mirror of
https://github.com/DISARMFoundation/DISARMframeworks.git
synced 2024-12-21 21:45:27 -05:00
22abaf93d8
Took a copy of the current AMITT github repository - we'll be updating this and merging the SPICE branch back in Rebranded to DISARM Moved generated pages to their own folder, to make looking at the repository less confusing
283 lines
10 KiB
Python
283 lines
10 KiB
Python
''' Manage AMITT counters
|
|
|
|
Create a page for each of the AMITT counter objects.
|
|
Don't worry about creating notes etc for these - they'll be in the generating spreadsheet
|
|
|
|
Reads 1 excel file: ../AMITT_MASTER_DATA/AMITT_Counters_MASTER.xlsx with sheets
|
|
* AMITT_objects: tactics, responses, actors, techniques
|
|
* Countermeasures
|
|
*
|
|
|
|
Creates markdown files
|
|
* ../counter_tactic_counts.md
|
|
* ../counter_tactics/{}counters.md
|
|
* ../counter_metatag_counts.md
|
|
* ../counter_metatag/{1}counters.md
|
|
* ../counter_resource_counts.md
|
|
* ../counter_resource/{1}counters.md
|
|
*
|
|
* {}/{}counters.md
|
|
*
|
|
|
|
'''
|
|
|
|
import pandas as pd
|
|
import numpy as np
|
|
import os
|
|
from sklearn.feature_extraction.text import CountVectorizer
|
|
|
|
|
|
class Counter:
|
|
def __init__(self, infile = '../AMITT_MASTER_DATA/AMITT_Counters_MASTER.xlsx'):
|
|
|
|
|
|
# Create counters cross-tables
|
|
crossidtechs = self.splitcol(self.dfcounters[['ID', 'Techniques']],
|
|
'Techniques', 'Techs', '\n')
|
|
crossidtechs = crossidtechs[crossidtechs['Techs'].notnull()]
|
|
crossidtechs['TID'] = crossidtechs['Techs'].str.split(' ').str[0]
|
|
crossidtechs.drop('Techs', axis=1, inplace=True)
|
|
self.idtechnique = crossidtechs
|
|
|
|
crossidres = self.splitcol(self.dfcounters[['ID', 'Resources needed']],
|
|
'Resources needed', 'Res', ',')
|
|
crossidres = crossidres[crossidres['Res'].notnull()]
|
|
self.idresource = crossidres
|
|
|
|
|
|
def analyse_counter_text(self, col='Title'):
|
|
# Analyse text in counter descriptions
|
|
alltext = (' ').join(self.dfcounters[col].to_list()).lower()
|
|
count_vect = CountVectorizer(stop_words='english')
|
|
word_counts = count_vect.fit_transform([alltext])
|
|
dfw = pd.DataFrame(word_counts.A, columns=count_vect.get_feature_names()).transpose()
|
|
dfw.columns = ['count']
|
|
dfw = dfw.sort_values(by='count', ascending=False)
|
|
return(dfw)
|
|
|
|
|
|
def splitcol(self, df, col, newcol, divider=','):
|
|
# Thanks https://stackoverflow.com/questions/17116814/pandas-how-do-i-split-text-in-a-column-into-multiple-rows?noredirect=1
|
|
return (df.join(df[col]
|
|
.str.split(divider, expand=True).stack()
|
|
.reset_index(drop=True,level=1)
|
|
.rename(newcol)).drop(col, axis=1))
|
|
|
|
|
|
# Print list of counters for each square of the COA matrix
|
|
# Write HTML version of framework diagram to markdown file
|
|
def write_counters_tactics_markdown(self, outfile = '../counter_tactic_counts.md'):
|
|
|
|
coacounts = pd.pivot_table(self.dfcounters[['Tactic', 'Response',
|
|
'ID']], index='Response', columns='Tactic', aggfunc=len, fill_value=0)
|
|
|
|
html = '''# AMITT Courses of Action matrix:
|
|
|
|
<table border="1">
|
|
<tr>
|
|
<td> </td>
|
|
'''
|
|
#Table heading = Tactic names
