Current File : //lib64/python3.9/site-packages/setools/infoflow.py
# Copyright 2014-2015, Tresys Technology, LLC
#
# SPDX-License-Identifier: LGPL-2.1-only
#
import itertools
import logging
from contextlib import suppress
from typing import cast, Iterable, List, Mapping, Optional, Union
try:
import networkx as nx
from networkx.exception import NetworkXError, NetworkXNoPath, NodeNotFound
except ImportError:
logging.getLogger(__name__).debug("NetworkX failed to import.")
from .descriptors import EdgeAttrIntMax, EdgeAttrList
from .permmap import PermissionMap
from .policyrep import AVRule, SELinuxPolicy, TERuletype, Type
__all__ = ['InfoFlowAnalysis']
InfoFlowPath = Iterable['InfoFlowStep']
class InfoFlowAnalysis:
"""Information flow analysis."""
_exclude: List[Type]
_min_weight: int
_perm_map: PermissionMap
def __init__(self, policy: SELinuxPolicy, perm_map: PermissionMap, min_weight: int = 1,
exclude: Optional[Iterable[Union[Type, str]]] = None,
booleans: Optional[Mapping[str, bool]] = None) -> None:
"""
Parameters:
policy The policy to analyze.
perm_map The permission map or path to the permission map file.
minweight The minimum permission weight to include in the analysis.
(default is 1)
exclude The types excluded from the information flow analysis.
(default is none)
booleans If None, all rules will be added to the analysis (default).
otherwise it should be set to a dict with keys corresponding
to boolean names and values of True/False. Any unspecified
booleans will use the policy's default values.
"""
self.log = logging.getLogger(__name__)
self.policy = policy
self.min_weight = min_weight
self.perm_map = perm_map
self.exclude = exclude # type: ignore # https://github.com/python/mypy/issues/220
self.booleans = booleans
self.rebuildgraph = True
self.rebuildsubgraph = True
try:
self.G = nx.DiGraph()
self.subG = self.G.copy()
except NameError:
self.log.critical("NetworkX is not available. This is "
"requried for Information Flow Analysis.")
self.log.critical("This is typically in the python3-networkx package.")
raise
@property
def min_weight(self) -> int:
return self._min_weight
@min_weight.setter
def min_weight(self, weight: int) -> None:
if not 1 <= weight <= 10:
raise ValueError(
"Min information flow weight must be an integer 1-10.")
self._min_weight = weight
self.rebuildsubgraph = True
@property
def perm_map(self) -> PermissionMap:
return self._perm_map
@perm_map.setter
def perm_map(self, perm_map: PermissionMap) -> None:
self._perm_map = perm_map
self.rebuildgraph = True
self.rebuildsubgraph = True
@property
def exclude(self) -> List[Type]:
return self._exclude
@exclude.setter
def exclude(self, types: Optional[Iterable[Union[Type, str]]]) -> None:
if types:
self._exclude: List[Type] = [self.policy.lookup_type(t) for t in types]
else:
self._exclude = []
self.rebuildsubgraph = True
def shortest_path(self, source: Type, target: Type) -> Iterable[InfoFlowPath]:
"""
Generator which yields one shortest path between the source
and target types (there may be more).
Parameters:
source The source type.
target The target type.
Yield: generator(steps)
steps Yield: tuple(source, target, rules)
source The source type for this step of the information flow.
target The target type for this step of the information flow.
rules The list of rules creating this information flow step.
"""
s = self.policy.lookup_type(source)
t = self.policy.lookup_type(target)
if self.rebuildsubgraph:
self._build_subgraph()
self.log.info("Generating one shortest information flow path from {0} to {1}...".
format(s, t))
with suppress(NetworkXNoPath, NodeNotFound):
# NodeNotFound: the type is valid but not in graph, e.g.
# excluded or disconnected due to min weight
# NetworkXNoPath: no paths or the target type is
# not in the graph
# pylint: disable=unexpected-keyword-arg, no-value-for-parameter
yield self.__generate_steps(nx.shortest_path(self.subG, source=s, target=t))
def all_paths(self, source: Union[Type, str], target: Union[Type, str], maxlen: int = 2) \
-> Iterable[InfoFlowPath]:
"""
Generator which yields all paths between the source and target
up to the specified maximum path length. This algorithm
tends to get very expensive above 3-5 steps, depending
on the policy complexity.
