diff mbox

tpm: require to compile as part of the kernel

Message ID 20180629151005.10899-1-jarkko.sakkinen@linux.intel.com (mailing list archive)
State New, archived
Headers show

Commit Message

Jarkko Sakkinen June 29, 2018, 3:10 p.m. UTC
Do not allow to compile TPM core as a module. TPM defines a root of
trust for integrity and keyring subsystems and should be always
available and not be loaded from the user space. There is no a
reasonable use case for a loadable module existing.

Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
---
 drivers/char/tpm/Kconfig | 2 +-
 include/linux/tpm.h      | 3 +--
 2 files changed, 2 insertions(+), 3 deletions(-)

Comments

Jason Gunthorpe June 29, 2018, 3:31 p.m. UTC | #1
On Fri, Jun 29, 2018 at 06:10:02PM +0300, Jarkko Sakkinen wrote:
> Do not allow to compile TPM core as a module. TPM defines a root of
> trust for integrity and keyring subsystems and should be always
> available and not be loaded from the user space. There is no a
> reasonable use case for a loadable module existing.
> 
> Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
> ---
>  drivers/char/tpm/Kconfig | 2 +-
>  include/linux/tpm.h      | 3 +--
>  2 files changed, 2 insertions(+), 3 deletions(-)

This doesn't really make sense..

The kconfig method is that if IMA requires TPM it should declare so
and TPM will become non-modular because IMA is non-modular.

There are lots of legitimate use cases for TPM that don't involve IMA
or keyring.

> diff --git a/drivers/char/tpm/Kconfig b/drivers/char/tpm/Kconfig
> index 18c81cbe4704..9728771aecbd 100644
> --- a/drivers/char/tpm/Kconfig
> +++ b/drivers/char/tpm/Kconfig
> @@ -3,7 +3,7 @@
>  #
>  
>  menuconfig TCG_TPM
> -	tristate "TPM Hardware Support"
> +	bool "TPM Hardware Support"
>  	depends on HAS_IOMEM
>  	select SECURITYFS
>  	select CRYPTO
> diff --git a/include/linux/tpm.h b/include/linux/tpm.h
> index 4609b94142d4..cefa61b12891 100644
> --- a/include/linux/tpm.h
> +++ b/include/linux/tpm.h
> @@ -50,8 +50,7 @@ struct tpm_class_ops {
>  	void (*clk_enable)(struct tpm_chip *chip, bool value);
>  };
>  
> -#if defined(CONFIG_TCG_TPM) || defined(CONFIG_TCG_TPM_MODULE)
> -
> +#if defined(CONFIG_TCG_TPM)

Huh. This new version is certainly right

Jason
Jarkko Sakkinen June 29, 2018, 5:43 p.m. UTC | #2
On Fri, Jun 29, 2018 at 09:31:41AM -0600, Jason Gunthorpe wrote:
> On Fri, Jun 29, 2018 at 06:10:02PM +0300, Jarkko Sakkinen wrote:
> > Do not allow to compile TPM core as a module. TPM defines a root of
> > trust for integrity and keyring subsystems and should be always
> > available and not be loaded from the user space. There is no a
> > reasonable use case for a loadable module existing.
> > 
> > Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
> > ---
> >  drivers/char/tpm/Kconfig | 2 +-
> >  include/linux/tpm.h      | 3 +--
> >  2 files changed, 2 insertions(+), 3 deletions(-)
> 
> This doesn't really make sense..
> 
> The kconfig method is that if IMA requires TPM it should declare so
> and TPM will become non-modular because IMA is non-modular.
> 
> There are lots of legitimate use cases for TPM that don't involve IMA
> or keyring.

In what context would it make sense to have TPM core as a module? I
forgot to add RFC tag this patch. Did not meant to push it to
mainline but more to rise up the discussion.

/Jarkko
Jarkko Sakkinen June 29, 2018, 5:47 p.m. UTC | #3
On Fri, Jun 29, 2018 at 09:31:41AM -0600, Jason Gunthorpe wrote:
> > -#if defined(CONFIG_TCG_TPM) || defined(CONFIG_TCG_TPM_MODULE)
> > -
> > +#if defined(CONFIG_TCG_TPM)
> 
> Huh. This new version is certainly right

Hmm...