|
|
for col in coacounts.columns.get_level_values(1):
|
|
tid = self.create_tactic_file(col)
|
|
html += '<td><a href="counter_tactics/{0}counters.md">{1}</a></td>\n'.format(
|
|
tid, col)
|
|
html += '</tr><tr>\n'
|
|
|
|
# number of counters per response type
|
|
for response, counts in coacounts.iterrows():
|
|
html += '<td>{}</td>\n'.format(response)
|
|
for val in counts.values:
|
|
html += '<td>{}</td>\n'.format(val)
|
|
html += '</tr>\n<tr>\n'
|
|
|
|
# Total per tactic
|
|
html += '<td>TOTALS</td>\n'
|
|
for val in coacounts.sum().values:
|
|
html += '<td>{}</td>\n'.format(val)
|
|
html += '</tr>\n</table>\n'
|
|
|
|
with open(outfile, 'w') as f:
|
|
f.write(html)
|
|
print('updated {}'.format(outfile))
|
|
return
|
|
|
|
def create_tactic_file(self, tname):
|
|
if not os.path.exists('../counter_tactics'):
|
|
os.makedirs('../counter_tactics')
|
|
|
|
tid = tname[:tname.find(' ')]
|
|
html = '''# Tactic {} counters\n\n'''.format(tname)
|
|
|
|
html += '## by action\n\n'
|
|
for resp, counters in self.dfcounters[self.dfcounters['Tactic'] == tname].groupby('Response'):
|
|
html += '\n### {}\n'.format(resp)
|
|
|
|
for c in counters.iterrows():
|
|
html += '* {}: {} (needs {})\n'.format(c[1]['ID'], c[1]['Title'],
|
|
c[1]['Resources needed'])
|
|
|
|
html += '\n## by technique\n\n'
|
|
tactecs = self.techniques[self.techniques['phase'] == tid]['Id'].to_list()
|
|
for tech in [tid] + tactecs:
|
|
if tech == tid:
|
|
html += '\n### {}\n'.format(tech)
|
|
else:
|
|
techname = self.techniques[self.techniques['Id']==tech]['longname']
|
|
html += '\n### {}\n'.format(techname)
|
|
|
|
taccounts = self.idtechnique[self.idtechnique['TID'] == tech]
|
|
# html += '\n{}\n'.format(taccounts)
|
|
for c in self.dfcounters[self.dfcounters['ID'].isin(taccounts['ID'])].iterrows():
|
|
html += '* {}: {} (needs {})\n'.format(c[1]['ID'], c[1]['Title'],
|
|
c[1]['Resources needed'])
|
|
|
|
datafile = '../counter_tactics/{}counters.md'.format(tid)
|
|
print('Writing {}'.format(datafile))
|
|
with open(datafile, 'w') as f:
|
|
f.write(html)
|
|
f.close()
|
|
return(tid)
|
|
|
|
|
|
def create_object_file(self, index, rowtype, datadir):
|
|
|
|
oid = index
|
|
html = '''# {} counters: {}\n\n'''.format(rowtype, index)
|
|
|
|
html += '## by action\n\n'
|
|
for resp, clist in self.dfcounters[self.dfcounters[rowtype] == index].groupby('Response'):
|
|
html += '\n### {}\n'.format(resp)
|
|
|
|
for c in clist.iterrows():
|
|
html += '* {}: {} (needs {})\n'.format(c[1]['ID'], c[1]['Title'],
|
|
c[1]['Resources needed'])
|
|
|
|
datafile = '{}/{}counters.md'.format(datadir, oid)
|
|
print('Writing {}'.format(datafile))
|
|
with open(datafile, 'w') as f:
|
|
f.write(html)
|
|
f.close()
|
|
return(oid)
|
|
|
|
|
|
def write_counters_metacounts_markdown(self, outfile = '../counter_metatag_counts.md'):
|
|
|
|
coltype = 'Response'
|
|
rowtype = 'metatechnique'
|
|
rowname = 'metatag'
|
|
mtcounts = pd.pivot_table(self.dfcounters[[coltype, rowtype,'ID']],
|
|
index=rowtype, columns=coltype, aggfunc=len,
|
|
fill_value=0)
|
|
mtcounts['TOTALS'] = mtcounts.sum(axis=1)
|
|
|
|
html = '''# AMITT {} courses of action
|
|
|
|
<table border="1">
