Re: linux-next: Tree for Aug 22 (evm)

[Mimi Zohar <zohar@linux.vnet.ibm.com>]

On Mon, 2011-08-22 at 17:49 -0700, Randy Dunlap wrote:
> On Mon, 22 Aug 2011 20:47:00 -0400 Arnaud Lacombe wrote:
> 
> > Hi,
> > 
> > On Mon, Aug 22, 2011 at 3:53 PM, Randy Dunlap <rdunlap@xenotime.net> wrote:
> > > On Mon, 22 Aug 2011 14:53:04 +1000 Stephen Rothwell wrote:
> > >
> > >> Hi all,
> > >>
> > >> [The kernel.org mirroring is a bit low today]
> > >
> > > (on x86_64:)
> > >
> > > When CONFIG_EVM=y, CONFIG_CRYPTO_HASH2=m, CONFIG_TRUSTED_KEYS=m,
> > > CONFIG_ENCRYPTED_KEYS=m, the build fails with:
> > >
> > You did not provide the value of CONFIG_TCG_TPM, I'll assume it was
> > 'm'. That said, correct me if I'm wrong, but we currently have:
> 
> Yes, it was 'm'.
> 
> > menuconfig TCG_TPM
> >         tristate "TPM Hardware Support"
> > 
> > [...]
> > 
> > config EVM
> >         boolean "EVM support"
> >         depends on SECURITY && KEYS && TCG_TPM
> > 
> > which seems terribly broken to me... How can you have a built-in
> > feature, which depends on another potentially-not-built-in feature ?
> 
> Yup.

Easy, different use cases. The TPM has been around and used for a while,
not requiring it to be built-in.  EVM, a new use case, requires it to be
built-in.

> > If you change EVM to 'tristate', you will see that you are not allowed
> > to make it built-in if TCG_TPM is not built-in.
> 
> Right.

The TPM, crypto, trusted and encrypted keys are tristate.  Like the
LSMs, EVM is boolean, which when selected using 'make xconfig', converts
the tristates to built-in.  The tristate/boolean mismatches aren't
corrected, when .config is edited directly.

Mimi

> >  - Arnaud
> > 
> > > (.text+0x378aa): undefined reference to `key_type_encrypted'
> > > evm_crypto.c:(.text+0x37992): undefined reference to `crypto_alloc_shash'
> > > evm_crypto.c:(.text+0x37a24): undefined reference to `crypto_shash_setkey'
> > > evm_crypto.c:(.text+0x37ad9): undefined reference to `crypto_shash_update'
> > > evm_crypto.c:(.text+0x37aeb): undefined reference to `crypto_shash_final'
> > > (.text+0x37b4b): undefined reference to `crypto_shash_update'
> > > (.text+0x37c61): undefined reference to `crypto_shash_update'
> > > (.text+0x37cb9): undefined reference to `crypto_shash_update'
> > >
> > > even though EVM (Kconfig) selects ENCRYPTED_KEYS and TRUSTED_KEYS..
> > > and even after I add "select CRYPTO_HASH2".
> > >
> > > Is this because EVM is bool and kconfig is confused about 'select's
> > > when a bool is selecting tristates?  Shouldn't the tristates become
> > > 'y' instead of 'm' if they are selected by a bool that is 'y'?
> > >
> > >
> > > xconfig shows these symbol values:
> > >
> > > Symbol: EVM [=y]
> > > Type : boolean
> > > Prompt: EVM support
> > > Defined at security/integrity/evm/Kconfig:1
> > > Depends on: SECURITY [=y] && KEYS [=y] && TCG_TPM [=m]
> > > Location:
> > > -> Security options
> > > Selects: CRYPTO_HMAC [=m] && CRYPTO_MD5 [=m] && CRYPTO_SHA1 [=m] && CRYPTO_HASH2 [=m] && ENCRYPTED_KEYS [=m] && TRUSTED_KEYS [=m]
> > >
> > >
> > > Hm, changing TCG_TPM to =y also changes TRUSTED_KEYS and ENCRYPTED_KEYS and
> > > lots of CRYPTO_ symbols from =m to =y.  There must be some kind of min/max
> > > symbol checking that is confused?
> > >
> > there is definitively an underlying min/max, but I would not point
> > finger too fast.
> 
> 
> Thanks for your help.
> 
> ---
> ~Randy
> *** Remember to use Documentation/SubmitChecklist when testing your code ***