def _resolve_conflicts(self, conf):
# Overrides
if conf_has_op(conf, 'rwindow'):
conf.max_cache = 0 # rwindow randomly shifts traces, so we cannot cache these traces during training
# Sanity checks
if conf.refset and not conf_has_op(conf, 'align'):
raise EMMAConfException("Refset specified, but no align action")
if conf.key_low >= conf.key_high:
raise EMMAConfException("key_low should be < key_high")
if self.dataset_val is not None and str(self.dataset.id) == str(
self.dataset_val.id):
raise EMMAConfException(
"Validation set should never be the same as the training set.")
if conf_has_op(conf, 'keyplot'):
if not conf_has_op(
conf, 'groupkeys'
): # Transparently add a groupkeys op, which is required for keyplot
conf.actions = conf.actions[0:-1] + [Action('groupkeys')
] + conf.actions[-1:]
if conf_has_op(conf, 'corrtrain'):
if conf.loss_type == 'softmax_crossentropy' and conf.key_high - conf.key_low != 1:
raise EMMAConfException(
"Softmax crossentropy with multiple key bytes is currently not supported"
)
def __perform_plot(emma, *params):
trace_sets_to_get = max(
int(emma.conf.plot_num_traces / emma.dataset.traces_per_set), 1)
em_result = submit_task(
ops.work, # Op
emma.dataset.trace_set_paths[0:trace_sets_to_get],
emma.conf,
keep_trace_sets=True,
keep_scores=False, # Op parameters
remote=emma.conf.remote,
message="Performing actions")
visualizations.plot_trace_sets(
em_result.reference_signal,
em_result.trace_sets,
params=params,
no_reference_plot=emma.conf.plot_no_reference,
num_traces=emma.conf.plot_num_traces,
title=emma.conf.plot_title,
xlabel=emma.conf.plot_xlabel,
ylabel=emma.conf.plot_ylabel,
colorbar_label=emma.conf.plot_colorbar_label,
time_domain=(not (conf_has_op(emma.conf, 'spec')
or conf_has_op(emma.conf, 'fft')))
or emma.conf.plot_force_timedomain,
sample_rate=1.0)
def __perform_keyplot(emma, message="Grouping keys..."):
if emma.conf.remote:
async_result = parallel_work(emma.dataset.trace_set_paths, emma.conf)
em_result = wait_until_completion(async_result, message=message)
else:
em_result = ops.work(emma.dataset.trace_set_paths, emma.conf)
em_result = ops.merge(em_result, emma.conf)
visualizations.plot_keyplot(
em_result.means,
time_domain=(not (conf_has_op(emma.conf, 'spec')
or conf_has_op(emma.conf, 'fft')))
or emma.conf.plot_force_timedomain,
sample_rate=1.0,
show=True)
def __perform_keyplot(emma, message="Grouping keys..."):
for subkey in range(emma.conf.key_low, emma.conf.key_high):
emma.conf.subkey = subkey # Set in conf, so the workers know which subkey to attack
if emma.conf.remote:
async_result = parallel_work(emma.dataset.trace_set_paths,
emma.conf)
em_result = wait_until_completion(async_result, message=message)
else:
em_result = ops.work(emma.dataset.trace_set_paths, emma.conf)
em_result = ops.merge(em_result, emma.conf)
visualizations.plot_keyplot(
em_result.means,
time_domain=(not (conf_has_op(emma.conf, 'spec')
or conf_has_op(emma.conf, 'fft')))
or emma.conf.plot_force_timedomain,
sample_rate=1.0,
show=True)
def merge(self, to_merge, conf):
if type(to_merge) is EMResult:
to_merge = [to_merge]
# Is it useful to merge?
if len(to_merge) >= 1:
result = EMResult(task_id=self.request.id)
# If we are attacking, merge the correlations
# TODO this can be cleaned up
if conf_has_op(conf, 'attack') or conf_has_op(
conf, 'memattack') or conf_has_op(conf, 'spattack'):
# Get size of correlations
shape = to_merge[
0].correlations._n.shape # TODO fixme init hetzelfde als in attack
# Init result
result.correlations = CorrelationList(shape)
# Start merging
for m in to_merge:
result.correlations.merge(m.correlations)
elif conf_has_op(
conf, 'dattack'
): # TODO just check for presence of to_merge.distances instead of doing this
shape = to_merge[0].distances._n.shape
result.distances = DistanceList(shape)
for m in to_merge:
result.distances.merge(m.distances)
elif conf_has_op(conf, 'pattack'):
shape = to_merge[0].probabilities.shape
result.probabilities = np.zeros(shape)
for m in to_merge:
result.probabilities += m.probabilities
elif conf_has_op(conf, 'keyplot'):
result.means = {}
tmp = defaultdict(lambda: [])
for m in to_merge:
for key, mean_traces in m.means.items():
tmp[key].extend(mean_traces)
for key, mean_traces in tmp.items():
all_traces = np.array(mean_traces)
print("Merging %d traces for subkey value %s" %
(all_traces.shape[0], key))
result.means[key] = np.mean(all_traces, axis=0)
# Clean up tasks
if conf.remote:
for m in to_merge:
logger.warning("Deleting %s" % m.task_id)
app.AsyncResult(m.task_id).forget()
return result
else:
return None
def _setup(self, emma_conf):
"""
Get a list of relative trace set paths for the dataset identifier and retrieve
a reference signal for the entire dataset.
Example trace set paths:
em-arduino/trace1.npy
em-arduino/trace2.npy
...
em-arduino/tracen.npy
Where trace1.npy is loaded as the reference signal.
At a later time, the relative paths need to be resolved to absolute paths
on the workers.
"""
settings = configparser.RawConfigParser()
settings.read('settings.conf')
self.root = settings.get("Datasets", "datasets_path")
# Assign trace set paths
if self.format == "cw": # .npy
path = join(self.root, self.id)
self.trace_set_paths = sorted([
join(self.id, f) for f in listdir(path)
if isfile(join(path, f)) and '_traces.npy' in f
])
elif self.format == "sigmf": # .meta
self.trace_set_paths = None
raise NotImplementedError
elif self.format == "gnuradio": # .cfile
self.trace_set_paths = None
raise NotImplementedError
elif self.format == "ascad": # ASCAD .h5
# Hack to force split between validation and training set in ASCAD
validation_set = join(
self.root,
'ASCAD/ASCAD_data/ASCAD_databases/%s.h5-val' % self.id)
training_set = join(
self.root,
'ASCAD/ASCAD_data/ASCAD_databases/%s.h5-train' % self.id)
# Make sure we never use training set when attacking or classifying
if emma_conf is not None and (
conf_has_op(emma_conf, 'attack')
or conf_has_op(emma_conf, 'classify')
or conf_has_op(emma_conf, 'dattack')
or conf_has_op(emma_conf, 'spattack')
or conf_has_op(emma_conf, 'pattack')):
self.trace_set_paths = [validation_set]
else:
self.trace_set_paths = [validation_set, training_set]
else:
raise Exception("Unknown input format '%s'" % self.format)
assert (len(self.trace_set_paths) > 0)
# Assign reference signal
reference_trace_set = emio.get_trace_set(join(self.root,
self.trace_set_paths[0]),
self.format,
ignore_malformed=False,
remote=False)
self.traces_per_set = len(reference_trace_set.traces)
self.reference_signal = reference_trace_set.traces[
self.reference_index].signal