def get_backend(domain, listener, syscall_filtering):
"""Return backend based on libvmi configuration.
If analyze if False, returns a dummy backend
that does not analyze system calls.
Returns None if the backend is missing
"""
libvmi = Libvmi(domain.name())
os_type = libvmi.get_ostype()
try:
return BACKENDS[os_type](domain, libvmi, listener, syscall_filtering)
except KeyError:
raise BackendNotFoundError('Unable to find an appropritate backend for'
'this OS: {}'.format(os_type))
def get_backend(domain, listener, syscall_filtering):
"""
Return a suitable backend based on guest operating system.
:param domain: libvirt domain
:returns: new backend instance
:rtype: Backend
:raises: BackendNotFoundError
"""
libvmi = Libvmi(domain.name())
os_type = libvmi.get_ostype()
try:
return BACKENDS[os_type](domain, libvmi, listener, syscall_filtering)
except KeyError:
raise BackendNotFoundError('Unable to find an appropritate backend for'
'this OS: {}'.format(os_type))
class LibVMIStub(GDBStub):
def __init__(self, conn, addr, vm_name, process):
super().__init__(conn, addr)
self.vm_name = vm_name
self.process = process
self.cmd_to_handler = {
GDBCmd.GEN_QUERY_GET: self.gen_query_get,
GDBCmd.GEN_QUERY_SET: self.gen_query_set,
GDBCmd.SET_THREAD_ID: self.set_thread_id,
GDBCmd.TARGET_STATUS: self.target_status,
GDBCmd.READ_REGISTERS: self.read_registers,
GDBCmd.WRITE_REGISTERS: self.write_registers,
GDBCmd.DETACH: self.detach,
GDBCmd.READ_MEMORY: self.read_memory,
GDBCmd.WRITE_MEMORY: self.write_memory,
GDBCmd.WRITE_DATA_MEMORY: self.write_data_memory,
GDBCmd.CONTINUE: self.cont_execution,
GDBCmd.SINGLESTEP: self.singlestep,
GDBCmd.IS_THREAD_ALIVE: self.is_thread_alive,
GDBCmd.REMOVE_XPOINT: self.remove_xpoint,
GDBCmd.INSERT_XPOINT: self.insert_xpoint,
GDBCmd.BREAKIN: self.breakin,
GDBCmd.V_FEATURES: self.v_features,
GDBCmd.KILL_REQUEST: self.kill_request,
}
self.features = {
b'multiprocess': False,
b'swbreak': True,
b'hwbreak': False,
b'qRelocInsn': False,
b'fork-events': False,
b'vfork-events': False,
b'exec-events': False,
b'vContSupported': True,
b'QThreadEvents': False,
b'QStartNoAckMode': True,
b'no-resumed': False,
b'xmlRegisters': False,
b'qXfer:memory-map:read': True
}
def __enter__(self):
# init LibVMI
self.vmi = Libvmi(self.vm_name,
init_flags=INIT_DOMAINNAME | INIT_EVENTS,
partial=True)
self.vmi.init_paging(flags=0)
# catch every exception to force a clean exit with __exit__
# where vmi.destroy() must be called
try:
# determine debug context
if not self.process:
self.ctx = RawDebugContext(self.vmi)
else:
self.vmi.init_os()
ostype = self.vmi.get_ostype()
if ostype == VMIOS.WINDOWS:
self.ctx = WindowsDebugContext(self.vmi, self.process)
elif ostype == VMIOS.LINUX:
self.ctx = LinuxDebugContext(self.vmi, self.process)
else:
raise RuntimeError('unhandled ostype: {}'.format(
ostype.value))
self.ctx.attach()
self.attached = True
except:
logging.exception('Exception while initializing debug context')
return self
def __exit__(self, type, value, traceback):
try:
self.ctx.detach()
self.attached = False
self.ctx.bpm.restore_opcodes()
except:
logging.exception('Exception while detaching from debug context')
finally:
self.vmi.destroy()
