[RFC PATCH v2 03/13] ima: invalidate unsupported PCR banks

[Mimi Zohar]

On Wed, 2025-03-26 at 10:01 +0100, Nicolai Stange wrote:
> Mimi Zohar <zohar at linux.ibm.com> writes:
> 
> > > diff --git a/security/integrity/ima/ima_crypto.c b/security/integrity/ima/ima_crypto.c
> > > index 6f5696d999d0..a43080fb8edc 100644
> > > --- a/security/integrity/ima/ima_crypto.c
> > > +++ b/security/integrity/ima/ima_crypto.c
> > > @@ -625,26 +625,43 @@ int ima_calc_field_array_hash(struct ima_field_data *field_data,
> > >  	u16 alg_id;
> > >  	int rc, i;
> > >  
> > > +#if IS_ENABLED(CONFIG_IMA_COMPAT_FALLBACK_TPM_EXTEND)
> > >  	rc = ima_calc_field_array_hash_tfm(field_data, entry, ima_sha1_idx);
> > >  	if (rc)
> > >  		return rc;
> > >  
> > >  	entry->digests[ima_sha1_idx].alg_id = TPM_ALG_SHA1;
> > > +#endif
> > >  
> > >  	for (i = 0; i < NR_BANKS(ima_tpm_chip) + ima_extra_slots; i++) {
> > > +#if IS_ENABLED(CONFIG_IMA_COMPAT_FALLBACK_TPM_EXTEND)
> > >  		if (i == ima_sha1_idx)
> > >  			continue;
> > > +#endif
> > >  
> > >  		if (i < NR_BANKS(ima_tpm_chip)) {
> > >  			alg_id = ima_tpm_chip->allocated_banks[i].alg_id;
> > >  			entry->digests[i].alg_id = alg_id;
> > >  		}
> > >  
> > > -		/* for unmapped TPM algorithms digest is still a padded SHA1 */
> > > +		/*
> > > +		 * For unmapped TPM algorithms, the digest is still a
> > > +		 * padded SHA1 if backwards-compatibility fallback PCR
> > > +		 * extension is enabled. Otherwise fill with
> > > +		 * 0xfes. This is the value to invalidate unsupported
> > > +		 * PCR banks with. Also, a non-all-zeroes value serves
> > > +		 * as an indicator to kexec measurement restoration
> > > +		 * that the entry is not a violation and all its
> > > +		 * template digests need to get recomputed.
> > > +		 */
> > >  		if (!ima_algo_array[i].tfm) {
> > > +#if IS_ENABLED(CONFIG_IMA_COMPAT_FALLBACK_TPM_EXTEND)
> > >  			memcpy(entry->digests[i].digest,
> > >  			       entry->digests[ima_sha1_idx].digest,
> > >  			       TPM_DIGEST_SIZE);
> 
>                                ^
> That's been here before, just for the record for the below.

And it is correct.

> 
> > > +#else
> > > +			memset(entry->digests[i].digest, 0xfe, TPM_DIGEST_SIZE);
> > > +#endif
> > 
> > Using TPM_DIGEST_SIZE will result in a padded 0xfe value.
> 
> Yes, but as the sysfs files for unsupported algos are gone, this will be
> used only for extending the PCR banks. tpm[12]_pcr_extend()
> (necessarily) truncate the digests to the correct size before sending
> them to the TPM.
> 
> But if you prefer I can absolutely replace TPM_DIGEST_SIZE by
> hash_digest_size[ima_algo_array[i].algo].

Unlike violations, which are the full digest size, a padded sha1 is extended
into the unsupported algos TPM banks.  I assume you'd want it to be the full
digest size like violations.

Mimi