crypto: jitterentropy - Hide esoteric Kconfig options under FIPS and EXPERT

On Thu, Nov 02, 2023 at 08:32:36PM -1000, Linus Torvalds wrote:
> 
> I think that would help the situation, but I assume the sizing for the
> jitter buffer is at least partly due to trying to account for cache
> sizing or similar issues?
> 
> Which really means that I assume any static compile-time answer to
> that question is always wrong - whether you are an expert or not.
> Unless you are just building the thing for one particular machine.
> 
> So I do think the problem is deeper than "this is a question only for
> experts". I definitely don't think you should ask a regular user (or
> even a distro kernel package manager). I suspect it's likely that the
> question is just wrong in general - because any particular one buffer
> size for any number of machines simply cannot be the right answer.
> 
> I realize that the commit says "*allow* for configuration of memory
> size", but I really question the whole approach.

Yes I think these are all valid points.  I just noticed that I
forgot to cc the author so let's see if Stephan has anything to
add.

> But yes - hiding these questions from any reasonable normal user is at
> least a good first step.

OK here's the patch:

---8<---
As JITTERENTROPY is selected by default if you enable the CRYPTO
API, any Kconfig options added there will show up for every single
user.  Hide the esoteric options under EXPERT as well as FIPS so
that only distro makers will see them.

Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Am Montag, 6. November 2023, 11:00:08 CET schrieb Herbert Xu:

Hi Herbert,

> On Thu, Nov 02, 2023 at 08:32:36PM -1000, Linus Torvalds wrote:
> > I think that would help the situation, but I assume the sizing for the
> > jitter buffer is at least partly due to trying to account for cache
> > sizing or similar issues?
> > 
> > Which really means that I assume any static compile-time answer to
> > that question is always wrong - whether you are an expert or not.
> > Unless you are just building the thing for one particular machine.
> > 
> > So I do think the problem is deeper than "this is a question only for
> > experts". I definitely don't think you should ask a regular user (or
> > even a distro kernel package manager). I suspect it's likely that the
> > question is just wrong in general - because any particular one buffer
> > size for any number of machines simply cannot be the right answer.
> > 
> > I realize that the commit says "*allow* for configuration of memory
> > size", but I really question the whole approach.
> 
> Yes I think these are all valid points.  I just noticed that I
> forgot to cc the author so let's see if Stephan has anything to
> add.

I concur that these questions are more for experts. 
> 
> > But yes - hiding these questions from any reasonable normal user is at
> > least a good first step.
> 
> OK here's the patch:
> 
> ---8<---
> As JITTERENTROPY is selected by default if you enable the CRYPTO
> API, any Kconfig options added there will show up for every single
> user.  Hide the esoteric options under EXPERT as well as FIPS so
> that only distro makers will see them.
> 
> Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
> 
> diff --git a/crypto/Kconfig b/crypto/Kconfig
> index bbf51d55724e..70661f58ee41 100644
> --- a/crypto/Kconfig
> +++ b/crypto/Kconfig
> @@ -1297,10 +1297,12 @@ config CRYPTO_JITTERENTROPY
> 
>  	  See https://www.chronox.de/jent.html
> 
> +if CRYPTO_JITTERENTROPY
> +if CRYPTO_FIPS && EXPERT
> +
>  choice
>  	prompt "CPU Jitter RNG Memory Size"
>  	default CRYPTO_JITTERENTROPY_MEMSIZE_2
> -	depends on CRYPTO_JITTERENTROPY
>  	help
>  	  The Jitter RNG measures the execution time of memory accesses.
>  	  Multiple consecutive memory accesses are performed. If the memory
> @@ -1344,7 +1346,6 @@ config CRYPTO_JITTERENTROPY_OSR
>  	int "CPU Jitter RNG Oversampling Rate"
>  	range 1 15
>  	default 1
> -	depends on CRYPTO_JITTERENTROPY
>  	help
>  	  The Jitter RNG allows the specification of an oversampling rate (OSR).
>  	  The Jitter RNG operation requires a fixed amount of timing
> @@ -1359,7 +1360,6 @@ config CRYPTO_JITTERENTROPY_OSR
> 
>  config CRYPTO_JITTERENTROPY_TESTINTERFACE
>  	bool "CPU Jitter RNG Test Interface"
> -	depends on CRYPTO_JITTERENTROPY
>  	help
>  	  The test interface allows a privileged process to capture
>  	  the raw unconditioned high resolution time stamp noise that
> @@ -1377,6 +1377,28 @@ config CRYPTO_JITTERENTROPY_TESTINTERFACE
> 
>  	  If unsure, select N.
> 
> +endif	# if CRYPTO_FIPS && EXPERT
> +
> +if !(CRYPTO_FIPS && EXPERT)
> +
> +config CRYPTO_JITTERENTROPY_MEMORY_BLOCKS
> +	int
> +	default 64
> +
> +config CRYPTO_JITTERENTROPY_MEMORY_BLOCKSIZE
> +	int
> +	default 32
> +
> +config CRYPTO_JITTERENTROPY_OSR
> +	int
> +	default 1
> +
> +config CRYPTO_JITTERENTROPY_TESTINTERFACE
> +	bool
> +
> +endif	# if !(CRYPTO_FIPS && EXPERT)
> +endif	# if CRYPTO_JITTERENTROPY
> +
>  config CRYPTO_KDF800108_CTR
>  	tristate
>  	select CRYPTO_HMAC