@lru_cache(maxsize=None)
def get_memory_map_xml(self):
# retrieve list of maps
root = etree.Element('memory-map')
for page_info in self.vmi.get_va_pages(self.ctx.get_dtb()):
# <memory type="ram" start="addr" length="length"/>
addr = str(hex(page_info.vaddr))
size = str(hex(page_info.size))
region = etree.Element('memory',
type='ram',
start=addr,
length=size)
root.append(region)
doctype = '<!DOCTYPE memory-map ' \
'PUBLIC "+//IDN gnu.org//DTD GDB Memory Map V1.0//EN"' \
' "http://sourceware.org/gdb/gdb-memory-map.dtd">'
xml = etree.tostring(root,
xml_declaration=True,
doctype=doctype,
encoding='UTF-8')
return xml
# commands
def gen_query_get(self, packet_data):
if re.match(b'Supported', packet_data):
reply = self.set_supported_features(packet_data)
pkt = GDBPacket(reply)
self.send_packet(pkt)
return True
if re.match(b'TStatus', packet_data):
# Ask the stub if there is a trace experiment running right now
# reply: No trace has been run yet
self.send_packet(GDBPacket(b'T0;tnotrun:0'))
return True
if re.match(b'TfV', packet_data):
# TODO
return False
if re.match(b'fThreadInfo', packet_data):
reply = b'm'
for thread in self.ctx.list_threads():
if reply != b'm':
reply += b','
reply += b'%x' % thread.id
self.send_packet(GDBPacket(reply))
return True
if re.match(b'sThreadInfo', packet_data):
# send end of thread list
self.send_packet(GDBPacket(b'l'))
return True
m = re.match(b'ThreadExtraInfo,(?P<thread_id>.+)', packet_data)
if m:
tid = int(m.group('thread_id'), 16)
thread = self.ctx.get_thread(tid)
if not thread:
return False
self.send_packet(GDBPacket(thread.name.encode()))
return True
if re.match(b'Attached', packet_data):
# attach existing process: 0
# attach new process: 1
self.send_packet(GDBPacket(b'0'))
return True
if re.match(b'C', packet_data):
# return current thread id
self.send_packet(GDBPacket(b'QC%x' % self.ctx.cur_tid))
return True
m = re.match(b'Xfer:memory-map:read::(?P<offset>.*),(?P<length>.*)',
packet_data)
if m:
offset = int(m.group('offset'), 16)
length = int(m.group('length'), 16)
xml = self.get_memory_map_xml()
chunk = xml[offset:offset + length]
msg = b'm%s' % chunk
if len(chunk) < length or offset + length >= len(xml):
# last chunk
msg = b'l%s' % chunk
self.send_packet(GDBPacket(msg))
return True
return False
def gen_query_set(self, packet_data):
if re.match(b'StartNoAckMode', packet_data):
self.no_ack = True
self.send_packet(GDBPacket(b'OK'))
# read last ack
c = self.sock.recv(1)
if c == b'+':
return True
else:
return False
return False
def set_thread_id(self, packet_data):
m = re.match(b'(?P<op>[cg])(?P<tid>([0-9a-f])+|-1)', packet_data)
if m:
op = m.group('op')
tid = int(m.group('tid'), 16)
self.log.debug('Current thread: %s', tid)
self.ctx.cur_tid = tid
# TODO op, Enn
self.send_packet(GDBPacket(b'OK'))
return True
return False
def target_status(self, packet_data):
msg = b'S%.2x' % GDBSignal.TRAP.value
self.send_packet(GDBPacket(msg))
return True
def kill_request(self, packet_data):
self.attached = False
return True
def read_registers(self, packet_data):
addr_width = self.vmi.get_address_width()
if addr_width == 4:
pack_fmt = '@I'
else:
pack_fmt = '@Q'
cur_thread = self.ctx.get_thread()
