test/sys/opencrypto: Fix NIST KAT parser iterator

[kd]

Summary

When "yield" a.k.a a "generator" iterator is used we need to return all data using "yield", before actually returning from the function.
Because of that only encryption tests were run for AES-CBC.
I believe that other modes suffered from this bug as well.
Add one more loop to the iterator "next" routine to fix that.
This unveiled a problem in the GCM AEAD parser logic, which didn't correctly handle tests cases with empty plaintext, i.e. AAD only.
Include the fix in this patch as it's a rather trivial one.

Test Plan

Do ./runtests.sh on an x86 machine

Before:

kern.crypto.allow_soft: 1 -> 1
1..1
..................
----------------------------------------------------------------------
Ran 18 tests in 0.868s

After:

kern.crypto.allow_soft: 1 -> 1
1..1
..................
----------------------------------------------------------------------
Ran 18 tests in 4.102s

OK
ok 1
kern.crypto.allow_soft: 1 -> 1

Note the x5 difference in execution time.

[jhb]

FWIW, here are the before/after stats for runs on ccr(4) which give a more detailed breakdown of the change in stats.

Before:

dev.ccr.0.stats.port.1.completed: 2601
dev.ccr.0.stats.port.1.queued: 2601
dev.ccr.0.stats.port.1.active_sessions: 0
dev.ccr.0.stats.port.0.completed: 2609
dev.ccr.0.stats.port.0.queued: 2609
dev.ccr.0.stats.port.0.active_sessions: 0
dev.ccr.0.stats.sw_fallback: 390
dev.ccr.0.stats.process_error: 0
dev.ccr.0.stats.sglist_error: 0
dev.ccr.0.stats.pad_error: 0
dev.ccr.0.stats.mac_error: 240
dev.ccr.0.stats.inflight: 0
dev.ccr.0.stats.wr_nomem: 0
dev.ccr.0.stats.ccm_decrypt: 360
dev.ccr.0.stats.ccm_encrypt: 2130
dev.ccr.0.stats.gcm_decrypt: 0
dev.ccr.0.stats.gcm_encrypt: 0
dev.ccr.0.stats.eta_decrypt: 0
dev.ccr.0.stats.eta_encrypt: 0
dev.ccr.0.stats.cipher_decrypt: 0
dev.ccr.0.stats.cipher_encrypt: 1639
dev.ccr.0.stats.hmac: 180
dev.ccr.0.stats.hash: 901

After:

dev.ccr.0.stats.port.1.completed: 7404
dev.ccr.0.stats.port.1.queued: 7404
dev.ccr.0.stats.port.1.active_sessions: 0
dev.ccr.0.stats.port.0.completed: 7410
dev.ccr.0.stats.port.0.queued: 7410
dev.ccr.0.stats.port.0.active_sessions: 0
dev.ccr.0.stats.sw_fallback: 2190
dev.ccr.0.stats.process_error: 0
dev.ccr.0.stats.sglist_error: 0
dev.ccr.0.stats.pad_error: 0
dev.ccr.0.stats.mac_error: 696
dev.ccr.0.stats.inflight: 0
dev.ccr.0.stats.wr_nomem: 0
dev.ccr.0.stats.ccm_decrypt: 360
dev.ccr.0.stats.ccm_encrypt: 2130
dev.ccr.0.stats.gcm_decrypt: 900
dev.ccr.0.stats.gcm_encrypt: 6300
dev.ccr.0.stats.eta_decrypt: 0
dev.ccr.0.stats.eta_encrypt: 0
dev.ccr.0.stats.cipher_decrypt: 1639
dev.ccr.0.stats.cipher_encrypt: 1639
dev.ccr.0.stats.hmac: 945
dev.ccr.0.stats.hash: 901

From this it seems like none of the GCM tests were actually run previously. The total number of tests run went from 5210 to 14814.

[ngie]

tests/sys/opencrypto/cryptodev.py

291

Why not just use a truthiness statement instead?

377–386

didread is only used once; this logic can be compressed, avoiding the need for an intermediate variable.

389–393

This seems better expressed with a regular expression, e.g.,

# `field_re` should be initialized outside the loops for performance reasons.
field_re = re.compile(r"\[(?P<field>[^]]+)\]")
matches = field_re.match(i)
if matches is None:
    raise ValueError("Unknown line: %r" % (i))
yield matches.group("field"), self.fielditer()

[kd]

Update the patch based on ngie@ comments.

@jhb Could you please run the test suite again on ccr?
I just want to make sure that the numbers of tests run hasn't decreased in this version of the patch.

[kd]

tests/sys/opencrypto/cryptodev.py

291

I wanted to match the condition in line 261.
I've changed both of them instead.