def _memory_access(self, state):
if state.scratch.sim_procedure is not None: return
key = (state.scratch.bbl_addr, state.scratch.stmt_idx)
#l.debug("hit access (%#x, %d)...", *key)
try:
cur_access = self.accesses[key]
except KeyError:
l.error("There's an access I know nothing about!!!!!!! (%#x, %d)", *key)
import ipdb; ipdb.set_trace()
return
if cur_access.is_write:
pointer = state.inspect.mem_write_address
#length = state.inspect.mem_write_length
data = state.inspect.mem_write_expr
else:
pointer = state.inspect.mem_read_address
#length = state.inspect.mem_read_length
data = state.inspect.mem_read_expr
ptr_ty = self.ty_backend.convert(pointer).ty
if type(ptr_ty) is sim_type.SimTypePointer:
#l.info("...got em!")
offset = ptr_ty.offset
subty = ptr_ty.pts_to
if type(subty) is SimStructAbstract:
subty.offsets[offset.cache_key].add(cur_access)
cur_access.source = ValueSource(subty, offset)
if data.op == 'BVS' and data.args[0].startswith('mem_') and len(data.annotations) == 0:
# this is a fresh read! we need to mark its source.
newty = sim_type.SimTypeTop(label=[ValueSource(subty, offset)])
data = data.annotate(TypeAnnotation(newty))
state.memory.store(pointer, data, inspect=False, endness=state.arch.memory_endness)
else:
l.warning('...pointer is to %s?', repr(subty))
elif not pointer.symbolic:
#l.info("...global data!")
self.global_struct.offsets[pointer.cache_key].add(cur_access)
cur_access.source = ValueSource(self.global_struct, state.se.eval(pointer))
else:
#l.info("...don't got em!")
if ptr_ty.label is not None and len(ptr_ty.label) > 0:
if len(ptr_ty.label) > 1:
import ipdb; ipdb.set_trace()
print 'not sure how this case can arise but it needs special handling if it does'
#l.info("...but we have a source!")
self.pass_results.append(('SOURCE', ptr_ty.label[0]))
if cur_access.is_write:
#state.inspect.mem_write_address = pointer
#state.inspect.mem_write_length = length
state.inspect.mem_write_expr = data
else:
#state.inspect.mem_read_address = pointer
#state.inspect.mem_read_length = length
state.inspect.mem_read_expr = data
def _init_analysis(self, func):
mark_addr = None
for node in func.nodes:
if type(node) is not angr.knowledge.codenode.BlockNode:
continue
block = self.project.factory.block(node.addr, size=node.size)
for idx, stmt in enumerate(block.vex.statements):
if stmt.tag == 'Ist_IMark':
mark_addr = stmt.addr
if stmt.tag == 'Ist_WrTmp' and stmt.data.tag == 'Iex_Load':
self.accesses[(block.addr,
idx)] = Access(block.addr, idx, False,
stmt.data.result_size >> 3,
mark_addr)
elif stmt.tag == 'Ist_Store':
self.accesses[(block.addr,
idx)] = Access(block.addr, idx, True,
stmt.data.result_size >> 3,
mark_addr)
elif stmt.tag == 'Ist_Dirty' and stmt.cee.name == 'x86g_dirtyhelper_loadF80le':
self.accesses[(block.addr,
idx)] = Access(block.addr, idx, False, 10,
mark_addr)
elif stmt.tag == 'Ist_Dirty' and stmt.cee.name == 'x86g_dirtyhelper_storeF80le':
self.accesses[(block.addr,
idx)] = Access(block.addr, idx, True, 10,
mark_addr)
frame_struct = SimStructAbstract(label='stack_%x' % func.addr)
self._register_addr(frame_struct)
sp_bv = self.pointer_to_abstruct(frame_struct)
self.function_initial_regs[func.addr] = {'sp': sp_bv}
self.initial_state.memory.store(
sp_bv,
claripy.BVV(0x1234, self.project.arch.bits),
endness=self.project.arch.memory_endness,
inspect=False)
for reg in self.project.arch.default_symbolic_registers:
if self.project.arch.registers[reg][
0] == self.project.arch.registers['sp'][0]:
continue
val = claripy.BVS('reg_%s' % reg, self.project.arch.bits)
val = val.annotate(
TypeAnnotation(
sim_type.SimTypeTop(label=[ValueSource('register', reg)])))
self.function_initial_regs[func.addr][reg] = val
self._analyze(func.addr)
def _runtime_unify(self,
state,
one,
two,
stack_frame=False,
overwrite=True):
"""
decide if one and two need to be unified, if so add a 'UNIFY' tag
:param state: The analysis state that holds intermediate results
:param one: The first value to unify
:param two: The second value to unify
:param stack_frame: If we're only allowed to look at offsets in front of the pointer
:param overwrite: Whether to use the semantics that one is "overwriting" two
"""
one_ty = self.ty_backend.convert(one).ty
two_ty = self.ty_backend.convert(two).ty
# if both of them are pointers!!! this gets very tricky
if type(one_ty) is type(two_ty) is sim_type.SimTypePointer:
one_subty = one_ty.pts_to
two_subty = two_ty.pts_to
one_offset = one_ty.offset
two_offset = two_ty.offset
if one_offset.symbolic or two_offset.symbolic:
import ipdb
ipdb.set_trace()
print('yikes! (jesus christ)')
if one_subty is two_subty:
if one_offset is not two_subty:
import ipdb
ipdb.set_trace()
print('yikes? (arrays maybe. recursion probably)')
else:
import ipdb
ipdb.set_trace()
print('yikes. (no object identity but yes equality)')
