%PDF- %PDF-
Direktori : /lib/python3/dist-packages/sos/cleaner/parsers/ |
Current File : //lib/python3/dist-packages/sos/cleaner/parsers/mac_parser.py |
# Copyright 2020 Red Hat, Inc. Jake Hunsaker <jhunsake@redhat.com> # This file is part of the sos project: https://github.com/sosreport/sos # # This copyrighted material is made available to anyone wishing to use, # modify, copy, or redistribute it subject to the terms and conditions of # version 2 of the GNU General Public License. # # See the LICENSE file in the source distribution for further information. from sos.cleaner.parsers import SoSCleanerParser from sos.cleaner.mappings.mac_map import SoSMacMap import re # aa:bb:cc:fe:ff:dd:ee:ff IPV6_REG_8HEX = ( r'((?<!([0-9a-fA-F\'\"]:)|::)([^:|-])?([0-9a-fA-F]{2}(:|-)){7}' r'[0-9a-fA-F]{2}(\'|\")?(\/|\,|\-|\.|\s|$))' ) # aabb:ccee:ddee:ffaa IPV6_REG_4HEX = ( r'((?<!([0-9a-fA-F\'\"]:)|::)(([^:\-]?[0-9a-fA-F]{4}(:|-)){3}' r'[0-9a-fA-F]{4}(\'|\")?(\/|\,|\-|\.|\s|$)))' ) # aa:bb:cc:dd:ee:ff avoiding ipv6 substring matches IPV4_REG = ( r'((?<!([0-9a-fA-F\'\"]:)|::)' r'(([^:\-])?(([0-9a-fA-F]{2}([:\-\_])){5,6}([0-9a-fA-F]{2}))))' ) class SoSMacParser(SoSCleanerParser): """Handles parsing for MAC addresses""" name = 'MAC Parser' regex_patterns = [ IPV6_REG_8HEX, IPV6_REG_4HEX, IPV4_REG ] obfuscated_patterns = ( '53:4f:53', '534f:53' ) skip_files = [ 'sos_commands/.*/modinfo.*' ] map_file_key = 'mac_map' compile_regexes = False def __init__(self, config): self.mapping = SoSMacMap() super(SoSMacParser, self).__init__(config) def reduce_mac_match(self, match): """Strips away leading and trailing non-alphanum characters from any matched string to leave us with just the bare MAC addr """ while not (match[0].isdigit() or match[0].isalpha()): match = match[1:] while not (match[-1].isdigit() or match[-1].isalpha()): match = match[0:-1] # just to be safe, call strip() to remove any padding return match.strip() def _parse_line(self, line): count = 0 for pattern in self.regex_patterns: matches = [m[0] for m in re.findall(pattern, line, re.I)] if matches: count += len(matches) for match in matches: stripped_match = self.reduce_mac_match(match) if stripped_match.startswith(self.obfuscated_patterns): # avoid double scrubbing continue new_match = self.mapping.get(stripped_match) line = line.replace(stripped_match, new_match) return line, count