tf_fuzz: add new crypto key generation model
* Add a new model of what it means for a crypto policy to be valid. This
fixes the simulation of psa_generate_key() calls, which TF-Fuzz
currently cannot accurately predict.
The PSA Crypto properties modelled are whether an algorithm and key
type are compatible, whether a key type/ algorithm are disabled, and
what key sizes are allowed.
Although the PSA Crypto specification has additional, more complicated
requirements for policies and keys, this commit only implements what
is necessary for predicting the outcome of psa_generate_key(), and not
requirements that make a key useless but still valid or involve
features not yet in TF-M or MbedTLS.
* Improve the way in which policies are filled in, and allow policies to
be updated at simulation time using the value of a named policy asset.
Information about other assets and calls is not accessible during
parse time when the key policy is usually filled in. However, this is
required for the improved simulation of psa_generate_key calls. This
is because policy information for generate key calls come from a named
policy created earlier in the test file.
* Add valid flag to set-policy calls, allowing the creation of a random
valid policy. For example, see demo/36.test.
* Add demo/36.test. This test generates a policy with a (roughly) even
chance of being valid or invalid and then tries to generate a key
using it.
Running this test a large number of times (~300) succeeds with the
changes in this commit, showing that TF-Fuzz can now accurately
predict the outcome of psa_generate_key calls.
Change-Id: Ia40ff893db50b8d2c579d975aa23341b7aab004d
Signed-off-by: Nik Dewally <Nik.Dewally@arm.com>
23 files changed