Parameters:
source The source type.
target The target type.
maxlen Maximum length of paths.
Yield: generator(steps)
steps Yield: tuple(source, target, rules)
source The source type for this step of the information flow.
target The target type for this step of the information flow.
rules The list of rules creating this information flow step.
"""
if maxlen < 1:
raise ValueError("Maximum path length must be positive.")
s = self.policy.lookup_type(source)
t = self.policy.lookup_type(target)
if self.rebuildsubgraph:
self._build_subgraph()
self.log.info("Generating all information flow paths from {0} to {1}, max length {2}...".
format(s, t, maxlen))
with suppress(NetworkXNoPath, NodeNotFound):
# NodeNotFound: the type is valid but not in graph, e.g.
# excluded or disconnected due to min weight
# NetworkXNoPath: no paths or the target type is
# not in the graph
for path in nx.all_simple_paths(self.subG, s, t, maxlen):
yield self.__generate_steps(path)
def all_shortest_paths(self, source: Union[Type, str], target: Union[Type, str]) \
-> Iterable[InfoFlowPath]:
"""
Generator which yields all shortest paths between the source
and target types.
Parameters:
source The source type.
target The target type.
Yield: generator(steps)
steps Yield: tuple(source, target, rules)
source The source type for this step of the information flow.
target The target type for this step of the information flow.
rules The list of rules creating this information flow step.
"""
s = self.policy.lookup_type(source)
t = self.policy.lookup_type(target)
if self.rebuildsubgraph:
self._build_subgraph()
self.log.info("Generating all shortest information flow paths from {0} to {1}...".
format(s, t))
with suppress(NetworkXNoPath, NodeNotFound):
# NodeNotFound: the type is valid but not in graph, e.g.
# excluded or disconnected due to min weight
# NetworkXNoPath: no paths or the target type is
# not in the graph
for path in nx.all_shortest_paths(self.subG, s, t):
yield self.__generate_steps(path)
def infoflows(self, type_: Union[Type, str], out: bool = True) -> Iterable['InfoFlowStep']:
"""
Generator which yields all information flows in/out of a
specified source type.
Parameters:
source The starting type.
Keyword Parameters:
out If true, information flows out of the type will
be returned. If false, information flows in to the
type will be returned. Default is true.
Yield: generator(steps)
steps A generator that returns the tuple of
source, target, and rules for each
information flow.
"""
s = self.policy.lookup_type(type_)
if self.rebuildsubgraph:
self._build_subgraph()
self.log.info("Generating all information flows {0} {1}".
format("out of" if out else "into", s))
with suppress(NetworkXError):
# NetworkXError: the type is valid but not in graph, e.g.
# excluded or disconnected due to min weight
if out:
flows = self.subG.out_edges(s)
else:
flows = self.subG.in_edges(s)
for source, target in flows:
yield InfoFlowStep(self.subG, source, target)
def get_stats(self) -> str: # pragma: no cover
"""
Get the information flow graph statistics.
Return: str
"""
if self.rebuildgraph:
self._build_graph()
return f"Graph nodes: {nx.number_of_nodes(self.G)}\n" \
f"Graph edges: {nx.number_of_edges(self.G)}"
#
# Internal functions follow
#
def __generate_steps(self, path: List[Type]) -> InfoFlowPath:
"""
Generator which returns the source, target, and associated rules
for each information flow step.
Parameter:
path A list of graph node names representing an information flow path.
Yield: tuple(source, target, rules)
source The source type for this step of the information flow.
target The target type for this step of the information flow.
rules The list of rules creating this information flow step.