If the option is kept as tristate, shouldn't this be actually:

#if defined(CONFIG_TCG_TPM) && !defined(CONFIG_TCG_TPM_MODULE)

?

/Jarkko
Jason Gunthorpe June 29, 2018, 6:09 p.m. UTC | #4
On Fri, Jun 29, 2018 at 08:47:43PM +0300, Jarkko Sakkinen wrote:
> On Fri, Jun 29, 2018 at 09:31:41AM -0600, Jason Gunthorpe wrote:
> > > -#if defined(CONFIG_TCG_TPM) || defined(CONFIG_TCG_TPM_MODULE)
> > > -
> > > +#if defined(CONFIG_TCG_TPM)
> > 
> > Huh. This new version is certainly right
> 
> Hmm...
> 
> If the option is kept as tristate, shouldn't this be actually:
> 
> #if defined(CONFIG_TCG_TPM) && !defined(CONFIG_TCG_TPM_MODULE)
> 
> ?

Er, yes, it should be writte as

#if IS_ENABLED(CONFIG_TCG_TPM)

these days

Jason
Jason Gunthorpe June 29, 2018, 6:11 p.m. UTC | #5
On Fri, Jun 29, 2018 at 08:43:28PM +0300, Jarkko Sakkinen wrote:
> On Fri, Jun 29, 2018 at 09:31:41AM -0600, Jason Gunthorpe wrote:
> > On Fri, Jun 29, 2018 at 06:10:02PM +0300, Jarkko Sakkinen wrote:
> > > Do not allow to compile TPM core as a module. TPM defines a root of
> > > trust for integrity and keyring subsystems and should be always
> > > available and not be loaded from the user space. There is no a
> > > reasonable use case for a loadable module existing.
> > > 
> > > Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
> > >  drivers/char/tpm/Kconfig | 2 +-
> > >  include/linux/tpm.h      | 3 +--
> > >  2 files changed, 2 insertions(+), 3 deletions(-)
> > 
> > This doesn't really make sense..
> > 
> > The kconfig method is that if IMA requires TPM it should declare so
> > and TPM will become non-modular because IMA is non-modular.
> > 
> > There are lots of legitimate use cases for TPM that don't involve IMA
> > or keyring.
> 
> In what context would it make sense to have TPM core as a module? I
> forgot to add RFC tag this patch. Did not meant to push it to
> mainline but more to rise up the discussion.

The usual reasons for modules, embedded that wants minimize kernel
image size to minimize boot time - load modules after the system has
started.. Developers that wish to use module-reload to test the code
they are working on, etc.

Jason
diff mbox

Patch

diff --git a/drivers/char/tpm/Kconfig b/drivers/char/tpm/Kconfig
index 18c81cbe4704..9728771aecbd 100644
--- a/drivers/char/tpm/Kconfig
+++ b/drivers/char/tpm/Kconfig
@@ -3,7 +3,7 @@ 
 #
 
 menuconfig TCG_TPM
-	tristate "TPM Hardware Support"
+	bool "TPM Hardware Support"
 	depends on HAS_IOMEM
 	select SECURITYFS
 	select CRYPTO
diff --git a/include/linux/tpm.h b/include/linux/tpm.h
index 4609b94142d4..cefa61b12891 100644
--- a/include/linux/tpm.h
+++ b/include/linux/tpm.h
@@ -50,8 +50,7 @@  struct tpm_class_ops {
 	void (*clk_enable)(struct tpm_chip *chip, bool value);
 };
 
-#if defined(CONFIG_TCG_TPM) || defined(CONFIG_TCG_TPM_MODULE)
-
+#if defined(CONFIG_TCG_TPM)
 extern int tpm_is_tpm2(struct tpm_chip *chip);
 extern int tpm_pcr_read(struct tpm_chip *chip, int pcr_idx, u8 *res_buf);
 extern int tpm_pcr_extend(struct tpm_chip *chip, int pcr_idx, const u8 *hash);