regs = cur_thread.read_registers()
gen_regs_32 = [
X86Reg.RAX, X86Reg.RCX, X86Reg.RDX, X86Reg.RBX, X86Reg.RSP,
X86Reg.RBP, X86Reg.RSI, X86Reg.RDI, X86Reg.RIP
]
gen_regs_64 = [
X86Reg.R9, X86Reg.R10, X86Reg.R11, X86Reg.R12, X86Reg.R13,
X86Reg.R14, X86Reg.R15
]
# not available through libvmi
seg_regs = [x + 1 for x in range(0, 6)]
# write general registers
msg = b''.join(
[hexlify(struct.pack(pack_fmt, regs[r])) for r in gen_regs_32])
if addr_width == 8:
msg += b''.join(
[hexlify(struct.pack(pack_fmt, regs[r])) for r in gen_regs_64])
# write eflags
msg += hexlify(struct.pack(pack_fmt, regs[X86Reg.RFLAGS]))
# write segment registers
msg += b''.join([hexlify(struct.pack(pack_fmt, r)) for r in seg_regs])
self.send_packet(GDBPacket(msg))
return True
def write_registers(self, packet_data):
addr_width = self.vmi.get_address_width()
if addr_width == 4:
pack_fmt = '@I'
else:
pack_fmt = '@Q'
gen_regs_32 = [
X86Reg.RAX, X86Reg.RCX, X86Reg.RDX, X86Reg.RBX, X86Reg.RSP,
X86Reg.RBP, X86Reg.RSI, X86Reg.RDI, X86Reg.RIP
]
gen_regs_64 = [
X86Reg.R9, X86Reg.R10, X86Reg.R11, X86Reg.R12, X86Reg.R13,
X86Reg.R14, X86Reg.R15
]
# TODO parse the entire buffer
# regs = Registers()
regs = self.vmi.get_vcpuregs(0)
iter = struct.iter_unpack(pack_fmt, unhexlify(packet_data))
for r in gen_regs_32:
value, *rest = next(iter)
logging.debug('%s: %x', r.name, value)
regs[r] = value
# 64 bits ?
if addr_width == 8:
for r in gen_regs_64:
value, *rest = next(iter)
logging.debug('%s: %x', r.name, value)
regs[r] = value
# eflags
value, *rest = next(iter)
regs[X86Reg.RFLAGS] = value
# TODO segment registers
try:
self.vmi.set_vcpuregs(regs, 0)
except LibvmiError:
return False
else:
self.send_packet(GDBPacket(b'OK'))
return True
def detach(self, packet_data):
# detach
self.attached = False
try:
self.vmi.resume_vm()
except LibvmiError:
pass
self.send_packet(GDBPacket(b'OK'))
return True
def read_memory(self, packet_data):
m = re.match(b'(?P<addr>.*),(?P<length>.*)', packet_data)
if m:
addr = int(m.group('addr'), 16)
length = int(m.group('length'), 16)
# TODO partial read
try:
buffer, bytes_read = self.vmi.read(
self.ctx.get_access_context(addr), length)
except LibvmiError:
return False
else:
self.send_packet(GDBPacket(hexlify(buffer)))
return True
return False
def write_memory(self, packet_data):
m = re.match(b'(?P<addr>.+),(?P<length>.+):(?P<data>.+)', packet_data)
if m:
addr = int(m.group('addr'), 16)
length = int(m.group('length'), 16)
data = unhexlify(m.group('data'))
# TODO partial write
try:
bytes_written = self.vmi.write(
self.ctx.get_access_context(addr), data)
except LibvmiError:
return False
else:
self.send_packet(GDBPacket(b'OK'))
return True
return False
def write_data_memory(self, packet_data):
# ‘X addr,length:XX…’
m = re.match(b'(?P<addr>.+),(?P<length>.+):(?P<data>.*)', packet_data)
if m:
addr = int(m.group('addr'), 16)
length = int(m.group('length'), 16)
data = m.group('data')
# TODO partial write
try:
bytes_written = self.vmi.write(
self.ctx.get_access_context(addr), data)
except LibvmiError:
return False
else:
self.send_packet(GDBPacket(b'OK'))
return True
return False
def cont_execution(self, packet_data):
# TODO resume execution at addr
addr = None