Reviewed-by: Stephan Mueller <smueller@chronox.de>


Ciao
Stephan

Hi Herbert, Yamada-san,

On Mon, Nov 6, 2023 at 11:00 AM Herbert Xu <herbert@gondor.apana.org.au> wrote:
> On Thu, Nov 02, 2023 at 08:32:36PM -1000, Linus Torvalds wrote:
> > I think that would help the situation, but I assume the sizing for the
> > jitter buffer is at least partly due to trying to account for cache
> > sizing or similar issues?
> >
> > Which really means that I assume any static compile-time answer to
> > that question is always wrong - whether you are an expert or not.
> > Unless you are just building the thing for one particular machine.
> >
> > So I do think the problem is deeper than "this is a question only for
> > experts". I definitely don't think you should ask a regular user (or
> > even a distro kernel package manager). I suspect it's likely that the
> > question is just wrong in general - because any particular one buffer
> > size for any number of machines simply cannot be the right answer.
> >
> > I realize that the commit says "*allow* for configuration of memory
> > size", but I really question the whole approach.
>
> Yes I think these are all valid points.  I just noticed that I
> forgot to cc the author so let's see if Stephan has anything to
> add.
>
> > But yes - hiding these questions from any reasonable normal user is at
> > least a good first step.
>
> OK here's the patch:
>
> ---8<---
> As JITTERENTROPY is selected by default if you enable the CRYPTO
> API, any Kconfig options added there will show up for every single
> user.  Hide the esoteric options under EXPERT as well as FIPS so
> that only distro makers will see them.
>
> Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
> Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

Thanks for your patch, which is now commit e7ed6473c2c8c4e4 ("crypto:
jitterentropy - Hide esoteric Kconfig options under FIPS and EXPERT").

> --- a/crypto/Kconfig
> +++ b/crypto/Kconfig
> @@ -1297,10 +1297,12 @@ config CRYPTO_JITTERENTROPY
>
>           See https://www.chronox.de/jent.html
>
> +if CRYPTO_JITTERENTROPY
> +if CRYPTO_FIPS && EXPERT
> +
>  choice
>         prompt "CPU Jitter RNG Memory Size"
>         default CRYPTO_JITTERENTROPY_MEMSIZE_2
> -       depends on CRYPTO_JITTERENTROPY
>         help
>           The Jitter RNG measures the execution time of memory accesses.
>           Multiple consecutive memory accesses are performed. If the memory
> @@ -1344,7 +1346,6 @@ config CRYPTO_JITTERENTROPY_OSR
>         int "CPU Jitter RNG Oversampling Rate"
>         range 1 15
>         default 1
> -       depends on CRYPTO_JITTERENTROPY
>         help
>           The Jitter RNG allows the specification of an oversampling rate (OSR).
>           The Jitter RNG operation requires a fixed amount of timing
> @@ -1359,7 +1360,6 @@ config CRYPTO_JITTERENTROPY_OSR
>
>  config CRYPTO_JITTERENTROPY_TESTINTERFACE
>         bool "CPU Jitter RNG Test Interface"
> -       depends on CRYPTO_JITTERENTROPY
>         help
>           The test interface allows a privileged process to capture
>           the raw unconditioned high resolution time stamp noise that
> @@ -1377,6 +1377,28 @@ config CRYPTO_JITTERENTROPY_TESTINTERFACE
>
>           If unsure, select N.
>
> +endif  # if CRYPTO_FIPS && EXPERT
> +
> +if !(CRYPTO_FIPS && EXPERT)
> +
> +config CRYPTO_JITTERENTROPY_MEMORY_BLOCKS
> +       int
> +       default 64
> +
> +config CRYPTO_JITTERENTROPY_MEMORY_BLOCKSIZE
> +       int
> +       default 32
> +
> +config CRYPTO_JITTERENTROPY_OSR
> +       int
> +       default 1
> +
> +config CRYPTO_JITTERENTROPY_TESTINTERFACE
> +       bool

This duplicates the symbols in the CRYPTO_FIPS && EXPERT section above,
which is fragile.

For the int and bool symbols, this can be handled without duplication
using:

     config CRYPTO_JITTERENTROPY_OSR
    -       int "CPU Jitter RNG Oversampling Rate"
    +       int "CPU Jitter RNG Oversampling Rate" if CRYPTO_FIPS && EXPERT

     config CRYPTO_JITTERENTROPY_TESTINTERFACE
    -       bool "CPU Jitter RNG Test Interface"
    +       bool "CPU Jitter RNG Test Interface" if CRYPTO_FIPS && EXPERT

Unfortunately the following does not work for the choice statement,
although kconfig does not report an error:

     choice
    -       prompt "CPU Jitter RNG Memory Size"
    +       prompt "CPU Jitter RNG Memory Size" if CRYPTO_FIPS && EXPERT
             default CRYPTO_JITTERENTROPY_MEMSIZE_2

Unlike for other symbol types, which just become silent if
!(CRYPTO_FIPS && EXPERT), the choice is skipped completely if
!(CRYPTO_FIPS && EXPERT), and CRYPTO_JITTERENTROPY_MEMSIZE_2 is not set.

Yamada-san: Do you know why choice behaves differently?
Is this easy to fix?

Thanks!

> +
> +endif  # if !(CRYPTO_FIPS && EXPERT)
> +endif  # if CRYPTO_JITTERENTROPY
> +
>  config CRYPTO_KDF800108_CTR
>         tristate
>         select CRYPTO_HMAC

Gr{oetje,eeting}s,

                        Geert