# these are two different structures that we might have to unify.
# possible cases:
# - two structures are actually the same structure.
# - two stack frames. flag tells us this. only deal with the argument parts
# - a structure is present in another structure
# TODO: do some type checking on the two structs to make sure we're not making
# a huge mistake!!!
else:
if claripy.is_true(
one_offset == two_offset) and claripy.is_true(
one_offset == 0):
self.pass_results.append(('UNIFY', (one, two)))
elif stack_frame:
for ckey, _ in two_subty.offsets.iteritems():
offset = ckey.ast
if not claripy.is_true(claripy.SGT(offset, 0)):
continue
two_value = state.memory.load(
two + offset,
size=state.arch.bytes,
inspect=False,
endness=state.arch.memory_endness)
one_value = state.memory.load(
one + offset,
size=state.arch.bytes,
inspect=False,
endness=state.arch.memory_endness)
# one last edge case consideration: if one_value doesn't have a source
# (since we disabled inspect, this might happen)
# we should manually give it a source since this is something we know
one_value_ty = self.ty_backend.convert(one_value).ty
if type(one_value_ty) is not sim_type.SimTypePointer and \
len(one_value_ty.label) == 0:
one_value = one_value.annotate(
TypeAnnotation(
sim_type.SimTypeTop(
label=[ValueSource(two_subty, offset)
])))
self._runtime_unify(state, one_value, two_value)
else:
import ipdb
ipdb.set_trace()
# look through the structs and check offset by offset
# do we want to check in the initial state or elsewhere?
# think we should check in the current state. everything from the initial state wil be there still?
print('okay??')
# when only one of them is a pointer!
# if a source is available we should toss it to SOURCE, she'll just drop the other in!
# if a source is not available, wait. eventually one will be available :)
elif type(two_ty) is sim_type.SimTypePointer:
if len(one_ty.label) > 0:
if len(one_ty.label) > 1:
import ipdb
ipdb.set_trace()
print('????????????')
self.pass_results.append(('SOURCE', one_ty.label[0]))
# this is where overwrite semantics comes into play.
# if one overwrites two, then we can't mark two as a pointer just because 1 is a pointer.
# otherwise, this is the same as the previous case I guess?
# I'm not sure what good this does
elif type(one_ty) is sim_type.SimTypePointer:
import ipdb
ipdb.set_trace()
if not overwrite:
if len(two_ty.label) > 0:
if len(two_ty.label) > 1:
import ipdb
ipdb.set_trace()
print('????????????')
self.pass_results.append(('SOURCE', two_ty.label[0]))
# If neither of them are pointers bail out. this is not a general type inference :)
else:
pass
def _exit_taken(self, state):
jk = state.inspect.exit_jumpkind
target = state.inspect.exit_target
if target.symbolic: # shit lmao
all_targets = tuple(state.solver.any_n_int(target, 257))
if len(all_targets) > 256:
import ipdb
ipdb.set_trace()
print('shit!! lmao')
else:
all_targets = (state.solver.eval(target), )
if jk == 'Ijk_Call' or jk.startswith('Ijk_Sys'):
if jk == 'Ijk_Call':
ret_addr = state.memory.load(state.regs.sp,
size=state.arch.bytes,
endness=state.arch.memory_endness,
inspect=False)
state.regs.sp += state.arch.bytes
state.inspect.exit_target = ret_addr
else:
sys_num = state.regs.eax
if sys_num.symbolic:
import ipdb
ipdb.set_trace()
l.error("SHIT. F**K. SHIT F**K.")
try:
all_targets = (self.syscall_mapping[state.solver.any_int(
sys_num)][0], )
except KeyError:
# ????????????????????
all_targets = (0x1234678d, )
for target in all_targets:
if target in self.function_initial_regs:
for reg, stored in self.function_initial_regs[
target].iteritems():
self._runtime_unify(state,
state.registers.load(reg),
stored,
stack_frame=reg == 'sp')
if all(target in self.cfg.functions
and not self.cfg.functions[target].returning
for target in all_targets):
state.inspect.exit_jumpkind = 'Ijk_Ret'
else:
state.inspect.exit_jumpkind = 'Ijk_FakeRet'
# okay.
# go through all the targets. if any of them have return values available,
# take one of them and drop it into the state, take the rest and unify them
# if only some of them have return values available, the ones without just
# inherit the first value. if none of them are available, make a fresh value,
# give it a source taint, and let the 'SOURCE' tag handle it
are_any = False
for target in all_targets:
if target in self.function_return_vals:
if not are_any:
are_any = True
state.regs.eax = self.function_return_vals[target]
else:
import ipdb
ipdb.set_trace()
# I don't THINK this should ever happen.... pdb to make sure assumptions are good
# reason: there are only multiple call targets if it's a call table. entries in a call table
# are typically not reused, and function_return_vals entries are only set from analyses
# of the caller.
self._runtime_unify(
state,
state.regs.eax,
self.function_return_vals[target],
overwrite=False)
if are_any:
for target in all_targets:
if target not in self.function_return_vals:
self.function_return_vals[target] = state.regs.eax
else:
state.regs.eax = claripy.BVS('retval', 32).annotate(
TypeAnnotation(
sim_type.SimTypeTop(
label=[ValueSource('return', all_targets)])))