m = re.match(b'(?P<addr>.+)', packet_data)
if m:
addr = int(m.group('addr'), 16)
return False
self.action_continue()
self.send_packet(GDBPacket(b'OK'))
# TODO race condition if listen thread started by action_continue
# sends a packet before our 'OK' reply
return True
def singlestep(self, packet_data):
# TODO resume execution at addr
addr = None
m = re.match(b'(?P<addr>.+)', packet_data)
if m:
addr = int(m.group('addr'), 16)
return False
self.ctx.bpm.wait_process_scheduled()
self.ctx.bpm.singlestep_once()
msg = b'S%.2x' % GDBSignal.TRAP.value
self.send_packet(GDBPacket(msg))
return True
def is_thread_alive(self, packet_data):
m = re.match(b'(?P<tid>.+)', packet_data)
if m:
tid = int(m.group('tid'), 16)
thread = self.ctx.get_thread(tid)
if not thread:
# TODO Err XX
return False
reply = None
if thread.is_alive():
reply = b'OK'
else:
# TODO thread is dead
reply = b'EXX'
self.send_packet(GDBPacket(reply))
return True
return False
def remove_xpoint(self, packet_data):
# ‘z type,addr,kind’
m = re.match(b'(?P<type>[0-9]),(?P<addr>.+),(?P<kind>.+)', packet_data)
if not m:
return False
btype = int(m.group('type'))
addr = int(m.group('addr'), 16)
# kind -> size of breakpoint
kind = int(m.group('kind'), 16)
if btype == 0:
# software breakpoint
self.ctx.bpm.del_swbp(addr)
self.send_packet(GDBPacket(b'OK'))
return True
return False
def insert_xpoint(self, packet_data):
# ‘Z type,addr,kind’
m = re.match(b'(?P<type>[0-9]),(?P<addr>.+),(?P<kind>.+)', packet_data)
if not m:
return False
btype = int(m.group('type'))
addr = int(m.group('addr'), 16)
# kind -> size of breakpoint
kind = int(m.group('kind'), 16)
if btype == 0:
# software breakpoint
cb_data = {
'stub': self,
'stop_listen': self.ctx.bpm.stop_listen,
}
self.ctx.bpm.add_swbp(addr, kind, self.ctx.cb_on_swbreak, cb_data)
self.send_packet(GDBPacket(b'OK'))
return True
return False
def breakin(self, packet_data):
# stop event thread
self.ctx.bpm.stop_listening()
self.ctx.attach()
msg = b'S%.2x' % GDBSignal.TRAP.value
self.send_packet(GDBPacket(msg))
return True
def v_features(self, packet_data):
if re.match(b'MustReplyEmpty', packet_data):
# reply empty string
# TODO refactoring, this should be treated as an unknown packet
self.send_packet(GDBPacket(b''))
return True
if re.match(b'Cont\?', packet_data):
# query the list of supported actions for vCont
# reply: vCont[;action…]
# we do not support continue or singlestep with a signal
# but we have to advertise this to GDB, otherwise it won't use vCont
self.send_packet(GDBPacket(b'vCont;c;C;s;S'))
return True
m = re.match(b'Cont(;(?P<action>[sc])(:(?P<tid>.*?))?).*', packet_data)
if m:
# vCont[;action[:thread-id]]…
# we don't support threads
action = m.group('action')
if action == b's':
self.ctx.bpm.wait_process_scheduled()
self.ctx.bpm.singlestep_once()
self.send_packet_noack(
GDBPacket(b'T%.2x' % GDBSignal.TRAP.value))
return True
if action == b'c':
self.action_continue()
return True
if re.match(b'Kill;(?P<pid>[a-fA-F0-9]).+', packet_data):
# vKill;pid
# ignore pid, and don't kill the process anyway
# just detach from the target
# sent when GDB client has a ^D
self.attached = False
self.send_packet(GDBPacket(b'OK'))
return True
return False
