recommendation to use string hash to Kconfig help text

v2:
  - Adjustment to ECDSA key detector string in 2/2
  - Rephrased cover letter and patch descriptions with Mimi


Stefan Berger (2):
  certs: Trigger creation of RSA module signing key if it's not an RSA
    key
  certs: Add support for using elliptic curve keys for signing modules

 certs/Kconfig                         | 26 ++++++++++++++++++++++++++
 certs/Makefile                        | 21 +++++++++++++++++++++
 crypto/asymmetric_keys/pkcs7_parser.c |  8 ++++++++
 3 files changed, 55 insertions(+)

-- 
2.29.2

Add support for using elliptic curve keys for signing modules. It uses
a NIST P384 (secp384r1) key if the user chooses an elliptic curve key
and will have ECDSA support built into the kernel.

Note: A developer choosing an ECDSA key for signing modules should still
delete the signing key (rm certs/signing_key.*) when building an older
version of a kernel that only supports RSA keys. Unless kbuild automati-
cally detects and generates a new kernel module key, ECDSA-signed kernel
modules will fail signature verification.

Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Cc: David Howells <dhowells@redhat.com>
Cc: David Woodhouse <dwmw2@infradead.org>
---
 certs/Kconfig                         | 26 ++++++++++++++++++++++++++
 certs/Makefile                        | 13 +++++++++++++
 crypto/asymmetric_keys/pkcs7_parser.c |  8 ++++++++
 3 files changed, 47 insertions(+)

diff --git a/certs/Kconfig b/certs/Kconfig
index f4e61116f94e..916cbb1af928 100644
--- a/certs/Kconfig
+++ b/certs/Kconfig

@@ -15,6 +15,32 @@ config MODULE_SIG_KEY
          then the kernel will automatically generate the private key and
          certificate as described in Documentation/admin-guide/module-signing.rst
 
+choice
+	prompt "Type of module signing key to be generated"
+	default MODULE_SIG_KEY_TYPE_RSA
+	help
+	 The type of module signing key type to generate. This option
+	 does not apply if a #PKCS11 URI is used.
+
+config MODULE_SIG_KEY_TYPE_RSA
+	bool "RSA"
+	depends on MODULE_SIG || (IMA_APPRAISE_MODSIG && MODULES)
+	help
+	 Use an RSA key for module signing.
+
+config MODULE_SIG_KEY_TYPE_ECDSA
+	bool "ECDSA"
+	select CRYPTO_ECDSA
+	depends on MODULE_SIG || (IMA_APPRAISE_MODSIG && MODULES)
+	help
+	 Use an elliptic curve key (NIST P384) for module signing. Consider
+	 using a strong hash like sha256 or sha384 for hashing modules.
+
+	 Note: Remove all ECDSA signing keys, e.g. certs/signing_key.pem,
+	 when falling back to building Linux 5.11 and older kernels.
+
+endchoice
+
 config SYSTEM_TRUSTED_KEYRING
 	bool "Provide system-wide ring of trusted keys"
 	depends on KEYS

diff --git a/certs/Makefile b/certs/Makefile
index 72758684d254..4eb69bdadc79 100644
--- a/certs/Makefile
+++ b/certs/Makefile

@@ -66,9 +66,21 @@ ifeq ($(CONFIG_MODULE_SIG_KEY),"certs/signing_key.pem")
 
 ifeq ($(openssl_available),yes)
 X509TEXT=$(shell openssl x509 -in $(CONFIG_MODULE_SIG_KEY) -text)
+endif
 
+# Support user changing key type
+ifdef CONFIG_MODULE_SIG_KEY_TYPE_ECDSA
+keytype_openssl = -newkey ec -pkeyopt ec_paramgen_curve:secp384r1
+ifeq ($(openssl_available),yes)
+$(if $(findstring id-ecPublicKey,$(X509TEXT)),,$(shell rm -f $(CONFIG_MODULE_SIG_KEY)))
+endif
+endif # CONFIG_MODULE_SIG_KEY_TYPE_ECDSA
+
+ifdef CONFIG_MODULE_SIG_KEY_TYPE_RSA
+ifeq ($(openssl_available),yes)
 $(if $(findstring rsaEncryption,$(X509TEXT)),,$(shell rm -f $(CONFIG_MODULE_SIG_KEY)))
 endif
+endif # CONFIG_MODULE_SIG_KEY_TYPE_RSA
 
 $(obj)/signing_key.pem: $(obj)/x509.genkey
 	@$(kecho) "###"

