On Fri, May 17, 2024 at 07:48:48PM +0300, Jarkko Sakkinen wrote:
> On Fri May 17, 2024 at 7:22 PM EEST, Nícolas F. R. A. Prado wrote:
> > On Fri, May 17, 2024 at 07:25:40AM -0700, James Bottomley wrote:
> > > On Fri, 2024-05-17 at 15:43 +0200, Ard Biesheuvel wrote:
> > > > On Fri, 17 May 2024 at 15:35, James Bottomley
> > > > <James.Bottomley@xxxxxxxxxxxxxxxxxxxxx> wrote:
> > > [...]
> > > > > Thanks for the analysis. If I look at how CRYPTO_ECC does it, that
> > > > > selects CRYPTO_RNG_DEFAULT which pulls in CRYPTO_DRBG, so the fix
> > > > > would be the attached. Does that look right to you Ard?
> > > >
> > > > No it doesn't - it's CRYPTO_RNG_DEFAULT not CRYTPO_RNG_DEFAULT :-)
> > > >
> > > > With that fixed,
> > > >
> > > > Acked-by: Ard Biesheuvel <ardb@xxxxxxxxxx>
> > >
> > > Erm, oops, sorry about that; so attached is the update.
> > >
> > > James
> > >
> > > ---8>8>8><8<8<8---
> > >
> > > From 2ac337a33e6416ef806e2c692b9239d193e8468f Mon Sep 17 00:00:00 2001
> > > From: James Bottomley <James.Bottomley@xxxxxxxxxxxxxxxxxxxxx>
> > > Date: Fri, 17 May 2024 06:29:31 -0700
> > > Subject: [PATCH] tpm: Fix sessions cryptography requirement for Random Numbers
> > > MIME-Version: 1.0
> > > Content-Type: text/plain; charset=UTF-8
> > > Content-Transfer-Encoding: 8bit
> > >
> > > The ECDH code in tpm2-sessions.c requires an initial random number
> > > generator to generate the key pair. If the configuration doesn't have
> > > CONFIG_RNG_DEFAULT, it will try to pull this in as a module (which is
> > > impossible for the early kernel boot where the TPM starts). Fix this
> > > by selecting the required RNG.
> > >
> > > Reported-by: Nícolas F. R. A. Prado <nfraprado@xxxxxxxxxxxxx>
> > > Fixes: 1b6d7f9eb150 ("tpm: add session encryption protection to tpm2_get_random()")
> > > Acked-by: Ard Biesheuvel <ardb@xxxxxxxxxx>
> > > Signed-off-by: James Bottomley <James.Bottomley@xxxxxxxxxxxxxxxxxxxxx>
> > > ---
> > > drivers/char/tpm/Kconfig | 1 +
> > > 1 file changed, 1 insertion(+)
> > >
> > > diff --git a/drivers/char/tpm/Kconfig b/drivers/char/tpm/Kconfig
> > > index 4f83ee7021d0..ecdd3db4be2b 100644
> > > --- a/drivers/char/tpm/Kconfig
> > > +++ b/drivers/char/tpm/Kconfig
> > > @@ -31,6 +31,7 @@ config TCG_TPM2_HMAC
> > > bool "Use HMAC and encrypted transactions on the TPM bus"
> > > default y
> > > select CRYPTO_ECDH
> > > + select CRYPTO_RNG_DEFAULT
> > > select CRYPTO_LIB_AESCFB
> > > select CRYPTO_LIB_SHA256
> > > help
> > > --
> > > 2.35.3
> > >
> > >
> >
> > Hi James,
> >
> > thanks for the patch. But I actually already had that config enabled builtin. I
> > also had ECDH and DRBG which have been suggested previously:
> >
> > CONFIG_CRYPTO_RNG_DEFAULT=y
> >
> > CONFIG_CRYPTO_DRBG_MENU=y
> > CONFIG_CRYPTO_DRBG_HMAC=y
> > # CONFIG_CRYPTO_DRBG_HASH is not set
> > # CONFIG_CRYPTO_DRBG_CTR is not set
> > CONFIG_CRYPTO_DRBG=y
> >
> > CONFIG_CRYPTO_ECDH=y
> >
> > I've pasted my full config here: http://0x0.st/XPN_.txt
> >
> > Adding a debug print I see that the module that the code tries to load is
> > "crypto-hmac(sha512)". I would have expected to see
> >
> > MODULE_ALIAS_CRYPTO("hmac(sha512)");
> >
> > in crypto/drbg.c, but I don't see it anywhere in the tree. Maybe it is missing?
>
This is "normal" behavior when the crypto API instantiates a template:
1. drbg.c asks for "hmac(sha512)"
2. The crypto API looks for a direct implementation of "hmac(sha512)".
This includes requesting a module with alias "crypto-hmac(sha512)".
3. If none is found, the "hmac" template is instantiated instead.
There are two possible fixes for the bug. Either fix ecc_gen_privkey() to just
use get_random_bytes() instead of the weird crypto API RNG, or make
drbg_init_hash_kernel() pass the CRYPTO_NOLOAD flag to crypto_alloc_shash().
Or if the TPM driver could be changed to not need to generate an ECC private key
at probe time, that would also avoid this problem.
- Eric