# helpers
def action_continue(self):
self.vmi.resume_vm()
# start listening on VMI events, asynchronously
self.ctx.bpm.listen(block=False)
def set_supported_features(self, packet_data):
# split string and get features in a list
# trash 'Supported
req_features = re.split(b'[:|;]', packet_data)[1:]
for f in req_features:
if f[-1:] in [b'+', b'-']:
name = f[:-1]
value = True if f[-1:] == b'+' else False
else:
groups = f.split(b'=')
name = groups[0]
value = groups[1]
# TODO check supported features
reply_msg = b'PacketSize=%x' % PACKET_SIZE
for name, value in self.features.items():
if isinstance(value, bool):
reply_msg += b';%s%s' % (name, b'+' if value else b'-')
else:
reply_msg += b';%s=%s' % (name, value)
return reply_msg
class ProcessChecker:
whlte_list = ['sshd', 'insmod', 'kworker/0:2', 'kworker/u2:2']
def __init__(self, vm, callback=None, interval=10):
self._vm = vm
self._callback = callback
self._init_vmi()
self.logger = setup_logger(
self._vm.name, self._vm.name + '/' + self._vm.name + '.log',
logging.INFO)
self._ori_ps_list = self._get_process_list()
self._ori_ps_set = set(self._ori_ps_list.keys())
self._timer = RepeatableTimer(interval, self.check_process)
self._dump_enabled = False
def _init_vmi(self):
""" Initialize LibVMI """
self._vmi = Libvmi(self._vm.name)
# get ostype
self._os = self._vmi.get_ostype()
# init offsets values
self._tasks_offset = None
self._name_offset = None
self._pid_offset = None
if self._os == VMIOS.LINUX:
self._tasks_offset = self._vmi.get_offset("linux_tasks")
self._name_offset = self._vmi.get_offset("linux_name")
self._pid_offset = self._vmi.get_offset("linux_pid")
elif self._os == VMIOS.WINDOWS:
self._tasks_offset = self._vmi.get_offset("win_tasks")
self._name_offset = self._vmi.get_offset("win_pname")
self._pid_offset = self._vmi.get_offset("win_pid")
else:
self.logger.error("Unknown OS")
return 0
return 1
def _get_process_list(self):
"""
Get the process list inside the VM
:return: process list of getting the page successfully or None.
"""
# pause vm
with pause(self._vmi):
# demonstrate name and id accessors
name = self._vmi.get_name()
id = self._vmi.get_vmid()
self.logger.debug("Process listing for VM %s (id: %s)", name, id)
if self._os == VMIOS.LINUX:
list_head = self._vmi.translate_ksym2v("init_task")
list_head += self._tasks_offset
elif self._os == VMIOS.WINDOWS:
list_head = self._vmi.read_addr_ksym("PsActiveProcessHead")
else:
self.logger.error("Unknown OS")
return None
process_list = dict()
cur_list_entry = list_head
next_list_entry = self._vmi.read_addr_va(cur_list_entry, 0)
while True:
current_process = cur_list_entry - self._tasks_offset
pid = self._vmi.read_32_va(current_process + self._pid_offset,
0)
procname = self._vmi.read_str_va(
current_process + self._name_offset, 0)
process_list[pid] = (procname, hex(current_process))
self.logger.debug("[%s] %s (struct addr:%s)", pid, procname,
hex(current_process))
cur_list_entry = next_list_entry
next_list_entry = self._vmi.read_addr_va(cur_list_entry, 0)
if self._os == VMIOS.WINDOWS and next_list_entry == list_head:
break
elif self._os == VMIOS.LINUX and cur_list_entry == list_head:
break
return process_list
def check_process(self):