|
|
<tr>
|
|
<td> </td>
|
|
'''.format(rowtype)
|
|
|
|
# Table heading row
|
|
for col in mtcounts.columns.get_level_values(1)[:-1]:
|
|
html += '<td>{}</td>\n'.format(col)
|
|
html += '<td>TOTALS</td></tr><tr>\n'
|
|
|
|
# Data rows
|
|
datadir = '../counters_{}'.format(rowname)
|
|
if not os.path.exists(datadir):
|
|
os.makedirs(datadir)
|
|
for index, counts in mtcounts.iterrows():
|
|
tid = self.create_object_file(index, rowtype, datadir)
|
|
html += '<td><a href="counter_{0}/{1}counters.md">{2}</a></td>\n'.format(
|
|
rowname, tid, index)
|
|
for val in counts.values:
|
|
html += '<td>{}</td>\n'.format(val)
|
|
html += '</tr>\n<tr>\n'
|
|
|
|
# Column sums
|
|
html += '<td>TOTALS</td>\n'
|
|
for val in mtcounts.sum().values:
|
|
html += '<td>{}</td>\n'.format(val)
|
|
html += '</tr>\n</table>\n'
|
|
|
|
with open(outfile, 'w') as f:
|
|
f.write(html)
|
|
print('updated {}'.format(outfile))
|
|
|
|
return
|
|
|
|
|
|
def create_resource_file(self, index, rowtype, datadir):
|
|
oid = index
|
|
counterrows = self.idresource[self.idresource['Res'] == index]['ID'].to_list()
|
|
html = '''# {} counters: {}\n\n'''.format(rowtype, index)
|
|
html += '## by action\n\n'
|
|
omatrix = self.dfcounters[self.dfcounters['ID'].isin(counterrows)].groupby('Response')
|
|
for resp, clist in omatrix:
|
|
html += '\n### {}\n'.format(resp)
|
|
for c in clist.iterrows():
|
|
html += '* {}: {} (needs {})\n'.format(c[1]['ID'], c[1]['Title'],
|
|
c[1]['Resources needed'])
|
|
|
|
datafile = '{}/{}counters.md'.format(datadir, oid)
|
|
print('Writing {}'.format(datafile))
|
|
with open(datafile, 'w') as f:
|
|
f.write(html)
|
|
f.close()
|
|
return(oid, omatrix)
|
|
|
|
|
|
def write_counters_resource_markdown(self, outfile = '../counter_resource_counts.md'):
|
|
|
|
coltype = 'Response'
|
|
rowtype = 'resource'
|
|
rowname = 'resource'
|
|
|
|
html = '''# AMITT {} courses of action
|
|
|
|
<table border="1">
|
|
<tr>
|
|
<td> </td>
|
|
'''.format(rowtype)
|
|
|
|
# Table heading row
|
|
colvals = self.dfcounters[coltype].value_counts().sort_index().index
|
|
for col in colvals:
|
|
html += '<td>{}</td>\n'.format(col)
|
|
html += '<td>TOTALS</td></tr><tr>\n'
|
|
|
|
# Data rows
|
|
datadir = '../counter_{}'.format(rowname)
|
|
if not os.path.exists(datadir):
|
|
os.makedirs(datadir)
|
|
for index in self.idresource['Res'].value_counts().sort_index().index:
|
|
(oid, omatrix) = self.create_resource_file(index, rowtype, datadir) #self
|
|
row = pd.DataFrame(omatrix.apply(len), index=colvals).fillna(' ')
|
|
html += '<td><a href="counter_{0}/{1}counters.md">{2}</a></td>\n'.format(
|
|
rowname, oid, index)
|
|
if len(row.columns) > 0:
|
|
for val in row[0].to_list():
|
|
html += '<td>{}</td>\n'.format(val)
|
|
html += '<td>{}</td></tr>\n<tr>\n'.format('')
|
|
|
|
html += '</tr>\n</table>\n'
|
|
|
|
with open(outfile, 'w') as f:
|
|
f.write(html)
|
|
print('updated {}'.format(outfile))
|
|
|
|
return
|
|
|
|
|
|
|
|
def main():
|
|
counter = Counter()
|
|
counter.write_counters_tactics_markdown()
|
|
counter.write_counters_metacounts_markdown()
|
|
counter.write_counters_resource_markdown()
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|