@@ -83,6 +95,7 @@ $(obj)/signing_key.pem: $(obj)/x509.genkey
 		-batch -x509 -config $(obj)/x509.genkey \
 		-outform PEM -out $(obj)/signing_key.pem \
 		-keyout $(obj)/signing_key.pem \
+		$(keytype_openssl) \
 		$($(quiet)redirect_openssl)
 	@$(kecho) "###"
 	@$(kecho) "### Key pair generated."

diff --git a/crypto/asymmetric_keys/pkcs7_parser.c b/crypto/asymmetric_keys/pkcs7_parser.c
index 967329e0a07b..6592279d839a 100644
--- a/crypto/asymmetric_keys/pkcs7_parser.c
+++ b/crypto/asymmetric_keys/pkcs7_parser.c

@@ -269,6 +269,14 @@ int pkcs7_sig_note_pkey_algo(void *context, size_t hdrlen,
 		ctx->sinfo->sig->pkey_algo = "rsa";
 		ctx->sinfo->sig->encoding = "pkcs1";
 		break;
+	case OID_id_ecdsa_with_sha1:
+	case OID_id_ecdsa_with_sha224:
+	case OID_id_ecdsa_with_sha256:
+	case OID_id_ecdsa_with_sha384:
+	case OID_id_ecdsa_with_sha512:
+		ctx->sinfo->sig->pkey_algo = "ecdsa";
+		ctx->sinfo->sig->encoding = "x962";
+		break;
 	default:
 		printk("Unsupported pkey algo: %u\n", ctx->last_oid);
 		return -ENOPKG;

-- 
2.29.2

Address a kbuild issue where a developer created an ECDSA key for signing
kernel modules and then builds an older version of the kernel, when bi-
secting the kernel for example, that does not support ECDSA keys.

If openssl is installed, trigger the creation of an RSA module signing
key if it is not an RSA key.

Fixes: cfc411e7fff3 ("Move certificate handling to its own directory")
Signed-off-by: Stefan Berger <stefanb@linux.ibm.com>
Cc: David Howells <dhowells@redhat.com>
Cc: David Woodhouse <dwmw2@infradead.org>
---
 certs/Makefile | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/certs/Makefile b/certs/Makefile
index 359239a0ee9e..72758684d254 100644
--- a/certs/Makefile
+++ b/certs/Makefile

@@ -57,11 +57,19 @@ endif
 redirect_openssl	= 2>&1
 quiet_redirect_openssl	= 2>&1
 silent_redirect_openssl = 2>/dev/null
+openssl_available       = $(shell openssl help 2>/dev/null && echo yes)
 
 # We do it this way rather than having a boolean option for enabling an
 # external private key, because 'make randconfig' might enable such a
 # boolean option and we unfortunately can't make it depend on !RANDCONFIG.
 ifeq ($(CONFIG_MODULE_SIG_KEY),"certs/signing_key.pem")
+
+ifeq ($(openssl_available),yes)
+X509TEXT=$(shell openssl x509 -in $(CONFIG_MODULE_SIG_KEY) -text)
+
+$(if $(findstring rsaEncryption,$(X509TEXT)),,$(shell rm -f $(CONFIG_MODULE_SIG_KEY)))
+endif
+
 $(obj)/signing_key.pem: $(obj)/x509.genkey
 	@$(kecho) "###"
 	@$(kecho) "### Now generating an X.509 key pair to be used for signing modules."

-- 
2.29.2

On Wed, Jun 02, 2021 at 10:35:35AM -0400, Stefan Berger wrote:

This series adds support for ECDSA-signed kernel modules. It also
attempts to address a kbuild issue where a developer created an ECDSA
key for signing kernel modules and then builds an older version of the
kernel, when bisecting the kernel for example, that does not support
ECDSA keys.

The first patch addresses the kbuild issue of needing to delete that
ECDSA key if it is in certs/signing_key.pem and trigger the creation
of an RSA key. However, for this to work this patch would have to be
backported to previous versions of the kernel but would also only work
for the developer if he/she used a stable version of the kernel to which
this patch was applied. So whether this patch actually achieves the
wanted effect is not always guaranteed.

The 2nd patch adds the support for the ECSDA-signed kernel modules.

This patch depends on the ECDSA support series currently queued here:
https://git.kernel.org/pub/scm/linux/kernel/git/herbert/cryptodev-2.6.git/log/?h=ecc

  Stefan

v5:
  - do not touch the key files if openssl is not installed; likely
    addresses an issue pointed out by kernel test robot

v4:
  - extending 'depends on' with MODULES to (IMA_APPRAISE_MODSIG && MODULES)
  
v3:
  - added missing OIDs for ECDSA signed hashes to pkcs7_sig_note_pkey_algo
  - added recommendation to use string hash to Kconfig help text

v2:
  - Adjustment to ECDSA key detector string in 2/2
  - Rephrased cover letter and patch descriptions with Mimi


Stefan Berger (2):
  certs: Trigger creation of RSA module signing key if it's not an RSA
    key
  certs: Add support for using elliptic curve keys for signing modules

 certs/Kconfig                         | 26 ++++++++++++++++++++++++++
 certs/Makefile                        | 21 +++++++++++++++++++++
 crypto/asymmetric_keys/pkcs7_parser.c |  8 ++++++++
 3 files changed, 55 insertions(+)

-- 
2.29.2

Please instead send a fix.