"""
for s in range(1, len(path)):
yield InfoFlowStep(self.subG, path[s - 1], path[s])
#
#
# Graph building functions
#
#
# 1. _build_graph determines the flow in each direction for each TE
# rule and then expands the rule. All information flows are
# included in this main graph: memory is traded off for efficiency
# as the main graph should only need to be rebuilt if permission
# weights change.
# 2. _build_subgraph derives a subgraph which removes all excluded
# types (nodes) and edges (information flows) which are below the
# minimum weight. This subgraph is rebuilt only if the main graph
# is rebuilt or the minimum weight or excluded types change.
def _build_graph(self) -> None:
self.G.clear()
self.G.name = "Information flow graph for {0}.".format(self.policy)
self.perm_map.map_policy(self.policy)
self.log.info("Building information flow graph from {0}...".format(self.policy))
for rule in self.policy.terules():
if rule.ruletype != TERuletype.allow:
continue
(rweight, wweight) = self.perm_map.rule_weight(cast(AVRule, rule))
for s, t in itertools.product(rule.source.expand(), rule.target.expand()):
# only add flows if they actually flow
# in or out of the source type type
if s != t:
if wweight:
edge = InfoFlowStep(self.G, s, t, create=True)
edge.rules.append(rule)
edge.weight = wweight
if rweight:
edge = InfoFlowStep(self.G, t, s, create=True)
edge.rules.append(rule)
edge.weight = rweight
self.rebuildgraph = False
self.rebuildsubgraph = True
self.log.info("Completed building information flow graph.")
self.log.debug("Graph stats: nodes: {0}, edges: {1}.".format(
nx.number_of_nodes(self.G),
nx.number_of_edges(self.G)))
def _build_subgraph(self) -> None:
if self.rebuildgraph:
self._build_graph()
self.log.info("Building information flow subgraph...")
self.log.debug("Excluding {0!r}".format(self.exclude))
self.log.debug("Min weight {0}".format(self.min_weight))
self.log.debug("Exclude disabled conditional policy: {0}".format(
self.booleans is not None))
# delete excluded types from subgraph
nodes = [n for n in self.G.nodes() if n not in self.exclude]
self.subG = self.G.subgraph(nodes).copy()
# delete edges below minimum weight.
# no need if weight is 1, since that
# does not exclude any edges.
if self.min_weight > 1:
delete_list = []
for s, t in self.subG.edges():
edge = InfoFlowStep(self.subG, s, t)
if edge.weight < self.min_weight:
delete_list.append(edge)
self.subG.remove_edges_from(delete_list)
if self.booleans is not None:
delete_list = []
for s, t in self.subG.edges():
edge = InfoFlowStep(self.subG, s, t)
# collect disabled rules
rule_list = []
# pylint: disable=not-an-iterable
for rule in edge.rules:
if not rule.enabled(**self.booleans):
rule_list.append(rule)
deleted_rules: List[AVRule] = []
for rule in rule_list:
if rule not in deleted_rules:
edge.rules.remove(rule)
deleted_rules.append(rule)
if not edge.rules:
delete_list.append(edge)
self.subG.remove_edges_from(delete_list)
self.rebuildsubgraph = False
self.log.info("Completed building information flow subgraph.")
self.log.debug("Subgraph stats: nodes: {0}, edges: {1}.".format(
nx.number_of_nodes(self.subG),
nx.number_of_edges(self.subG)))
class InfoFlowStep:
"""
A graph edge. Also used for returning information flow steps.
Parameters:
graph The NetworkX graph.
source The source type of the edge.
target The target type of the edge.
Keyword Parameters:
create (T/F) create the edge if it does not exist.
The default is False.
"""
rules = EdgeAttrList('rules')
# use capacity to store the info flow weight so
# we can use network flow algorithms naturally.
# The weight for each edge is 1 since each info
# flow step is no more costly than another
# (see below add_edge() call)
weight = EdgeAttrIntMax('capacity')
def __init__(self, graph, source: Type, target: Type, create: bool = False) -> None:
self.G = graph
self.source = source
self.target = target
if not self.G.has_edge(source, target):
if create:
self.G.add_edge(source, target, weight=1)
self.rules = None
self.weight = None
else:
raise ValueError("InfoFlowStep does not exist in graph")
def __getitem__(self, key):
# This is implemented so this object can be used in NetworkX
# functions that operate on (source, target) tuples
if isinstance(key, slice):
return [self._index_to_item(i) for i in range(* key.indices(2))]
else:
return self._index_to_item(key)
def _index_to_item(self, index):
"""Return source or target based on index."""
if index == 0:
return self.source
elif index == 1:
return self.target
else:
raise IndexError("Invalid index (edges only have 2 items): {0}".format(index))
Mr. DellatioNx196 GaLers xh3LL Backd00r 1.0, Coded By Mr. DellatioNx196 - Bogor BlackHat