self._timer.start()
ps_list = self._get_process_list()
#ps_list = self._filter_white_list(ps_list)
ps_set = set(ps_list.keys())
new_ps = ps_set - self._ori_ps_set
reduce_ps = self._ori_ps_set - ps_set
if new_ps:
for pid in new_ps:
if ps_list[pid][0] not in self.whlte_list:
if not self._vm.dump_enabled:
self._vm.dump_enabled = True
self._callback()
self.logger.warning(
"detected new process. [%s] %s (struct addr:%s)", pid,
ps_list[pid][0], ps_list[pid][1])
if reduce_ps:
for pid in reduce_ps:
if self._ori_ps_list[pid][0] not in self.whlte_list:
self.logger.warning(
"detected process leave. [%s] %s (struct addr:%s)",
pid, self._ori_ps_list[pid][0],
self._ori_ps_list[pid][1])
self._ori_ps_list = ps_list
self._ori_ps_set = ps_set
def start(self):
"""Start a thread that compares both of origin process list and a new one"""
self._timer.start()
def stop(self):
"""stop process checker"""
self._timer.cancel()
self._vmi.destroy()
class DebugContext:
def __init__(self, vm_name, process_name):
self.log = logging.getLogger(__class__.__name__)
self.vm_name = vm_name
self.full_system_mode = False
self.target_name = process_name
self.target_pid = None
if process_name is None:
self.full_system_mode = True
self.target_name = 'kernel'
# kernel space is represented by PID 0 in LibVMI
self.target_pid = 0
self.target_dtb = None
self.vmi = Libvmi(self.vm_name, INIT_DOMAINNAME | INIT_EVENTS)
self.kernel_base = self.get_kernel_base()
if self.kernel_base:
logging.info('kernel base address: %s', hex(self.kernel_base))
def get_kernel_base(self):
if self.vmi.get_ostype() == VMIOS.LINUX:
return self.vmi.translate_ksym2v('start_kernel')
if self.vmi.get_ostype() == VMIOS.WINDOWS:
# small hack with rekall JSON profile to get the kernel base address
# LibVMI should provide an API to query it
profile_path = self.vmi.get_rekall_path()
if not profile_path:
raise RuntimeError('Cannot get rekall profile from LibVMI')
with open(profile_path) as f:
profile = json.load(f)
ps_head_rva = profile['$CONSTANTS']['PsActiveProcessHead']
ps_head_va = self.vmi.translate_ksym2v('PsActiveProcessHead')
return ps_head_va - ps_head_rva
return None
def attach(self):
self.log.info('attaching on %s', self.target_name)
# VM must be running
self.vmi.pause_vm()
if self.full_system_mode:
# no need to intercept a specific process
regs = self.vmi.get_vcpuregs(0)
self.target_dtb = regs[X86Reg.CR3]
return
cb_data = {'interrupted': False}
def cb_on_cr3_load(vmi, event):
pname = dtb_to_pname(vmi, event.cffi_event.reg_event.value)
self.log.info('intercepted %s', pname)
pattern = re.escape(self.target_name)
if re.match(pattern, pname, re.IGNORECASE):
vmi.pause_vm()
self.target_dtb = event.cffi_event.reg_event.value
self.target_pid = vmi.dtb_to_pid(self.target_dtb)
cb_data['interrupted'] = True
reg_event = RegEvent(X86Reg.CR3, RegAccess.W, cb_on_cr3_load)
self.vmi.register_event(reg_event)
self.vmi.resume_vm()
while not cb_data['interrupted']:
self.vmi.listen(1000)
# clear queue
self.vmi.listen(0)
# clear event
self.vmi.clear_event(reg_event)
def detach(self):
try:
logging.info('resuming VM execution')
self.vmi.resume_vm()
except LibvmiError:
# already in running state
pass
self.vmi.destroy()