/Jarkko

On 6/3/21 2:47 AM, Jarkko Sakkinen wrote:

quoted

-- 
2.29.2

Please instead send a fix.

We have a Fixes tag in 1/2, so we want this to propagate to older 
kernels and need the fix in 1/2 for that reason.

    Stefan

On Thu, Jun 03, 2021 at 08:32:59AM -0400, Stefan Berger wrote:

On 6/3/21 2:47 AM, Jarkko Sakkinen wrote:

quoted

quoted

-- 
2.29.2

Please instead send a fix.

We have a Fixes tag in 1/2, so we want this to propagate to older kernels
and need the fix in 1/2 for that reason.

   Stefan

So please do an additional fix and send it.

/Jarkko

On 6/9/21 8:44 AM, Jarkko Sakkinen wrote:

On Thu, Jun 03, 2021 at 08:32:59AM -0400, Stefan Berger wrote:

quoted

On 6/3/21 2:47 AM, Jarkko Sakkinen wrote:

quoted

quoted

-- 
2.29.2

Please instead send a fix.

We have a Fixes tag in 1/2, so we want this to propagate to older kernels
and need the fix in 1/2 for that reason.

    Stefan

So please do an additional fix and send it.

1/2 is supposed to propagate to older kernels and needs to change as 
posted here in v5 (assuming that this does indeed fix what the build bot 
was complaining about). 2/2 also changes. A fix on top of v4 would fix 
2/2 but won't apply cleanly to 1/2 as cannot easily propagate to older 
kernels. Is that what we want? Why can you not remove v4 from your queue 
and replace it with v5?

    Stefan

On Wed, Jun 09, 2021 at 09:58:29AM -0400, Stefan Berger wrote:

On 6/9/21 8:44 AM, Jarkko Sakkinen wrote:

quoted

On Thu, Jun 03, 2021 at 08:32:59AM -0400, Stefan Berger wrote:

quoted

On 6/3/21 2:47 AM, Jarkko Sakkinen wrote:

quoted

quoted

-- 
2.29.2

Please instead send a fix.

We have a Fixes tag in 1/2, so we want this to propagate to older kernels
and need the fix in 1/2 for that reason.

    Stefan

So please do an additional fix and send it.

1/2 is supposed to propagate to older kernels and needs to change as posted
here in v5 (assuming that this does indeed fix what the build bot was
complaining about). 2/2 also changes. A fix on top of v4 would fix 2/2 but
won't apply cleanly to 1/2 as cannot easily propagate to older kernels. Is
that what we want? Why can you not remove v4 from your queue and replace it
with v5?

   Stefan

What you can do is to send fix or fixes with appropriate fixes tags and
I can then squash them for appropriate patches. That's less work for me.

/Jarkko

On 6/10/21 5:03 AM, Jarkko Sakkinen wrote:

On Wed, Jun 09, 2021 at 09:58:29AM -0400, Stefan Berger wrote:

quoted

On 6/9/21 8:44 AM, Jarkko Sakkinen wrote:

quoted

On Thu, Jun 03, 2021 at 08:32:59AM -0400, Stefan Berger wrote:

quoted

On 6/3/21 2:47 AM, Jarkko Sakkinen wrote:

quoted

quoted

-- 
2.29.2

Please instead send a fix.

We have a Fixes tag in 1/2, so we want this to propagate to older kernels
and need the fix in 1/2 for that reason.

     Stefan

So please do an additional fix and send it.

1/2 is supposed to propagate to older kernels and needs to change as posted
here in v5 (assuming that this does indeed fix what the build bot was
complaining about). 2/2 also changes. A fix on top of v4 would fix 2/2 but
won't apply cleanly to 1/2 as cannot easily propagate to older kernels. Is
that what we want? Why can you not remove v4 from your queue and replace it
with v5?

    Stefan

What you can do is to send fix or fixes with appropriate fixes tags and
I can then squash them for appropriate patches. That's less work for me.

Once you squash a fix on top of existing 1/2 , existing 2/2 will not 
apply anymore. I am not sure what to send you. I think it would take 
less time to remove the existing 2 patches and replace them with v5.

    Stefan